Design and Optimization of Anti-amyloid Domain Antibodies Specific for β-Amyloid and Islet Amyloid Polypeptide*

Antibodies with conformational specificity are important for detecting and interfering with polypeptide aggregation linked to several human disorders. We are developing a motif-grafting approach for designing lead antibody candidates specific for amyloid-forming polypeptides such as the Alzheimer peptide (Aβ). This approach involves grafting amyloidogenic peptide segments into the complementarity-determining regions (CDRs) of single-domain (VH) antibodies. Here we have investigated the impact of polar mutations inserted at the edges of a large hydrophobic Aβ42 peptide segment (Aβ residues 17–42) in CDR3 on the solubility and conformational specificity of the corresponding VH domains. We find that VH expression and solubility are strongly enhanced by introducing multiple negatively charged or asparagine residues at the edges of CDR3, whereas other polar mutations are less effective (glutamine and serine) or ineffective (threonine, lysine, and arginine). Moreover, Aβ VH domains with negatively charged CDR3 mutations show significant preference for recognizing Aβ fibrils relative to Aβ monomers, whereas the same VH domains with other polar CDR3 mutations recognize both Aβ conformers. We observe similar behavior for a VH domain grafted with a large hydrophobic peptide from islet amyloid polypeptide (residues 8–37) that contains negatively charged mutations at the edges of CDR3. These findings highlight the sensitivity of antibody binding and solubility to residues at the edges of CDRs, and provide guidelines for designing other grafted antibody fragments with hydrophobic binding loops.

The misfolding and assembly of peptides and proteins into prefibrillar oligomers and amyloid fibrils is linked to several neurodegenerative diseases (1)(2)(3)(4). To understand the contributions of polypeptide aggregation to such disorders, it is important to characterize the biochemical properties of aggregates. However, pre-amyloid and amyloid aggregates are difficult to characterize using many traditional biochemical methods that are routinely used for soluble proteins. High-resolution structural analysis of such aggregates is particularly challenging and must be performed using specialized methods such as solid state NMR (5)(6)(7)(8)(9)(10)(11)(12)(13) or x-ray crystallography of small peptide fragments (14 -16). These powerful structural methods generally lack the time resolution to probe pre-amyloid intermediates and oligomers unless they can be kinetically trapped.
Given these challenges, antibodies with conformational specificity for prefibrillar oligomers and amyloid fibrils have proven valuable for biochemical characterization (17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28). An obvious strength of such antibodies is their ability to detect specific types of aggregates formed both in vivo and in vitro. The ability of conformational antibodies to bridge in vivo and in vitro studies is important for understanding the biochemical mechanisms that contribute to protein misfolding disorders.
Several approaches have been used to generate conformational antibodies. The most widely used one is immunization, which has yielded a wide range of important antibodies specific for different types of prefibrillar oligomers and amyloid fibrils (17)(18)(19)(20)(25)(26)(27). Another powerful approach that has proven useful is to employ in vitro display methods such as phage and yeast surface display (21-24, 29 -37). These display methods afford more control over antigen presentation and have been used primarily to identify antibody fragments (rather than fulllength antibodies) specific for oligomers and fibrils of several amyloid-forming polypeptides.
Nevertheless, it remains a challenge to generate antibodies with sequence and conformational specificity for different types of amyloid aggregates, which is required for many applications. To address this challenge, we are investigating the potential of designing conformational antibody fragments by mimicking the natural process of amyloid formation (38). Our approach is to graft amyloidogenic peptide segments from polypeptides such as the Alzheimer A␤ 2 peptide into the complementarity-determining regions (CDRs) of single-domain (V H ) antibodies. We find that these Grafted AMyloid-Motif AntiBODIES (gammabodies) recognize their cognate amyloid aggregates via homotypic interactions with modest (submicromolar to micromolar) affinity, and weakly recognize disaggregated conformers as well as amyloid aggregates formed by other polypeptides (38 -40).
The Alzheimer A␤ peptide contains two key amyloidogenic peptide segments that mediate aggregation and are structured within the amyloid core (6,7,41,42). In the more amyloidogenic variant of A␤ (A␤42), these segments typically include residues 17 LVFFA 21 and 31 IIGLMVGGVVIA 42 . We posited that a V H antibody grafted with an A␤ peptide fragment containing both segments (A␤ residues 17-42) would have improved binding properties relative to shorter 10-mer variants that we have previously reported (38 -40). Moreover, our previous results (39,40) suggested that adding hydrophilic residues to the edges of CDR3 would enhance the solubility of V H domains displaying this large amyloidogenic peptide. We also posited that these flanking polar residues would influence the conformational specificity of the A␤ V H domains as well. Therefore, we have investigated the solubility and conformational specificity of A␤17-42 V H domains with different hydrophilic residues inserted at the edges of CDR3. We have also evaluated the generality of these findings by applying them to design antibody domains specific for islet amyloid polypeptide (IAPP), which is the polypeptide responsible for amyloid formation in type 2 diabetes.

Experimental Procedures
Antibody Engineering, Expression, and Purification-Genes for V H expression were created using polymerase chain reaction-based gene synthesis (43) and ligated into a pET-17b bacterial expression vector (Novagen) between the NdeI and XhoI restriction sites. Restriction sites for BamHI and NotI were inserted at the edges of the CDR3 loop for introducing different sequences via ligation of synthetic primers. A PelB leader sequence was added to the N terminus of the V H gene to direct it to the periplasm. A triple FLAG tag was added to the C terminus of the V H domain, followed by a His 7 tag.
Vectors containing the V H gene were transformed into BL21(DE3) pLysS cells (200132, Agilent Technologies). Transformed cells were incubated on LB-ampicillin plates for 2 days at room temperature. Next, V H expression was performed for 48 h at 20°C and 225 rpm in 1-liter shaker flasks containing 200 ml of autoinduction media (44) supplemented with ampicillin (100 g/ml) and chloramphenicol (35 g/ml). Cells were sedimented by centrifugation and then discarded. Nickel-agarose resin (30230, Qiagen) was added to the clarified supernatant and incubated overnight.
The protein of interest was isolated by batch purification. The resin was washed with 100 mM imidazole at pH 7.4 to remove impurities, prior to elution with 250 mM imidazole at pH 7.4. Eluted samples were then refolded by denaturation in 6 M GuHCl at pH 7.4. Following an overnight incubation, the samples were then refolded at 4°C by buffer exchanging into PBS, filtered, and stored at Ϫ80°C. Refolded samples were analyzed via SDS-PAGE (WG1203BX10, Life Technologies) for samples that were reduced and boiled. The gels were then stained using Coomassie dye (24615, Thermo Scientific). The concentration of the V H domains was measured using a Bradford assay (1856209, Thermo Fisher Scientific).
Expression of V H Domains in Pichia pastoris-V H domains were also expressed in P. pastoris. The V H genes were inserted into the pJ912 Pichia expression vector (DNA2.0, Inc.), which was linearized at the SacI restriction site to direct integration of the plasmid into chromosomal DNA at the AOX1 promoter. Cells were transformed by electroporation and then grown on plates (yeast extract peptone dextrose medium with sorbitol) supplemented with 1 mg/ml of zeocin. Colonies were picked and grown overnight in a 5-ml seed culture at 30°C and 225 rpm. The seed cultures were then added to baffled shake flasks with 50 ml of minimal media containing glucose. The cells were grown for 60 h at 30°C and 225 rpm, and then induced with methanol (0.5% final concentration). Expression of V H domains was detected in supernatant samples via SDS-PAGE (WG1203BX10, Life Technologies) and silver staining (LC6100, Life Technologies).
Yeast Surface Display Analysis of Aga2-V H Fusions-Yeast surface display was performed with the EBY100 Saccharomyces cerevisiae strain using methods that we have described previously (45). Aga2-V H fusions were subcloned from existing pET-17b constructs using PCR amplification to introduce overhangs complementary to the pCTCON2 yeast display plasmid. The PCR products were mixed with pCTCON2 plasmid that was digested with NheI and SalI, and transformed into the EBY100 strain for homologous recombination and selection on SD-CAA agar plates.
Overnight yeast cell cultures were miniprepped (D2004, Zymoprep II yeast miniprep kit, Zymo Research) and subsequently transformed into electroporation competent XL1-blue bacterial cells (200228, Agilent Technologies). Individual bacterial colonies were miniprepped (27106, Qiagen) and sequenced. Cells were then transformed with pCTCON2 plasmids encoding Aga2-V H fusions and grown overnight at 30°C with agitation in low pH SD-CAA medium (20 g/liter of dextrose, 6.7 g/liter of yeast nitrogen base, 5 g/liter of casamino acids, 14.7 g/liter of sodium citrate, and 4.3 g/liter of citric acid) to an A 600 of 1-2. Surface display of the Aga2-V H fusions was induced by pelleting the cells and resuspending them in SG-CAA media (20 g/liter of galactose, 6.7 g/liter of yeast nitrogen base, 5 g/liter of casamino acids, 8.56 g/liter of NaH 2 PO 4 ⅐H 2 O, and 5.4 g/liter of Na 2 HPO 4 ⅐2H 2 O). The cells were then incubated overnight at 30°C with agitation. Both the SD-CAA and SG-CAA media were supplemented with 100 g/ml of ampicillin and kanamycin as well as a 1ϫ solution of penicillin/streptomycin. Display levels were analyzed using flow cytometry. For each Aga2-V H fusion, ϳ10 6 cells were collected and washed twice with 1 ml of PBS-BX (PBS with 10 mg/ml of BSA and 1% (v/v) Triton X-100). A C-terminal epitope (myc) tag was then probed with anti-myc IgY antibody (1:200 dilution; A-21280, Life Technologies) for 1-5 h at 25°C. After this incubation, the cells were washed with 1 ml of PBS-BX before labeling with a secondary antibody (1:100 dilution, Alexa Fluor 488-conjugated goat anti-chicken IgG, A-11039, Life Technologies) for 5 min. Fluorescently labeled cells were washed again in 1 ml of PBS-BX and the level of Aga2-V H display (as judged using the anti-myc antibody) was analyzed on a BD LSRII flow cytometer using 100,000 events.
Circular Dichroism Analysis-The secondary structure of V H domains was characterized by collecting their circular dichroism (CD) spectra on a Jasco J-815 spectrometer in the far-UV region of 200 -260 nm. Samples were diluted to 0.2 g/liter in PBS, and their spectra were measured using a cuvette with a 1-mm path length. Scans were collected at a rate of 50 nm/min (1 nm bandwidth and 4 s response time). For each V H domain, 10 spectra were collected and averaged. Mean residue ellipticity values were calculated from the background-subtracted spectra.
For each V H domain, the reversibility of unfolding after thermal denaturation was monitored by CD. V H domains were diluted to 0.1 g/liter in PBS in a cuvette with a 1-cm path length. Samples were heated from 25 to 95°C at a rate of 1°C/min, while monitoring the ellipticity at 235 nm. Following the first melt, samples were cooled to 25°C for at least 20 min before repeating the melt. The fraction of V H domain that remained folded over the course of the melt was calculated as ( T Ϫ U )/ ( F Ϫ U ). T is the ellipticity signal as a function of temperature, and U and F are the ellipticity signals for the unfolded and folded samples, respectively. The apparent melting temperature (T m ) was calculated from the initial melt.
Protein A Binding Assay-The binding of V H domains to Protein A was measured in well plates using immobilized Protein A. Recombinant Protein A (77673, Pierce) was diluted to 5 M in PBS, and 100 l was added to a Nunc-Immuno 96-well plate with MaxiSorp surface coating (442404, Thermo Fisher Scientific). Protein A was incubated in the plate at room temperature overnight. Milk was then added to the plate at a concentration of 10% (w/v) and incubated at room temperature for several hours. The plate was then washed with PBS, and V H domains at varying concentrations in PBST (PBS with 0.1% (v/v) Tween 20) were added (100 l).
The V H domains were incubated in the wells overnight at room temperature. The plate was then washed with PBS, and 100 l of monoclonal anti-FLAG tag antibody produced in mouse (F1804-1 mg, Sigma) was added that had been diluted 1000-fold in PBST. Following a 1 h incubation of the anti-FLAG antibody, the plate was washed again with PBS, and 100 l of an HRP-conjugated goat anti-mouse secondary antibody (32430, Pierce) was added that had been diluted 1000-fold in PBST. After 1 h of incubation, the plate was washed a final time with PBS, and 100 l of substrate (1-Step Ultra TMB-ELISA Substrate Solution, 34028, Life Technologies) was added. The plates were developed for 1 h, and the reaction quenched by adding 2 M H 2 SO 4 (100 l). Signals were determined by measuring the absorbance at 450 nm for each well using a Tecan Safire 2 plate reader. IC 50 values were calculated from the doseresponse curve that was generated for each V H domain.
Antibody Solubility Analysis-The relative solubility of V H domains as a function of temperature was analyzed. Samples were diluted to ϳ13 M in PBS, and divided into multiple aliquots. Each aliquot was held at a given temperature (50 -95°C) for 20 min, and then cooled to room temperature. The concentration of protein remaining in solution was measured by first centrifuging the cooled sample for 15 min at 21,000 RCF. After-ward, the top 50% of the supernatant volume was removed to determine the soluble V H concentration using the Bradford assay.
The relative solubility of V H domains at 25°C was also analyzed by monitoring their monomeric content over a week. Samples were prepared by centrifuging them at 208,000 RCF for 1 h at 4°C, and then recovering the top 80% of the supernatant. This procedure aimed to remove aggregated species at the beginning of the experiment. The samples were then concentrated to ϳ1 g/liter (ϳ50 M) using ultracentrifugal concentrators (10-kDa molecular weight cut-off membrane, UFC501096, Millipore). The monomeric content of the initial and 1 week samples was analyzed by size exclusion chromatography. A Waters 600F HPLC instrument was used to inject 10 l of sample at ϳ1 g/liter onto a TSKgel G3000SW XL column (08541, Tosoh Bioscience) with an in-line guard column (08543, Tosoh Bioscience). The mobile phase was PBS with 0.2 M arginine (pH 7.4) at a flow rate of 0.5 ml/min. V H elution was monitored via absorbance measurements at 280 nm.
Thioflavin T Fluorescence-Thioflavin T, a fluorescent dye that recognizes amyloid and some non-amyloid aggregates, was added to V H samples (0.2 g/liter) to achieve a final thioflavin T concentration of 88 M. Samples were excited at 450 nm, and emission was detected at 482 nm. The background buffer was also measured and its signal was subtracted from that of the samples. Samples were measured initially and after they had been heated to 95°C and cooled to room temperature.
Antibody Binding Analysis-The binding of V H domains was tested for multiple antigens. These antigens include A␤42 fibril, freshly disaggregated A␤42 (also referred to as A␤ monomer*), IAPP fibril, and freshly disaggregated IAPP (also referred to as IAPP monomer*). Disaggregated A␤42 was generated by first incubating the lyophilized peptide (62-0-80, American Peptide) in hexafluoroisopropanol. Following overnight incubation, the hexafluoroisopropanol was removed by evaporation. The A␤42 peptide was dissolved in 50 mM NaOH at a peptide concentration of 1 g/liter (220 M), and then centrifuged at 208,000 RCF for 1 h at 4°C. The top 80% of the supernatant was recovered and passed through a 0.22-m filter (SLGV004SL, Millipore). The sample was then diluted into acidified PBS to achieve a final pH of 7.4 and a final peptide concentration of 25 M. A␤42 fibrils were formed by incubating disaggregated peptide at 37°C without agitation for at least 2 days. Prior to using the A␤42 fibrils for binding assays, they were sonicated on ice for 3 cycles of 10 s at 100% power with 30 s of rest between cycles (FB-120 Sonic Dismembrator, Thermo Fisher Scientific).
Disaggregated IAPP was prepared by dissolving the peptide in hexafluoroisopropanol. After lyophilization, the dried peptide was dissolved in 20 mM Tris (pH 7.4) to achieve a final concentration of 32 M. IAPP fibrils were formed by incubating disaggregated IAPP at 37°C (300 rpm) for at least 5 days. The IAPP fibrils were sonicated on ice for 10 s at 75% power (3 times) with 30 s of cooling between rounds of sonication.
The binding of V H domains to A␤ and IAPP was characterized by an ELISA assay conducted in 96-well plates. The antigens were diluted to 2.5 M, and 100 l was added to a Nunc-Immuno 96-well plate with MaxiSorp surface coating (442404, Thermo Fisher Scientific) and incubated overnight at room temperature. Control wells without antigen were also included to determine the background signals. Milk dissolved in PBS (10% w/v) was then added to each well and incubated at room temperature for several hours before washing the wells with PBS. V H domains that were labeled with a biotin tag (EZ-Link Sulfo-NHS-LC-Biotin, 21335, Thermo Fisher Scientific) were diluted to varying concentrations. At each concentration, 100 l of the V H domain sample was added to wells with immobilized antigen or background. The diluent for the V H domains was 5% (w/v) milk in PBST supplemented with 0.02% (w/v) NaN 3 to reduce nonspecific interactions.
Following overnight incubation at room temperature, the biotinylated V H domains were washed from the plate. HRPconjugated streptavidin (21126, Pierce) was diluted 30,000-fold in PBST and 100 l was added to the wells for incubation for 1 h at room temperature. The plate was washed a final time, and 100 l of the substrate (1-Step Ultra TMB-ELISA Substrate Solution, 34028, Life Technologies) was added to each well. The plates were developed for 1-2 h and quenched with 2 M H 2 SO 4 (100 l). The absorbance signal at 450 nm was measured for each well using a Tecan Safire 2 plate reader. Signals were then normalized by subtracting the background value at each V H concentration, and then dividing by the background. IC 50 values were calculated from the resulting dose-response curve that was generated for each V H domain.
The binding of V H domains to different antigens was also characterized using an immunoblotting assay. The antigens (A␤42 fibril and disaggregated A␤42, IAPP fibril, and disaggregated IAPP) were immobilized on nitrocellulose membranes (10600004, GE Healthcare) by spotting 2 l of each antigen at different concentrations. The membranes were then blocked in 10% (w/v) milk (dissolved in PBS) for 2 h at room temperature. After the blots were washed, solutions of biotinylated V H domains (75 nM) in 5% (w/v) milk and PBST (supplemented with 0.02% w/v NaN 3 ) were added to the blots. After overnight incubation at room temperature with mild rocking, the blots were washed three times in PBST (10 min per wash). Next, HRP-conjugated streptavidin (21126, Pierce) was added that was diluted 45,000-fold in PBST, and the blots were incubated at room temperature for 1 h. The blots were then washed again with PBST (four times for 10 min per wash) before adding the substrate (Pierce ECL Western blotting Substrate, 32106, Thermo Fisher Scientific) and imaging the blots using an x-ray film developer (Konica Minolta SRX-101A).
Alanine Mutational Analysis of A␤17-42 V H Domain-Sitedirected mutagenesis was used to generate variants of the V H domain containing A␤17-42 in CDR3 along with DDD at each edge of the grafted peptide. V H domains were created in which three consecutive residues in CDR3 were replaced with three alanines. The variants examined were those in which residues in the linker region of the A␤ peptide (A␤ residues 22-29) were mutated to alanine.
Site-directed mutagenesis reactions were carried out using PfuUltra II Hotstart PCR Master Mix (600850, Agilent Technologies), which was added to a mixture of the template DNA, and forward and reverse primers. The PCR involved 19 cycles of denaturation at 95°C for 30 s, followed by 1 min of annealing at 55°C, and 12 min of extension at 68°C. Following the PCR, the template DNA was digested by the addition of DpnI (R0176L, New England BioLabs Inc) for 2 h at 37°C.

Polar CDR3 Mutations Strongly Impact Expression Levels of A␤ V H Domains-To
examine the effect that different types of residues have on the solubility and conformational specificity of V H domain antibodies grafted with a large amyloidogenic A␤ peptide (residues 17-42), we first cloned variants that contained the A␤17-42 peptide flanked by three identical residues at each edge of CDR3 (Fig. 1). Variants studied included those with six identical residues that were negatively charged (DDD and EEE), positively charged (KKK and RRR), and non-charged (NNN, QQQ, SSS, TTT, and AAA).
After cloning the V H domains, we expressed them in bacteria to evaluate how different CDR3 mutations affect their final yield after expression and purification. Our previous studies revealed that increased expression of A␤ V H domains generally correlates with increased solubility (39,40). Therefore, the relative amounts of expressed and purified V H domains can potentially be used to evaluate their relative solubility. We found that the highest expressing A␤17-42 V H domains fol- A, amino acid sequence of the wild-type V H domain, as defined using Kabat numbering. The C terminus also contained a 3ϫ FLAG tag and a His 7 tag, whereas the N terminus contains residues Met-Ser-Lys-Leu (not shown). B, amino acid sequence of CDR3 for A␤17-42 V H domains. CDR3 positions flanking the grafted A␤17-42 peptide (designated as X) were systematically varied with six identical residues that were charged (Asp, Glu, Arg, and Lys) or non-charged (Gln, Asn, Ser, Thr, and Ala). C, crystal structure of the parent wild-type V H domain (Protein Data Bank 3B9V) with CDR3 highlighted in red.
lowing purification and refolding were those with negatively charged mutations (DDD and EEE; Fig. 2).
In contrast, variants with positively charged residues (KKK and RRR) or small non-polar residues (AAA) expressed poorly (Fig. 2). SDS-PAGE analysis revealed that purified KKK and AAA V H domains could only be detected using overloaded gels, whereas the RRR variant could not be detected (data not shown). Of the V H domains with polar mutations, the TTT variant also expressed poorly (Fig. 2) and could only be detected via an overloaded SDS-PAGE gel (data not shown). Interestingly, V H domains with NNN, QQQ, or SSS mutations in CDR3 were expressed at significant levels. Although the yields of these variants were lower than the DDD and EEE variants, the ability of these non-charged residues to promote expression of the hydrophobic A␤17-42 V H domain suggested that they increased antibody solubility.
The V H domains were next refolded in 6 M GuHCl after purification. The recovery of the V H domains after refolding was high (86 -95%). Importantly, the refolding step resulted in improvements in the fraction of monomeric V H domains, especially for the NNN, QQQ, and SSS variants (data not shown).
We next sought to evaluate whether the inability to express certain A␤17-42 V H domains was specific to our bacterial expression system or general to other expression hosts. This is a particularly important question for the KKK and RRR V H domains because their low expression in bacteria may simply be due to their inability to translocate across the cytosolic membrane (46,47) and not due to low solubility. Therefore, we also evaluated the expression levels of the V H domains in S. cerevisiae (Fig. 3). To simplify the expression analysis, we cloned these domains as fusions to the C terminus of a yeast cell-surface protein (Aga2) and detected their display on the surface of yeast via a C-terminal myc tag.
Flow cytometry analysis revealed that the expression trends on the surface of yeast (Fig. 3) were similar to those observed for bacterial expression (Fig. 2). The expression levels of the negatively charged variants (DDD and EEE) as well as some of the non-charged variants (NNN, QQQ, and SSS) were higher than the other variants (KKK, RRR, TTT, and AAA; Fig. 3). Finally, we also evaluated expression of the KKK, AAA, and DDD vari-ants as autonomous V H domains in P. pastoris because this host is well known to produce recombinant proteins at high levels. However, we found that expression of only the DDD variant could be detected (data not shown). These findings collectively suggest that negatively charged and some non-charged residues (asparagine, glutamine, and serine) are best at promoting expression of the hydrophobic A␤17-42 V H domain.
Non-charged Polar CDR3 Mutations Promote High Conformational Stability of V H Domains-We next sought to evaluate the secondary structures and stabilities of the A␤17-42 V H domains with negatively charged (DDD and EEE) or noncharged (NNN, QQQ, and SSS) residues. We reasoned that the conformational stability of the non-charged polar variants may be higher due to the potentially unfavorable impact of the negatively charged residues on the stability of the grafted V H domains. Circular dichroism analysis revealed that the secondary structures of the charged and uncharged variants were typical of ␤-sheet-rich V H domains (data not shown).
To evaluate the potential impact of negatively charged mutations relative to non-charged ones on V H conformational sta-  bility, we first analyzed the relative affinity of the A␤17-42 domains for Protein A. This analysis is motivated by the fact that folded V H 3 domains (such as the A␤17-42 domains) have a Protein A binding site on their scaffold, and the relative binding of Protein A to V H 3 domains is correlated with V H stability (45,48,49). Using an ELISA assay to measure V H binding to immobilized Protein A, we found that the NNN, QQQ, and SSS variants had similar IC 50 values of ϳ1-2 nM, whereas the DDD and EEE variants had higher IC 50 values (ϳ5-10 nM; Fig. 4 and Table 1). The higher IC 50 values for the negatively charged variants suggested they are less stable than the other polar variants, which may be due to intramolecular repulsion.
We further examined the conformational stability of the A␤17-42 V H domains by measuring their apparent melting temperature (T m ) values using circular dichroism. The DDD and EEE variants had lower T m values (ϳ69 and ϳ71°C, respectively) than the NNN, QQQ, and SSS variants (ϳ75°C; Fig. 5 and Table 1). These findings are consistent with the Protein A results (Fig. 4) and also suggest that the V H domains with negatively charged mutations in CDR3 are less conformationally stable than the variants with non-charged mutations.
Aspartate, Glutamate, and Asparagine Increase Solubility and Prevent Irreversible Unfolding of V H Domains-We have previously found that negatively charged mutations in CDR3 of grafted V H domains prevent irreversible unfolding of some hydrophobic V H domains but not others (39,40). Thus, we sought to evaluate if the DDD and EEE variants promoted reversible unfolding and refolding of the A␤17-42 V H domains as well as to what extent non-charged mutations could promote such reversible unfolding. Circular dichroism was used to monitor the ellipticity of the samples at a fixed wavelength (235 nm) as a function of temperature. The DDD and EEE variants showed largely reversible unfolding and refolding behavior (Fig.  5), consistent with our previous findings (39,40,50). Surprisingly, the NNN variant also displayed unfolding behavior that was largely reversible, whereas the QQQ and SSS variants unfolded irreversibly. These differences were also apparent in terms of the circular dichroism spectra at 25°C for V H domains before and after being heating to 95°C. The DDD, EEE, and NNN V H variants largely regained their initial secondary structure after being heated to 95°C and cooled, whereas the secondary structure of the QQQ and SSS variants was significantly perturbed (data not shown).
These results suggested that the QQQ and SSS V H domains aggregated at high temperature. To investigate this in more detail, we performed solubility measurements after heating the A␤17-42 V H domains to different temperatures (50 -95°C) and cooling them to 25°C (Fig. 6). The solubility of the DDD and EEE variants was high and unaffected by heat stress (Fig.  6A). The NNN V H variant was modestly less soluble than the DDD and EEE variants, but significantly more soluble than the QQQ and SSS variants. These solubility differences appear related to the formation of amyloid-like aggregates (Fig. 6B). The QQQ and SSS V H domains displayed high thioflavin T fluorescence after heating, whereas the other polar variants displayed similar (DDD) or modestly increased (EEE and NNN) thioflavin T fluorescence after heating.
We also investigated the relative solubility of the A␤17-42 V H domains at room temperature (25°C) and a higher antibody concentration (ϳ50 M) over the course of a week (Fig. 7) to evaluate similarities or differences with the non-native solubility data (Fig. 6). Samples were evaluated for their fractional monomer content using size exclusion chromatography. The monomer content of the DDD and EEE variants remained constant at ϳ90% over the course of 1 week at 25°C. The NNN variant had a lower initial monomer content (ϳ65%), which remained roughly constant after 1 week. The other variants that initially contained 66% (QQQ) and 53% (SSS) monomer displayed further aggregation over the course of 1 week. These findings reveal that the relative native and non-native solubilities of the A␤17-42 domains display similar trends, which is consistent with our hypothesis that the solvent-exposed A␤17-42 peptide in CDR3 mediates aggregation of both unfolded and folded V H domains.
To evaluate the generality of our findings, we also sought to design a grafted V H domain specific for IAPP. IAPP is extremely amyloidogenic and several small peptide segments derived from residues 8 -37 are known to form fibrils (51)(52)(53)(54)(55). Thus, we chose to graft this large peptide segment (residues 8 -37) into CDR3 along with negatively charged residues (DED) at each edge of CDR3. This IAPP8 -37 V H domain expressed well (94 mg/liter) and showed similar secondary structure as the DDD  A␤17-42 V H domain (data not shown). Upon thermal denaturation, it was also able to unfold in a nearly reversible manner (Fig. 8A), its melting temperature (71°C) was similar to those of the DDD and EEE A␤17-42 V H domains (69 -71°C), and it was largely monomeric like the DDD A␤17-42 V H domain (Fig.  8B). In contrast, the same IAPP8 -37 V H domain with NNN residues at each edge of CDR3 showed reduced expression (15 mg/liter), suggesting that the negatively charged residues enhanced solubility of this antibody domain.

Charged and Non-charged Polar Mutations Differentially
Impact the Conformational Specificity of A␤ V H Domains-The significant effect of the polar mutations flanking the A␤17-42grafted peptide on V H solubility led us to also evaluate how such mutations influence antigen binding. Using an ELISA assay to measure concentration-dependent binding, we found that the DDD, NNN, QQQ, and SSS variants all recognized A␤42 fibrils. The EEE variant was not evaluated in detail because it displays similar binding activity as the DDD variant (data not shown). The fibril IC 50 values were lowest for the DDD (43 Ϯ 15 nM) and NNN (43 Ϯ 12 nM) variants, and highest for the QQQ (114 Ϯ 75 nM) and SSS (346 Ϯ 64 nM) variants ( Fig. 9A and Table 2). Interestingly, the most significant difference between the A␤17-42 V H domains was their ability to recognize disaggregated A␤. The negatively charged (DDD) variant showed little binding to disaggregated A␤42 (Fig. 9A). In contrast, V H variants with non-charged mutations were able to bind to disaggregated A␤42. The SSS and QQQ variants displayed increased binding signals for disaggregated A␤ relative to fibrils. The IC 50 values for binding of the NNN, QQQ, and SSS variants to disaggregated A␤ were similar (200 -300 nM; Table 2).
Given that some of the V H domains form higher order species (Fig. 7A), we evaluated whether such multimers may influence the observed conformational specificity or IC 50 values of The ellipticity signal at 235 nm was monitored as a function of temperature using circular dichroism. Each V H domain was cooled to 25°C after its initial melt, and then melted a second time to evaluate reversibility. the A␤17-42 V H domains. To accomplish this, we enriched the fraction of V H monomer in the purified samples for the QQQ and SSS V H domains and evaluated their binding. A rigorous centrifugation step (208,000 RCF for 1 h) reduced the fraction of high molecular weight species from 15 Ϯ 1 to 9 Ϯ 1% for the QQQ V H domain and from 17 Ϯ 1 to 11 Ϯ 1% for the SSS V H domain.
For the QQQ and SSS V H domains, samples with reduced aggregate content had IC 50 values that were within error of the control samples for binding to disaggregated A␤ (data not shown). This suggests that the presence of V H multimers did not significantly affect binding. Interestingly, we observed that the reduction of V H multimers modestly enhanced the normalized binding signals (data not shown). Collectively these findings suggest that the enhanced binding of the QQQ and SSS V H domains (relative to the DDD variant) to disaggregated A␤ is not strongly linked their higher aggregate content.
We also found that the polar residues at the edges of CDR3 influenced the sequence specificity of the A␤17-42 V H domains. The DDD variant displayed high specificity for A␤ fibrils relative to IAPP fibrils (Fig. 9A). The NNN, QQQ, and SSS variants bound preferentially to A␤ fibrils but also displayed modest binding to IAPP fibrils. We confirmed that the specificity of the A␤17-42 V H domains for A␤ fibrils was not simply due to low reactivity of IAPP fibrils by evaluating binding of the IAPP8 -37 V H domain (Fig. 9B). This V H domain with DED residues at the edges of CDR3 recognized IAPP fibrils (IC 50 value of 126 Ϯ 68 nM) and weakly bound to either A␤ conformer.
We also evaluated the binding activity of the A␤ and IAPP V H domains using an immunoblotting assay (Fig. 10). A␤42 and IAPP fibrils and disaggregated peptides were deposited on nitrocellulose membranes at a range of concentrations and probed with V H domains as well as with conventional antibodies. The DDD A␤17-42 V H domain displayed significant conformational specify for A␤ fibrils relative to disaggregated A␤  and little cross-reactivity with IAPP conformers. The NNN, QQQ, and SSS variants displayed similar binding to fibrils. However, they also weakly recognized disaggregated A␤ (Fig.  10), which is generally consistent with the ELISA results (Fig. 9). Moreover, the background signals for the NNN V H domain and especially for the QQQ and SSS V H domains were higher than for the DDD variant. This suggests that the non-charged variants were stickier and less well suited for immunoblotting applications. Interestingly, the specificity of conventional antibodies (LOC and OC) with conformational specificity for A␤ fibrils was only observed at the lowest amounts of deposited A␤ (0.5 pmol). Moreover, these antibodies that recognize fibrils of different amyloid-forming polypeptides (19,56,57) either failed to recognize IAPP fibrils or recognized both IAPP conformers after long exposure times (LOC, data not shown). As expected, conventional antibodies that recognize sequencespecific epitopes within A␤42 (NAB 228, 12F4, and 4G8) recognized both A␤ fibrils and disaggregated A␤.
Conversely, we observed that the IAPP8 -37 V H domain selectively recognized IAPP fibrils relative to disaggregated IAPP, and showed low cross-reactivity with A␤ conformers (Fig. 10). This is in contrast to a conventional IAPP sequencespecific antibody (R10/99) that recognized both IAPP conformers. These results collectively demonstrate that V H domains grafted with amyloidogenic peptides can be generated with sequence and conformational specificity for amyloid fibrils.
Conformational Specificity of the DDD A␤17-42 V H Appears to be Mediated by Repulsive Electrostatic Interactions-The fact that the DDD variant of the A␤17-42 V H weakly recognizes disaggregated A␤ relative to the other A␤ V H variants without charged mutations suggests that repulsive electrostatic interactions may mediate the conformational specificity of the DDD variant. This is further supported by the fact that A␤42 is predicted to have an acidic isoelectric point (ϳ5) and contains six negatively charged residues, some of which are adjacent to the most amyloidogenic regions. To test this hypothesis, we compared the binding of the DDD A␤17-42 V H to A␤ and IAPP conformers at normal (0.14 M) and elevated (1 M) concentrations of sodium chloride (Fig. 11A). Importantly, this V H domain displayed significantly enhanced binding to disaggre- (referred to as monomers*) relative to IAPP fibrils. B, binding of the IAPP8 -37 V H domain (defined in Fig. 8) to IAPP fibrils relative to A␤42 fibrils and monomers*. The normalized binding values are reported as signal minus background divided by background. For each experiment, two replicates were collected for each point, and the standard deviations are reported. The experiments were performed at least three times for each V H domain, and data from one representative experiment is shown.

Design of Anti-amyloid Domain Antibodies
gated A␤ at high salt, and did not show increased binding to IAPP conformers. The DDD A␤17-42 V H also displayed modest increases in its binding signals for A␤ fibrils. We also evaluated whether the enhanced binding of the DDD antibody variant to A␤ at high salt was mediated by salt-induced V H oligomerization. To test this possibility, we used size exclusion chromatography to analyze the potential oligomerization of the DDD A␤17-42 V H as a function of salt concentration. However, the levels of the high molecular weight species of the DDD V H domain were a weak function of salt (5% at 0.14 M NaCl, 6% at 1 M NaCl). Thus, we conclude that the enhanced affinity of the A␤17-42 V H for disaggregated A␤ is FIGURE 10. Evaluation of the conformational and sequence specificity of the A␤17-42 and IAPP8 -37 V H domains relative to conventional antibodies using an immunoblotting assay. A␤42 and IAPP fibrils (F) and freshly disaggregated peptide (referred to as monomers*, M) were deposited at various concentrations on nitrocellulose membranes. The binding of A␤17-42 V H domains (75 nM) was evaluated for variants with negatively charged (DDD) or non-charged (NNN, QQQ, and SSS) residues at each edge of the grafted amyloid peptide. An IAPP8 -37 V H domain (75 nM) with DED residues at each edge of CDR3 was also tested. Their behavior was compared with that of conventional antibodies with conformational specificity (LOC and OC; ϫ2500 dilution) or sequence specificity for A␤ (NAB 228, 12F4, and 4G8; ϫ500 -2000 dilution) or IAPP (R10/99; ϫ100 dilution). A loading control blot was developed using silver colloidal stain. FEBRUARY 5, 2016 • VOLUME 291 • NUMBER 6 likely due to the screening of repulsive electrostatic interactions rather than salt-induced oligomerization of the V H domain.

Design of Anti-amyloid Domain Antibodies
The enhanced binding for the DDD A␤17-42 V H to A␤42 monomer (pI ϳ5) at high salt suggests that repulsive electrostatic interactions influence the corresponding conformational specificity. In contrast, IAPP is a basic polypeptide (theoretical pI of Ͼ10) with only three residues that potentially could be charged (Lys 1 , Arg 11 , and His 18 ) and which are peripheral to the most amyloidogenic region (IAPP residues [22][23][24][25][26][27]. Thus, we expected that the binding of the IAPP8 -37 V H with negatively charged residues (DED) at the edges of CDR3 would be weakly impacted by changes in salt concentration. Indeed, we find that high salt does not alter IAPP8 -37 V H binding to IAPP or A␤ conformers (Fig. 11B). These results reveal important differences in the mechanisms of conformational specificity for the A␤ and IAPP V H domains.
Alanine Mutational Analysis Reveals That A␤ Residues 22-29 Modulate Expression and Binding of the DDD A␤17-42 V H -We next generated and analyzed a series of alanine mutations in CDR3 of the DDD A␤17-42 V H to understand how residues within the grafted A␤ peptide modulate antibody expression and binding. We focused on A␤ residues 22 EDVG-SNKG 29 that link the two major aggregation-prone regions (A␤ residues 17 LVFFA 21 and 31 IIGLMVGGVVIA 42 ) within A␤42. Mutants were generated in which three consecutive residues were replaced by three alanines.
We find that A␤ residues 21 AED 23 are critical for expression, as mutating them to alanines significantly reduced V H expression (Fig. 12A). Likewise, mutation of residues 27 NKG 29 to three alanines greatly reduced expression. However, mutation of A␤ residues 23 DVG 25 or 25 GSN 27 to three alanines resulted in A␤17-42 V H domains with similar or modestly lower expression levels relative to wild-type V H (Fig. 12A). Size exclusion chromatograms and far-UV CD spectra for the mutants and wild-type were also similar (data not shown). Their binding behavior, however, was significantly different (Fig. 12B). The 23 AAA 25 and 25 AAA 27 V H mutants bound to disaggregated A␤ better than wild-type. These findings indicate that hydrophilic A␤ residues 22-29 generally reduce A␤ self-association and increase solubility, and that A␤ residues 23-27 reduce A␤17-42 V H binding to disaggregated A␤.

Discussion
Our studies aimed at designing anti-amyloid antibody fragments were originally motivated by a previous report using a similar motif-grafting approach to generate full-length antibodies specific for aggregated forms of the mammalian prion protein (PrP) (58). These antibodies were generated by grafting PrP peptides containing residues 89 -112 (24-mer) or 136 -158 (23-mer) into heavy chain CDR3 of an antibody originally specific for an HIV protein. These peptides were identified based on their importance in mediating conversion from PrP C to PrP Sc (29, 59 -63). The resulting grafted antibodies recognized the aggregated, infectious form of PrP (PrP Sc ) but not the soluble, non-infectious form (PrP C ).
Notably, the grafted PrP peptides each contain four positively charged residues (58,64). Mutating these residues to alanine eliminates binding for the grafted variant with PrP residues 89 -112 in heavy chain CDR3 (this analysis was not reported for the other variant). A third PrP peptide (residues 19 -33) was later identified that also mediates binding to aggregated forms of PrP when grafted into heavy chain CDR3 (64). This peptide also contains four positively charged residues and binding is eliminated when such residues are mutated to alanine. Collectively these results suggest that the binding mechanism of these PrP conformational antibodies involves electrostatic interactions.
Our results suggest that A␤ and IAPP-grafted antibody domains employ non-electrostatic interactions to mediate binding to amyloid aggregates. This is supported by the fact that fibril binding for the A␤17-42 and IAPP8 -37 V H domains with negatively charged CDR3 mutations is similar or better at high salt than at low salt (Fig. 11). Moreover, fibril binding for A␤17-42 V H domains was similar for CDR3 mutations that have side chains of similar size (aspartate and asparagine; Fig. 9).
Nevertheless, we find that repulsive electrostatic interactions are important for determining the affinity of A␤17-42 V H binding to disaggregated A␤. Addition of negatively charged residues (DDD) at the edges of the grafted A␤17-42 peptide reduced V H binding to disaggregated A␤, whereas removal of negative charge in the grafted A␤17-42 peptide ( 23 DVG 25 to 23 AAA 25 ) increased V H binding. These findings suggest that the conformational specificity of the A␤17-42 V H for binding to A␤ fibrils relative to disaggregated A␤ is strongly mediated by repulsive electrostatic interactions.
Collectively our results suggest that the mechanism for the conformational specificity of the DDD A␤17-42 V H domain is due to multiple contributions. One factor may be the entropic penalty for binding of the grafted A␤17-42 peptide to disaggregated A␤, the latter of which is largely unstructured. We reason that the interaction of this grafted V H domain with A␤ fibrils is more favorable because of the lower entropic penalty for binding to structured fibrils. A second factor that appears to contribute to the conformational specificity of the DDD A␤17-42 V H domain is electrostatic repulsion between negatively charged residues in CDR3 and those within A␤. We speculate that the DDD variant binds to A␤ fibrils because amyloidogenic (homotypic) interactions involving A␤ residues 17-42 are more significant than the repulsive electrostatic interactions that discourage binding to disaggregated A␤. It is also possible that the effective net charge of A␤ within structured fibrils is lower than for disaggregated A␤, which may also contribute to the observed conformational specificity. Moreover, avidity effects due to the polyvalent nature of aggregates relative to monomers may also contribute to the observed conformational specificity (65).
Our findings for the DED IAPP8 -37 V H domain provide further insight into the mechanisms of conformational specificity for such grafted antibodies. Notably, this V H domain binds well to IAPP fibrils and poorly to disaggregated IAPP. The fact that the IAPP peptide contains little charge and that the conformational specificity of the DED IAPP8 -37 V H domain is insensitive to salt concentration suggests that its binding mechanism is largely mediated by non-electrostatic interactions. The conformational specificity of this V H domain may be due to the reduced entropic penalty for binding to structured fibrils relative to disordered peptide and/or avidity effects due to polyvalent aggregates. These speculative conclusions will need to be tested for additional grafted V H domains and using different types of aggregates.
The fact that the expression levels of the A␤17-42 V H domains were strongly influenced by elimination of charged residues in the grafted A␤17-42 peptide also deserves further consideration. Our alanine mutational data revealed that A␤ residues 22 ED 23 are critical for expression, which is consistent with previous studies that have highlighted the importance of these residues for A␤ solubility (66 -68). Deletion or substitu- tion of these negatively charged residues for non-charged or positively charged residues, as occurs for A␤ variants associated with familial forms of Alzheimer disease, has been found in some cases to significantly increase the aggregation rates of A␤ (69 -71).
Our findings related to the impact of CDR3 mutations on V H solubility also deserve further consideration. We have previously shown that negatively charged mutations are effective at increasing the solubility of V H domains grafted with shorter (10-mer) amyloidogenic peptides (39,40), which is consistent with our findings here. Moreover, we have also previously found that positively charged mutations are less effective at increasing expression and solubility of grafted V H domains unless the antibody scaffold has a highly basic isoelectric point (39). Our current findings for a scaffold with a near-neutral isoelectric point (theoretical pI of ϳ8.5 when omitting the CDRs and FLAG tags) are consistent with our previous observations given the low expression levels of the A␤17-42 V H domains with positively charged mutations in both Escherichia coli and P. pastoris, as well as the low display levels of such variants on the surface of S. cerevisiae.
It is notable that the polar residues asparagine, glutamine, and serine enabled V H expression at sufficient levels for purification and analysis, whereas threonine did not. Moreover, asparagine was found to be most solubilizing, followed by glutamine and serine. We originally posited that serine would be the most solubilizing of these polar residues for multiple reasons. First, we suspected that the propensity of polyglutamine and polyasparagine to self-associate and aggregate (72,73) would make these residues less solubilizing than multiple serines when introduced into CDR3. Moreover, previous work by others demonstrated that serine is the most solubilizing noncharged residue when introduced into a solvent-exposed loop within Ribonuclease Sa (74). However, there are several key differences between this previous study and our study, including the number of mutations (one for Ribonuclease Sa and six for V H domains) and the solution conditions (1-2 M ammonium sulfate at pH 4.25 for Ribonuclease Sa and PBS at pH 7.4 for our V H domains) that make it difficult to compare them.
Nevertheless, our findings are consistent with several hydropathy indices that report asparagine as the most hydrophilic non-charged residue followed by glutamine, serine, and threonine (in that order) (75,76). Thus, it is possible that residues such as asparagine and glutamine are simply increasing the local hydrophilicity adjacent to the grafted hydrophobic peptides and thereby increasing V H solubility. Moreover, it is likely that six asparagine or glutamine residues in CDR3 are unable to mediate the aggregation observed for longer repeat lengths of polyasparagine and polyglutamine (73,77). It is also notable that hydropathy indices do not describe the overall solubility trends we observed for charged and non-charged residues. In particular, positively charged mutations are typically predicted to be more hydrophilic than non-charged polar residues and, in some cases, more hydrophilic than negatively charged residues (75,76).
We have also identified a number of factors that influence the performance of our grafted V H domains. First, we find that the longer grafted A␤ (26-mer) and IAPP (30-mer) peptides lead to improved binding relative to the shorter (10-mer) grafted peptides that we have previously reported (38 -40, 78). Our findings as well as those for PrP antibodies (58,64) suggest that peptide lengths of 23-30 residues are a logical starting point for designing antibodies and antibody fragments specific for other amyloid-forming polypeptides. Second, we find that our V H domains display some propensity to interact non-specifically with well plates and immunoblotting membranes unless the experiments are performed in milk. Thus, it is important to block well plates and membranes with 10% milk, and conduct the binding experiments in at least 5% milk using our V H domains. This also suggests that the A␤ and IAPP V H domains are not appropriate for use in assay formats that are incompatible with milk.
We also find that the detection sensitivity is higher using biotin-based detection (streptavidin-HRP) than peptide tagbased detection (anti-FLAG and secondary antibody-HRP). Thus, it is recommended to biotinylate the V H domains at low levels (ϳ4 -7 biotins per domain) to maximize specificity and detect them using standard biotin-specific reagents. Moreover, we find that sequence specificity of the V H domains is reduced as the amount of immobilized antigen is increased in immunoblotting assays. For the NNN, QQQ, and SSS A␤17-42 V H domains, we observe some cross-reactivity with IAPP fibril in well plates at V H concentrations Ͼ100 nM (Fig. 9). We also sometimes observe cross-reactivity of the A␤17-42 (DDD) and IAPP8 -37 (DED) V H domains with non-cognate antigens such as ␣-synuclein and Tau for immunoblots loaded with Ͼ200 ng of protein (data not shown). Thus, it is important to evaluate the binding of such V H domains over a range of amounts of immobilized antigen to identify suitable conditions for maximizing sequence specificity.
We also find that the assay format influences the extent of conformational specificity of the V H domains. The A␤17-42 (DDD) and IAPP8 -37 (DED) V H domains show significant conformational specificity both in well plates and immunoblot assays (Figs. 9 -11). However, some differences were observed for the QQQ and SSS A␤17-42 V H domains. The higher binding signals of these domains for disaggregated A␤ in the well plate assay and for fibrils in the immunoblotting assay may be due to the different mechanisms of antigen immobilization and/or the impact of varying antigen concentration (immunoblotting) versus varying antibody concentration (well plates). Our findings demonstrate the importance of standardizing assays with well defined fibril and disaggregated peptide samples to properly interpret the results. Indeed, we find that conventional antibodies with conformational specificity for fibrils (LOC and OC) only show fibril specificity at low amounts of immobilized antigen (Fig. 10). Interestingly, we also observe a similar level of conformational specificity for a sequence-specific A␤ antibody (NAB 228) at low amounts of immobilized antigen. This demonstrates that the observed conformational specificity of many antibodies (including conventional ones) is dependent on assay conditions and must be interpreted carefully (65).
A key next step will be to identify the residues within the grafted A␤17-42 and IAPP8 -37 peptides that contribute most to the affinity and conformational specificity of the V H domains. This is important to further optimize the antibody domains as well as to guide the design of anti-amyloid antibodies specific for other amyloid-forming polypeptides. We are currently expanding our use of alanine-scanning mutagenesis to the entire CDR3 to identify the most important residues for expression, solubility, and binding. It will also be important to further increase the affinity and conformational specificity of these grafted domains using directed evolution methods. We are currently randomizing positions within and near CDR3 of similar grafted V H domains, and selecting variants with improved affinity and conformational specificity. We expect that these and related approaches will lead to the rapid design and optimization of antibodies with high conformational and sequence specificity for amyloid aggregates.