Structure and biochemical analysis of the heparin-induced E1 dimer of the amyloid precursor protein

Abstract

The amyloid precursor protein (APP) is the key player in Alzheimer’s disease pathology, yet APP and its analogues are also essential for neuronal development and cell homeostasis in mammals. We have determined the crystal structure of the entire N-terminal APP-E1 domain consisting of the growth factor like and the copper binding domains at 2.7-Å resolution and show that E1 functions as a rigid functional entity. The two subdomains interact tightly in a pH-dependent manner via an evolutionarily conserved interface area. Two E1 entities dimerize upon their interaction with heparin, requiring 8–12 sugar rings to form the heparin-bridged APP-E1 dimer in an endothermic and pH-dependent process that is characterized by a low micromolar dissociation constant. Limited proteolysis confirms that the heparin-bridged E1 dimers obtained in solution correspond to a dimer contact in our crystal, enabling us to model this heparin-[APP-E1]₂ complex. Correspondingly, the APP-based signal transduction, cell–cell- and/or cell–ECM interaction should depend on dimerization induced by heparin, as well as on pH, arguing that APP could fulfill different functions depending on its (sub)cellular localization.

Alzheimer’s disease (AD) is the most frequent form of progressive dementia, occurring predominantly in the elderly population. The short 40–42 amino acid peptide Aβ is generally believed to be the causative agent of the disease. Aβ is proteolytically derived from the amyloid precursor protein (APP), the founding member of a highly conserved gene family with redundant function (1), consisting of the members APP, APLP1, and APLP2. Generated by sequential β- and γ-secretase cleavage of APP, Aβ forms small heavily neurotoxic aggregates and finally accumulates in the brain as disease-typical senile plaques. Alternatively, cleavage of APP by α- and γ-secretase releases the nontoxic peptide p3. Although this general pathological pathway of AD is well documented, only little is understood regarding the physiological function of APP, how the proteolytic cleavage events relate to it, and which structural features of APP provide the required functionalities (for recent reviews see, e.g., refs. 2–6).

Several splice variants have been reported for APP. In neuronal cells, the protein is expressed predominantly as a 695-residue type-I transmembrane protein, showing a large ectodomain, that consists of several subdomains (Fig. S1) (7). Experimental 3D structures are known for the N-terminal growth factor like (GFLD) (8) and the copper binding domains (CuBD) (9, 10) as well as for the C-terminal helical central APP domain (CAPPD) (11–13) and its Caenorhabditis elegans homologue (14). These two regions of defined secondary structure, also called E1 and E2 domain, are connected by a potentially flexible acidic domain (AcD). A linker of unknown structure containing the α- and β-secretase cleavage sites connects the ectodomain to a helical transmembrane segment containing the γ-secretase cleavage site, which is followed by a short cytoplasmic domain (AICD) of weak secondary structure (15). Integration of the structural information on isolated fragments into a comprehensive overall picture has been attempted by small angle x-ray scattering (SAXS) (16). However, the resulting SAXS model has not been corroborated by other structural or functional studies to date.

Many potential physiologic functions have been proposed and discussed for APP. The binding of metal ions [such as copper and zinc (17, 18)] as well as components of the extracellular matrix [ECM; (19)] like, e.g., heparin to its ectodomain, suggests a function in cell–cell and/or cell–ECM adhesion and metal transport (3). The two heparin binding sites in the APP ectodomain (19, 20) could hereby act either independently or simultaneously. APP has been extensively associated with a receptor-like signaling function that might result in a regulated intramembrane proteolysis series of events in analogy to the well known Notch signaling (21), possibly depending on dimerization. Different proteins are discussed for the intracellular relay of an external signal (recently reviewed in ref. 22) and several proteins have been described to interact with the ectodomain of APP, therefore representing potential APP ligands (23–25). Besides, secreted soluble APP-ectodomain fragments have been shown to interact in a ligand-like manner with membrane-bound APP (26, 27).

The physiologic oligomerization state of APP that could be linked to its receptor function has been elusive to date. In addition to monomeric protein, dimers and higher oligomers are found for both isolated subdomains and full length APP (13, 16, 28). The ordered dimerization of APP has also been implicated in cell–cell interactions (29). It has been reported that dimerization depends on zinc binding (18, 30), heparin binding (16), or the simultaneous interaction of several sites within the APP molecule (31) including its transmembrane domain (29) and that it is functionally associated with the concurrent dimerization of the β-site APP cleaving enzyme (28, 31–33). Recently, Gralle et al. corroborated the physiological importance of APP dimerization using single molecule FRET methods (27). However, it is currently unclear by which molecular mechanism the oligomerization of APP is realized.

In addition, the soluble, membrane shed extracellular domain of APP has been extensively associated with trophic functions (see, e.g., refs. 19, 26, 34, 35), probably acting as a ligand for one or more unknown receptors. In contrast, Nikolaev et al. reported recently that the soluble APP N terminus resulting from β-secretase cleavage (APPsβ) acts as ligand for the orphan death receptor six (DR6), leading to axon pruning and neuronal death after deprivation of other trophic factors (36).

The segmented structural data available so far cannot explain most of the proposed functions of APP. Consequently, structural analyses of larger constructs as well as detailed studies of their interactions with APP binding molecules are essential for a deeper understanding of the physiologic function of APP. We report the crystal structure of the entire N-terminal APP E1 domain, revealing a rigid independent entity and demonstrate tight dimer formation upon interaction with heparin.

Results

Structure Determination and Overall Structure.

The independently folded N-terminal E1 segment of the APP was expressed in Escherichia coli and crystallized, resulting in rhombohedral crystals containing eight molecules per asymmetric unit. We have determined its molecular structure using a complex molecular replacement scheme (SI Text). The structure was refined at 2.7-Å resolution showing good R factors and stereochemistry (Table 1). In the final model, all eight APP chains are well defined in the electron density map from Leu28 to Ala198 including a large portion of the C-terminal histidine tag.

Table 1.

Data collection and refinement statistics

Data collection statistics
Beamline	In-house	BL14.1
Wavelength, Å	1.54	0.92
Resolution limit, Å	3.4	2.7
Unit cell parameters:*a, c, Å	143.3, 350.0	144.0, 351.2
Completeness, %†	99.2 (100.0)	99.8 (99.7)
Rmerge, %†	11.8 (30.9)	6.6 (40.3)
〈I/σI〉†	17.2 (5.9)	10.8 (2.1)
Refinement statistics
Resolution range, Å	29.0-2.7
Unique reflections, work/free	70,672/3,735
Rwork/Rfree, %†	21.5 (35.2)/25.0 (39.1)
Atoms (non-H):	Protein/water/other	11,073/282/89
B factors:	Overall/Wilson plot, Å2	48.6/69.2
rms deviations:	From standard bond length, Å	0.008
	Of bonded B factors, Å2	3.3

^*Space group H3, hexagonal setting of R3.

^†Highest resolution shell in parentheses (BL14.1, 2.85–2.7 Å; in-house, 3.58–3.4 Å; refinement, 2.82–2.7 Å).

The polypeptide chain of all eight molecules folds into two separate but abutting domains (Fig. 1), corresponding to the GFLD (Leu28-Val123) and CuBD (Cys132-Leu189). The slightly larger GFLD consists of a central antiparallel β-sheet (Gβ1, Gβ2, Gβ3, Gβ4, and Gβ9), one α-helix (Gα1), and a short two-stranded β-sheet (Gβ5 and Gβ8), cross-connected by three disulfide bridges (Cys32-Cys68, Cys73-Cys117, and Cys98-Cys105). The region between Gβ5 and Gβ8 described previously as β-hairpin (8) is flexible and adopts alternative conformations in our crystals. The CuBD consists of a central three-stranded antiparallel β-sheet (Cβ1, Cβ2, and Cβ3) and the long α-helix Cα1, catenated by three additional disulfide bridges (Cys133-Cys187, Cys144-Cys174, and Cys158-Cys186). Both domains are interconnected by a well-defined interdomain linker (Gly120—Leu132). A more detailed analysis of the eight molecules in the asymmetric unit revealed two sets of four molecules each (set 1: MOLA, MOLD, MOLF and MOLG; set 2: MOLB, MOLC, MOLE, MOLH), sharing a much higher degree of similarity (rmsd of their C^α-atoms: 0.10 and 0.08 Å, respectively) than between the sets (rmsd 0.37 Å). In the following, only the best defined molecules, MOLB and MOLD will be described, showing larger differences for the N- and C-terminal residues and for the two loops Lys99-His108 and Val129-Cys133. All variable regions are characterized by elevated crystallographic B factors.

Fig. 1.

Cartoon representation of the APP E1 domain, depicting the GFLD in blue and the CuBD in green. Dark colors represent α-helices, light colors β-sheets; connecting irregular structures are shown in gray. The interdomain linker is shown in red and the side chains of disulfide forming cysteine residues are illustrated as spheres. Secondary structure elements are labeled using the prefixes “G” for the GFLD and “C” for the CuBD.

Domain Interface.

The highly conserved, well-defined interdomain linker does not adopt any standard secondary structure but connects both subdomains like a zipper and contributes to the mostly hydrophobic interdomain interface area of 496 Å². The interface is stabilized by one salt bridge and several hydrogen bond networks such as the interactions formed between Glu131 and Arg116 of the GFLD and Asp125 of the CuBD as well as the GFLD Asn89⋯CuBD His147/Tyr168 and the GFLD Glu87⋯CuBD Asp177 interaction, which is further stabilized by the GFLD residues Tyr115 and Lys66.

The interface residues show low B factors comparable to those of internal residues that are decreased compared to the independently solved structures of the single subdomains. The low B factors and the strong interdomain interactions clearly show that the E1 domain forms one rigid entity and does not consist of two independent domains connected by a flexible linker.

We found a high degree of evolutionary conservation for the interacting residues of the GFLD and the CuBD among APP and APLP2 proteins (Fig. 2A) arguing for a special functional importance of this interface. In contrast, we could not observe a similar conservation when we compared APP to the APLP1 proteins (Fig. 2B). Within this subfamily, the GFLD is more diverse, whereas the CuBD is as conserved as in the APP and APLP2 proteins. Accordingly, the interface residues are less well conserved in the GFLD of APLP1. However, the interacting residues of its CuBD are also quite variable. This implicates a loss of evolutionary constraints of this site in APLP1 compared to the APP and APLP2 family members, possibly because of (an) altered function(s) requiring this interdomain interaction. When we compared APP to nonvertebrate APPs (organisms where APLP1 and APLP2 proteins are not found), we also found a much higher variability of the interface area (Fig. S2). This points to a possible divergence in structure and function also between vertebrate and nonvertebrate APPs. Altogether, different interdomain interaction properties of distinct APP family proteins could be an explanation for functional differences of these proteins (31, 37).

Fig. 2.

Interdomain interface of E1. Surface representation of the GFLD and CuBD subdomains, colored according to conservation scores calculated for the respective APP family members ranging from blue (no conservation) to yellow (conserved throughout all sequences). Surface patches belonging to the interaction interface are in light colors; other surface regions shaded with gray. Both subdomains were rotated 90° against each other with respect to Fig. 1. Conservation of surface residues are shown among (A) vertebrate APPs and APLP2s, and (B)  vertebrate APPs and APLP1s.

Interestingly, the Cu-binding site (9, 10) is found in proximity to the interface between the GFLD and the CuBD such that the side chains of Glu87 and Asn89 belonging to the GFLD could contribute to copper binding. Furthermore, the known Cu-binding residues of the CuBD (38) show a conservation among the APP family members similar to the entire CuBD-GFLD interface. However, no bound copper ions could be detected in our crystals, grown in the absence of metal ions.

Dimer Interface.

The analysis of the interaction of the eight molecules in the asymmetric unit with one another and with crystallographic symmetry mates revealed only one dimeric interaction of possible physiologic relevance (dimer I; see SI Text). In this dimer, formed between molecule MOLB and MOLD (one molecule of each set), the two positively charged surface patches, which have been attributed to heparin binding before (8), come together obeying roughly a 2-fold symmetry, consistent with a heparin-mediated dimerization of APP (see below). The two molecules interact via a 618 Å² interface area mainly stabilized by extensive hydrogen bond networks involving several well-defined water molecules buried in the interface. A salt bridge between Lys60 of molecule MOLD and Glu79 of molecule MOLB, as well as strong hydrophobic contacts including residues Pro56 and Val76 of molecule MOLD and Pro56, Val76, and Pro78 of molecule MOLB, further stabilize the interaction.

Limited Proteolysis Confirms the Observed Interdomain and Dimer Interfaces.

To analyze the importance of the domain–domain and dimer interactions observed in our structure, limited proteolysis experiments were performed. Incubating the E1 protein with different proteases resulted in distinct fragmentation bands (Fig. 3A and Fig. S3), typical for the proteolytic digestion of well-folded proteins. The resulting fragments were analyzed by Edman sequencing and by MALDI-MS, thereby determining the exact cleavage sites (Table 2). All proteases employed generated fragments in the range of 20–22 kDa, which result from the cleavage of the C-terminal His₆-tag at neighboring residues. Only few additional fragments were observed, resulting from “internal” protease cuts after Arg102 and Ile94, being located in and near the interface of dimer I, respectively, as well as after Asp131 and Glu121 (Fig. 3A). All cleavage sites are located at surface loops accessible to protease action in the folded state of the protein. In particular no cleavage was found within areas that are protected by the interface between the GFLD and CuBD domains, showing that the rigid interaction observed in our crystals also occurs in solution. Next, we added different heparin molecules of defined length to the reactions. Although no effect was observed for the cleavage after Asp131 and Glu121, the two cleavage sites residing at the dimer-interface I were protected against proteolysis (Fig. 3B). Both, a heparin tetramer (dp4) and an octamer (dp8), protected from cleavage at Ile94. However, only dp8 could significantly reduce cleavage at Arg102, indicating that the heparin chain length influences dimerization (see below). Control digestions employing α-Casein as substrate showed that addition of heparin oligosaccharides itself did not influence the activity of the proteases (Fig. S4A–D).

Fig. 3.

Limited proteolysis of APP-E1. (A–C) Representative SDS-PAGE gels including the most characteristic time points of the proteolytic reactions employing trypsin (T), elastase (E), and V8 protease (V8). The ◂ and ◃ indicate full length APP-E1 and species without His₆-tag, respectively. (A) Stable proteolytic fragments as analyzed by Edman sequencing and MS (Table 2 and Text). (B) Effect of heparin tetramers (dp4) and octamers (dp8) on proteolysis. (C) Effect of pH on cleavage by V8 at different protease concentrations. (D) Protease cleavage sites mapped onto the APP-E1 structure. Thick and red tubes indicate regions of high B factors (high flexibility), thin and blue tubes represent low B factors (low flexibility).

Table 2.

Limited proteolysis data.

	N-terminal	Mass,		pH	Heparin
Fragment*	sequence	kDa	Cleavage site	sensitivity	protection by
1	KQCKT	10.5	R102↓K103	−	(dp8)
2	MLEVP	9.6	R102↓K103	−	(dp8)
3	QNWCK	11.7	I94↓Q95	−	(dp4/dp8)
4	MLEVP	11.9	E121↓F122	+	-
5	MLEVP	13.2	D131↓K132	+	-
6	FVSDA	8.1	E121↓F122	+	-

^*The cleavage fragments are numbered as in Fig. 3 (arrowheads).

As these limited proteolysis data were obtained at pH 8.0 and our crystals were grown at more acidic pH, we investigated the pH-dependence of the proteolysis by V8 protease that shows two pH optima and is hence optimally suited for this comparison. Fragments 4, 5, and 6 resulting from cleavage after Glu121 and Asp131 were observed at pH 8.0, whereas no proteolysis occurred at those sites when the reaction was performed at pH 5.7, even at a fivefold higher protease concentration (Fig. 3C). Formation of the ∼20 kDa band as a result of His₆-tag cleavage proves the activity of the protease (Fig. 3 and Fig. S4E). The observed pH-dependent cleavage can be explained by different accessibilities resulting from a more open and flexible structure at pH 8.0 and stronger subdomain interactions at pH 5.7, as also evidenced by the very low B factors of the domain interface residues in our crystal structure (Fig. 3D). However, the very weak bands resulting from cleavage after Glu121 and the absence of any cleavage within the interface area show that, even at pH 8.0, the two domains interact tightly, showing only a slight plasticity.

Biophysical Analyses Prove Heparin-Induced Dimerization of E1.

Prompted by our structural and limited proteolysis data, we investigated the formation of E1 dimers in the presence and absence of heparin (Fig. 4 and Table 3). Analytical gel permeation chromatography (GPC) indeed revealed a twofold higher apparent molecular weight (MW_rh) in the presence of heparin. In order to distinguish between an increase in the hydrodynamic radius (rh) that might result from changes in shape upon heparin binding and true dimerization, we also investigated the absolute molecular weight (MW_abs) in static light scattering (SLS) experiments, where at fixed scattering-angle the SLS signal is proportional to MW_abs. Keeping the protein concentration constant (equal UV₂₈₀), we observed a doubling of the SLS signal upon addition of heparin, as expected for a true dimer (Fig. 4A). The detailed MWs as calculated using both GPC and SLS are summarized in Table 3 and show almost the exact values for the monomer and the dimer in the absence and presence of heparin, respectively. In addition, we tried intensely to investigate MW_abs and MW_rh at pH 5.7. However, at this pH, a minimum of 500 mM NaCl had to be added to obtain a proper size-dependent retention volume on several GPC column materials tested, which unfortunately interfered with heparin binding.

Fig. 4.

Oligomerisation analysis by GPC, SLS, and ITC. (A) GPC-chromatogram showing the UV₂₈₀ absorbance and right angle light scattering (RALS) signals. The data were normalized with respect to the maximum UV absorbance (UV280) to correct for slight differences in concentration. (B) Heparin-binding curves at pH 5.7 and 7.2. Peak integrals are given as ▴ and ● ; curves fitted to a model representing a single set of independent sites are plotted as ─ and - - lines, respectively.

Table 3.

GPC, SLS, and ITC.

Molecular weights* calculated from GPC and SLS measurements
	pH	MWrh (kDa)	MWabs (kDa)
E1-HP	8.0	29	22.9 ± 0.2
E1+HP	8.0	60	42 ± 2
Binding parameters measured by ITC
	pH	Kd, μM	ΔH, kJ/mol	ΔS, J/K·mol	n†
E1+dp12	5.7	3.0 ± 0.1	+54 ± 4	+280±16	1.0 ± 0.1
E1+dp12	7.2	5 ± 1	+62 ± 6	+301±40	0.41 ± 0.06

^*MW_th = 21.7 kDa

^†The stoichiometric factor n corresponds to the molar ratio of heparin: APP–E1 in the complex.

To study the energetics of the interaction between the E1 domain of APP and heparin, we performed isothermal titration calorimetry (ITC) studies at close to physiological conditions. As expected from our MW calculations, we observed a heparin-to-protein binding ratio of about 1∶2 at pH 7.2 (Fig. 4B and Table 3). Surprisingly, this stoichiometry changed to 1∶1 (corresponding to a heparin:protein ratio of, e.g., 1∶1 or 2∶2) at the lower pH of 5.7. For easy comparison of the obtained stoichiometries, the peak integrals representing reaction enthalpies were always fitted to an independent site model yielding single values for ΔH, K_d, and the stoichiometry factor n. However, a more detailed inspection of the experimental data indicates a possible two-step process at acidic pH (Fig. 4B). The observed low micromolar dissociation constants of heparin, as well as the enthalpy and entropy values are quite similar at both pH values (Table 3). Interestingly, acidic pH values are found, e.g., in endosomes, indicating a potential physiological relevance of our observation of a pH-sensitive oligomerization (see Discussion).

Dynamic light scattering analyses performed at different protein concentrations show that the E1 domain dimerizes also without heparin at concentrations above 100 μM (Fig. S5A). Additional ITC measurements revealed that the measurable enthalpy contributions originate almost exclusively from the protein–protein dimerization, as increased protein concentrations resulted in strongly decreasing reaction enthalpies (Table S2).

Discussion

We have recombinantly produced and crystallized the autonomously folded heparin binding domain E1 of the large APP ectodomain and solved the first crystal structure of the whole E1 entity at 2.7-Å resolution. The successful structural investigation and its binding to heparin show that the protein is of native fold and no glycosylation sites that possibly could interfere with ligand binding are known within the E1 domain. Unlike previously assumed, the two constituting subdomains GFLD and CuBD interact tightly, showing an overall arrangement different from those predicted from SAXS and modeling analyses (16, 38). Despite the rather small interface area (496 Å²), the two subdomains form a stable unit in the crystal lattice, confirmed by low B factors of the interface residues, and in solution (evidenced by our limited proteolysis experiments). The latter data indicate, in addition, a pH-dependent variation in rigidity of the APP–E1 domain.

The oligomerization state of APP and its relation to various suggested physiologic functions is controversially discussed [e.g., (3, 7, 13, 16, 29, 32)]. Following the reductive approach that purified protein fragments corresponding to functional units of the holoprotein must show the properties of the whole protein, we have demonstrated the heparin dependent dimerization of APP. We identified a dimeric arrangement in our crystals (dimer I) consistent with heparin-mediated dimerization that we confirmed by limited proteolysis, enabling us to model this heparin–[APP-E1]₂ complex (Fig. 5A). About 10 sugar rings are required to span the positively charged surface area made up by the two opposing GFLDs which is in excellent agreement with our biochemical data: Initial GPC experiments of APP–E1 premixed with heparin chains of different length indicated that the molecular weight increases with heparin chain length, showing a transition to the dimeric size at a chain length of 10–12 sugar residues (Fig. S5B). Because the coinjection of heparin and APP leads to the separation of both molecules during GPC, we turned to supplement the buffer with heparin, using a heparin preparation of ∼3.6 kDa (about 11 sugar rings). By this method, the heparin-dependent dimerization was clearly demonstrated. Furthermore, binding of heparin to E1 is characterized by low micromolar dissociation constants as measured by ITC. This reaction is driven by high entropy contributions of around 300 J/mol·K, probably resulting from the release of associated ions and water molecules upon heparin binding, as also observed for other heparin-binding proteins (reviewed, e.g., in ref. 39). Using heparin molecules of defined chain lengths, we also investigated their effect by limited proteolysis. Again readily explained by our model (Fig. 5), only cleavage sites close to the positive surface patch were affected and we observed a strong dependence on heparin chain length. Ile94 can be directly protected by the binding of any heparin molecule to the positively charged surface. However, Arg102 is buried in the dimer interface, and only heparin chains long enough to enforce dimer formation can block cleavage at this site.

Fig. 5.

Model of the heparin-induced APP-E1 dimer. (A) Surface representation of the APP–E1 MOLB–MOLD dimer colored according to its negative (Red, -10 e^-/kT) and positive (Blue, 10 e^-/kT) electrostatic surface potential shown together with a modeled heparin decamer (stick model). (B) Close-up view of the interacting molecules MOLB (cartoon representation, Blue) and MOLD (surface representation, Green) together with the heparin decamer (stick model). Side chains of Arg102 and Ile94 preceding the protease cleavage sites sensitive to heparin binding are represented by spheres.

Unfortunately, we were not able to see any electron density for these heparin molecules in crystals grown in the presence of, or soaked with, defined heparin oligomers, even though heparin binding should be possible because the respective region is not involved in any crystal contacts. However, the large dimensions of the positive surface patch might result in differently bound heparin chains. The resulting averaged electron density would then be too weak to be visible at the current resolution limit.

Altogether, our results resolve several apparently contradictory functional features of APP’s complex biology (Fig. 6): At first, functions associated with the E1 domain such as dimerization, signal reception, cell–cell interaction, as well as binding to metal ions and ECM components, are likely to be structurally interrelated. Each one might be modulated by pH and differentially conserved among the APP family members. In fact, we have proven the heparin-dependent dimerization of E1. Extending those data to membrane-bound APP, the majority of extracellular APP should be bound in dimeric form to the highly abundant heparan sulphate proteoglycans (HSPGs) at the ECM or at the cell surface (Fig. 6A), whereas intracellular APP should be monomeric (Fig. 6B). This model is in excellent agreement with recent findings of Soba et al. showing the APP–E1-based dimeric interaction of APP in cellular assays (29). As our data concern the E1 domain, we cannot conclude whether the AcD, the E2/CAPPD domain, and/or the transmembrane helix of APP take part in dimerization and how far this affects cleavage by the secretases. Further studies are also needed to resolve the question whether the two protomers interact in cis, in trans, or both. However, our data strongly disfavor dimers, where E1 and E2/CAPPD interact crosswise as previously suggested (13, 16). The dimerization of the entire APP ectodomain has been observed only for an 18 kDa (∼70 sugar rings) but not for a 3 kDa heparin preparation (16) and the participation of the CAPPD in heparin-induced dimerization has been suggested (13, 16). Correspondingly, heparin chains longer than 10 sugar residues should be required for the simultaneous interaction of HSPGs/heparin with both binding sites of APP, essentially similar to the known ternary complex of heparin, thrombin, and antithrombin (40).

Fig. 6.

APP⁶⁹⁵-dimerization models within their biological context. The E1 domain is shown in structural detail as surface representation, and other parts are represented schematically. (A) HSPG induced homodimerization of APP solely based on the interaction of two E1 domains. The participation of other domains in the dimeric contact(s) is not part of this work but has been concluded from other data. (B) Monomeric APP, also illustrating the known processing and glycosylation sites. The α-, β-, and γ-secretase cleavage sites, a proposed fourth site of proteolysis (36) and the known O- and N-glycosylation sites are represented by red arrows and brown spheres, respectively. (C) Fragments resulting from β/γ-secretase processing. (D) HSPG induced dimerization of soluble APP-ectodomain fragments. The identity of R depends on the previous proteolytic reactions and stands, e.g., for the AcD only or the entire rest of ectodomain including AcD and E2/CAPPD.

Upon proteolytical processing of membrane-bound APP (Fig. 6B) various shed fragments are generated. AICD, responsible for signal transduction as well as the pathological Aβ are released together with various soluble ectodomain species (Fig. 6C and D). The interaction with HSPGs should also dimerize such soluble species that are reported to participate in axonal growth (34) and axonal pruning (36). They might also dimerize with membrane-bound APP and ECM binding of soluble forms could increase the local concentration of secreted APP leading to a modulation of its function as demonstrated, e.g., for growth factors (41).

Interestingly, the stoichiometry of the E1–heparin interaction as well as the rigidity of the CuBD–GFLD-interface depend on the pH value. If APP is shuttled between the cell surface (pH 7.2–7.4) and endosomes (pH 5.7; see, e.g., ref. 42), one might expect a switch in function dependent on the respective compartment. Taking this argument further, different pools of defined APP oligomers such as APP monomers, heparin-bridged dimers, ECM-linked APP, secreted fragments of APP, etc. might exist in different cellular compartments such as ER, Golgi, cell surface, and endosomes, but also in different tissues (different HSPG expression), resulting in its apparent diverse functions.

Last but not least, the symmetrical dimerization of receptors is a typical molecular event in signal transduction (43). The described heparin-bridged E1 dimer could hence represent an essential (sub)structure of the proposed signal transduction activity of APP.

Materials and Methods

Recombinant E1 domain comprising amino acids 18–190 of APP₆₉₅ was expressed in E. coli OrigamiB(DE3) and the His₆-tagged protein was purified by HisTrap HP, heparin affinity, and gel-filtration (all GE-Healthcare) chromatography, yielding up to 2 mg pure protein per liter bacterial culture.

Limited proteolysis was performed at pH 8.0 and pH 5.7 using elastase, V8 proteinase, and trypsin, stopped by addition of PMSF and analyzed by SDS-PAGE. Bands were characterized by Edman sequencing (Procise 494A, Applied Biosystems) and by MALDI-MS (Ultraflex II, Bruker Daltonics).

Analytical GPC and SLS studies were performed in 150 mM NaCl, 5 mM Tris·HCl, pH 8.0, on calibrated (MWGF70, Sigma) Superdex S200 columns (GE-Healthcare) using 90 μM (GPC) or 70 μM (GPC/SLS) APP–E1. Where applicable, a ∼3.6 kDa heparin preparation (Reviparin-Sodium, Abott) was added in 50-fold excess to the protein and the buffer was supplemented with 460 μM heparin. SLS data were recorded using VE 3580 RI and 270 Dual detectors (Viscotek).

ITC experiments were performed on a NANO ITC calorimeter (TA-Instruments) in 10 mM sodium phosphate, 150 mM NaCl at pH 5.7 or pH 7.2, 37 °C adding stepwise 0.5 mM heparin dodecasaccharide dp12 (Dextra Laboratories) to 50 μM APP–E1.

Crystals were grown by the sitting drop vapor diffusion method at 20 °C from 0.1 M acetate, pH 5.0, 2 M ammonium sulphate, 2% PEG 400, and 4% (±)-1,3-butanediol, adding 25% glycerol for cryoprotection. They belong to space group R3 with eight molecules in the asymmetric unit (solvent content 74%). Native x-ray data were collected at 100 K to 3.4 Å inhouse (Nonius FR591 generator, XENOCS FOX Optics, MAR Research MAR 345 image plate) and to 2.7 Å at the synchrotron [Berliner Elektronen-Speicherring Gesellschaft für Synchrotronstrahlung (BESSY/BL14.1)] and processed with programs of the CCP4 suite (44) (Table 1). The structure was solved by molecular replacement, using the Protein Data Bank structures 1mwp (8) and 2fjz (10) as search models (see SI Text), manually completed in MAIN (45) and refined in CNS v1.2 (46). All main chain angles fall into the most favored and additionally allowed regions of the Ramachandran plot (47).

PYMOL (DeLano Scientific LLC, www.pymol.org) was used for molecular graphics and rmsd calculations. Dimer and domain interfaces were analyzed using the PISA server (48) and are given as averaged accessible surface areas per protomer. For the calculation of the interdomain interaction surface, chain MOLD was divided in a GFLD part (Leu28Gly120) and a CuBD part (Glu121-Ala190), grouping the interdomain linker to the latter. The heparin–[APP-E1]₂ model was constructed manually onto the extended positively charged surface of E1 dimer.