Structural basis of Alzheimer β-amyloid peptide recognition by engineered lipocalin proteins with aggregation-blocking activity

Crystallographic analysis of engineered lipocalins in complex with Aβ₄₀ and the minimal epitope peptide

Initially, the anticalin US7 was co-crystallized with the synthetic hexapeptide Ac-VFFAED-NH₂, which represents the central Aβ epitope as identified in a SPOT epitope screen (Rauth et al. 2016). Subsequently, US7 was also crystallized isomorphously in 1:1 complex with the synthetic full length Aβ₄₀ peptide. The two crystal structures, each with one protein•peptide complex in the asymmetric unit (a.u.), were solved at resolutions of 1.7 and 1.5 Å, respectively, using a synchroton X-ray source (Table 1). Finally, we also obtained crystals of the anticalin H1GA in complex with the full length Aβ₄₀ peptide at 2.3 Å resolution, this time showing a different morphology with four protein•peptide complexes in the a.u.

In all these crystal structures the anticalins US7 and H1GA (Figure 1) revealed the canonical lipocalin fold (Schiefner and Skerra 2015; Skerra 2000), as expected, comprising an eight-stranded β-barrel with an adjacent α-helix (Figures 2 and 3). Despite the considerable number of 17 mutated amino acids (plus two fixed mutations) that were distributed across the four structurally variable loops of the Lcn2 scaffold (PDB ID: 1L6M) (Achatz et al. 2022; Goetz et al. 2002) both anticalins showed a high degree of three-dimensional conservation, revealing root mean square deviations (RMSD) versus Lcn2 of 0.44 Å for US7•Aβ₄₀ (0.45 Å for US7•VFFAED) and of 0.42 Å (chain A) for H1GA•Aβ₄₀ upon superposition via a set of 58 conserved Cα positions within the β-barrel (Skerra 2000).

Figure 1:

Amino acid sequence alignment of wtLcn2 with the Anticalins US7 and H1GA.

Residues that contact the Aβ₄₀-peptide in the crystal structures (see Table 2) are highlighted using a gray background.

Figure 2:

Crystal structure of an anticalin in complex with the Aβ₄₀ peptide.

(A) H1GA•Aβ₄₀ side view (loops orange, peptide carbon atoms yellow); the disordered N- and C-termini (P1–P16 and P28–P40) of Aβ₄₀ are colored light blue. (B) View into the binding site of the H1GA•Aβ₄₀ complex, rotated by 90° around a horizontal axis. The first and last residues of the visible part of the Aβ₄₀ peptide are labelled.

Figure 3:

Structural comparison of the two anticalins selected against Aβ₄₀ with wtLcn2.

(A) Superposition of H1GA•Aβ₄₀ (loops orange, peptide yellow) with US7•Aβ₄₀ (loops blue, peptide cyan). The central part of the Aβ₄₀ peptide, including the Phe/Phe moiety, shows high overlap, whereas the flanking parts exhibit deviations. Loops #1, #3 and #4 reveal significant conformational changes as discussed in the text. (B) Superposition of US7•Aβ₄₀ (loops blue, peptide cyan) with US7•VFFAED (loops pink, peptide magenta). The minimal epitope peptide structurally coincides with the central moiety of the Aβ₄₀ peptide in the two different complexes while the loops of the anticalin show high structural similarity. (C) Superposition of US7•VFFAED (loops pink, peptide magenta) with wtLcn2 in complex with its natural ligand, Fe^III•enterobactin (loops green, ligand forest; PDB ID: 3CMP). The VFFAED peptide in the complex with the anticalin only partially occupies the cavity where the natural ligand of wtLcn2 is bound. Loops #1, #3 and #4 of the anticalin US7 deviate to a similar extent from the wtLcn2 structure as from the anticalin H1GA (see panel (A)). (D) Distinct interactions between the pair of side chains Trp¹⁰⁶/Glu¹²⁵ in H1GA versus Ala¹⁰⁶/Val¹²⁵ in US7 and the central residues Asp^P23/Val^P24 in the bound Aβ peptide. (E) Interaction between the conserved pair of positively charged residues, Lys⁷⁹ and Arg⁸¹, in both anticalins and the two anionic side chains at the center of the bound Aβ peptide, Glu^P22 and Asp^P23. (F) The side chain of Cys³⁶ in the original anticalin version H1G1 modelled in the crystal structure of H1GA with its most plausible rotamer, revealing close contacts (∼3 Å) to Phe^P20 and Ala^P21.

Compared with the wild type Lcn2, the loop conformations in the anticalins differ to varying degrees. Loop #2 in the US7•Aβ₄₀ complex as well as loop #4 in the H1GA•Aβ₄₀ complex show virtually identical backbone structure. The conformations of the longer loops #1 and #3, on the other hand, exhibit deviations up to 7.8 Å (position 43 in loop #1 of H1GA•Aβ₄₀) and 6.8 Å (position 99 in loop #3 of US7•Aβ₄₀). While, generally, the side chains both of the conserved and the mutated residues in the anticalins show similar orientations as in wtLcn2, position 41 reveals a notable exception: the side chain of the Ile residue in wtLcn2 points into the ligand pocket whereas the mutated side chains of Ser in US7•Aβ₄₀ and of Leu in H1GA•Aβ₄₀ are oriented outward.

The resulting reshaped ligand pocket of the anticalin US7, for example, is moulded by two pairs of opposing loops, with loops #1 and #2 on one side and loops #3 and #4 on the other (Figure 2), and has roughly rectangular dimensions of 22 Å by 23 Å. Due to the bulky side chains of residues Tyr⁵² and Lys⁷⁹ and of Lys⁴⁰ and Gln¹²⁷ on both sides of the cleft, the binding site of US7 is rather slim if compared with the one of the natural Lcn2 for its ligand enterobactin, a siderophore complexing Fe^III (Goetz et al. 2002), and appears ideally shaped to accommodate the linear Aβ peptide ligand.

In the crystallized complex of US7 with the full length Aβ₄₀ peptide, a stretch of 13 peptide residues in its central region, P16–P28 (KLVFFAEDVGSNK, minimal epitope underlined), is well defined in the electron density whereas the N- and C-terminal peptide segments are disordered. Compared with the structure of the complex between US7 and the shorter peptide VFFAED the diffraction quality was only slightly inferior (cf. Table 1). In line with the isomorphous crystallization, the conformations of both the anticalin and the bound epitope peptide are highly similar in the US7•Aβ₄₀ complex, with an RMSD of merely 0.099 Å for the 58 conserved Cα positions and of 0.373 Å for all 177 common Cα atoms of the protein and peptide chains. In particular, the conformations of the structurally variable loops are virtually identical in both structures; the largest deviation occurs in loop #3 with 2.1 Å at residue Lys⁹⁸ and in loop #1 with 1.8 Å at Lys⁴⁰.

Interestingly, the visible Aβ peptide segment is compactly folded in three to four turns (see below). Its conformation is stabilized by several intramolecular hydrogen bonds, including a hydrogen bond network with 5 water molecules that appear well defined in the electron density (Figure 4). This peptide stretch wriggles deeply into the cavity of the engineered lipocalin, which in this manner gets almost completely filled. Nine peptide residues (VFFAEDVGS) are fully buried within the ligand pocket. The tight interaction with the anticalin is dominated by a multitude of hydrogen bonds, salt bridges and intimate hydrophobic interactions (Table 2).

Figure 4:

Conformation of the Aβ₄₀ peptide and its detailed interactions with the anticalin H1GA.

(A) Aβ₄₀ peptide conformation in the anticalin complex with intramolecular hydrogen bonds shown as black dashed lines (carbon atoms yellow). The three turns formed by the central segment of the Aβ peptide are marked by ellipses in different colors and displayed individually to the right. The hydrogen bond between the main chain C=O(i) and N-H(i + 3) of turn 3 is highlighted by a green dashed line whereas the corresponding Cα/Cα distances in the other two turns are indicated by gray dashed lines. (B) Aβ₄₀ bound to H1GA (top view). Protein residues that form hydrogen bonds (black dashed lines) or salt bridges (red dashed lines) with the peptide ligand are depicted as sticks. (C) Surface representation of the binding site of H1GA with the bound Aβ₄₀ peptide, whose hydrophobic side chains occupy different subpockets, in the same orientation as shown in panel (B). Protein residues that form contacts to Aβ₄₀ are colored light pink (for residue labelling see panel (B)).

Due to the extraordinary target affinity and the pronounced Aβ₄₀ aggregation-blocking activity of the anticalin H1GA its interactions with the bound peptide are of particular interest. H1GA was selected in an independent phage display campaign and differs in 13 amino acid positions from US7 (Rauth et al. 2016). This explains its mode of crystallisation in a different crystal form with four protein•peptide complexes in the a.u. (Table 1). Polypeptide chain A showed the lowest crystallographic B factors overall and, thus, was chosen for detailed analysis. Compared with the structure of the complex between US7 and the Aβ₄₀ peptide the conformations of both anticalins with the bound peptide are highly similar, resulting in a very low RMSD of 0.306 Å for the 58 conserved Cα positions and of 1.903 Å for all 172 common Cα atoms of the polypeptide chains. Similar to US7, the central segment of the Aβ₄₀ peptide ligand (P17–P27: LVFFAEDVGSN, minimal epitope underlined) showed well defined electron density also in complex with the anticalin H1GA (Figure 2).

A striking feature of this bound Aβ segment is the cluster of five strongly hydrophobic side chains, LVFF^P17–P20 and Val^P24, with a negatively charged Glu-Asp dipeptidyl moiety in between (Figure 4). This coiled stretch, with the hydrophobic side chains pointing into different directions, constitutes the experimentally verified minimal epitope (Rauth et al. 2016). Notably, the LVFF moiety forms the first in a series of altogether three (four in the US7 complex) adjacent tight turns (Figure 4A): Leu^P17–Phe^P20, turn 1; Ala^P21–Val^P24, turn 2; Asp^P23–Ser^P26, turn 3. Turn 3 is a type II β-turn with the canonical H-bond between the main chain C=O(i) and N-H(i + 3) – as well as a Gly residue at position i + 2 – whereas turns 1 and 2 belong to the "open" class of β-turns with a distance between Cα(i) and Cα(i + 3) of less than 7 Å (Chou 2000; Wilmot and Thornton 1988). Interestingly, in the US7•Aβ₄₀ complex, where Lys^P28 is also visible in the higher resolution electron density map (cf. Table 1), a fourth β-turn becomes evident, formed by residues Gly^P25–Lys^P28 and belonging to type I, though with a geometrically non-ideal intra-backbone H-bond (not shown). Overall, this leads to a peculiar zigzag-bend conformation for the central Aβ segment in both anticalin complexes.

Analysis with PISA (Krissinel and Henrick 2007) revealed an interface area of 886 Å² for the visible part of the Aβ₄₀ peptide, which constitutes 65% of the total solvent accessible surface of the peptide stretch P17–P27. Within this interface, 9 hydrogen bonds and 3 salt bridges are formed (Table 2). A detailed interaction analysis using CONTACT (CCP4 1994) revealed that 11 peptide residues interact (within a radius of 4.0 Å) with 22 residues of the anticalin H1GA. The majority of anticalin residues that play a role for complex formation (18 of 22) interact via their side chains. Of note, 11 of these 18 residues had been mutated during the generation of H1GA (cf. Table 2 and Figure 1). As expected from the biochemical epitope mapping analysis (Rauth et al. 2016), the majority of interactions involve the central Aβ positions Phe^P19, Phe^P20, Ala^P21, Glu^P22 and Asp^P23 (see also further below).

While the first visible N-terminal residues of the bound Aβ peptide, Leu^P17 and Val^P18, are located at the entrance of the ligand pocket and still interact with the solvent, the following residue, Phe^P19, forms tight contacts with the side chains of Lys⁷⁹ and Ser⁶⁸ of the anticalin. Interestingly, the peculiar residue pair Phe^P19 and Phe^P20 within the central epitope of the Aβ peptide rests with almost oppositely oriented aromatic side chains in two distinct hydrophobic subpockets (Figure 4C). The phenyl group of Phe^P19 is sandwiched between residues Lys⁷⁹ and Tyr⁵² while Phe^P20 is wedged between Leu⁴⁹ and Tyr⁵², involving an edge-to-face contact with the aromatic side chain (Figure 4B).

The following small residue Ala^P21 is placed in a canyon formed by the hydrophobic side chains Val³³, Ala³⁶, Thr¹³⁶ and Tyr¹³⁸ of the engineered lipocalin (Figure 4B and C). The deepest point of this canyon is occupied by Glu^P22, which is anchored through an extended polar network between its carboxylate group and the guanidinium group of Arg⁸¹, comprising numerous hydrogen bonds and one salt bridge (Table 2). Beyond that, this residue forms a hydrogen bond to the only water molecule within the binding site, which itself is engaged in two hydrogen bonds to Tyr⁵² and Lys⁷⁹.

The next peptide residue, Asp^P23, is also deeply buried and forms two salt bridges and three hydrogen bonds with the side chains of Lys⁷⁹ and Arg⁸¹. Val^P24 is situated further up in the cavity again, close to Leu^P17, and interacts with residues Trp¹⁰⁶ and Asn¹³⁴ of the anticalin. While Gly^P25 is not engaged in a close interaction with H1GA – in contrast with the anticalin US7 (cf. Table 2) – Ser^P26 forms a final contact with Trp¹⁰⁶ and leads into the solvent-exposed disordered C-terminal peptide segment of Aβ. The last visible peptide residue in the H1GA•Aβ₄₀ complex, Asn^P27, just loosely interacts with the anticalin.

Similarities and differences between the individual anticalin•Aβ complexes

Although the anticalin US7 differs in 13 amino acid positions from H1GA (cf. Figure 1) and it crystallized non-isomorphously, the structure of its binding site and the mode of complex formation with Aβ, including the conformation of the bound peptide, are surprisingly similar. Nevertheless, there are some notable structural differences (Figure 3A). In particular, the binding pocket of H1GA is slightly larger, mostly because loop #4 is bent more outward and the side chain of Leu at position 49 occupies less space than the corresponding Trp side chain in US7.

According to an analysis with CONTACT, there are altogether 23 residues in US7 that intertact with the Aβ peptide (Table 2) while 14 contact residues are identical in both anticalins. Further analysis of the buried surface area of the Aβ peptide with PISA revealed that predominant interactions with both anticalins are made by residues Val^P18 to Ser^P26 (Table 2 and Figure 4B). On the side of the anticalin, most interactions arise from identical residues between US7 and H1GA. Contributions by differing residues are confined to the less tightly bound ends of the central peptide epitope, e.g. His⁷⁷ in US7 versus Leu⁷⁷ in H1GA as well as Trp⁴⁹/Leu⁴⁹ and Gln¹²⁷/Ala¹²⁷, or to the periphery of the ligand pocket, e.g. Val³⁶/Ala³⁶. An interesting exception is the linked pair of distinct residues Ala¹⁰⁶/Val¹²⁵ in US7 versus Trp¹⁰⁶/Glu¹²⁵ in H1GA (Figure 3D), which will be discussed below.

Of the 8 differing residues that form contacts with the Aβ peptide in US7 and H1GA (see Table 2), some are associated with significant deviations in the peptide binding mode and/or altered loop conformations. Trp at position 49 in US7 has a slightly smaller contact area compared to Leu in H1GA (42.1 versus 43.1 Å²; Figure 4B) but is involved in a π–π stacking of its aromatic ring with the side chain of Phe^P20 (not shown). In H1GA, Asp⁷² makes a van der Waals contact to Val^P18, which is not the case for Gly⁷² in US7. The two linked mutations Tyr¹⁰⁶ to Ala/Trp and Lys¹²⁵ to Val/Glu are responsible for the most prominent structural differences between the complexes of US7 and H1GA with the Aβ₄₀ peptide. Trp¹⁰⁶ in the H1GA structure provides a much larger hydrophobic interface with the peptide residues Asp^P23 and Val^P24 than the Ala residue in US7 (53.2 versus 7.0 Å²). This Trp side chain also adopts the role of the Val¹²⁵ side chain which is involved in a contact with the bound peptide in US7 (Figure 3D). Conversely, in H1GA the corresponding Glu¹²⁵ side chain is moved away from the binding site by the large indole group of Trp¹⁰⁶, which is reflected by the much larger contact surface of 32.6 Å² for Val¹²⁵ in US7 compared with just 1.8 Å² for Glu¹²⁵ in H1GA. Apart from the kinked side chain conformation of Glu¹²⁵, space for the bulky side chain of Trp¹⁰⁶ is further provided in H1GA by pushing outward the first two residues of β-strand F (Thr¹⁰⁴ and Ser¹⁰⁵) and the last three residues of β-strand G (Lys¹²⁴ to Val¹²⁶) (Figure 3D).

The conformation of the flanking loops is affected, too. Loop #2 (Gly⁷²–Lys⁷⁵ in US7, Asp⁷³–Lys⁷⁴ in H1GA) is bent outward in US7, with deviations up to 2.5 Å at the Cα atom of Asp⁷³. Furthermore, loop #3 in H1GA has a considerably different conformation compared to its counterpart in US7: the N-terminal loop segment, Gly⁹⁵–Pro¹⁰¹, is bent toward the binding site, whereas the C-terminal segment, Gly¹⁰²–Thr¹⁰⁴, leans outward (Figure 3A). Finally, loop #4 is also bent outward by up to 7.0 Å in H1GA (at the Cα atom of Asn¹²⁹) compared to the US7 structure. The conformational deviation of loop #4 ends at the differing residues Ser¹³²/Thr¹³² (US7/H1GA) where the Ser hydroxyl group adopts a distinct rotamer and forms an additional water-mediated hydrogen bond to the peptide at residue Val^P24.

In spite of these structural differences between the two anticalins, the position and conformation of the bound Aβ₄₀ peptide in the complexes with H1GA and US7 is surprisingly similar. Superposition of the 11 Cα atoms of the visible peptide segment Leu^P17–Asn^P27 results in a very low RMSD value of 0.83 Å. There are just a few smaller deviations either due to intrinsic conformational plasticity of the peptide or as a result of distinct contacts with the two anticalins. For example, the Cζ carbon atoms of the equivalent Phe^P20 residues are mutually shifted by 0.7 Å. In US7, this residue is in contact with the indole side chain of Trp⁴⁹ through face-to-face-stacking, as mentioned above, while in H1GA this is not possible with the corresponding Leu residue (Figure 4B).

A sequence alignment with wtLcn2 (Rauth et al. 2016) revealed mutually identical amino acids in 7 of the 20 randomized positions (Tyr⁵², Ser⁶⁸, Gly⁷⁰, Lys⁷⁹, Arg⁸¹, Arg⁹⁶, Asn¹³⁴) between H1GA and US7 (Figure 1), of which 3 are unchanged from wtLcn2. According to the crystal structure of the H1GA•Aβ₄₀ complex (see Table 2), each of these residues, except for Arg⁹⁶, is involved in the binding of the Aβ peptide, which explains their high conservation amongst the set of all three independently selected anticalins, including S1A4 (Rauth et al. 2016). The two basic residues Lys⁷⁹ and Arg⁸¹ seem particularly important in this regard since they form hydrogen bonds as well as salt bridges with the central pair of acidic residues within the Aβ epitope, Glu^P22 and Asp^P23 (Figure 3E).

In another aspect, the anticalin H1GA was derived from the initially selected anti-Aβ anticalin H1G1 by replacing an unpaired Cys residue that emerged at the randomized position 36 on the N-terminal side of loop #1 (Rauth et al. 2016). This residue was rationally exchanged by Ala or Val mainly to prevent non-physiological crosslinking and/or protein misfolding under oxidizing conditions. While, as anticipated, both exchanges resulted in much improved biochemical protein properties the Ala mutation, unexpectedly, led to a 70-fold higher affinity towards the Aβ₄₀ peptide. To explore the structural environment of Cys³⁶, we modelled this residue into the H1GA crystal structure (Figure 3F). The thiol side chain in the most probable rotamer gets directed toward the Aβ peptide, leading to van der Waals contacts with Phe^P20 and Ala^P21. However, it seems that this contact is too tight such that removal of the bulky sulfur atom via replacement by Ala – in contrast to Val – results in a sterically more favourable situation.

Comparison between the bound Aβ epitope and its conformation in fibril structures or other peptide complexes

A search through the protein data bank (PDB) for structurally related peptides with the Aβ peptide segment Leu^P17–Asn^P27 from the US7 complex using the SPASM server (Madsen and Kleywegt 2002) revealed only poor hits without obvious similarities. Even the first hit (PDB ID: 1AK0) showed a rather high RMSD value of 1.88 Å for a totally unrelated peptide sequence (ILGSSSSSYLASI) as part of a larger protein structure, nuclease P1 (Romier et al. 1998). While this best matching peptide comprises several turns, its conformation differs significantly from the one of the Aβ peptide segment. Hence, this analysis indicated a unique conformation of the central Aβ peptide moiety as seen in the complexes with the two different anticalins. Considering that the ligand pockets of H1GA and US7 differ by 13 residues, as described above, and that both anticalins were selected using different procedures and even different target molecules – the biotin-labelled full length synthetic Aβ₄₀ peptide in one case and a recombinant thioredoxin-Aβ₂₈ hybrid protein in the other (Rauth et al. 2016) – suggests that the peculiar structure of the central region of the Aβ peptide observed in the bound state represents a preferred conformation that may preexist in solution.

The structural properties of Aβ peptides have been analysed in many studies before, both in the free monomeric state, in complexes with antibodies and also with alternative binding proteins and as part of macromolecular fibrils. Furthermore, molecular dynamics simulations for Aβ peptides and their oligomers have been performed (Ciudad et al. 2020), however, with no hint on a peculiar conformation of its central moiety. To investigate the functional significance of the zigzag-bend seen here for the Aβ peptide when bound to the anticalins, we compared its conformation with several relevant coordinate sets for Aβ peptides or their complexes: the NMR structure of Aβ₄₀ bound to a dimeric affibody protein (PDB ID: 2OTK; Aβ residues 16–40 resolved) (Hoyer et al. 2008), the high resolution cryo-EM structure of native Aβ amyloid fibrils (PDB ID: 6SHS) (Kollmer et al. 2019) and the X-ray structure of Aβ in complex with solanezumab (PDB ID: 4XXD; Aβ residues 16–26 resolved) (Crespi et al. 2015). All these Aβ structures were compared with the peptide moiety in the higher-resolution US7•Aβ₄₀ complex using as reference the characteristic central Phe^P19/Phe^P20 motif, which was resolved in all experimental structures (Figure 5).

Figure 5:

Distinct conformations of the Aβ₄₀ peptide in different structural environments:

(i) In complex with the anticalins from this study, H1GA and US7, (ii) in complex with solanezumab (PDB ID: 4XXD), (iii) in complex with a dimeric affibody protein (PDB ID: 2OTK) and (iv) as part of an amyloid fibril purified from Alzheimer’s brain tissue and elucidated by cryo-EM (PDB ID: 6SHS). The peptides are shown as Cα traces, side chains of the central residues Phe^P19 and Phe^P20 are depicted as sticks.

Remarkably, none of the superimposed structures share the unique backbone conformation seen for the anticalin-bound Aβ peptide. Only the feature of the two Phe side chains pointing into more or less opposite directions is seen in the amyloid fibril structure and also in the complex with the dimerized affibody. However, the peptide backbone surrounding these Phe residues is extended in both structures and not compactly folded as in the anticalin complexes. Nevertheless, the feature of the two exposed large hydrophobic side chains appears to contribute to the formation of macromolecular β-amyloid fibrils (Kollmer et al. 2019), where Phe^P20 in one Aβ subunit nestles among residues Phe^P4 and Val^P18 of its own N-terminal peptide segment whereas Phe^P19 is engaged in close contacts to residues Leu^P17, Ala^P21, Val^P24 and Ile^P31 of a neighbouring strand. Obviously, the anticalins US7 und H1GA can provide an alternative hydrophobic environment for the pair of Phe side chains and, thus, compete with this kind of Aβ aggregation.

Comparison between the Aβ₄₀ peptide in complex with anticalins and with a clinical-stage anti-Aβ antibody

The crystal structure of solanezumab, one of the leading antibodies targeting Aβ that has been tested in multiple phase III clinical trials, has been published (PDB ID: 4XXD) (Crespi et al. 2015). Similar to the anticalins described here, solanezumab recognizes a central epitope in the Aβ peptide. This mAb was shown to reduce brain Aβ burden in transgenic mouse models of AD by sequestering plasma Aβ and favouring efflux from the CNS, which is accompanied by a decrease of Aβ levels in the cerebrospinal fluid (DeMattos et al. 2002). In the crystal structure of solanezumab, the residues P16–P26 (KLVFFAEDVGS) exhibit unambiguous and continuous electron density in the most complete of the two macromolecular assemblies present in the a.u.

At first glance, there are similarities between this Fab•Aβ complex and the ones of the anticalins. First, a linear peptide epitope of almost the same sequence and length (11–13 residues) is recognized by both types of binding proteins. Second, the buried surface of the bound Aβ segment has a similar size in all three complexes, with 886 Å² and 868 Å² for the anticalins H1GA and US7, respectively, and 960 Å² in the case of solanezumab. Third, despite differing backbone conformations, the dominant residues Phe^P19 and Phe^P20 are both deeply buried at the center of each binding site.

On the other hand, the conformation of the Aβ peptide bound to the solanezumab Fab fragment is quite different from both anticalin complexes. In the Fab complex the N-terminal peptide moiety, i.e. residues Lys^P16–Phe^P19, adopts an extended conformation, whereas the C-terminal segment, Phe^P20–Ser^P26, exhibits a helical structure and projects out of the binding site. The core of the Aβ epitope is formed by the Phe^P19/Phe^P20 motif, however, this time with both aromatic side chains pointing into the same direction. This dipeptide moiety is deeply buried at the bottom of the paratope, within the cleft between the pair of variable immunoglobulin domains. This is a very different Aβ conformation from the one seen in the anticalin complexes, where the two aromatic side chains point into almost opposite directions and each of them rests in a distinct hydrophobic subpocket. In fact, superposition of the Cα-atoms of residues Leu^P17–Ser^P26 in the Fab complex with the corresponding peptide segment bound to US7 or H1GA results in rather high RMSD values of ∼2.7 Å. For comparison, the corresponding RMSD value between the two anticalin complexes is as low as 0.27 Å.

Furthermore, the individual interactions with the respective binding protein and, in particular, the pattern of Aβ peptide residues that form hydrogen bonds or salt bridges are markedly different. In solanezumab only the N-terminal peptide residues, Lys^P16, Leu^P17, Val^P18, Phe^P19, Ala^P21 and Asp^P23, form tight interactions, mainly via the main chain. In the anticalin complexes, however, hydrogen bonds and salt bridges are formed primarily by the C-terminal residues: Ala^P21, Glu^P22 and Asp^P23 in the H1GA complex and Phe^P20, Ala^P21, Glu^P22, Asp^P23, Gly^P25, Ser^P26 and Lys^P28 in the US7 complex (see Table 2). Whereas the negatively charged Aβ residue Asp^P23 is involved in the formation of a salt bridge also in solanezumab, only the anticalin complexes comprise an additional salt bridge with the neighbouring Glu^P22.

Cross-reactivity analysis of the anticalin H1GA between Aβ peptides and related sequences occurring in abundant plasma proteins

Solanezumab, as well as other anti-Aβ antibodies such as crenezumab, are known to cross-react with human plasma proteins, which is thought to lead to partial saturation of the biopharmaceutical upon systemic administration to patients and, thus, to negatively affect the therapy of AD (Crespi et al. 2015; Watt et al. 2014). Consequently, it was of interest whether the anticalins selected against Aβ₄₀, in particular H1GA, might show a similar interaction with those proteins. The 12 plasma proteins in question (Watt et al. 2014) (see Table 3) share high sequence similarity, or even complete identity, with Aβ around the LVFF motif at the core of the linear epitope seen both in the solanezumab complex and also in our anticalin structures, as discussed above.

Hence, we investigated the potential recognition of those related epitope sequences present in the human plasma proteins by the Aβ-specific anticalin H1GA using the SPOT technique. This method, which employs peptides synthesized on a hydrophilic cellulose membrane as solid support (Frank 2002), was successfully applied before in order to precisely identify the minimal epitope sequence recognized by our selected anticalins (Rauth et al. 2016). Now, a set of 12 peptide sequences occurring in those relevant human plasma proteins and having similarity to the linear epitope ¹⁸VFFAED²³ recognized by H1GA were synthesized with varying lengths, as octamer, dodecamer and hexadecamer peptides, on the membrane along with the target peptide, Aβ_16–23 (Figure 6).

Figure 6:

Investigation of potential cross-reactivity of the Aβ-specific anticalin H1GA with related epitope sequences in human plasma proteins (Table 3) using the SPOT technique. Twelve protein sequences (2–13) with similarity to the linear epitope ¹⁸VFFAED²³ recognized by H1GA were synthesized as octamer (8 AA), dodecamer (12 AA) and hexadecamer (16 AA) peptides onto a hydrophilic cellulose membrane together with the original target Aβ_16–23 (1). The reverse sequence of Aβ_16–23 served as negative control (−1). The membranes were incubated with the anti-Aβ anticalin, followed by detection via the Strep-tag II. The synthetic Strep-tag II peptide itself was present on the same membrane as a positive control (0) and also as reverse sequence (−0). Apart from the very low non-specific background signals that were also seen without anticalin, or for wtLcn2, H1GA revealed binding activity exclusively to its target sequence Aβ_16–23.

After incubation with the anti-Aβ anticalin, only the peptide fragments derived from Aβ itself gave rise to pronounced signals (for all three lengths). The very low signals detected for the 12 plasma protein sequences were in the same range as seen for the reverse Aβ peptide fragment, serving as a negative control, or for the wtLcn2 protein, which was applied in parallel, or with the secondary detection reagent alone (see Figure 6). These data demonstrate that the anticalin H1GA binds to its Aβ target sequence with unparalleled specificity, which is explained by the tight structural complex formation with this peptide segment, also involving flanking residues of the central LVFF motif as described in detail above.

Conclusions

Using X-ray crystallographic analysis we have elucidated the structural mechanism how the central region of the Aβ₄₀ peptide is specifically recognized and tightly bound by two different anticalins. An unexpected finding was that, although selected in independent phage display campaigns and employing the target peptide in different molecular formats, both anticalins, H1GA and US7, bind to the same peptide epitope, which in both complexes adopts an unprecedented compact zigzag-bend conformation. Together with the earlier observation that in particular the anticalin H1GA suppresses Aβ aggregation already at substoichiometric levels, this feature supports the hypothesis that these anticalins selectively bind a peculiar conformation of Aβ₄₀ which might represent a relevant intermediate during oligomerization followed by aggregate and/or fibril formation. Hence, apart from a simple stoichiometric scavenging effect – as aspired in the light of the peripheral sink hypothesis – our structural analysis indicates that these anticalins may prevent Aβ oligomerization via a conformational selection mechanism, thus blocking the presumably most toxic event in the early molecular pathogenesis of Alzheimer’s disease. Together with the experimentally proven high target specificity, our study illustrates the potential of generating alternative binding proteins, beyond conventional antibodies, via advanced protein design as drug candidates for the treatment of neurodegenerative diseases.

Protein crystallization and X-ray data collection

The anticalins H1GA and US7 as well as the recombinant wild-type lipocalin (wtLcn) were prepared as soluble proteins, carrying the C-terminal Strep-tag II, by secretion in Escherichia coli and purified to homogeneity from the periplasmic cell extract via StrepTactin affinity chromatography (Schmidt and Skerra 2007) and size exclusion chromatography as previously described (Rauth et al. 2016). The Aβ₄₀ peptide was acquired from the Keck Biotechnology Resource Laboratory (Yale University, New Haven, CT). The short hexapeptide VFFAED (N-terminally acetylated and C-terminally amidated) was purchased from Centic Biotec (Weimar, Germany).

The monomeric Aβ₄₀ peptide dissolved in water (Rauth et al. 2016) was co-crystallized with the anticalins at 1.2:1 molar ratio using solutions of 0.8 mg/ml H1GA in 10 mM Tris/HCl pH 8.0, 50 mM NaCl and 2 mg/ml US7 in 10 mM Tris/HCl pH 8.0, 115 mM NaCl. After incubation for 1 h at 4 °C the mixtures were concentrated to 17 and 11 mg/ml, respectively, using amicon Ultra-4 centrifugal filter units (10 kDa cutoff; Merck Millipore, Darmstadt, Germany). After successful search for suitable precipitation conditions using an in-house nanodrop precipitant screen, diffraction quality crystals were obtained at 20 °C using the hanging drop vapour diffusion technique from the following mixtures: 1 μl anticalin•Aβ₄₀ solution with 1 μl 24% (w/v) PEG 3350 for H1GA and with 1 μl 30% (w/v) PEG 4000, 100 mM Na-acetate pH 5.25 for US7. For co-crystallization of US7 with the short VFFAED peptide, 1 μl of the purified anticalin (12 mg/ml) was mixed with 1 μl of the synthetic peptide (5 mM), both dissolved in 10 mM Tris/HCl pH 8.0, 115 mM NaCl, followed by the addition of 1 μl reservoir solution consisting of 27% (w/v) PEG 8000, 100 mM MES/NaOH pH 6.5. Crystals were transferred into their respective precipitant buffer supplemented with 20% (v/v) or 10% (v/v) PEG 200 as cryoprotectant for the US7 and H1GA complexes, respectively, and immediately frozen in liquid nitrogen. X-ray diffraction data were collected at beamlines 14.1 and 14.2 of the BESSY synchroton (Helmholtz-Zentrum Berlin, Germany). All anticalin•peptide complexes crystallized in the space group P2₁2₁2₁ (Table 1), however, with different numbers of protein•peptide complexes in the asymmetric unit (a.u.).

Crystal structure determination

X-ray diffraction data were processed with MOSFLM and scaled with SCALA (Winn et al. 2011). Molecular replacement was carried out with MOLREP for the US7 complexes and with Phaser for the H1GA complexes using the coordinates of the natural Lcn2 (PDB ID: 1L6M) (Goetz et al. 2002) as search model. The electron density of H1GA•Aβ₄₀ was initially averaged for non-crystallographic symmetry with RESOLVE (Terwilliger 2000). Atomic models were built with Coot (Emsley et al. 2010) and refined with Refmac5 (Murshudov et al. 1997) in iterative cycles using manual correction. Water molecules were added to all models using ARP/wARP (Langer et al. 2008) while rotamers of Asn and Gln residues were adjusted with NQ-flipper (Weichenberger and Sippl 2007). The refined structural models were validated with PROCHECK (Laskowski et al. 1993) and WHAT_CHECK (Hooft et al. 1996) and via the MolProbity server (Williams et al. 2018). Secondary structures were assigned using DSSP (Kabsch and Sander 1983), and protein-ligand contact surfaces were calculated with PISA (Krissinel and Henrick 2007). Molecular graphics and structural superpositions were prepared with PyMOL (DeLano 2002).

For the US7•Aβ₄₀ complex, electron density was resolved for 172 of the 188 protein residues, namely Asp⁶–Asp¹⁷⁷, whereas the Aβ₄₀ peptide was only defined in its central region for residues Lys^P16–Lys^P28. The electron density of the structurally variable loops #1 and #3 at the entrance to the ligand pocket, in particular positions Ser⁴¹–Asp⁴⁵ and Ile⁹⁷–Arg¹⁰³, had low quality and these segments were tentatively modelled with plausible stereochemistry. The electron density map of the US7•VFFAED complex allowed model building for residues Leu⁷–Asp¹⁷⁷ together with the entire epitope peptide. The N-terminal acetyl group as well as the C-terminal amide nitrogen of the peptide were clearly visible in the electron density. Again, however, poor density was seen for residues Ser⁴¹–Asp⁴⁷ and Lys⁹⁸–Gly¹⁰² in the loop regions. In case of the H1GA•Aβ₄₀ complex there was continuous electron density in all four polypeptide chains within the a.u. (A, B, C, D) for residues Ser⁵–Asp¹⁷⁷, Asp⁶–Asp¹⁷⁷, Ser⁵–Gly¹⁷⁸, and Leu⁷–Asp¹⁷⁷, respectively. The four copies of the Aβ₄₀ peptide were defined throughout residues Leu^P17–Asn^P27 (chains E and F bound to chains A and B, respectively) and Val^P18–Ser^P26 (chains G and H bound to chains C and D, respectively). Due to their overall lower B-factors and the better definition of electron density both for the anticalin and the bound peptide ligand, chains A and E were chosen for the structural analysis of H1GA as described in the text.

The atomic coordinates and structure factors of the refined crystal structures US7•Aβ₄₀, US7•VFFAED and H1GA•Aβ₄₀ have been deposited at the PDB, Research Collaboratory for Structural Bioinformatics (Rutgers University, New Brunswick, NJ), under accession codes 4MVI, 4MVK and 4MVL, respectively.

Epitope-peptide binding assay

To investigate the specificity of the anticalin H1GA towards its prescribed target Aβ₄₀, 12 human plasma proteins showing sequence homology with the central region of the linear Aβ epitope, KLVFFAED (Watt et al. 2014) (Table 3), were tested for possible cross-reactivity using the SPOT technique (Frank 2002). The relevant sequences were prepared as octamers, dodecamers and hexadecamers via fluorenylmethoxycarbonyl (FMOC) solid-phase peptide synthesis on a MultiPep RS instrument (Intavis, Köln, Germany) using an amino-PEG500-derivatized cellulose membrane (Intavis) as support for C-terminal immobilisation. After final acetylation of the N-termini, side chains were deprotected with trifluoroacetic acid as described (Zander et al. 2007). The Strep-tag II 10mer peptide (Schmidt and Skerra 2007) was synthesized on the same membrane with a C-terminal Gly–Gly spacer and served as an internal positive control for detection. Both the Aβ peptide fragments and the Strep-tag II were also synthesized with reverse sequences as negative controls. After completion of the synthesis, the membrane was washed once with ethanol, three times with phosphate-buffered saline (PBS), blocked with 3% (w/v) bovine serum albumin in PBS/Tween for 1 h and washed 3 times with PBS/T. Three copies of the membrane were prepared in this manner. The first and second membrane were probed with the anticalin H1GA or wtLcn2, respectively (100 nM each), in PBS/T for 1 h, followed by washing three times with PBS/T. The membrane was subsequently incubated with a 1:1000 dilution of Chromeo546 labeled StrepTactin (IBA, Göttingen, Germany) in PBS/T for 1 h to detect the bound recombinant lipocalin protein via its C-terminal Strep-tag II. To quantify any background signal arising from the use of this secondary reagent, the third membrane was incubated in the same manner while skipping the incubation with the lipocalin protein. Finally, the membranes were washed three times with PBS/T prior to signal detection on an Ettan DIGE system (GE Healthcare, Freiburg Germany) using the Cy3 channel. Quantification of the fluorescence signals including background subtraction was performed using Quant software (TotalLab, Newcastle-Upon-Tyne, UK).