doi:10.1016/S0022-2836(02)01237-8
`
`J. Mol. Biol. (2003) 325, 531–553
`
`Biophysical Properties of Human Antibody
`Variable Domains
`
`Stefan Ewert, Thomas Huber, Annemarie Honegger and
`Andreas Plu¨ ckthun*
`
`Biochemisches Institut
`Universita¨t Zu¨rich
`Winterthurerstr. 190
`CH-8057 Zu¨ rich, Switzerland
`
`There are great demands on the stability, expression yield and resistance
`to aggregation of antibody fragments. To untangle intrinsic domain effects
`from domain interactions, we present first a systematic evaluation of the
`isolated human immunoglobulin variable heavy (VH) and light (VL) germ-
`line family consensus domains and then a systematic series of VH– VL
`combinations in the scFv format. The constructs were evaluated in terms
`of their expression behavior, oligomeric state in solution and denaturant-
`induced unfolding equilibria under non-reducing conditions. The seven
`VH and seven VL domains represent the consensus sequences of the
`major human germline subclasses, derived from the Human Combina-
`torial Antibody Library (HuCALw). The isolated VH and VL domains
`with the highest thermodynamic stability and yield of soluble protein
`were VH3 and Vk3, respectively. Similar measurements on all domain com-
`binations in scFv fragments allowed the scFv fragments to be classified
`according to thermodynamic stability and in vivo folding yield. The scFv
`fragments containing the variable domain combinations H3k3, H1bk3,
`H5k3 and H3k1 show superior properties concerning yield and stability.
`Domain interactions diminish the intrinsic differences of the domains.
`ScFv fragments containing Vl domains show high levels of stability, even
`though Vl domains are surprisingly unstable by themselves. This is due
`to a strong interaction with the VH domain and depends on the amino
`acid sequence of the CDR-L3. On the basis of these analyses and model
`structures, we suggest possibilities for further improvement of the bio-
`physical properties of individual frameworks and give recommendations
`for library design.
`
`q 2003 Elsevier Science Ltd. All rights reserved
`
`*Corresponding author
`
`Keywords: antibody engineering; protein stability; expression;
`scFv fragment
`
`Introduction
`
`Because of their high degree of specificity and
`broad target range, antibodies have found numer-
`
`Present address: S. Ewert, ESBA Tech AG, Wagistr. 21,
`CH-8952 Zu¨ rich-Schlieren, Switzerland.
`Abbreviations used: CDR, complementary-
`determining region; GdnHCl, guanidine hydrochloride;
`HuCALw, Human Combinatorial Antibody Library;
`IMAC, immobilized metal ion affinity chromatography;
`SB, super broth; scFv, single-chain antibody fragment
`consisting of the variable domains of the heavy and of
`the light chain connected by a peptide linker; VH,
`variable domain of the heavy chain of an antibody; VL,
`variable domain of the light chain of an antibody.
`E-mail address of the corresponding author:
`plueckthun@biocfebs.unizh.ch
`
`ous applications in basic research, medicine and
`the biotechnology industry, where they serve as
`tools to recognize selectively virtually any kind of
`target molecule. However, despite their versatility,
`only a subset of antibodies has the biophysical
`properties ideally suited for such applications. For
`example, therapeutic or in vivo diagnostic antibody
`fragments require a long serum half-life in human
`patients to accumulate at the desired target, and
`they must therefore be resistant to aggregation,
`precipitation and degradation by proteases.1,2
`Industrial applications often demand antibodies
`that have a long half-life or can function in organic
`solvents, surfactants or at high temperatures.3,4
`Applications in functional genomics will require
`stability against surface denaturation and drying
`in miniaturized devices. Furthermore,
`the high
`
`0022-2836/03/$ - see front matter q 2003 Elsevier Science Ltd. All rights reserved
`
`Ex. 1045 - Page 1 of 23
`
`AMGEN INC.
`Exhibit 1045
`
`

`

`532
`
`Biophysical Properties of Human Antibody Domains
`
`demand for and the increasing number of appli-
`cations of
`antibodies
`require more
`efficient
`methods for their high-level production and the
`availability of molecules with favorable properties
`for every antigen. Single-chain Fv (scFv) fragments
`are a particularly popular antibody format.5 – 7 The
`size of these molecules is reduced to the antigen-
`binding part of the antibody, and they contain the
`variable domains of the heavy and light chain
`connected via a flexible linker. Most scFv frag-
`ments can be obtained easily from recombinant
`expression in Escherichia coli in sufficient amounts.8
`As production yields of
`these fragments are
`influenced by their stability as well as their folding
`efficiency, considerable efforts have been made to
`identify positions in the variable domains critical
`for influencing the expression behavior of various
`antibody molecules.9 – 12 The factors influencing the
`biophysical properties of antibody molecules have
`been studied mostly with scFv fragments.13 The
`overall stability of scFv fragments depends on the
`intrinsic structural stability of VH and VL as well
`as on the extrinsic stabilization provided by their
`interaction.14 For some scFvs,
`the stabilities of
`isolated VH and VL domains as well as of the
`whole scFv fragment have been measured and
`compared.14 – 16
`In this study, we present the first fully systematic
`biophysical characterization of isolated human VH
`and VL domains as well as an investigation of
`the effect of combinations of VH and VL in the
`scFv format. The analyzed variable antibody
`domains are family consensus domains taken
`from the Human Combinatorial Antibody Library
`(HuCALw).17 In this library, a consensus sequence
`of
`the
`framework sequences of
`the major
`VH-families (VH1, VH2, VH3, VH4, VH5, and VH6)
`and VL-families (Vk1, Vk2, Vk3, Vk4 and Vl1, Vl2,
`Vl3) was generated. CDR-1 and CDR-2 sequences
`were taken from the germline gene with the
`highest usage in a family as calculated from
`rearranged sequences. The VH1 family is further
`divided into VH1a and VH1b because of different
`CDR-H2 conformations. We chose these seven VH
`and VL domains for our study, because they ideally
`represent the structural repertoire of human VH
`they were engineered to
`and VL frameworks,
`contain identical CDR-3s and their individual
`performance in recombinant
`libraries has been
`studied extensively.17,18
`Human VH domains occur in nature only in
`complex with VL. Most isolated VH domains have
`a pronounced tendency to aggregate, due pri-
`marily to the large hydrophobic surface area
`exposed in the absence of VL.16,19,20 This makes
`their expression and purification laborious, and
`the study of their biophysical properties in iso-
`lation
`almost
`impossible.
`The
`experiments
`reported here became possible, nevertheless,
`because we recently discovered by a novel selec-
`tion strategy a 17 amino acid residue long
`CDR-H3, which increases the solubility of VH
`domains, possibly by partially covering the
`
`hydrophobic interface region to VL (J. Burmester
`et al., unpublished results).
`In contrast to VH domains, isolated VL domains
`usually show only a small
`tendency to aggre-
`gate,20,21 apart from some Bence –Jones proteins, of
`which the fibril-forming ones may be identifiable
`risk factors”.22 However,
`by “structural
`the
`CDR-L3 is again of critical importance, because it
`differs from Vk to Vl domains. The majority of the
`CDR-L3 (70%) of human Vk-domains has a cis-pro-
`line residue at position 136 (the nomenclature is
`described by Honegger & Plu¨ ckthun23), which
`defines a rigid V-loop conformation, while a cis-
`proline residue is observed only rarely in Vl
`domains (4%). The CDR-L3 V loop conformation
`of Vk is not compatible with the CDR-L1 confor-
`mation usually observed in Vl domains, as the
`combination of a rigid Vk CDR-L3 with a Vl
`CDR-L1 leads to steric clashes, destabilizing the
`chimeric domains
`severely. Taking this
`into
`account, we used two different CDR-L3s for Vk
`and Vl domains.
`The seven human consensus VH domains with
`the
`solubility-enhancing CDR-H3,
`the
`seven
`human consensus VL domains with a k-like
`CDR-L3 for k domains and l-like CDR-L3 for the
`l domains, were expressed and purified, and a
`classification of human VH and VL frameworks is
`given, on the basis of their yield of soluble peri-
`plasmic protein, examination of their oligomeric
`state in solution, denaturant-induced unfolding
`and refolding equilibria under non-reducing
`conditions. We further examined scFv fragments
`with systematic combinations of the most stable
`VH domain (VH3) with each VL domain and with
`the most stable VL domain (Vk3) with each VH
`domain.
`With this study it is now possible to make pre-
`dictions about scFv properties such as yield, oligo-
`meric state and thermodynamic stability,
`just by
`identifying the VH and VL family. Most important,
`however, the detailed analysis of the individual
`properties of
`the consensus variable domains
`and comparisons of sequence alignments and
`model structures enables a further improvement
`of properties of any immunoglobulin variable
`domain and of libraries on the basis of incorpora-
`ting any of the frameworks, to maintain the full
`diversity of antibodies.
`It gives some insight
`into interacting groups of
`residues and their
`co-evolution seen in the variable domain families.
`
`Results
`
`Expression and protein purification of
`VH fragments
`
`The seven HuCALw consensus VH domains
`representing the major framework subclasses were
`expressed with the same CDR-H3 to enable the
`comparison of
`their biophysical properties. We
`first investigated VH domains with the CDR-H3
`
`Ex. 1045 - Page 2 of 23
`
`

`

`Biophysical Properties of Human Antibody Domains
`
`533
`
`Table 1. Summary of biophysical characterization of isolated VH and VL domains
`
`Human family
`
`CDR-3
`
`Soluble yield
`(mg ml21 OD550 ¼ 10)
`
`Oligomeric state
`
`Midpoint [GdnHCl]
`(M)
`
`DGN-U
`(kJ mol21)
`
`m
`(kJ M21 mol21)
`
`VH
`
`VL
`
`1a
`1b
`2
`3
`3e
`4
`5
`6
`
`k1
`k2
`k3
`k4
`l1
`l2
`l3
`
`Longa
`Long
`Long
`Long
`Shortf
`Long
`Long
`Long
`
`k-likeg
`k-like
`k-like
`k-like
`l-likei
`l-like
`l-like
`
`1.0
`1.2
`Refold.c
`2.4
`2.1
`Refold.
`Refold.
`Refold.
`
`4.5
`14.2
`17.1
`9.6
`0.3
`1.9
`0.8
`
`Mb
`M
`NDd
`M
`ND
`ND
`M
`ND
`
`M
`M
`M
`D, Mh
`M
`M
`D, Mh
`
`1.5
`2.1
`1.6
`3.0
`2.7
`1.8
`2.2
`0.8
`
`2.1
`1.5
`2.3
`1.5
`2.1
`1.0
`0.9
`
`13.7
`26.0
`ND
`52.7
`39.7
`ND
`16.5
`ND
`
`29.0
`24.8
`34.5
`ND
`23.7
`16.0
`15.1
`
`10.1
`12.7
`ND
`17.6
`14.6
`ND
`7.0
`ND
`
`14.1
`16.1
`14.8
`ND
`11.1
`16.2
`15.9
`
`a Long CDR-H3, sequence: YNHEADMLIRNWLYSDV.
`b Monomer (M) in 50 mM sodium phosphate (pH 7.0) and 500 mM NaCl, in case of VH1a with 0.9 M GdnHCl.
`c No soluble protein obtained, purification via refolding of inclusion bodies.
`d Not determined.
`e Data from Ewert et al.25
`f Short CDR-H3, sequence: WGGDGFYAMDY.
`g k-Like CDR-L3, sequence: QQHYTTPPT.
`h Dimer (D) and monomer (M) equilibrium.
`i l-Like CDR-L3, sequence: QSYDSSLSGVV.
`
`(WGGDGFYAMDY) from the antibody hu4D5-8
`(PDB entry 1FVC24), but only the VH3 domain
`could be purified from the soluble fraction of the
`bacterial lysate,25 while the other VH domains by
`themselves were
`insoluble
`and gave
`small
`inclusion body pellets (data not shown). This was
`not surprising, as many if not most isolated VH
`domains have been found to be insoluble upon
`periplasmic expression,16,19,20,26 since they contain
`an exposed large hydrophobic interface that
`is
`usually covered by VL. However, three isolated VH
`domains from the HuCALw (with framework
`classes VH1a, VH1b, and VH3) were obtained
`recently from a metabolic selection experiment
`(J. Burmester et al., unpublished results). These
`could be expressed in the periplasm of E. coli and
`purified from the soluble fraction of
`the cell
`extracts. The main feature of
`the selected VH
`domains is the length of the CDR-H3, as all three
`selected and soluble VH fragments contain a
`CDR-H3 that is longer than 15 residues. This long
`CDR-H3 may partially cover the hydrophobic
`interface of VH, thereby preventing aggregation.
`After introducing the long CDR-H3 from one
`of
`the selected VH3 domains (YNHEADMLIR-
`NWLYSDV), VH1a, VH1b and VH3 could be
`expressed in soluble form in the periplasm of
`E. coli and purified from the soluble fraction of the
`cell extracts with a yield of 2.4 mg l21 in the case
`of VH3 and of 1.0 mg l21 and 1.2 mg l21 in the case
`of VH1a and VH1b, respectively (Table 1).
`In contrast, VH2, VH4, VH5 and VH6 with the long
`CDR-H3 were still
`insoluble in the E. coli peri-
`plasm. Protein production was induced with a
`final concentration of IPTG of 0.05 mM, followed
`by an incubation time of 15 hours. The low concen-
`
`tration of IPTG caused a slower protein expression
`that was necessary, because standard concen-
`trations of
`IPTG between 0.1 mM and 1 mM
`resulted in cell lysis after two to three hours (data
`not shown). The domains were purified from the
`insoluble fraction with immobilized metal
`ion
`affinity chromatography (IMAC) under denaturing
`conditions, and the eluted fractions were sub-
`jected to in vitro refolding. Approximately 1 mg of
`soluble, refolded VH5 domain could be obtained
`from one liter of E. coli culture using an oxidizing
`glutathione redox shuffle, while VH2, VH4 and VH6
`gave disulfide-linked multimers under the same
`conditions (data not shown). The even-numbered
`VH domains could be refolded only by using a
`redox shuffle with an excess of reduced gluta-
`thione followed by extensive dialysis against
`50 mM sodium phosphate (pH 7.0), 100 mM NaCl
`and slow oxidation by dissolved air-oxygen.
`Finally, about 0.2 mg of soluble, refolded protein
`could be obtained from one liter of E. coli culture.
`VH1a, VH1b, VH3 and VH5 remained in solution at
`4 8C and no degradation was observed. In contrast,
`VH2, VH4 and VH6 have a great
`tendency to
`aggregate during storage at 4 8C. Therefore, all
`subsequent experiments were performed with
`freshly purified proteins.
`
`Analytical gel-filtration of VH fragments
`
`To analyze the oligomeric state of the purified VH
`domains in solution, analytical gel-filtration experi-
`ments were performed. VH1b, VH3, and VH5 elute
`at the expected size of a monomer (Figure 1(a)
`with VH3 as an example for monomeric VH
`domains). VH1a elutes under native conditions in
`
`Ex. 1045 - Page 3 of 23
`
`

`

`534
`
`Biophysical Properties of Human Antibody Domains
`
`Figure 1. Determination of apparent molecular mass of isolated VH and VL domains. Gel-filtration runs were
`performed in 50 mM sodium phosphate (pH 7.0), 500 mM NaCl of (a) isolated human consensus VH domains (5 mM)
`on a Superdex-75 column with VH3 (continuous line) and VH1a (dotted line) and VH1a in the presence of 0.9 M
`GdnHCl (long-dash line); (b) isolated Vk domains (50 mM) on a Superose-12 column with Vk1 (continuous line), Vk2
`(long-dash line), Vk3 (dotted line) and Vk4 (short-dash line); and (c) isolated Vl domains (5 mM) on a TSK column
`with Vl1 (continuous line), Vl2 (long-dash line) and Vl3 (dotted line). Arrows indicate elution volumes of molecular
`mass standards: carbonic anhydrase (29 kDa), and cytochrome c (12.4 kDa). (d) Equilibrium sedimentation of Vk3 at
`19,000 rpm (29,100 g) with a detection wavelength of 280 nm. The continuous line was obtained from fitting of the
`data to a single species, and a molecular mass of 13,616 Da was calculated. The residuals of the fit are scattered ran-
`domly, indicating that the assumption of the monomeric state is valid.
`
`three peaks that could not be assigned. We there-
`fore investigated whether
`small amounts of
`denaturant might break up the aggregates. Using
`an elution buffer containing 0.5 M GdnHCl, the
`unassigned peaks decrease and a peak at the size
`of a monomer was found. With 0.9 M GdnHCl,
`
`VH1a elutes in a single peak corresponding to a
`monomer (Figure 1(a) with the elution profile of
`VH1a at 0 M and 0.9 M GdnHCl). It should be
`noted that this concentration of GdnHCl is below
`the
`transition in equilibrium unfolding (see
`below). VH2, VH4 and VH6 did not elute from the
`
`Ex. 1045 - Page 4 of 23
`
`

`

`Biophysical Properties of Human Antibody Domains
`
`535
`
`which is consistent with two-state behavior. The
`VH1a domain starts to unfold at 0.9 M GdnHCl,
`where VH1a is monomeric in solution as indicated
`by gel-filtration analysis (Figure 1(a)). Therefore,
`the transition is influenced by only the stability of
`the monomeric VH1a domain and is not affected
`by multimerization equilibria. For the determi-
`nation of the free energy of unfolding, the pre-
`transition region of VH1a, whose actual slope is
`influenced by the spectral changes caused by dis-
`sociation, was assumed to have the same slope
`and intercept as the VH1b domain. VH3 displays
`the highest change in free energy upon unfolding
`(DGN-U) with 52.7 kJ mol21 and an unfolding
`cooperativity (m value) of 17.6 kJ mol21 M21. VH1b
`intermediate stability with a DGN-U of
`is of
`26.0 kJ mol21 and an m value of 12.7 kJ mol21 M21.
`VH1a and VH5 are less stable and have DGN-U
`values of 13.7 kJ mol21 and 19.1 kJ mol21, and m
`values of 10.1 kJ mol21 M21 and 8.6 kJ mol21 M21,
`respectively (Table 1). The range of m values can
`be compared to that expected for proteins of this
`size (14– 15 kDa), and indicate that at least VH1a,
`VH1b, and VH3 have the cooperativity expected for
`a two-state transition.27 The transition curves of
`VH2, VH4 and VH6 in Figure 2(b) show poor
`cooperativity, which indicates that no two-state
`behavior is followed during GdnHCl equilibrium
`denaturation. As the monomeric state of these
`VH domains could not be ascertained, it is likely
`that part of this complicated transition involves
`the dissociation of multimers. The broad transition
`of VH2 and VH4 occurred between 1.0 M and 2.5 M
`GdnHCl with a midpoint of 1.6 M and 1.8 M
`GdnHCl,
`respectively. VH6 shows a transition
`between 0.5 M and 1.4 M GdnHCl with a midpoint
`of 0.8 M. This is the lowest midpoint of
`the
`domains examined, which indicates that VH6 is
`the least stable human VH domain.
`
`Expression and protein purification of
`VL fragments
`
`The four human consensus Vk domains (Vk1,
`Vk2, Vk3 and Vk4) carrying the k-like CDR-L3
`(sequence: QQHYTTPPT)
`from the
`antibody
`hu4D5-8 (PDB entry 1FVC24) were expressed in
`soluble form in the periplasm of E. coli. After puri-
`fication with IMAC followed by a cation-exchange
`column the Vk domains could be obtained in large
`amounts, ranging from 17.1 mg l21 of bacterial
`culture normalized to an A550 of 10 for Vk3 to
`4.5 mg for Vk1 (Table 1).
`The k-like CDR-L3 has a cis-proline residue at
`position 136 (the numbering scheme for
`the
`variable domain residues is described by Honegger
`& Plu¨ ckthun23). While 70% of
`the human Vk
`domains have cis-Pro in position L136, only 4% of
`the human Vl domains show Pro in this position.
`Therefore, we used a human consensus l-like
`CDR-L3 for the Vl domains (sequence QSYDS-
`SLSGVV; for details, see Materials and Methods).
`The three human consensus Vl domains (Vl1, Vl2
`
`Figure 2. Overlay of GdnHCl denaturation curves of
`VH domains (a) VH1a (filled circles), VH1b (open squares),
`VH3 (filled squares) and VH5 (open circles). (b) VH2 (filled
`circles), VH4 (open squares) and VH6 (filled squares).
`(a) and (b) All unfolding transitions were measured by
`following the change in emission maximum as a function
`of denaturant concentration at an excitation wavelength
`of 280 nm. Relative maxima refers to a scaling in which
`the lowest emission maximum wavelength is set to 0,
`the highest to 1. Note that this procedure does not flatten
`the pre-transition or post-transition baseline.
`
`column under native conditions. Even addition of
`1.7 M GdnHCl to the elution buffer did not prevent
`these domains from sticking to the column. Elution
`could be achieved only with 1 M NaOH.
`
`Equilibrium transition experiments of
`VH fragments
`
`The thermodynamic stability of the seven human
`consensus VH domains with the long CDR-H3 was
`examined by GdnHCl equilibrium denaturation
`experiments. Unfolding of the VH domains was
`monitored by the shift of the fluorescence emission
`maximum as a function of denaturant
`con-
`centration. Figure 2(a) shows an overlay of the
`equilibrium denaturation curves of VH1a, VH1b,
`VH3, and VH5. The equilibrium denaturation of
`these domains under non-reducing conditions is
`cooperative and reversible (data not
`shown),
`
`Ex. 1045 - Page 5 of 23
`
`

`

`536
`
`Biophysical Properties of Human Antibody Domains
`
`and Vl3) were expressed in soluble form in the
`periplasm of E. coli, but the yield after purification
`with IMAC and anion-exchange column was
`much less than for the Vk domains, ranging from
`1.9 mg l21 of bacterial culture normalized to an
`A550 of 10 for Vl2 to 0.3 mg l21 for Vl1 (Table 1).
`
`Analytical gel-filtration of VL fragments
`
`While the monomeric VH fragments elute at the
`expected molecular mass around 13 kDa (Figure
`1(a)), VL domains in 50 mM sodium phosphate
`(pH 7.0), 500 mM NaCl bind to gel-filtration
`column materials, resulting in an elution profile in
`which the “affinity separation” of the different
`oligomeric species on the column material over-
`shadows the gel-filtration effect. In the case of Vk
`domains, the best results could be obtained with
`a Superose-12 column (Figure 1(b)). At a protein
`concentration of 50 mM, Vk3 and Vk2 elute at an
`apparent molecular mass of only 2 kDa, Vk4 at
`12 kDa and Vk1 elutes as a broad peak at a volume
`larger
`than the total volume of
`the column.
`Changing the concentration of Vk4 from 5 mM to
`50 mM, the peak shifts from an apparent molecular
`mass of 2 –12 kDa,
`indicating a concentration
`dependent monomer dimer equilibrium, assuming
`that Vk domains eluting at 2 kDa are indeed mono-
`meric and at 12 kDa are dimeric (see below).
`Addition of 1 M GdnHCl or adjusting the NaCl
`concentration to 2 M did not alter the elution pro-
`file. Vl domains at concentrations of 5 mM show
`the weakest unspecific interaction with silica-
`based TSK columns (Figure 1(c)), and Vl1 and Vl2
`elute at a molecular mass of 7 kDa and Vl3 elutes
`at an apparent molecular mass of 12 kDa.
`To interpret these results from analytical gel-
`filtration, we analyzed the samples by equilibrium
`ultracentrifugation. We used this method to cali-
`brate the elution volumes of the different columns
`for VL domains: Vk3 (shown in Figure 1(d) as an
`example) and Vl2 give results consistent with a
`monomer, while Vl3 shows a monomer-dimer
`equilibrium. Therefore, we can conclude that the
`VL domains Vk2, Vk3, Vl1 and Vl2, eluting at an
`apparent molecular mass at 6 kDa and 2 kDa,
`respectively, are indeed monomeric and the VL
`domains Vk4 and Vl3 eluting at 12 kDa are dimeric.
`Vk1, which elutes even after the total volume of the
`column, indicating a strong interaction with the
`column material, behaves in the ultracentrifugation
`as a monomer (Table 1).
`
`Equilibrium transition experiments of
`VL fragments
`
`Most VL domains have only one tryptophan
`residue (the highly conserved Trp L43), which is
`buried in the core in the native state. Under native
`conditions, no emission maxima could be deter-
`mined, because the fluorescence is quenched fully
`by the disulfide bond Cys L23-Cys L106. During
`GdnHCl-induced unfolding, the tryptophan resi-
`
`Figure 3. Overlay of GdnHCl denaturation curves of
`VL domains (a) Vk domains with Vk1 (filled circles), Vk2
`(filled squares), Vk3 (open squares) and Vk4 (open circles)
`and (b) Vl domains with Vl1 (filled squares), Vl2 (filled
`circles), Vl3 (open squares). (a) and (b) All unfolding
`transitions were measured by following the change of
`fluorescence intensity as a function of denaturant
`concentration at an excitation wavelength of 280 nm.
`
`due becomes solvent-exposed, giving a steep
`increase in fluorescence intensity. Therefore, the
`thermodynamic parameters were calculated using
`the six-parameter fit28 on the plot of concentration
`of GdnHCl versus fluorescence intensity, giving
`curves consistent with two-state behavior. All VL
`domains
`show reversible unfolding behavior
`under non-reducing conditions (data not shown).
`Figure 3(a) and (b) show relative fluorescence
`intensity plots against GdnHCl concentration of
`Vk and Vl domains. Vk3 is the most stable VL
`domain with a DGN-U of 34.5 kJ mol21, followed by
`Vk1 with 29.0 kJ mol21 and Vk2 and Vl1 with
`24.8 kJ mol21 and 23.7 kJ mol21, respectively (Table
`1). The least stable VL domains are Vl2 and Vl3
`with a DGN-U of 16.0 kJ mol21 and 15.1 kJ mol21.
`show m values
`between
`All VL domains
`11.1 kJ mol21 M21 and 16.2 kJ mol21 M21, indicating
`that they have the cooperativity expected for a
`two-state transition.27 The human consensus Vk4
`carries an exposed tryptophan residue at position
`
`Ex. 1045 - Page 6 of 23
`
`

`

`Biophysical Properties of Human Antibody Domains
`
`537
`
`L58 in addition to the conserved Trp L43, which is
`not quenched in the native state. The denaturation
`curve is fully reversible, but shows a steep pre-
`transition baseline followed by a non-cooperative
`transition. Because of this uncertainly, we report
`no DGN-U value for Vk4, only the midpoint of
`transition, which is at 1.5 M GdnHCl. For the Vk4
`domain Len, a stability of 32 kJ mol21 has been
`reported.29
`
`Analysis of primary sequence and
`model structures
`
`Large differences are seen in the group of iso-
`lated VH fragments: VH3 shows the highest yield
`of soluble protein and thermodynamic stability;
`VH1a, VH1b and VH5 show intermediate yield and
`intermediate or low stability; while VH2, VH4
`and VH6 show more aggregation-prone behavior
`and low cooperativity during denaturant-induced
`unfolding. The properties of Vk and Vl domains
`are more homogeneous. The
`thermodynamic
`stabilities differ by only approximately 10 kJ mol21
`in the group of Vk and in the group Vl domains.
`In general, the stability and soluble yield is higher
`in isolated Vk domains than in Vl domains. To
`analyze possible structural reasons for this dif-
`ferent behavior of the variable antibody domains,
`we analyzed the primary sequence and the
`modeled structures of the seven human consensus
`VH and VL domains. These models have been
`published17 (PDB entries: 1DHA (VH1a), 1DHO
`(VH1b), 1DHQ (VH2), 1DHU (VH3), 1DHV (VH4),
`1DHW (VH5), and 1DHZ (VH6)) and VL domains
`(PDB entries: 1DGX (Vk1), 1DH4 (Vk2), 1DH5
`(Vk3), 1DH6 (Vk4), 1DH7 (Vl1), 1DH8 (Vl2), 1DH9
`(Vl3)). The quality of the models varies for the
`different domains. Many antibody structures in
`the Protein Data Bank (PDB) use, for example, the
`VH3 framework, and the chosen template structure
`for building the model shares 86% sequence
`identity, excluding the CDR-H3 region (PDB entry:
`1IGM) and the structural differences between
`templates could be traced to distinct sequence
`differences. In the case of VH6, the closest tem-
`plates were human VH4 and murine VH8 domains,
`since no crystal structure of a member of the VH6
`germline family is available in the PDB. Both
`germline families encode a different framework 1
`structural subtype (I) than VH6 (III).30 The chosen
`template for VH6 (PDB entry: 7FAB) shares 62%
`sequence identity, excluding the CDR-H3 region
`and belongs to human VH4. For each domain
`model, experimental structures of the ten most
`closely related domains (human, murine or engi-
`neered) available at the time were superimposed
`by a least-squares fit of the Ca coordinates of
`residues L3 – L6, L20 –L24, L41 – L47, L51– L57,
`L78 –L82, L89 –L93, L102 –L108 and L138 – L144 in
`the VL domain or H3– H6, H20 – H24, H41 – H47,
`H51 – H57, H78 – H82, H89 – H93, H102 –H108 and
`H138 – H144 in the VH domain, representing the
`structurally least variable positions. The aligned
`
`structures were scrutinized for the potential struc-
`tural effects of sequence differences, if necessary
`consulting additional structures, before templates
`were assigned for modeling. Currently, experi-
`mental structures of VH domains belonging to
`human germline families VH1, VH3 and VH4 can
`be found in the PDB. Modeling of the VL domains
`was easier, since for all of the four Vk and three Vl
`families representative structures can be found,
`although many of them only in the form of light
`chain dimers (Bence– Jones proteins), not as scFv,
`Fv or Fab fragments. The analyses of structural
`subtypes, core packing pattern, conserved hydro-
`gen bonding pattern, charge interactions, influence
`of mutations on stability etc. are on the basis of
`the analysis of all antibody variable domains with
`resolutions better than 3 A˚ represented in the PDB
`at the time of the analysis†. We wanted to address
`three questions regarding the domains in isolation:
`why is VH3 so extraordinarily stable; why do
`VH2, VH4 and VH6 behave comparatively poorly
`concerning expression and aggregation; and why
`did Vk domains give higher yields and are more
`stable than Vl domains?
`
`Salt-bridges
`
`Salt-bridges between oppositely charged amino
`acid residues and repulsions between equally
`charged amino acid residues play an important
`role in protein stability.31 Figure 4(a) shows a repre-
`sentation of an scFv fragment consisting of a Vk3
`and VH3 domain with its characteristic secondary
`structure. In Figure 4(b), positively charged resi-
`dues at pH 7.0 are colored in blue and negatively
`charged residues are colored in red. There is an
`accumulation of charged residues at the base of
`each domain. A hydrogen bond from the side-
`chain OH of
`tyrosine H104 to the main-chain
`carbonyl group of H100 orients residue H100,
`which forms a highly conserved buried salt-bridge
`with residue H77 (Figure 5(a)). In VH3, the two
`terminal nitrogen atoms of
`the Arg side-chain
`form a hydrogen bond to the two oxygen atoms of
`the side-chain carboxylate group of Asp H100.
`Arg H45 interacts with H100, as well as with Glu
`H53 and with the side-chain of Tyr H100. The
`more accessible and less highly conserved residues
`H99 and H97 complete the charge cluster in a
`germline family and subtype-dependent manner,
`with Glu H99 either interacting with an Arg in
`H97, as seen in the structure with PDB entry
`1IGM, or with Arg H45 as seen in structures with
`the PDB entries 1BJ1, 1INE, 2FB4 and 1VGE. Gln
`H77 in VH5 can form only a single hydrogen bond
`to Asp H100. In addition, to compensate for the
`shorter Gln side-chain, the side-chain Asp H100
`has to move slightly towards H77 and can no
`longer accept a hydrogen bond from Arg H45,
`
`† Presented in the AAAAA website at http://www.
`biochem.unizh.ch/antibody
`
`Ex. 1045 - Page 7 of 23
`
`

`

`538
`
`Biophysical Properties of Human Antibody Domains
`
`Figure 4. Model structure of an
`scFv fragment consisting of human
`consensus Vk3 (PDB entry: 1DH517)
`and VH3 domain (PDB entry:
`1DHU17).
`(a) Secondary structure
`with Vk3 on the left and VH3 on the
`right side (b) Colored for charged
`residues (blue: Arg, Lys and His;
`red: Asp and Glu). At the base of
`each domain is an accumulation
`of charged residues,
`the charge
`clusters of VL and VH domains.
`(c) Hydrophobic
`core
`residues:
`above the conserved Trp43 (green)
`is the upper core (red) and below
`the lower core (blue), see the text
`for details. (d) Positions possibly
`influencing folding efficiency are
`shown in orange; see the text for
`details. All images were generated
`using the program MOLMOL.33
`
`hydrogen bond to each of the two side-chains. In
`addition, the side-chain of Arg L77 can form a
`hydrogen bond to the main-chain carbonyl group
`of Glu L97 (Figure 5(b)). The least stable Vk domain
`Vk2 carries Leu at position L45, which is unable to
`form the side-chain to side-chain hydrogen bond
`to Tyr L104 conserved in the other VL domains
`and in VH domains (Figure 5(a) and (b)).
`
`Hydrophobic core packing
`
`Another important stabilizing factor is hydro-
`phobic core packing.32 All model structures were
`checked for
`cavities, which would indicate
`imperfect packing, leading to fewer van der Waals
`interactions and reduced thermodynamic stability.
`A van der Waals contact surface was generated
`for a water radius of 1.4 A˚ with the program
`MOLMOL (data not shown).33 When cavities were
`found,
`the surrounding residues were checked
`whether they would contribute hydrophobic sur-
`face area to the cavity. A cavity lined with hydro-
`phobic residues would be less desirable, as a
`water molecule would be energetically unfavorable
`at such a position. On the basis of these cavities
`and sequence comparisons between the dif-
`ferent variable domain frameworks, we identified
`positions in the hydrophobic core that may lead to
`sub-optimal packing. An overview of the analyzed
`core residues is given in Figure 4(c). The core
`residues can be divided into two regions:
`the
`upper and lower core according to the orientation
`shown in Figure 4(a). A layer of invariant residues,
`including the core Trp in position 43, the conserved
`disulfide bridge between Cys23 and Cys106, and
`Gln/Glu6 separate the upper from the lower core.
`The upper core is built of buried residues above
`(towards the CDRs) this central core of residues.
`
`opening up the network of charge interactions and
`hydrogen bonds.
`In VL domains (Figure 5(b)), the amino acid resi-
`due at position L45 is an uncharged Gln and those
`in positions L53 and L97 are either reversed com-
`pared to the amino acid residues at these positions
`in VH domains or are uncharged. Instead of two
`hydrogen bonds connecting the Arg L77 and Asp
`L10