A novel and conserved pocket of human κ-Fab fragments: Design, synthesis, and verification of directed affinity ligands

Abstract

Antibodies of type IgG may be divided into two classes, called λor κ, depending on the type of light chain. We have identified a conserved pocket between the two domains CH1 and CL of human IgG κ-Fab, which is not present in the λ type. This pocket was used as a target docking site with the purpose of exploring the possibilities of designing affinity ligands that could function as such even after immobilization to gel. The idea of the design arose mainly from the results of the saturated transfer difference (STD-NMR) screening of 46 compounds identified by means of virtual docking of 60 K diverse compounds from the Available Chemicals Directory (ACD). Surface plasmon resonance (SPR) was used as an alternative method to monitor binding in solution. A total of 24 compounds belonging to a directed library were designed, synthesized, and screened in solution. They consist essentially of an amino acid condensed to a N,N′-methylated phenyl urea. STD-NMR results suggest that a small hydrophobic side chain in the condensed amino acid promotes binding, whereas a hydroxyl-group–containing side chain implies absence of STD-NMR signals. Three compounds of the directed library were immobilized and evaluated as chromatographic probes. In one case, using D-Pro as the condensed amino acid, columns packed with ligand-coupled Sepharose (Amersham Biosciences) retained two different monoclonal samples of κ-Fab fragments with different variable regions, whereas a sample of monoclonal λ-Fab fragments was not retained under similar chromatographic conditions.

Demands for purified IgG and Fab fragments as biopharmaceuticals are high. About 30% of biotechnology-derived drugs under development are based on monoclonal antibodies of type G (Fassina et al. 2001), or Fab fragments thereof. They are used in cancer therapy and in the treatment of asthma, allergies, and other inflammatory diseases (Weiner 1999; Gelfand 2001).

Currently, IgGs and Fab fragments are generated in large quantities according to standard techniques using cellular expression systems. The most widely encountered production method today includes purification via affinity chromatography. For IgGs protein A from Staphylococcus aureus has gained a more established position as a highly specific affinity ligand than has protein G from group G Streptococcus for Fab fragments. Also, protein ligands are often relatively expensive to produce and amenable to different types of degradation (e.g., proteolysis, deamidation, isomerization, denaturation), and usually, these problems become aggravated at extreme pH conditions typically encountered in so-called cleaning in place protocols (CIP). It is therefore of interest to search for synthetic, low-molecular-weight ligands to Fab fragments that are not sensitive to proteolytic enzymes and, in general, are more stable than are proteins to extreme pH conditions.

Recently, the structure-based approach with in silico screening has been used to select (Westerfors et al. 2003) and design (Baumann et al. 2003) affinity ligands to a target protein. In this article, an attempt to explore the possibilities to develop low-molecular-weight ligands to Fab by using this approach is described.

The use of in silico screening for the identification of candidate generic affinity ligands to Fab fragments urges the search for a pocket or cleft, localized on a highly conserved region in these proteins. Variable domains (VL, VH) are ruled out because their sequences are generally not conserved among different human IgGs. The CH1 domain is a possible alternative. For instance, the binding site of protein G, located mainly in the CH1 domain, is highly conserved. However, it is also rather flat, as determined from the crystal structure of the complex (Derrick and Wigley 1994), and virtual screening is better suited for finding small ligands that bind to invaginations.

In this article, a conserved cavity in the form of a small pocket that is common to all antibodies of type IgG-κ is revealed. This potential site was found at the interface between the light chain and the heavy chain (CL and CH1 domains, respectively). This site is not present in Fab fragments with light chain of λ-type. Information concerning this pocket can be elicited indirectly from several structural descriptions in the Protein Data Bank. However, the existence of the pocket as such, as well as its structure and relevance, has, to the authors’ knowledge, never been described before. The pocket was used as a base for virtual screening to identify putative generic binders of κ-type antibodies. After testing the virtual screening hits for binding in solution, three compounds were immobilized to Sepharose and evaluated as chromatographic probes.

Results and Discussion

Sequence analysis

By sequence homology, heavy chains of IgGs can be classified into the four types 1, 2, 3, and 4, whereas light chains fall into two types called λ and κ. In humans, ~40% of the IgG molecules carry a light chain of λ type, whereas ~60% carry a light chain of κ type (Solomon 1976). IgGs and IgG fragments built up of both light and heavy chains inherit both types of partitioning. Light-chain sequences of the constant region are only moderately conserved (~30% identity), whereas they are highly conserved (~90% identity) within each of two different classes (λ or κ). As a consequence, a search for a novel conserved binding site should be carried out for each class separately.

A conserved pocket in κ-Fab

A partly hydrophobic pocket located between the constant domains (CH1 and CL) was identified. Compared to constant domains in IgG molecules with light chain of λ-type, which meet in a complementary fashion building a compact hydrophobic core, the CH1 and CL domains of κ-type antibodies leave a volume free from protein atoms that connects to the outside through a channel. The pocket is well defined and ~72 Å3 in size, as deduced from the number of equally spaced grid-points that could be accommodated without intercepting the van der Waals volume of the protein atoms by using an in-house–developed program (data not shown). The location of the pocket in relation to the entire Fab fragment is shown in Figure 1A ▶. Figure 1B ▶ illustrates how the compound AA4 (see below) has the potential to fit the pocket. Figure 1C ▶ shows a homology-based structural superimposition, made by using the program BIOPOLYMER (Tripos Inc.) focusing on the region with the pocket. The molecular sections superimposed are the constant parts (CH1 and CL) of eight κ-Fab structures available—1ad0 and 1ad9 (Banfield et al. 1997), 1dfb (He et al. 1992), 1gc1 (Kwong et al. 1998), 1bey (Cheetham et al. 1998), 1fvd (Eigenbrot et al. 1993), 1vge (Chacko et al. 1996), and 1b2w (Fan et al. 1999)—and they are all human (or humanized) Fabs belonging to at least two different subtypes (IgG1 and IgG4) according to the heavy chain partitioning. The figure illustrates the conserved occurrence of the empty volume of the pocket. The residues from the light chain that surround the pocket are Ser-131, Val-133, Ser-159, Gln-160, Glu-161, Ser-162, Ser-176, Thr-178, Leu-179, and Thr-180. The corresponding pocket-surrounding residues from the heavy chain are Leu-150, Lys-152, Phe-175, Pro-176, Ala-177, Val-178, Gln-180, Ser-186, Leu-187, and Ser-188. All these residues are strictly or highly conserved among all human IgGs of κ-type. The side chains from Val, Leu, and partially Phe residues point toward the interior of the pocket and contribute to its hydrophobic character. On the other hand, hydroxyl groups from Ser and Thr residues contribute to its hydrophilicity. However, due to the small volume of the pocket and to the high directionality of hydrogen bonding, the potential for H-bonding with small molecule ligands is probably limited. Apart from the above-mentioned, there are residues contributing with main-chain atoms with polar groups engaged in β-sheet–stabilizing hydrogen bonds.

Figure 1.

(A) Stereo-overview of the location of the pocket in Fab fragments with light chains of κ type. The domains are shown in ribbon diagram: light chain in cyan and heavy chain in orange. The pocket, shown in transparent blue solvent accessible surface (Lee and Richards style), is located between the CH1 and CL domains. Figure created with O. (B) Orthographic view of compound AA 4 in space-filling model docked with the program FlexX in the pocket in chicken-net Connolly channel created with MOLCAD. The surrounding residues are shown in stick model. (C) Stick model of eight overlaid structures of Fab fragments of κ type from the PDB in the region of the pocket. The surface of the pocket shown as Connolly-channel created with MOLCAD is color-coded according to hydrophobicity (polar to hydrophobic from blue to brown via green).

Screening in solution of the virtual screening hits

A total of 249 docked compounds satisfied a requirement of having one five- or six-member ring inside the pocket. After visual inspection, 84 were selected and ordered, 76 were readily available and purchased, and 46 were soluble enough for screening experiments.

The one-dimensional saturation transfer difference method (STD-NMR; Mayer and Meyer 1999), in combination with surface plasmon resonance (SPR), was used in the screening as previously reported (Baumann et al. 2003; Westerfors et al. 2003). Compounds were initially tested for binding using 0.5 μM (dilution 1) of the target protein Fab-K1, a monoclonal κ-Fab. The positive hits from this screen were tested for binding with 100 nM target protein (dilution 2). Only hits with a clear STD-NMR signal in the dilution 2 experiment were tested further with this method at a lower target protein concentration (20 nM, dilution 3) and in a control experiment with no protein to exclude false positives. Seven compounds, which were found positive in the dilution 2 STD-NMR experiment and were not rejected by the SPR assay, are considered to be the positive hits of the screening. Only these were given further consideration. Their structures are shown in Figure 2 ▶ and corresponding screening data in Table 1. The exclusion of all compounds found negative by STD-NMR was done at the risk of missing tight binders not detected by STD-NMR, however, giving positive signals with SPR.

Figure 2.

Positive hits from screening using monoclonal κ Fab, Fab K1. According to STD-NMR, compounds VSH2 and VSH6 are the strongest binders followed by VSH3. According to SPR, VSH7 dissociates slowly compared with the rest of the compounds tested with this technique. VSH1 and VSH6 were not tested with SPR.

Table 1.

The positive hits of screening in solution using monoclonal κ-Fab, Fab-K1

ID	Dilution 1 (500 nM of protein)	Dilution 2 (100 nM of protein)	Dilution 3 (20 nM of protein)	Control (no protein)	SPR
VSH1	Weak signal	Weak signal	n.d.	No signal	n.d.
VSH2	Signal	Signal	Weak signal	No signal	0.7
VSH3	Signal	Signal	No signal	No signal	1
VSH4	Weak signal	Weak signal	n.d.	n.d.	1.5
VSH5	Weak signal	Weak signal	n.d.	No signal	1
VSH6	Signal	Signal	Weak signal	No signal	n.d.
VSH7	Signal	Weak signal	n.d.	No signal	−2

Columns 2 through 4 give the results of the STD-NMR experiments at different protein concentrations. The fifth column gives the result of the control STD-NMR experiments (no protein). “Signal” means that a signal was detected in the STD-NMR spectra, indicating binding of ligand to protein. The sixth column gives the average SPR score. The SPR signals were interpreted qualitatively in the following way: 1, in case warfarin and/or buffer signal < SPR signal < 10 RU; 2, if SPR signal > 10 RU. The average of two or more injections was calcuated. A minus sign indicates slow dissociation (dissociation signal >50% of SPR signal). The structures of the compounds can be seen in Figure 2 ▶.

n.d. indicates not determined.

Compound VSH6 was not tested with SPR because the material was used up and there was none available for further study; instead, synthesized derivatives based on this structure were tested (see below). VSH1 was not tested with SPR due to limited solubility (slightly opaque solution).

From the point of view of the SPR-screening results, VSH7 dissociates slowly, with >25% of the compound being bound to the protein at the end of the dissociation period compared with VSH2, VSH3, VSH4, and VSH5, which dissociate readily. None of the seven compounds shown in Figure 2 ▶ produced an SPR response larger than the corresponding theoretical maximum value. Such high responses should be indicative of binding stoichiometries higher than one (Myszka and Rich 2000).

From the point of view of STD-NMR, the three most interesting compounds are VSH2, VSH3, and VSH6 showing clear STD-NMR signals at the two highest protein concentrations. At a lower protein concentration (dilution 3 in Table 1), only VSH2 and VSH6 were found positive (weak signals). These two were thus designated as the most promising ligands on the basis of the STD-NMR–based screening. Compound VSH2 was tested further with STD-NMR for binding toward the positive control Fab-K2 (monoclonal F(ab)′₂ κ fragment) giving a positive result, and that toward the negative control Fab-L2 (monoclonal Fab λ fragment) giving a negative result. These results do not contradict the hypothesis that the compound binds in the pocket. Corresponding control experiments using VSH6 were not carried out. Instead, such experiments were carried out with several derivatives of this compound containing a handle for immobilization to solid phase (see below). The somewhat weaker binder VSH3 is closely related to VSH6, both containing a benzylic accepting keto group with the phenyl ring substituted at para and meta positions. From binding data corresponding to compounds not considered positive, it could be deduced that an additional substitution at the ortho position most probably resulted in decreased binding (data not shown).

Strategy for ligand development

The structure of VSH6 offered a synthetic route for the introduction of a handle for immobilization to Sepharose in the form of the carboxylic acid functionality of an amino acid condensed to the urea (Fig. 3 ▶). Docking simulations of both VSH6 and AA4 (VSH6 with a spacer and a protected attachment point) into the pocket suggested that such a handle would be pointing to the outside of the protein (Fig. 1B ▶). A directed library was defined based on variation of (1) the substitution pattern around the aromatic ring, (2) the methylation versus nonmethylation of the urea, and (3) the type of amino acid condensed to the urea. The latter corresponds to the replacement of N-methyl amino butyric acid in the reaction shown in Figure 3 ▶ with an alternative amino acid type.

Figure 3.

Synthesis of directed library. In the construction of the directed library, the amino butyric acid moiety was replaced with alternative amino acid derivatives.

Ligand development was essentially divided in two stages. The first stage corresponded to the variation of the substitution pattern around the phenyl ring and the methylation versus nonmethylation of the urea, whereas the second stage was concentrated on the variation of the type of the condensed amino acid. All derivatives were tested for binding in solution as methyl esters in order to avoid nonspecific ionic interactions, which could lead to false positives, and also to best mimic the situation upon attachment to a gel.

Variation of substitution pattern

A group of 11 compounds around the structure of VSH6 were designed and synthesized in the first stage (Fig. 4 ▶). From the structure of VSH3, the idea arose of introducing a methoxy group at para position in some of the compounds. Fluoro substituents were introduced to expand hydrogen-bonding possibilities.

Figure 4.

First set of compounds of the directed library. Variation of this set is achieved by differences of the substitution pattern around the phenyl ring and the methylation vs. nonmethylation of the urea.

The screening results for these compounds are shown in the upper part of Table 2. The two different screening methods give somewhat different views of the binding. All 11 compounds bind to the target protein according to SPR, and none of them produced a SPR response larger than the maximum expected for a one-to-one binding complex. The best (and more or less equally good) binders are AA1, AA3, AA4, AA5, and AA8. Of these, AA8 and AA1 have slow dissociation. Especially, AA1 without substitutions on the ring is found to be one of the strongest binders with slow dissociation according to SPR.

Table 2.

Screening results for the directed library using monoclonal Fab κ, Fab K1

ID	Set	Dilution 1 (1000 nM protein)	Dilution 2 (100 nM protein)	SPR
AA1	1	No signal	No signal	−1.7
AA2	1	Weak signal	Weak signal	0.5
AA3	1	Signal	Weak signal	1.3
AA4	1	Signal	Signal	1.3
AA5	1	Signal	Weak signal	1.3
AA6	1	Signal	No signal	1
AA7	1	Weak signal	Weak signal	−1
AA8	1	Signal	n.d.	−1.2
AA9	1	Signal	n.d.	1
AA10	1	Weak signal	n.d.	0.2
AA11	1	No signal	n.d.	−0.3
AA12	2	No signal	No signal	0.5
AA13	2	Weak signal	No signal	0.5
AA14	2	Weak signal	Weak signal	0.8
AA15	2	Signal	Weak signal	0.5
AA16	2	Signal	Weak signal	0
AA17	2	Weak signal	No signal	1
AA18	2	Signal	Weak signal	1.3
AA19	2	Signal	Weak signal	1
AA20	2	No signal	No signal	1
AA21	2	No signal	No signal	n.d.
AA22	2	Weak signal	No signal	0.9
AA23	2	No signal	No signal	1
AA24	2	Weak signal	No signal	1

Column 2 is coded as 1 for set obtained by variation of substitution pattern and methylation vs. nonmethylation of urea (compounds shown in Fig. 4 ▶) and 2 for set obtained by variation of the condensed amino acid type (compounds shown in Fig. 5 ▶). Columns 3 and 4 give the results of the STD-NMR experiments at different protein concentrations. “Signal” means that a signal was detected in the STD-NMR spectra, indicating binding of ligand to protein. The fifth column gives the average SPR score. The SPR signals were interpreted qualitatively in the following way: 1, in case warfarin and/or buffer signal < SPR signal < 10 RU; 2, if SPR signal > 10 RU. The average of two or more injections was calculated. A minus sign indicates slow dissociation (dissociation signal >50% of SPR signal). n.d. indicates not determined.

According to STD-NMR, all compounds gave rise to STD-NMR signals, except for AA1 and AA11, both without substituents on the aromatic ring. A possible explanation to the apparent contradiction with the SPR result regarding AA1 could be that the affinity probably is too high to be detected in the STD-NMR affinity window. Further, comparing the four compounds with identical substitution pattern on the urea part, (AA2, AA3, AA4, and AA7) leads to the observation that pCl-mCl is the substitution pattern on the phenyl ring leading to the strongest signal at dilution 2 among those considered. In a parallel manner, comparing AA4 with AA5 and AA6, with identical substitution pattern on the phenyl ring but with different methylation pattern on the urea, indicates that methyl groups on both nitrogen atoms contribute to affinity. Thus, from the affinity point of view, AA4 was considered the best binder according to STD-NMR among the compounds of this group.

Selectivity was also tested with STD-NMR for five different compounds of this first set of the directed library. The results are shown in Table 3 together with the negative results of a control experiment without protein for those compounds. None of the tested compounds gave rise to STD-NMR signals when tested against 1 μM Fab-L2 (used as negative control), and the three compounds AA3, AA4, and AA5 showed weak signals against Fab-K2 (used as positive control). Among these three compounds, AA4 showed the best selectivity, binding one of two remaining negative control proteins contra two of two for both AA3 and AA5. Thus, even from the selectivity point of view, AA4 was considered the best among the compounds belonging to the first set based on the STD-NMR experiments.

Table 3.

Results of STD-NMR control experiments for some compounds of the directed library

		1 μM
ID	No protein	Fab-K2	Fab-L2	Protein G	3-amylase
AA1	No signal	No signal	No signal	No signal	No signal
AA2	No signal	No signal	No signal	Weak signal	No signal
AA3	No signal	Weak signal	No signal	Weak signal	Weak signal
AA4	No signal	Weak signal	No signal	Weak signal	No signal
AA5	No signal	Weak signal	No signal	Signal	Weak signal

Fab-K2 is monoclonal Fab of κ-type and Fab-L2 monclonal Fab of λ-type. Corresponding structures are shown in Fig. 4 ▶. Fab-K2 is a positive control, whereas the rest of the samples are negative controls.

Variation of condensed amino acid type

The next stage was to attempt to optimize the type of the condensed amino acid. This corresponded to replacing the linker region of AA4 with a structure with binding functionality. It was considered an advantage if this amino acid was methylated at the nitrogen because this would lead to products that are N-alkylated and therefore compatible with conclusions drawn previously. However, even nonmethylated alternatives were considered, allowing for more variants of amino acids to the possible price of an additional synthetic step.

The original strategy was to model all possible derivatives using the structures of commercially available amino acids. All these virtual products were then docked and analyzed. A molecular-weight filter was applied for the amino acids between 100 and 300 D to obtain compounds of the appropriate size from purely geometrical considerations of the space available at the entrance of the pocket. The amino acid suggestions obtained by this procedure were glycine, β-alanine, the dipeptide Gly-Pro, and the tetrapeptide Gly-Pro-Gly-Gly. Testing the glycine and the β-alanine derivatives would provide the possibility to test if the position of the keto group had any influence on affinity. Proline was chosen instead of Gly-Pro and Gly-Pro-Gly-Gly because of three reasons. First, from manual docking it was realized that there could be space available to the side chain, especially if some degree of induced fit occurs. Second, a pro-line derivative fulfills the N-alkylated urea motif obtained through structure/activity relationships of STD-NMR data. Finally, a proline side chain would provide a rigid alternative to the flexible linker and therefore a possibility to reduce entropy loss due to binding compared with AA4, as well as an expected increase in binding specificity. Additional suggestions, including variation in chirality of the side chain (L or D) and polarity of the amino acid, were tested in order to expand the expected structure/activity relationships further. One choice was to vary around the structure of the proline residue to include both L and D proline and four hydroxy proline variants (both L and D, both cis and trans). Another to include some nonbulky amino acids such as L-Ser, D-ser, L-Ala, and L-Thr. In total, 13 compounds were synthesized (Fig. 5 ▶), and the corresponding binding data are shown in Table 2.

Figure 5.

Second set of compounds of the directed library. Variation of this set arises from different types of the condensed amino acid (in brackets).

The best binders of the second set according to STD-NMR were AA14, AA15, AA16, AA18, and AA19, all of which show signals (however weak) at dilutions 1 and 2. All of these contain a hydrophobic side chain. The remaining compounds do not give rise to STD-NMR signals with two exceptions: AA13 and AA17, which show weak signals at dilution 1. Almost all of these contain a polar group; the only exception is AA13, a nonmethylated L-Pro variant. Thus, there exists a clear structure activity relationship. A small hydrophobic side chain (β-Ala, Gly, L-Pro, D-Pro, and L-Ala) promotes binding, whereas the inclusion of one hydroxyl group eliminates binding. Further the absence of the methyl group on the anilinic nitrogen reduces binding (from comparing AA13 to AA14), confirming previous results, and finally, even chirality of the proline side chain influences binding as AA18 has stronger STD-NMR signal than does the L-Pro counterpart AA14.

With the exception of AA16, all compounds of the second set bind to the target protein according to SPR, and none of them produced a SPR response larger than the maximum expected for a one-to-one binding complex. The best binder is AA18 according to this assay. In agreement with the STD-NMR results, it was found that compound AA18 containing a D-Pro residue gives a higher SPR signal than does the L-Pro–containing counterpart AA14.

Selection of compounds to immobilize

The binding data shown in Table 2 from STD-NMR and SPR were the base for the selection of which molecules that should be resynthesized in larger scale, immobilized to gel, and tested chromatographically. The two different experimental techniques give different views of the binding events as described (Westerfors et al. 2003), which make the results differ somewhat. The strategy was to choose after consensus. Among the compounds of the first set, several compounds (including AA4) share the best position according to SPR, whereas STD-NMR pointed to AA4 as the best binder. Therefore, AA4 was chosen for immobilization. In contrast to this situation, among the compounds of the second set (variations of the condensed amino acid) several compounds (including AA18) are equally good according to STD-NMR, whereas SPR points to AA18 as the best binder. Therefore AA18 was chosen for immobilization. Further, the L-Proline variant AA14 was chosen with the purpose to compare its performance upon immobilization to that of the D-Pro variant.

Determination of level of substitution

The selected compounds AA4, AA14, and AA18 were coupled to a Sepharose HP matrix to a concentration of ~10 to 11 μmole/mL gel.

Chromatographic runs

Chromatographic tests using PBS at neutral pH

All columns retained the target protein Fab-K1, but with different affinities. Columns AA4, AA14, and AA18 retained Fab-K1 10, 12, and 24 CV, respectively. The immobilized D-Pro–containing ligand was shown to bind the target protein better than did the L-Pro counterpart, which is in agreement with their relative affinities in solution as detected by STD-NMR and SPR. Two other monoclonal Fab fragments, one F(ab)′₂ κ fragment (Fab-K2, positive control) and one monoclonal Fab λ fragment (Fab-L2, negative control), were injected on the D-Pro–containing column, which resulted in binding of Fab-K2, whereas 96% of Fab-L2 ended up in the flow-through. Fab-K2 did not leave the column during the wash period, indicating that the ligand affinity toward the F(ab)′₂ fragment was stronger than was the affinity toward the Fab-K1 Fab fragment. This could probably be explained by the existence of twice as many binding sites in the F(ab)′₂ fragment.

No target protein (Fab-K1) or positive control (Fab-K2) was found to bind to the reference column with a capped spacer and no ligand.

Chromatographic tests adding 1M ammonium sulphate

Based on the results using PBS at neutral pH, efforts were made to improve the binding conditions and to find acceptable elution conditions for the AA18 column. Different pHs and different salt concentrations were tested. It turned out that using PBS with 1 M ammonium sulphate at pH 7.0 as binding buffer both Fab-K1 and Fab-K2 bound to the column, in contrast to Fab-L2 that came out directly in the flow-through. Appropriate elution conditions were, for instance, 50 mM acetate buffer (pH 4.0) containing 0.14 M NaCl, or PBS (pH 7.0) containing 10% n-propanol.

The addition of 1 M ammonium sulphate or 0.5 M sodium sulphate to the PBS resulted in improved binding of Fab-K1 to the AA18 column. In general, the addition of salt to the binding buffer is expected to increase the nonspecific hydrophobic interactions between the protein and, for instance, the spacer arm. The monoclonal Fab λ was, however, still found in the flow-through when run through the column under the same conditions.

Estimation of capacity and recovery

Ten milligrams of Fab-K1 were applied to 0.5 mL gel using PBS (pH 7.0) with addition of 1 M ammonium sulphate as binding buffer. It was found that all the protein was adsorbed to the column and could be eluted with acetate buffer (pH 4.0) containing 0.14 M NaCl (Fig. 6 ▶). The recovery was estimated to be ~100%. For Fab-K2, a breakthrough was observed when ~12 mg of protein had been applied to the column, and the capacity was therefore estimated to be >20 mg/mL. For Fab-K2 the recovery was ~92%.

Figure 6.

Capacity test of the affinity chromatography media containing ligand AA18 for binding of the monoclonal Fab κ fragment. Ten milligrams of Fab-K1 were applied to the column by using PBS (pH 7.0) with 1 M ammonium sulphate as binding buffer. Injection starts at 0 mL and ends at 10 mL. After this, a wash period of 6 CV followed. At ~15 mL, the protein was eluted with acetate buffer (pH 4.0) containing 0.14 M NaCl. The recovery was estimated to be ~100%.

Specificity test

Five milliliters of the Escherichia coli homogenate spiked with 5 mg Fab-K2 were injected onto the AA18 column. SDS-PAGE analysis indicated that the eluted fraction contained mainly Fab-K2. Thus, the binding ability remained in the presence of the protein mixture; however, some of the target protein was not retained, leaving the column before elution. Furthermore, a fraction of the E. coli proteins bound to the column and did not come off the column before CIP (data not shown).

Considerations regarding the generality of the results

The question of generality was addressed by testing the column AA18 for binding to polyclonal samples of intact human IgG1 of λ and κ type, respectively. It was found that the κ type polyclonal IgG1 sample was retained completely by the column, whereas the λ type polyclonal was not (data not shown). A partial binding of the λ-type sample was not surprising because it was indicated already in the specificity test that the column may bind other proteins. However, the capture of all the different proteins present in the polyclonal IgG1 sample of κ-type proves the generality of the column’s binding toward this class of proteins.

Conclusions

The first achievement of this work was the identification of a pocket between the two domains of the constant part of the Fab fragment. This pocket is novel in the sense that its existence has never been described before. The cavity is conserved for all human IgGs of κ-type and may prove to be useful for the discovery and improvements of new affinity ligands, which are specific for such targets. This can, for instance, be done by means of virtual screening combined with methods to detect binding in solution. We have attempted this procedure, and thus, virtual screening has been applied, producing a limited number of hits, which were purchased and tested for binding in solution using SPR and STD-NMR. Based on those results, a total of 24 compounds belonging to a directed library were designed, synthesized, and screened in solution. STD-NMR proved particularly valuable in relating the structures of ligands to their performance and helping to reveal even subtle improvements in the low-affinity range. Thanks to this, the best of variations in substitution pattern around the phenyl ring and methylation versus nonmethylation of the urea could be identified. In addition, a clear differentiation between the group of compounds with a small hydrophobic side chain in the condensed amino acid versus the group with small hydroxyl-containing side chains was made possible. SPR was useful in assisting in the choice of which of the hydrophobic side-chain–containing ligands to immobilize and in confirming binding to target protein.

All three experimental techniques—STD-NMR, SPR, and chromatography—differentiate the D-Pro derivative as a superior binder compared with the L-Pro counterpart, suggesting specificity in the interaction of these molecules and the target protein. In addition, five derivatives tested did not show STD-NMR signals when probed for binding to λ-Fab, which lacks the pocket, and at least three of the derivatives showed some kind of interaction with another λ-Fab with a different variable region. From the point of view of ligand design for affinity chromatography, four additional important goals have been fulfilled: (1) A linker with a protected attachment point was introduced without loosing affinity; (2) binding activity between three different immobilized ligands and the protein was retained even after immobilization; (3) one of the rationally designed columns can differentiate between monoclonal λ-Fab and monoclonal κ-Fab at a capacity of ~20 mg protein/mL and with a degree of immobilization about 10 to 11 μmole/mL; and (4) the binding activity was retained, although weakened, between the rationally designed column and the target protein when the latter was spiked into E. coli homogenate. The specificity of the column is, however, not ideal for affinity chromatography. It is possible that further improvements to this library would lead to an affinity ligand that can be more efficiently used in, for instance, affinity chromatography applications. Further improvements require, however, new information. Such information could be gained by a combinatorial approach in which a large number of variants of the present ligand are synthesized and properly evaluated. Another approach would be to attempt to determine the crystal structure with Fab fragment in complex with a ligand of this class and prepare a smaller directed library based on the information obtained.

Materials and methods

Molecular modeling

Modeling software

Versions 6.8 to 6.9 of the program package SYBYL (Tripos Inc.), running on an OCTANE (Silicon Graphics Inc.) workstation provided with two 195-MHz R10000 processors, have been used for most of the modeling including computations and visualization. Figure 1A ▶ has been produced with the program O, version 8.0 (Jones et al. 1991).

Filtering of data base for virtual screening

The program SELECTOR was used for filtering the molecules in ACD, allowing only for compounds with a molecular weight in the range from 200 to 500 D and a calculated logarithmic octanol/ water partition coefficient <4.0. The limit 500 D was used because it was assumed that the putative binding site was not appropriate to accommodate larger ligands. Also, smaller compounds with fewer degrees of freedom are more suitable for computational methods. Compounds containing triphosphate or tripeptide substructures were also rejected. After filtering, the number of ACD molecules was reduced to ~111 K. A distance-based algorithm, as implemented in the program Diverse Solutions, version 4.04 (Pearlman and Smith 1998), was used to select one diverse subset with 60 K molecules. The structures were then prepared for docking.

Bioinformatics

Sequence homology searches using BLAST (Altschul et al. 1990) followed by sequence alignments using CLUSTAL W (Thompson et al. 1994) were performed to identify the most invariant regions in the constant domains (CH1 and CL) of human IgG heavy chains of type 1, 2, 3, and 4 and light chains of κ-type, respectively. The MOLCAD (Tripos Inc.) multichannel surface tool in SYBYL was used to identify possible binding sites in the globular part formed when CH1 and CL bind to each other. The highest-resolution (2.0 Å) crystal structure of κ-Fab (Protein Data Bank accession code 1vge; Chacko et al. 1996) was investigated for this purpose.

Preparation of compounds for docking

The structures of the compounds were transformed into three dimensions by using the program CONCORD (Tripos Inc.), version 4.04, and ionized to reflect their most probable protonation state at pH 7.0. The coordinates were then subjected to 500 cycles of minimization by using the MMFF94 force field (Halgren 1996).

Docking simulations

The program FlexX (FhI SCAI / BioSolveIT GmbH; Rarey et al. 1996) was used for docking. Formal charges were turned on during the docking simulations. The protein structure used was the structure previously referred. Residues in the identified pocket together with some additional residues in the immediate surroundings were used in the active site file. Using one-letter code, these residues are Q124, S127, G128, T129, S131, V133, G157, N158, S159, Q160, E161, S162, S176, S177, T178, T180, and L181 from the light chain and K126, P128, S129, F131, L133, L150, K152, D153, F175, P176, V178, L179, Q180, S181, S182, L184, S186, L187, and S188 from the heavy chain. The terminal carboxamide group of H:Gln-180 was rotated 180 degrees so that the ɛ nitrogen of H:Gln-180 is at favorable hydrogen-bonding distance to one of the carboxyl oxygens of H:Asp153. Otherwise, defaults have been used and no special customizations have been done. Only the best-ranked conformation or each docked compound was retained for the next step in the virtual screening process (see next paragraph).

UNITY verify three-dimensional search

A query was constructed with the program UNITY, version 4.3 (Tripos Inc.), to select the docked compounds containing five- or six-membered rings with the center located within a sphere (radius, 2.0 Å) centered inside the identified pocket. This query was applied in verify mode to all the docked molecules. The hits obtained in this way were then subject to visual inspection.

Preparation of virtual combinatorial library

Searches for commercially available amino acids, natural and non-natural, were carried out with ISIS/Base. A molecular-weight filter between 100 and 300 D was used to obtain compounds of the appropriate size from purely geometrical considerations of the space available at the entrance of the putative binding pocket. Searches requiring carboxylic acids followed by searches requiring either primary amines or methylated nitrogens were then applied to obtain the amino acids. The corresponding derivatives were then virtually enumerated with the program LEGION (Tripos Inc.), prepared, docked to the putative binding site, and evaluated according to the procedures described above.

Synthesis of directed library

Preparation of methyl esters

The free carboxylic acid was dissolved in methanol, and thionyl chloride in catalytic amount was added drop by drop. The reaction mixture was stirred for 2 to 3 h at 0°C. Thereafter, the solvent was reduced in vacu, and the residue was purified by RP-HPLC or used without purification.

Synthesis of 3,4-dichloro-/(N-methyl)-aniline

3,4-Dichloro-aniline (40 mmole, 5 g) was dissolved in 400 mL of dichloromethane (DCM). To this solution was added iodo methane (40 mL), triethyl amine (5 mL), and NaH (40 mmole, 1.9 g, 50% in oil). The resulting mixture was stirred at ambient temperature overnight, after which small aliquots of water summing up to a total of 50 mL of water were added, followed by an additional hour of stirring. The reaction mixture was transferred to a separation funnel and extracted with 5% sodium thiosulphate, dried over magnesium sulphate, and concentrated in vacu to almost complete dryness. The remaining material was separated by silica chromatography (pentane/ether, 8:2); the appropriate fractions were collected and concentrated in vacu to almost complete dryness, yielding 3 g of material including some solvent. The correct material was indicated by liquid-chromatography mass spectroscopy (LC-MS) analysis. This material was directly used in the subsequent step.

Alternative synthesis of N-methylated aniline derivatives

The aniline derivative was dissolved in DCM and sodium hydride (in the case of AA3 sodium bis(trimethylsilyl) amide; 1.5 eq), and di-tertbutyl-di-carbonate (1.3 eq) was added, followed by stirring overnight at room temperature. The reaction mixture was transferred to a separatory funnel and extracted with water, dried over magnesium sulphate, and concentrated in vacu. The crude product was dissolved in tetrahydro furan (THF), lithium alumina hydride (1.2–2 eq) was added, and the reaction mixture was refluxed until completion as indicated by LC-MS. Thereafter, the mixture was filtered. This filtrate was used directly in the subsequent step.

Synthesis of urea derivatives

To a THF solution of the N-methylated aniline (or the non–N-methylated aniline derivative) was added phosgene (20% in toluene) in large excess, and the reaction mixture was stirred for 30 min at room temperature, concentrated in vacu, and redissolved in DCM. To this solution was added an excess of triethyl amine and the amino acid to be introduced (~1 eq). The reaction mixture was stirred for 3 h at room temperature, concentrated in vacu, and purified by RP-HPLC.

Hydrolysis of methyl esters

The methyl ester of the urea derivative (0.5 g) was dissolved in methanol (10 mL), and lithium hydroxide (0.25 g) was added. The resulting mixture was stirred at ambient temperature for 5 h, neutralized with 1 M HCl, and concentrated in vacu. The resulting material was purified by RP-HPLC.

Protein samples used in screening

The following protein samples have been used in screening: Fab-K1, humanized monoclonal κ-Fab (IgG1), Fab-K2, humanized monoclonal F(ab)′₂ (IgG3); and Fab-L2, humanized monoclonal λ-Fab (IgG1). All proteins were generously provided by Dr. Henry Heinsohn from Genentech Inc. The two different monoclonal κ-Fab samples have different variable parts. Some molecules were tested for binding to the negative control proteins porcine pancreas κ-amylase and protein G purchased from SIGMA.

STD-NMR screening

All NMR experiments were performed at 298 K on a Bruker Avance 500-MHz spectrometer.

In STD-NMR, ligands that either do not bind to or are very tightly bound to the protein cannot give any signal in the resulting difference spectrum. The detection limits can be tuned for binding by varying the protein concentration while keeping the ligand concentration constant. Under such conditions, at higher protein concentrations the weak to medium binders are detected, whereas at lower protein concentrations only medium binders are detected (Peng et al. 2001). In this study, different target protein (Fab-K1) concentrations were used, according to the desired dilution, for instance, 1.0 μM, 500 nM, 100 nM, and 20 nM. Proteins used as control (Fab-K2, Fab-L2, α-amylase and protein G) were used only at one concentration (1 μM). In all cases compounds were tested one by one. On-resonance irradiation was set at 0 ppm, and off-resonance irradiation was set at −40 ppm. Irradiation time in each scan was 2 sec, and 16 K data points were collected with 1024 scans in total. Compounds for testing were dissolved in deuterated dimethyl sulfoxide (DMSO) to a concentration of 50 mM, and 5 μL of the concentrated ligand solution was added to 495 μL buffer solution. The samples thus consisted of 0.5 mM ligand, 20 mM phosphate buffer, 100 mM NaCl, and 5% deuterated DMSO in D₂O at pD 7.5 uncorrected reading on pH-meter. A one-dimensional ¹H-spectrum was obtained first as reference spectrum, and subsequently, a saturation transfer difference (STD) spectrum was obtained.

Surface plasmon resonance

Experimental setup

SPR experiments were performed on a Biacore 2000 at 25 °C using PBS (pH 7.4) and 5% DMSO as a running buffer at a flow rate of 40 μL/min and a data collection rate of 2.5 Hz. Compounds to be tested were diluted from 50 mM stock solutions in DMSO and injected at 250 μM concentrations in running buffer. Both association and dissociation periods were set to 60 sec. Experiments were set up according to published recommendations (Myszka 1999) Compounds were screened for binding toward monoclonal κ-Fab Fab-K1 immobilized at levels ~6000 response units (RU) on CM5 chips via amine coupling. Control cycles in which protein G was injected over the Fab surfaces were used to monitor the quality of the Fab proteins. A decrease in protein G signal in the course of an SPR screening assay indicated a deterioration of the Fab protein surface. As a control for the used method and data evaluation procedures, warfarin was injected over immobilized human serum albumin surfaces in most screening experiments.

Data evaluation

BIAcore Control Software, version 3.1.1, and BIAevaluation, version 3.1, were used to run experiments and for data evaluation, respectively.

The response levels obtained were corrected by subtracting the correction factor from the reference-subtracted data according to reference (Frostell-Karlsson et al. 2000). A calibration cycle was run prior to the sample injections to be evaluated. The SPR signals were interpreted qualitatively in the following way: (1) in case warfarin and/or buffer signal < SPR signal < 10 RU; (2) if SPR signal > 10 RU. The average of two or more injections was calculated for each ligand. A minus sign was given to the average to indicate slow dissociation (dissociation signal >50% of SPR signal).

Immobilization to gel

Sepharose HP that had been derivatized with allyl glycidyl ether was activated with bromine and coupled with hexamethylene-di-amine according to an Amersham Biosciences standard protocol. The free amine content was determined to 17 μmole/mL gel according to an Amersham Biosciences standard protocol.

Two milliliters of this gel was transferred to a reaction vessel together with 2 mL N,N-dimethylformamide (DMF). 1-(3-Dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (0.15 mmole) and diisopropyl amine (0.1 mmole) was added, and the suspension was put on a shaker at 30°C. After 5 min the ligand to be coupled (0.1 mmole) was added, and the reaction was allowed to continue for 15 h.

Thereafter, the gel was transferred to a glass filter funnel and washed with a 1:1 mixture of DMF and acetic acid anhydride. The gel was allowed to be in contact with this solution for 30 min, after which it was washed with consecutively DMF, water, and 20% ethanol.

The amount of ligand coupled to the gel was determined with an NMR method using tri-methoxy benzene as internal reference.

Determination of level of substitution

High Resolution Magic Angle Spinn (HR-MAS) NMR spectroscopy for determination of the degree of ligand substitution on the affinity gel media was performed on a Bruker DRX 500 spectrometer equipped with a 4-mm HR-MAS probe. The assay was performed in DMSO at a temperature of 25°C. A gel sample consisting of 50 μL gel in deuterated DMSO was transferred to a 4-mm ceramic rotor containing 10 μL of a reference solution (0.1 M 1,3,5,-trimethoxybenzene). A one-dimensional ¹H NMR spectra was acquired collecting 64 scans, using 20 sec pulse delay, and a spinning speed of 5000 Hz. By comparing the integral of the aromatic signals of the reference with the integral of the aromatic signals of the ligand, the ligand density of the chromatographic media could be determined.

Chromatographic runs

General experimental details

Chromatography was performed by using two ÄKTA (Amersham Biosciences) explorer 10 systems, equipped with UV cells, pH meters, and conductivity cells. One system was equipped with an auto-injector, whereas samples were applied into the other system using a super-loop. Tricorn 5/20 or HR 5/5 columns were packed with 0.5 or 1 mL, respectively, of chromatography media to which the ligand to be evaluated was coupled to a concentration of 10–12 μmole/mL. Integration of peaks in the chromatograms was used to determine the amount of protein that was retarded, bound to the column, and/or ended up in the flow-through.

Chromatographic test of AA4

Three columns were used: two packed with 1 mL and one with 0.3 mL Sepharose HP with ligand AA4. Fab-K1 and Fab-K2 were injected on the columns at a concentration of 1 mg/mL in PBS, and 15 μg to 25 μg of proteins were injected at a flow rate of 0.5 mL/min; 20 mM citrate, 0.5 M NaCl (pH 4.0), and PBS (pH 7.0) were evaluated as binding buffers. The protein injections were followed by a wash period of at least 15 CV before a CIP was performed with 1 M NaOH for 5 CV.

Chromatographic tests of AA14, AA18, and a reference column

Fab-K1 was injected on five Tricorn columns packed with 0.5 mL Sepharose HP—two with ligand AA14, two with AA18, and one with a capped spacer—without ligand used as reference. In addition, Fab-K2 and Fab-L2 were injected on the columns with ligand AA18; 50 or 100 μg of protein at a concentration of 0.2 mg/mL were injected at a flow rate of 0.25 mL/min using PBS (pH 7.0) as binding buffer. Protein injections were followed by a wash period of 20 CV. CIP was performed with 5 CV 10 mM NaOH.

Test of different conditions for AA18

By using PBS (pH 7.0) with the addition of 1 M ammonium sulphate as binding buffer, several possible conditions were evaluated for elution of the Fab-K1 bound to the chromatographic media. PBS (pH 7.0) and no ammonium sulphate, as well as PBS (pH 7.0) with addition of 10%, 20%, and 30% of n-propanol and 50 mM acetate and 0.14 M NaCl (pH 4.0) were tested as elution buffers. The method for evaluation of different elution conditions included equilibration of the column with binding buffer, application of 100 μg of protein on column, buffer change, elution for 20 CV, CIP with 10 mM of NaOH, buffer change, and finally re-equilibration of the column with the binding buffer.

Estimation of capacity and recovery

Ten milligrams of protein Fab-K1 was dissolved in PBS with 1 M ammonium sulphate to a concentration of 1 mg/mL and applied to one of the AA18 columns by using PBS (pH 7.0) with addition of 1 M ammonium sulphate as binding buffer. Ten milligrams of Fab-K2 was applied, at a higher concentration (1.6 mg/mL), to the same column by using the same binding buffer. For both proteins, the loading was followed by a wash period of 6 CV. Elution was performed using pH 4.0 and low-salt concentration (0.14 M NaCl). Recovery was estimated by measuring the absorbance of the injected protein solutions and the eluted fraction.

Specificity test

A suspension of E. coli strain RV308, dry content 5%, was sonicated, centrifuged, and diluted five times in PBS with a final ammonium sulphate concentration of 1 M. Five milligrams of Fab-K2 was spiked into 5 mL of this E. coli homogenate and then injected onto the AA18 column, using PBS with 1 M ammonium sulphate at pH 7.0 as binding buffer. Elution was performed with acetate buffer (pH 4.0), 0.14 M NaCl.

Generality test

Polyclonal human IgG λ and κ (SIGMA) were diluted to the concentration of 0.2 mg/mL in PBS (pH 7.0). Fifty micrograms of the proteins were injected, using PBS as binding buffer. The flow rate was 0.25 mL/min. Protein injection was followed by a wash period of 20 CV before CIP was performed with 10 mM NaOH for 10 CV.

Abbreviations

• ACD, Available Chemicals Directory

• CH1, first constant domain of the heavy chain

• CIP, cleaning in place

• CL, constant domain of the light chain

• CV, column volume

• DCM, dichloromethane

• DMF, N,N-dimethylformamide

• DMSO, dimethyl sulfoxide

• LC-MS, liquid-chromatography mass spectroscopy

• MAS-HR, magic angle spin high resolution

• RP-HPLC, reversed-phase high-performance liquid chromatography

• RU, response units

• SPR, surface plasmon resonance

• STD, saturation transfer difference

• THF, tetrahydro furan

• VH, variable domain of the heavy chain

• VL, variable domain of the light chain