PEDS Advance Access originally published online on May 3, 2006
Protein Engineering Design and Selection 2006 19(7):325-336; doi:10.1093/protein/gzl016
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Engineering stability into Escherichia coli secreted Fabs leads to increased functional expression
Department of Protein Therapeutics, Diversa Corp. 4955 Directors Place San Diego, CA 92121, USA
5To whom correspondence should be addressed. Email: ghansen{at}lpath.com and stephen.demarest{at}biogenidec.com
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionAcknowledgementsREFERENCES |
|---|
The recombinant expression of immunoglobulin domains, Fabs and scFvs in particular, in Escherichia coli can vary significantly from antibody to antibody. We hypothesized that poor Fab expression is often linked to poor intrinsic stability. To investigate this further, we applied a novel approach for stabilizing a poorly expressing anti-tetanus toxoid human Fab with a predisposition for being misfolded and non-functional. Forty-five residues within the Fab were chosen for saturation mutagenesis based on residue frequency analysis and positional entropy calculations. Using automated screening, we determined the approximate midpoint temperature of thermal denaturation (TM) for over 4000 library members with a maximum theoretical diversity of 855 unique mutations. This dataset led to the identification of 11 residue positions, primarily in the Fv region, which when mutated enhanced Fab stability. By combining these mutations, the TM of the Fab was increased to 92°C. Increases in Fab stability correlated with higher expressed Fab yields and higher levels of properly folded and functional protein. The mutations were selected based on their ability to increase the apparent stability of the Fab and therefore the exact mechanism behind the enhanced expression in E.coli remains undefined. The wild-type and two optimized Fabs were converted to an IgG1 format and expressed in mammalian cells. The optimized IgG1 molecules demonstrated identical gains in thermostability compared to the Fabs; however, the expression levels were unaffected suggesting that the eukaryotic secretion system is capable of correcting potential folding issues prevalent in E.coli. Overall, the results have significant implications for the bacterial expression of functional antibody domains as well as for the production of stable, high affinity therapeutic antibodies in mammalian cells.
Keywords: affinity maturation/antibody engineering/molecular evolution/protein folding/stable expression
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsREFERENCES |
|---|
The extraordinary sequence diversity of immunoglobulins is essential for their ability to recognize a wide variety of molecular targets. Antibodies utilize a combination of two variable domains, the N-terminal domain of the heavy chain (VH) and the N-terminal domain of the light chain (VL), to create an antigen-binding site. The genomic diversity of antibodies is derived from gene duplications of the VH and VL domains. During differentiation and selection, B-cells combine various groups of genomic V,(D) and J regions to obtain mature antibody sequences. Upon antigen stimulation, the variable domains are further diversified through the process of somatic mutation. Thus, variable domain sequences may be less homologous to their original germ line sequences than antibody constant domains which are not under such selective pressure. While this diversity is necessary for strong and specific antigen recognition, variable domains may compromise stability to gain added functionality. Instability within the variable domains can result in a number of adverse behaviors in both natural and recombinant antibody products (Carter and Merchant, 1997
).
The stability of recombinant antibody products can have a major impact upon their functional expression especially in non-native hosts such as Escherichia coli (Blight et al., 1994; Verma et al., 1998). As the antigen-binding domain is often the major cause of instability within expressed antibody constructs, stability optimization of the variable domains has become a common step in the development of recombinant antibody fragments or recombinant antibody libraries (Brinkmann et al., 1995; Dooley et al., 1998; Harris et al., 1999; Willuda et al., 1999; Knappik et al., 2000; Presta, 2002). Without such considerations, poor expression, aggregation or protein degradation can inhibit the development of a recombinant antibody product for in vitro or in vivo testing. Recently, in vitro recombinant antibody libraries based on scFvs or Fabs have become an attractive means of generating high affinity antibodies. This approach is gaining popularity over traditional hybridoma technologies due to its ability to generate human antibodies against toxic, infectious or self-antigens. Fab or scFv libraries are based primarily on PCR amplification of B-cell repertoires from immunized or non-immunized individuals or upon de novo CDR engineering (Knappik et al., 2000; Sidhu et al., 2004). The ability to screen such libraries for soluble antibody constructs which recognize an antigen of interest can be hindered by low protein expression and the difficulty of discriminating soluble hits from poorly folded or aggregated library members (Griffiths et al., 1994; Barbas and Burton, 1996; Vaughan, et al., 1996; Sheets et al., 1998; De Haard et al., 1999). Antibody affinity maturation or antibody engineering for novel properties can also adversely affect stability resulting in misbehaved antibody products (Dall'Acqua et al., 1998).
As the stability of antibodies has become a key issue in the development of recombinant methodologies for their selection and expression, exhaustive efforts have been made to stabilize individual antibody domains. These approaches have included consensus engineering (Steipe et al., 1994; Ohage and Steipe, 1999; Knappik et al., 2000; Ewert et al., 2003b; Demarest et al., 2004), PCR generated mutagenesis (Proba et al., 1998), structural design (Wörn and Plückthun, 2001; Ewert et al., 2004) and phage/ribosome display (Jung et al., 1999; Jermutus et al., 2001). Using these various approaches, the stabilization of several unique scFv or antibody domains has been accomplished. However, the information derived from these studies does not always apply to novel antibody constructs generated against antigens of specific interest. In this report, we implement a saturation mutagenesis evolutionary approach for optimizing suboptimal residues within any unique antibody sequence demonstrating poor expression or low solubility/stability.
We chose to optimize a poorly expressing human anti-tetanus toxoid (
TT) Fab with the goal of increasing its stability and expressed functional yield. The human hybridoma was derived from a single B-cell from the plasma of a patient immunized with the tetanus toxoid (ATCC#HB-8501; Bizzini, 1979). The anti-tetanus toxoid IgG was cloned and converted to a Fab for expression in E.coli. The Fab construct was chosen for this study due to its low expressed yields (
560 µg/l) and the variability of its functional potency following purification. We hypothesized that the folded state of the secreted Fab was unstable leading to the development of various misfolded and non-functional forms even though the Fab could be purified to a single band on a non-reducing SDS–PAGE gel. Therefore, the TT Fab appeared to be an ideal system to implement a directed mutagenesis approach (described in the Results) for the evolution of molecular stability. Optimization of the TT Fab's stability led to enhanced secretion of the functional form and an increased expressed yield in E.coli. The approach can be generally applied to any human antibody even in the IgG format to improve its stability parameters and half-life as demonstrated by the conversion of two optimized TT FABs into full-length IgG1 molecules. These optimized IgG1s demonstrated identical increases in thermal tolerance over the wild-type IgG1 molecule as was observed for the Fabs.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsREFERENCES |
|---|
Cloning, expression and purification
TT antibody domains
The TT Fab was cloned from a hybridoma cell line derived from an individual immunized with tetanus toxoid (ATCC#HB-8501). Briefly, the hybridoma cells were grown to a density of 107 cells, harvested and the mRNA was collected using the RNeasy Mini kit (Qiagen). The poly(A)+ RNA fraction was purified using an Oligotex mRNA mini kit (Qiagen) and used to generate first strand cDNA (Clontech cDNA synthesis kit). Antibody variable domains were amplified by PCR using a set of degenerate primers (Larrick et al., 1992) and subcloned into an in-house bacterial periplasmic expression vector containing human CH1/CL domains (kappa/IgG1), a C-terminal hexa-histidine tag on the heavy chain and the OmpA signal sequence. Fab vectors were transformed into Ultra BL21(DE3)pLysS for expression (EdgeBiosystems). Transformed cultures were induced with 100 µM IPTG as the optical density (600 nm) reached 0.8–1.0 and allowed to culture overnight at 25°C. Cell cultures were spun down and functional Fab was collected from the supernatant.
The isolated VH, VL, CH1 and CL domains were individually amplified from the TT Fab vector and subcloned into the pSE420 vector (Invitrogen) containing a customized C-terminal hexa-histidine tag. Expression of the single domains was performed as described above for the Fab or in the E.coli cytoplasmic compartment of BL21trxB(DE3) (Novagen). The isolated Fab domain constructs were collected from BL21trxB(DE3) cytoplasm by sonication, pelleting of the cellular debris by centrifugation at 14 000 rpm and harvesting of the resulting supernatant.
Full-length TT IgG1 molecules were subcloned from the Fab vectors and expressed in HEK293F cells. Briefly, custom primers were designed for PCR amplification of the heavy or light chain DNA from wild-type and variant TT Fab vectors. These inserts were ligated into the pCEP4 vector system (Invitrogen). The modified heavy chain pCEP4 vector also contained the additional sequence of the hinge–CH2–CH3 domains for construction of a complete IgG1 domain. Large-scale plasmid DNA was prepared as described by the manufacturer (Qiagen, endotoxin-free MaxiPrep kit). Plasmids were transfected into the adenovirus-transformed human embryonic kidney (HEK) cell line 293F using 293fectin and 293F-FreeStyle Media for culture. Light and heavy chain plasmids were both transfected at 0.5 µg/ml. Transfections were performed at a cell density of 106 cells/ml. The transfected cells were grown for 4 days, centrifuged at 1200 rpm and resuspended in fresh media. Based on in-house optimization of expression protocols, antibody secretions into the media were found to reach maximal levels after 7 total days of culture (data not shown). Further incubation led to decreases in antibody concentrations in the supernatant due presumably to cellular apoptosis and proteolysis. Supernatants were collected after 7 days by centrifugation at 1200 rpm for 8 min at 25°C followed by filtration through 0.2 µm bottle filters. The thermostability of the IgG1 constructs was assessed using an ELISA after heat treatment of the cell culture supernatants for 10 min using a temperature gradient ranging from 70 to 94°C.
Hexa-histidine tagged TT Fab, VH, VL, CH1 and CL were isolated from bacterial supernatants or soluble bacterial cytoplasmic fractions by the addition of Ni-NTA resin (Qiagen), mixing for 30 min and collecting the resin over a gravity column. The resin was generally washed with 100 ml phosphate buffered saline (PBS) solution with 25 mM imidazole (pH 7–8), before the protein was eluted with 300 mM imidazole at pH 8.6. TT IgG1 proteins were purified from 293F culture supernatants by passage over a 5 ml HiTrap Protein G affinity column (Amersham Biosciences) using an AKTA FPLC. IgG1 was eluted from the column using 0.1 M glycine, pH 2.8 and neutralized immediately with 1 M Tris, pH 8.0. All purified proteins were dialyzed exhaustively against PBS, pH 7.4, and stored at 4°C. Sodium azide (0.05%) was added to reduce bacterial growth.
Statistical analysis of antibody sequences
Statistical analysis of the TT Fab was performed using custom designed IgG databases (described in the Results) and a modified PERL script (Demarest et al., 2004). The amino acid frequency of every residue within the TT Fab was calculated from the custom database. The residue frequency, pi(r), for each position, i, in an individual sequence was calculated by the number of times that particular residue-type (r = A,C,D, ... , V,W,Y) is observed within the data set divided by the total number of sequences. Table I reports the TT Fab residue frequency divided by that of the database consensus residue providing a broader description of both the overall frequency of the native residue and the variability observed at that position. The positional entropy, N(i), was calculated as a measure of every residue position's variability (Shenkin et al., 1991; Larson and Davidson, 2000) and is a function of the information theoretic Shannon entropy, H(i):
 |
|
Screening for thermotolerance
To create the 4050 member mutant library, 45 residue positions were individually randomized as described previously (Kretz et al., 2004) using the gene site saturation mutagenesis (GSSMTM) gene evolution technology. In brief, a library of variants was generated using 32-fold degenerate oligonucleotides to saturate each desired position in the light chain and heavy chain variable domains. The procedure was carried out with oligonucleotide primers containing mutations located at the center of each primer sequence. Each mutation was flanked on either side by 20 bases of matching sequence. The coding strand primer contained NNK at the position to be mutagenized. N denotes an equal mixture of all 4 nt and K an equal mixture of G and T. This approach yields a mixture of 32 different sequences coding for all 20 amino acids and one stop codon.
To incorporate mutations, individual reactions were carried out with 50 ng of double-stranded DNA template, 2.5 U of PfuTurbo DNA polymerase and its corresponding buffer (Stratagene, cat. no. 600256), 10 mM dNTP mix (Applied Biosystems, cat. no. N808-0260) and 2.5 µM of each of the mutagenic oligonucleotides resuspended in 5 mM Tris–HCl (pH 8.0), and 0.1 mM EDTA. The initial denaturation was carried out at 95°C for 30 s, followed by 20 cycles of amplification: 95°C for 45 s, 50°C for 45 s and 68°C for 2 min. Following temperature cycling, the final reaction was then digested with DpnI digest (NEB cat. no. 176L) at 37°C for 4–5 h to remove methylated parental DNA. The resultant GSSMTM library was transformed into competent BL21(DE3) star pLysS E.coli and plated on LB-agar containing 100 µg/ml carbenicillin and 30 µg/ml chloramphenicol.
Ninety variant colonies and three control colonies were picked from each transformation to inoculate 45 total deep-well 96-well plates and grown to maximum density overnight at 37°C. These plates were utilized for the creation of glycerol stocks. A volume of 100 µl from every well of these cultures was also used to inoculate fresh 96-well plates for induction. After 1 h of culture at 37°C, cultures were induced with 100 µg/ml IPTG and allowed to grow overnight at 25°C. All 45 plates were centrifuged and the supernatants were collected in fresh plates containing a protease inhibitor cocktail and 0.02% sodium azide. The supernatants were aliquoted further into three separate 96-well plates and challenged at 70, 72 and 74°C for 10 min. Subsequent to heat challenges at the various temperatures, untreated supernatants and heat-treated supernatants were combined into single 384-well streptavidin labeled plates (Sigma) pre-coated with a biotinylated anti-HisTag antibody (PENTA-His, Qiagen). Fab from these supernatants was detected on the plate surface using a HRP-labeled anti-human kappa antibody (Sigma). TMB substrate was added, allowed to react for 10–20 min and neutralized using 1% phosphoric acid. Quantification of Fab was based on AU450nm measurements taken in the linear portion of the ELISA curve using purified wild-type TT Fab as a standard. During the development of the screen, centrifugation/filtration of the heat challenged supernatants prior to the quantitative ELISA did not affect the obtained values. Therefore, this extra step was not included in the screen.
Possible thermotolerant ‘Hits’ from the primary screen were picked in duplicate into five secondary plates for reconfirmation. These cultures were screened in an identical fashion as the primary library. The five plates were also recultured for DNA sequencing. Subsequent to reconfirmation, 12 combinations of stabilizing mutations (listed in Table II) were created using the QuikChange Site-Directed mutagenesis kit (Stratagene). Each of the 12 upmutant combinations (Table III) were cultured in 1 liter shake flasks and purified as described above.
|
|
Thermostability measurements
Thermostability measurements for the wild-type TT Fab and 12 upmutant Fabs were performed using an Applied Thermodynamics Model N-DSCII differential scanning calorimeter (DSC) and an Aviv Model 215 circular dichroism (CD) spectrophotometer. The purified Fabs were dialyzed against PBS prior to analysis. For the DSC experiments, the PBS dialysate was used in the reference cell. DSC experiments were performed using wild-type and variant TT Fabs at protein concentrations ranging from 0.3 to 0.7 mg/ml under a constant positive pressure of N2 (g). Protein concentration can influence the melting curves when aggregation occurs; however, this relatively small range of Fab concentrations should allow for qualitative stability comparisons between Fab molecules. DSC scan rates were universally kept at 1°C/min. Data analysis for the DSC experiments was performed using the software provided by the manufacturer. Excess heat capacities were obtained after subtraction of the buffer baseline. 
HVH was estimated from the calorimetry curves and the absolute intensity of the excess heat capacity of each transition using the approximation:
 |
Hcal is the excess enthalpy measured for the reaction by the calorimeter (Fukada et al., 1983). CD experiments were all performed at 3 µM Fab concentrations using a 1 cm cuvette with constant stirring. The instrument was equipped with a thermoelectric cuvette-holder for controlling the temperature. Included in the 0.29°C/min scan rate of the CD temperature melts was a 30 s stasis period at 1°C intervals utilized for signal averaging at 235 nm. The Fabs all demonstrate a negative trend in the ellipticity measured at 235 nm upon unfolding. This trend is identical to what is observed at the absolute minimum of the CD spectrum, 217 nm. The 235 nm was utilized for the temperature melts to reduce the total protein absorbance in the 1 cm cuvette which often interfered with the ability to accurately measure the signal at 217 nm. The bandwidth was held at 2 nm. Subsequent data analysis was performed using KaleighdaGraphTM. Thermal unfolding of the Fabs was universally irreversible as measured by DSC and CD. When the unfolding properties of the Fabs were monitored by CD, each Fab began to undergo what appeared to be a sigmoidal-like change in ellipticity. However, during the transition, the ellipticity of all Fab constructs spiked abruptly and positively as the protein fell out of solution. The apparent TM of each Fab construct was determined by the temperature at which this spike occurred.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsREFERENCES |
|---|
Mutagenesis of Fab residues based on positional frequency and entropy
In a first step, TT Fab positions with the potential for stability optimization were identified. All four Fab domains (VH, CH1, VL and CL) were considered for optimization as it was unclear to what extent each individual domain limits or enhances the overall stability of the Fab molecule. Mammalian VH, CH1, VL and CL sequences were compiled from the NCBI non-redundant protein database and any sequence with greater than 95% identity to another sequence in this customized database was excluded to reduce bias. Stricter diversity requirements are necessary if the goal is to analyze residue covariation across different members of a structural subclass of protein such as the entire Ig superfamily (Larson and Davidson, 2000; Russ and Ranganathan, 2002). By restricting the database to mammalian IgG/kappa sequences as opposed to other members of the Ig superfamily, the complication of covarying residue pairs can be reduced (Demarest et al., 2004). The database was used to identify single residue positions within the Fab for saturation mutagenesis as opposed to simultaneous mutagenesis of multiple residue positions. Multiple single residue libraries leads to arithmetic increases in the number of variants for screening as opposed to the geometric increases observed for combinatorial mutagenesis.
Two independent criteria were used to identify residues within the TT Fab for mutagenesis. First, the VH, CH1, VL and CL sequence database was used to define highly variable residue positions within the Fv frameworks and constant domains. The variability of every Fab residue was scored based on the positional entropy calculated from our database of Fab sequences (see Materials and Methods). Highly variable positions should tolerate random mutagenesis and present an increased probability of finding alternative amino acids superior to the native residue. Additional sites for mutagenesis were selected if the wild-type residue at a given position was occupied by an amino acid infrequently observed in the sequence database. Based on these two statistical guidelines, 45 residues within the heavy and light chains of the Fab were chosen for mutagenesis (Table I).
Using saturation mutagenesis (Kretz et al., 2004), a library of variants was created which allowed for the substitution of all 19 other amino acids at each of the 45 Fab positions. The resulting Fab library consisted of 4050 members with a maximum theoretical diversity of 855 unique mutations. A single 96-well plate per residue position was created for the screening efforts with 90 positions used for the library and 6 positions used for wild-type and no vector controls. Using the ‘NNK’ degenerative codon strategy, each amino acid type was represented an average of two times at each randomized position and the theoretical number of unique amino acids (including the wild-type amino acid) that gets represented per plate was between 16 and 17. Six of the 45 plates were fully sequenced to insure that the expected mutagenesis ratio was intact.
Thermotolerance screening of the TT Fab library
In a first step, the expression level of each Fab variant was determined using a quantitative ELISA. The measured expression level of each Fab variant was used to determine the fraction of remaining Fab material subsequent to thermotolerance screening. This allowed us to eliminate expression level as an assay variable and focus on the stability of each construct. Next, all 4050 variants of the library were screened for stability by subjecting the supernatants to a thermotolerance screen. By challenging at three separate temperatures (70, 72 and 74°C) it was possible to determine approximate midpoints of the thermal unfolding transitions (TM) for every Fab variant. These temperatures were chosen based on the denaturation profile observed for the wild-type TT FAB after thermochallenging between a broad range of temperatures between 50 and 90°C. The majority of library members had reduced thermostability and expression. Variants that demonstrated enhanced thermostability were re-picked from master plates, re-expressed and re-subjected in duplicate to the thermotolerance screen to confirm their properties. Several examples of variant ‘Hits’ exhibiting superior properties are shown in Fig. 1A. A complete list of stabilizing mutations is provided in Table II. Approximately 1% of the variants in the library exhibited enhanced thermostability.
|
The majority (14) of non-optimal residue positions were located in the VH domain with the remaining few (4) in the VL domain (Table II and Fig. 2). This result suggests the stability of the native Fab was limited by the marginal stability of the VH. Surprisingly, no mutant within the
2000 member constant domain library was found to stabilize the Fab as a whole. We speculate that stabilizing mutations within the constant domains really do occur, but that the limited stability of the VH domain controls the temperature at which the Fab unfolds and denied our ability to observe such events (see DSC experiments below). Another explanation for the inability to discover stabilizing mutations within the CL/CH1 region may be that the 10 min unfolding period at elevated temperatures was not sufficient for fully unfolding the constant domains. Röthlisberger and coworkers (2005) found that unfolding of the CL/CH1 heterodimer is extremely slow and that the stability of all four Fab domains benefit from this kinetic stabilization. It is possible therefore that the constant domains may not unfold during the relatively short period of the thermochallenge (10 min).
|
,
or 
1,2,... dihedral angles. W50 in VL was found to sterically clash with CDR3 of VH. Mutation to His relieves this clash while still providing tight packing with neighboring VL residues and CDR3 of the VH. Mutation to His also allows for a potential hydrogen bond interaction with Y49 of VL. (C) Stabilizing mutations within the VH ß-sheets. Mutation of residue 72 to Asn may allow for better side chain packing and hydrogen bonding than the original residue, Thr. Supporting this notion, the isosteric Asp is the consensus residue at this position for most VH classes. Mutation of residues 78 and 24 to Thr and Ala, respectively, likely improve stability by increasing the ß-sheet propensity over the wild-type amino acids (S and G). The native Gly at residue 24 of Some of the stabilizing mutations found within the
TT Fab were changes to the human VH3 consensus. The VH3 subclass contains over 22 germline genes and represents the largest subclass of antibody variable domains (Cook et al., 1994; Knappik et al., 2000). The VH3 subclass has also been identified as the most stable and soluble of the VH sub-families based on consensus models of each family (Ewert et al., 2003a). However, this observation does not predicate that the stabilities of individual VH3 family members are equivalent. Unique VH3 sequences, including the TT VH domain, generally contain a number of non-consensus framework residues which may not be optimized for domain stability. Stabilizing mutations to VH3 consensus residues were V11L, G24A and S77T. Position 72 within the TT VH was also not optimized for stability. Its consensus residue in all germline VH subfamilies is Asp. After sequencing the library plate for residue 72, it was discovered that Asp was serendipitously not represented. Ewert and coworkers identified several VH positions important for VH3 domain stability (Ewert et al., 2003a,b). Asp at position 72 was one residue they demonstrated to be highly important for VH stability. Our mutation to the isosteric residue Asn was highly stabilizing over the native Thr. VH72N has been observed in human VH domains, but not at the overwhelming frequency observed for Asp (Chothia et al., 1998).
Several other stabilizing mutations identified by the screen did not conform to the VH3 consensus. A few of these stabilizing mutants were changes to the consensus residue found in other human VH subfamilies (Chothia et al., 1998; Knappik et al., 2000; Honegger and Plückthun, 2001). For instance, Ala23 is the consensus residue for VH3 domains, but mutation to Lys, the consensus residue for VH1a, VH1b and VH5, led to a greater than 2°C increase in the TM of the Fab. Other mutations that did not conform to the VH3 consensus were A84P and F89T. Thr is the consensus residue at position 89 for the VH2 subfamily while Pro is the consensus residue at position 84 for both the VH2 and VH6 subfamilies. Interestingly, Val is the overwhelming consensus residue at position 89 for most human VH subfamilies, but mutation to Val did not improve Fab stability. The isosteric VH2 consensus residue Thr was highly stabilizing in the screen (TM > 2°C). The similarly beta-branched Ile was one of the most stabilizing mutations discovered in the screen and is found occasionally at this position in human VH sequences, but is not the consensus residue for any VH subclass. Amazingly, all suboptimal positions within the TT VH domain could be optimized by replacement of an amino acid type that exists in a fraction of human VH sequences.
Four stabilizing mutants were discovered within the TT VL domain. Mutation away from the VK4 consensus residue, Trp, at residue 50 to the consensus for VK1, Ala, was highly stabilizing in the thermotolerance screen. A previous VL optimization study found that Gly at residue 50 was highly deleterious to VL stability and mutation to Glu was highly stabilizing (Ohage and Steipe, 1999). Based on our results, introduction of the single methyl group to obtain an Ala was enough to stabilize the TT Fab. Mutation at this residue to His was also highly stabilizing, even more so than Ala. His is rarely found at position 50 in human kappa variable domains, but is often found in human lambda variable domains. This residue is close to the VH/VL domain interface. We hypothesized that its contribution to Fab stability might be linked to potential buttressing of the VH domain since the VH domain in particular appears to limit the stability of the TT Fab. However, studies with the isolated VL domain suggest otherwise (described below). Mutation from Tyr to His at this position has been reported to improve the apparent stability and expression of a scFv lacking a disulfide (Proba et al., 1998). Thus, it appears that large aromatic groups at residue 50, the most C-terminal residue of VL framework 2, were not favorable for stability in multiple cases (Fig. 2B). In both the disulfide-free antibody study and in our study, mutation at this position did not negatively attenuate antigen binding and may actually have increased antigen recognition by stabilization of the folded/active form of the variable domains (described below). D9T and N22H appeared to be stabilizing in the thermotolerance screen. These two mutations increased the TM by <1°C. The V
consensus mutations at these two positions were not present after sequencing each of their 96-well libraries. This sequencing data indicates that these residue positions may be further optimized by substitution with V consensus mutations. Therefore, mutations to the V consensus were included in the combination study described below.
Stability measurement of Fab variants containing mutant combinations
Based on the results of the library thermotolerance screen, a total of 12 constructs were created which contained between 3 and 11 stabilizing mutations identified in the screen (Table III). The various combinations were derived rationally to enable us to determine the apparent contribution each mutant provides toward Fab stabilization. The assumption was made that each mutant would contribute to the overall stability of the Fab in an additive manner. We observed additive increases in stability when making consensus driven stabilizing mutations into the bovine CH3 domain in a previous study (Demarest et al., 2004). If this assumption holds true, then experimental determination of the relative stabilities of all 12 upmutant combinations will provide enough data for us to deconvolute the various contributions each mutation makes towards stabilizing the Fab. This can be done easily using the multiple solutions (i.e. experimentally derived Fab stabilities) to solve for multiple unknowns (i.e. the individual contributions each mutation makes to the overall Fab stability). All 12 constructs were expressed in bacteria and the secreted proteins were purified from the supernatant. Introduction of multiple stabilizing mutations to the TT Fab increased Fab expression. Bacterial expression of the wild-type Fab along with all 12 multiple mutant combinations was performed under identical induction/growth conditions. All stability optimized constructs yielded more protein than the wild-type Fab. The best constructs consistently exhibited a greater than 3-fold increase yield over wild-type (Table III).
The stability of each Fab was evaluated by differential scanning calorimetry (DSC; Fig. 1B) and CD. DSC and CD scans were universally irreversible and the solutions were highly turbid subsequent to heating. The scan rates for the CD and DSC experiments were 0.29 and 1°C/min, respectively. The midpoints of the thermal unfolding transitions (TM) of each Fab were consistently lower in the slower scanning CD experiments (–5.0 ± 0.3°C), a hallmark of irreversible unfolding transitions (Varea et al., 2004). Also in disagreement with a two-state approximation, the average Hcal/HVH ratio for the Fabs was 3.5 as opposed to unity. A similar irreversible unfolding behavior was reported previously for the 4D5 Fab construct (Kelley et al., 1992). Owing to the aggregation behavior of the Fabs, we did not attempt to interpret the unfolding data in terms of equilibrium-free energy differences between the folded and unfolded states. Instead we utilized the measured TM values to rank order the apparent stability of each Fab considering the ‘Stability’ of the Fabs to be a combination of multiple factors including the free energy of folding and the apparent propensity of each Fab to aggregate. However, the stabilized Fabs all maintain a 5°C TM difference between the two scan rates suggesting the mutations are generally more important for improving the free energy of folding and not for modifying the apparent aggregation behavior.
Cp° values between the folded and unfolded states of the TT Fab variants were on average 7.9 kcal mol–1 K–1 and very close to the theoretical number calculated based on the potential surface area difference between the folded and unfolded states of the Fab as a whole (7.7 kcal mol–1 K–1) and different from values expected for isolated unfolding of the VH, VL, CL or CH1 domains (Myers et al., 1995). Coupled unfolding of all four domains is a common although not universal feature of Fabs provided they are fully disulfide linked both intramolecularly and intermolecularly (Kelley et al., 1992; Shimba et al., 1995; Vermeer et al., 2000; Röthlisberger et al., 2005). Uncoupling of the unfolding reactions of multidomain proteins based on substantial differences between the intrinsic stabilities of the individual domains is not uncommon and depends on the overall affinity of the domains for one another and the total interaction surface between the domains (Ewert et al., 2003a; Varea et al., 2004). In the absence of any portion of the Fab, unfolding of the remaining domains may not be coupled. Rowe and Tanford (1973) have demonstrated that the VL and CL within a kappa light chain appear to denature in an uncoupled fashion when not attached to a heavy chain. It appears additional contacts with a heavy chain must be required for cooperativity. Röthlisberger and coworkers (2005) have recently reported exhaustive and elegant experiments demonstrating how VH and VL domains of differing stability can unfold separately from one another while those with high and similar stabilities can unite the unfolding transitions of all four domains of the Fab. The most stable Fab construct BD16198 (TM = 92°C) may be on the border of uncoupling the thermal unfolding reactions of its variable domains as well. This construct displayed an asymmetric excess heat capacity by DSC (Fig. 1B) perhaps due to the excess stability of the VH or Fv components. Wörn and Plückthun (1998) witnessed a similar disconnect in the equilibrium unfolding behaviors of the VH and VL of the ABPC48 scFv containing a disulfide stabilized VH domain.
The TMs of all 12 Fab variants were used to deconvolute the contribution each of the 11 mutations make to the stability of the Fab (i.e. multiple equations to solve multiple unknowns). Surprisingly, all mutations appeared to contribute to the stability of the Fab in an additive manner even though several of these residues are relatively close to one another both in primary sequence and within the tertiary fold of the Fab. The analysis indicated that four mutations account for 86% of the Fab stabilization. G24A, T72N and F89I mutations within the VH domain individually increased the TM of the Fab construct by 3.0, 2.8 and 2.8°C, respectively. W50H mutation within the VL domain increased the thermostability of the Fab by 3.5°C. Other substitutions at positions VH 89 and VL 50 within the Fab also led to enhanced stability although none as great as the VH F89I and VL W50H mutations listed in Table II. This was expected for residue VH 89 since the wild-type residue, Phe, is rarely observed in antibody VH sequences. The inherent instability of Trp at position 50 of the light chain was unexpected considering Trp was the most commonly observed residue in our kappa database.
Single domain studies
The majority of the stabilizing mutations were within the VH domain. Very few were found within the VL and we hypothesized that those that were found, particularly W50H/A which lies close to the VH/VL interface, stabilize the Fab by making direct contributions to VH stability. To test this hypothesis, we attempted to express all four TT Fab domains by themselves and assess the consequence of point mutations in these formats. Only the VL and CL expressed stably in E.coli. VL and CL were produced cytoplasmically using the oxidative strain BL21trxB(DE3) (Proba et al., 1995). Oxidative cytoplasmic expression using the BL21trxB(DE3) strain yielded correctly folded forms of both the VL and CL domains as judged by CD with a single unfolding transition observed by CD and DSC (see Fig. 1C for DSC data). There was no evidence for reduced VL or CL by SDS–PAGE or by heterogeneity in the thermal unfolding profiles of the two proteins. VL secreted from BL21(DE3)pLysS had an unusual CD spectrum and its thermal unfolding profile was non-sigmoidal and underwent multiple changes at temperatures well below the TM observed for the VL recovered from the cytoplasm of BL21trxB(DE3) (data not shown). The TT VH and CH1 domains did not yield enough soluble or insoluble protein in any bacterial strain for analysis.
The concept that Fab stabilizing mutations found in VL may be important only for scaffolding of the VH domain was too simplistic based on the effect these mutations had on the isolated VL domain. Mutation of VL W50 to Ala increased the TM of the isolated VL domain by 2.3°C, similar to the 2.1°C TM increase observed for this mutation within the TT Fab (data not shown). Thus, stabilizing mutations in both the TT VH and VL domains can increase the apparent stability of the entire Fab molecule.
The wild-type VL had a greater intrinsic stability than the CL domain (Fig. 1C). Additional interactions with the CH1 domain and with the variable domains appear to be necessary to generate a coupled unfolding reaction. The interaction strength between variable domains is believed to be somewhat weaker and thus these domains may not depend as highly upon heterodimerization for their stability as do the constant domains. This may explain why the TT VL as well as other VLs have a greater intrinsic stability than CL (Horne et al., 1982).
Observation of conformational heterogeneity in the least stable Fabs
All 12 upmutant constructs and 2 independent wild-type Fab preparations were analyzed by analytical size exclusion chromatography. All Fabs eluted at the expected 50 kDa molecular weight based on several molecular weight standards (Demarest et al., 2004). Wild-type and weakly stabilized TT Fabs appeared to have a second, broad peak at low (less than monomer) molecular weights (Fig. 3A). This peak likely reflects some misfolded fraction with exposed hydrophobics non-specifically associating with the gel filtration matrix. Interestingly, both wild-type preps contained equal proportions of monomer and ‘misfolded' material indicating that the protein may be in equilibrium between the two forms. Highly stabilized TT Fabs did not display this aberrant behavior and eluted uniformly as monodisperse proteins (Fig. 3B). All Fabs were isolated using the HisTag at the C-terminus of CH1 and therefore we do not believe the wild-type preparations contain isolated light chain as a contaminant. Additionally, the purified Fabs all run as single 48 kDa bands on non-reducing SDS–PAGE gels precluding other protein contaminants as a possible cause for the additional SEC peak.
|
Linkage between the Fab thermal stability and the functional protein fraction
One important consideration subsequent to framework mutagenesis was the potential effect this might have on the antigen-binding function of the TT Fab. The thermostabilizing mutations were derived from the original screen using a quantitative ELISA detecting both the CL domain and the HisTag at the C-terminus of CH1. It has often been speculated that the dynamics of an Fv domain can have a profound effect on antigen binding. We were concerned that the stabilizing mutations introduced into the Fv framework regions may interrupt a dynamic balance important for docking the antigen.
The thermostabilizing mutations did not attenuate the function of the molecule, but in fact enhanced the apparent affinity of the TT Fab in a functional ELISA using biotinylated tetanus toxoid (Fig. 3C and Table III). Subsequent to purification, the two independent wild-type preparations did not titrate equally with the antigen in a functional ELISA indicating that small variations during cell culture may subtly control the amount of functional Fab that gets expressed. Both wild-type preparations yielded less functional protein than most of the stabilized TT Fab constructs. The functional capacity of each Fab variant appeared to be directly correlated to its stability as described by the TM (Fig. 3D). Interestingly, each of the Fab variants was affinity purified to a single band on an SDS–PAGE gel before introduction into the functional ELISA. As the mutations were distant from the predicted binding loops, it is unlikely that these residues directly enhance contact with the antigen. Based on the behavior of the wild-type TT Fab, we believe that a fraction of the Fabs secreted from E.coli are in a non-functional or misfolded form. This interpretation correlates well with the fact that the isolated VL domain, when secreted from E.coli, was in a very different conformation than the VL expressed in the oxidizing cytosolic environment of BL21trxB(DE3). This also explains the functional variation observed between batch productions of Fab. By stabilizing the proper folded form of the molecule, the overall fraction of secreted functional Fab was likely increased.
Binding experiments using surface plasmon resonance up to 1 µM antigen concentrations demonstrated that the affinity between the two binding partners was fairly low (i.e. KD > 10–6 M). Owing to the limited availability of tetanus toxoid, determination of the absolute binding constants was not possible.
Thermostabilizing mutations also enhance the properties of full-length TT IgG1
As the unfolding of the Fab and Fc are generally independent events, it was anticipated that the enhanced stability of the variant Fabs would translate to the IgG format. However, we did not know whether the increase in expression and functional protein fraction would also apply to a full-length antibody expressed in mammalian cells. To test these unknowns, the BD16198, BD16200 and the wild-type Fabs were converted into separate full-length IgG1 and kappa vectors for expression in mammalian cells. The BD16198 and BD16200 Fabs were selected because they were the most stable and highly functional variants produced in E.coli. The mammalian vectors containing full-length antibody inserts were transfected into HEK293F cells for transient antibody expression. This expression system should contain all the correct processing machinery to secrete an antibody in its proper folded and functional form.
The thermostability enhancements observed for the BD16198 and BD16200 Fabs were preserved in the full-length IgG constructs. The wild-type TT IgG denatured at 72°C in the thermotolerance assay—exactly the same temperature as the wild-type TT Fab in the thermotolerance screen. IgG was detected in the assay using an anti-human IgG as opposed to the anti-human kappa which was used during the initial Fab screen. The BD16198 and BD16200 IgGs denatured at 86°C and 82°C, respectively (Fig. 4). These TM increases matched the stability increases observed for the optimized BD16198 and BD16200 Fabs. Thus, the apparent thermostability of the full-length TT antibody is also limited by the Fv portion of the molecule. The TMs of the CH2 and CH3 domains of a human IgG1 Fc fragment with a fully disulfide-linked hinge domain was determined by DSC under the same conditions used to measure the Fab TMs (S.J.D., unpublished results). The apparent TM of the CH2 domain (69.4°C) of human IgG1 was significantly lower than the apparent TM of the BD16198 and BD16200 Fabs. However, the ability of the antibody to persist in solution upon thermochallenge appears to be determined by the thermotolerance of the Fab as measured in our ELISA format. This result suggests that irreversible Fab unfolding dictates the thermotolerance of the TT IgG1 regardless of the fact that the CH2 domain has a lower TM value.
|
The expression level of the optimized
TT IgGs was no different from that of the wild-type IgG and the significant functional gains observed for the mutant Fabs were not reflected in their IgG counterparts. Apparently, the folding issues observed for the E.coli produced Fabs were alleviated using the mammalian expression system. The mammalian expression system appears to tolerate the low stability/poorer folding kinetics of the wild-type TT IgG. The native secretory pathway of the mammalian system including the chaperonins, isomerases and oxidative machinery must be attuned to dealing with the complicated assembly of antibody molecules, whereas the E.coli system may be unable to cope with certain kinetic or thermodynamic traps which can occur in non-bacterial proteins. Mutations within the frameworks have been shown to affect antibody secretion levels from mammalian cells. Saldanha et al. (1999) demonstrated a 40-fold increase in transient cos cell expression levels from 0.04 to 2.2 mg/l upon a single VL mutation at residue 9 from Asp to Ala. The transient expression level of our wild-type TT IgG1 was between 1.5 and 2.0 mg/l—similar to the highest secretion levels observed in the Saldanha et al. (1999) study. Thus, it may be difficult to improve the transient expression levels by making stabilizing mutations to the TT IgG.
The BD16198 and BD16200 IgGs were tested for increased binding using the tetanus toxoid functional ELISA (data not shown). While functional increases as observed for the E.coli produced Fabs (15- to 20-fold over the wild-type protein) were not observed for the optimized IgGs, a residual gain in function was possible for the BD16200 construct over wild-type (2-fold increase); however, this data was not statistically above the error of the measurement. The BD16200 Fab construct demonstrated the most significant gain in function over all the other optimized Fab variants and it is attractive to think that gains in stability may help solidify the antigen-binding conformation of the IgG. However, the data suggests if this really was an effect of Fv stabilization, it was fairly insignificant compared to the priming of the correctly folded Fab by the mammalian chaperonin/secretory system. Thus, while stability/folding engineering may affect the bacterial expression/secretion of functional antibodies, such functional gains due to stability/folding may not manifest themselves in mammalian expression systems.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionAcknowledgementsREFERENCES |
|---|
Optimization of antibody frameworks to increase Fv stability is an approach commonly used in antibody therapeutics (Carter and Merchant, 1997
; Presta, 2002). Significant work has been done to understand the differential stabilities of variable domains as well as pinpoint commonly found weaknesses in variable domain frameworks (Steipe et al., 1994; Knappik and Plückthun, 1995; Nieba et al., 1997; Proba et al., 1998). Problems with differential expression and activity within recombinant antibody libraries have led to the development of consensus frameworks for biasing antibodies away from poorly expressed, partially functional or ‘sticky' family members (Knappik et al., 2000). A biophysical study of single antibody domains derived from these consensus VH/VL families led to the discovery of non-optimized residues especially within the even numbered VH subclasses (VH2, VH4 and VH6, Ewert et al., 2003a). This information has been used to optimize the even numbered consensus derived frameworks for enhanced stability and potentially greater yield of active protein in bacterial or phage libraries (Ewert et al., 2003b). One general principle revealed in our study is that the individual VH subclasses (VH1-6) all appeared to have specific consensus residues within their frameworks optimal for Fv stability and the VH3 consensus residue was not always the most favorable. In fact, mutation away from the natural TT VH3 consensus residue in some cases was necessary to achieve optimal stability within the TT antibody. Native antibody sequences derived from the germlines with some level of hypersomatic mutation will exhibit unique weaknesses in framework residues compared to the consensus sequence of each variable domain subclass (Honegger and Plückthun, 2001). The approach described here can be applied to any antibody system, poorly behaved or not, without relying on previously defined mutagenesis and/or thermodynamic data. Consensus driven approaches towards protein stability are not new; however, unlike many earlier studies, the saturation mutagenesis approach described here also provides complete information pertaining to the suitability of nearly all 20 amino acids at every randomized position within a variable domain.
Randomizing residue positions based on high positional entropy enabled additional sites to be found for stability optimization over those discovered based on low residue frequencies of the wild-type residues. The most stabilizing mutation discovered in the entire TT Fab library was VL W50H. This residue was randomized based on its positional entropy even though the native Trp was the most commonly observed residue in our database. Subsequent to our discovery, we found two previous studies indicating that residue 50 was a critical site for two other VL domains (Proba et al., 1998; Ohage and Steipe, 1999). Both of these groups introduced the VL50 mutations into their constructs rationally. Ohage and Steipe (1999) used the myeloma protein REI as a model. No prior stability or structural information was used for choosing residue 50 as one of the 45 positions for optimization. If we had adhered to consensus modeling and not included positional entropy as a factor for mutagenesis, this valuable optimization would have been overlooked. Fersht and coworkers have used positional entropy to optimize the stability of both GroEL and the p53 DNA binding domain (Nikolova et al., 1998; Wang et al., 1999). While the success rate for identifying stabilizing residues was definitely much lower than using consensus modeling, they were also able to identify several stabilizing mutations which would have been missed using consensus modeling.
Initially, we were concerned that the dominant force determining protein stability, i.e. the hydrophobic effect, would bias the screen to introduce large hydrophobic residues adverse to fold specificity. The thermotolerance screen could also be biased towards the discovery of charged or polar amino acids using our screening methods. Since thermal denaturation is a non-equilibrium process, mutations which improve Fab aggregation properties or the solubility of Fab (un)folding intermediates may increase the apparent TMs (Nieba et al., 1997; Jespers et al., 2004). However, the difference between TMs measured at the two scan rates remains constant for all mutant combinations, suggesting real changes in the free energy difference between the folded and unfolded states and less significant changes in the solubility of folding intermediates or in aggregation rates. Mutations derived from the screen appeared to stabilize the Fab in many ways including enhancements of turns, ß-sheet propensities, core packing arrangements and the matching of electrostatic charge potentials. Thus, the library approach did not appear to bias towards any one method of stabilization.
Increased functional expression of antibody scFv and Fab fragments by mutagenesis of faulty CDR and/or framework residues is often necessary to obtain fully functional and soluble proteins. Wall and Plückthun (1999) have demonstrated that hierarchical mutations within scFv fragments can be necessary and important for improving functional expression—i.e. key residues must be ‘fixed' before further mutagenesis efforts result in improved functional expression (Wall and Plückthun, 1999). Three key positions were identified by Knappik and Plückthun (1995) in the turns of VH domains for increasing soluble Fv expression in the periplasm of bacteria (Kabat numbers Ala40, Ala60 and Asp61; Knappik and Plückthun, 1995). These mutations did not increase the in vitro stability of their construct, but appeared to increase the rate of oxidative folding, reducing the tendency to aggregate. The wild-type TT VH has the optimal amino acids at these critical positions. Therefore, these positions are unlikely contributors to the expression problems observed with the wild-type TT Fab. Nearly all the mutant combinations which were stabilizing to our Fab also increased the soluble expressed yield of bacterially produced TT Fab (Table III) arguing that hierarchical changes were not necessary in this particular case. Increases in ‘functional’ expression on the other hand correlated fairly well with increases in overall stability as shown in Fig. 3D. While we did not investigate the rate of oxidative folding for each of these constructs, an increase in the folding rate is one of two kinetic parameters which can lead to enhanced equilibrium stability of two-state folding proteins (i.e. KUnfolding = kunfolding/kfolding). As all stabilized Fabs in this report exhibited a similar irreversible aggregation behavior compared to the wild-type TT Fab, it is likely the majority of the mutations discovered in our screen contributes to the free energy difference between the folded and unfolded states of the Fab and therefore also affect the folding and/or unfolding rates of the variable domains.
In conclusion, we have used a mutagenesis approach to introduce all 20 amino acids at 45 different residue positions within an TT Fab yielding valuable data concerning the effect various amino acid substitutions throughout the framework have on Fab thermostability. Using an automated screening approach, we were able to approximate TM values for a 4050 member Fab library with a theoretical maximum of 855 unique mutations. The result of the study was the ability to provide an abundance of residue-specific information pertaining to Fv and Fab stability. While intolerant residues within the constant domains could be readily identified, no stabilizing mutations within these domains were found presumably due to the limiting stability of the Fv domain. Single mutants that increased the intrinsic stability of either the VL or VH domains were capable of enhancing the overall stability of the Fab. Interestingly, all of the stabilizing residues introduced into the TT Fab can be found in at least a fraction of native human antibody sequences (Chothia et al., 1998; Honegger and Plückthun, 2001) suggesting the approach may be used to stabilize antibodies without introducing non-human or immunogenic changes (Carter and Merchant, 1997). Fab expression correlated well with increased Fab stability. Combinations of these stabilizing mutations were found to increase the functional fraction of the expressed Fabs in an antigen-binding ELISA. Thus, we demonstrate that low apparent affinities (as observed for the TT Fab) may in some cases be due to the partitioning of the Fv domain into non-functional or misfolded forms. Application of these mutants to the full-length TT IgG stabilized the molecule to the same degree as the Fab. Increases in Fab stability in the IgG format expressed in mammalian cells did not additionally enhance the expression or functional activity of the antibody. The mammalian system appears capable of properly processing and secreting the unmodified antibody. This implies that the increased bacterial expression of stability optimized Fabs may be the result of enhanced folding rates and/or a decreased propensity for protein aggregation during or after folding. However, the mutations were selected based on their ability to increase the apparent stability of the Fab and therefore the exact mechanism behind the enhanced expression in E.coli remains undefined.
|
Footnotes |
|---|
1Present address: Biogen Idec, 5200 Research Place,San Diego, CA 92122, USA
2Present address: Ambrx, Inc., 10410 Science Center Drive, San Diego, CA 92121, USA
3Present address: E.O. Wilson Biodiversity Foundation, 10190 Telesis Court, San Diego, CA 92121, USA
4Present address: Lpath Therapeutics, Inc., 5335 Ferris Sq., Suite A. San Diego, CA 92121, USA
|
Acknowledgements |
|---|
|
TopAbstract |
|---|