Interpenetrating Cubes in the X-Ray Crystallographic Structure of a Peptide Derived from Medin_19–36

Abstract

Amyloidogenic peptides and proteins are rich sources of supramolecular assemblies. Sequences derived from well-known amyloids, including Aβ, human islet amyloid polypeptide, and tau have been found to assemble as fibrils, nanosheets, ribbons, and nanotubes. The supramolecular assembly of medin, a 50-amino acid peptide that forms fibrillary deposits in aging human vasculature, has not been heavily investigated. In this work we present an X-ray crystallographic structure of a cyclic β-sheet peptide derived from the 19–36 region of medin that assembles to form interpenetrating cubes. The edge of each cube is composed of a single peptide, and each vertex is occupied by a divalent metal ion. This structure may be considered a metal-organic framework (MOF) containing a large peptide ligand. This work demonstrates that peptides containing Glu or Asp that are preorganized to adopt β-hairpin structures can serve as ligands and assemble with metal ions to form MOFs.

Graphical Abstract

Introduction

Amyloidogenic peptides and proteins are rich sources of supramolecular assemblies. Sequences derived from Aβ (Alzheimer’s disease), human islet amyloid polypeptide (type II diabetes), and tau (Alzheimer’s disease and frontotemporal dementia) have been found to assemble as fibrils, nanosheets, ribbons, and nanotubes (Table 1).^1–20 Other types of assemblies have also been observed, including square channels from sequences derived from transthyretin (senile systemic amyloidosis) and cylindrins from sequences derived from αB-crystallin (cataracts).^21–22 Although these well characterized amyloidogenic peptides and proteins have yielded many novel supramolecular assemblies, the more recently discovered amyloidogenic peptide medin has not been as heavily studied and provides and exciting frontier for the discovery of interesting supramolecular assemblies.

Table 1.

Supramolecular Assemblies of Selected Sequences Derived from Amyloidogenic Peptides & Proteins

Peptide	Sequence	Assembly	Citation
Aβ(10–35)	YEVHHQKLVFFAEDVGSNKGAIIGLM	fibrila,b	1, 2, 3
Aβ(16–22)	Ac-KLVFFAE-NH2	fibril, helical ribbon, nanotubea,b,c,d	4, 5, 6
	KLVF(N-Me)FAE, KLVFFAE	nanotubee	7
Aβ(16–22), Italian mutant	KLVFFAK	nanosheetb,c	8
Aβ(16–20)	AAKLVFF	nanotubeb	9, 10
	βAβAKLVFF	helical ribbond	11
Aβ(19–20)	FF	nanotubeb	12
	Ac-FF-NH2, NH2-FF-NH2, Boc-FF-NH2	nanotubeb,c	13
	Fmoc-FF	flat ribbon, fibrilb,c,d	13, 14
hIAPP(20–29)	SNNFGAILSS	flat ribbon, helical ribbonc,d	15
	SNNFG(N-Bu)AI(N-Bu)LS(N-Bu)S	helical ribbonb	16
hIAPP(21–29)	NNFGAILSS	helical ribbonb,c	17
hIAPP(23–27)	FGAIL	flat ribbonb,c	18
tau(306–311)	Ac-VQIVYK-NH2	straight and twisted filamentb	19
transthyretin (106–112, 115–121)	TIAA(N-Me)LLS, SFSTTAV	square channele	20
αB-crystallin(90–100)	KVKVLGDVIEV	cylindrinb,e	21

^a.Solid-state NMR (ssNMR)

^b.Electron microscopy (EM)

^c.Atomic force microscopy (AFM)

^d.Cryo-EM

^e.X-ray crystallography

Medin, also referred to as aortic medial amyloid, is a 50-amino acid peptide that forms fibrillary deposits in aging human vasculature (Figure 1). These deposits have been implicated in the pathogenesis of thoracic aortic aneurysm and dissection. Westermark et al. found evidence of fibrillary medin in the aortic media of 97% of subjects over the age of 50.²² Although the amino acid sequence of medin was determined in 1999, the folding pattern of the monomer was not elucidated until 2017.²³ Using ¹³C NMR spectroscopy, Madine et al. identified that the monomer of medin contains three β-strand regions consisting of residues 7–13, 21–25, and 30–36.²⁴ The sequences 14–24 and 26–29 were suggested to form unstructured loops. Computational modeling using QUARK and ROSETTA consistently predicted the 21–35 region to fold as a β-hairpin.²⁴

Figure 1.

Amino acid sequence of medin. Amyloidogenic regions predicted by TANGO are highlighted in red. Underlined residues are β-strand regions identified by ¹³C NMR spectroscopy.

Medin is predicted by TANGO to have three amyloidogenic regions (Figure 1).²⁵ Westermark et al. studied the medin(1–25), medin(32–41), and medin(42–49) peptides. The researchers only observed the formation of fibrils from the latter two peptides by electron microscopy, and concluded that the amyloid forming motif of medin lies in its C-terminus. Gazit et al. also observed the formation of fibrils by the medin(42–49) peptide by electron microscopy.²⁶ Middleton et al. subsequently used solid-state NMR spectroscopy and X-ray fiber diffraction to establish that the fibrils from medin(42–49) consist of parallel, in-register β-sheets that assemble in a face-to-back manner.²⁷ To our knowledge, no additional structural information regarding the assembly of medin(32–41) has been elucidated.

Results

To gain further insights into the supramolecular assembly of medin, we set out to study the two β-strand regions in the medin monomer predicted to fold as a β-hairpin. To synthesize the β-hairpin mimic peptide 1, we connected two heptapeptide strands using δ-linked ornithine (^δOrn) turn units (blue) to form a macrocycle.^28,29 The two heptapeptide strands are derived from medin(19–25) DQWLQVD (top strand) and medin(31–37) EVTGIIT (bottom strand). N-Methylation (red) of the backbone was employed to attenuate uncontrolled aggregation.³⁰ This strategy to make β-hairpin mimics has been successfully used by our lab in the past to identify other amyloid assemblies using X-ray crystallography.³¹

We used X-ray crystallography to study the structure and assembly of peptide 1. We began our crystallization efforts by screening peptide 1 in 576 conditions in a 96-well plate format using crystallization kits from Hampton Research. Cube-shaped crystals (Figure S4) grew in abundance in drops containing sodium acetate, calcium chloride, and 2-methyl-2,4-pentanediol. Further optimization of the crystallization conditions afforded monocrystals suitable for X-ray diffraction (optimized conditions: 0.1 M NaOAc, 0.02 M CaCl₂, 30% MPD). The X-ray diffraction experiments were performed at 1.54 Å using an X-ray diffractometer. Despite the relatively high resolution of the acquired dataset (ca. 1.32 Å), we were unable to solve the structure by single wavelength anomalous diffraction (SAD). In an attempt to increase the magnitude of anomalous scattering, we crystallized peptide 1 in the optimized conditions substituting barium chloride for calcium chloride. The modified conditions afforded similar cube-shaped crystals that gave rise to enough measurable anomalous signal to facilitate SAD phasing and subsequent solution of the original dataset by isomorphous replacement. The structure was thus solved in the cubic I23 space group and refined to a final R_work of 0.1772 (PDB 7JRH).

The X-ray crystallographic structure of peptide 1 reveals a three-dimensional network of large interpenetrating cubes, ca. 4.3 nm in size (Figure 2). Each cube is composed of molecules of peptide 1 located at the edges and coordinated to calcium ions at the vertices. In the structure, peptide 1 folds to form a hydrogen-bonded β-sheet. The elongated β-strand conformation of the top strand places the two aspartic acid residues (D₁₉ and D₂₅) at the opposite ends of the β-sheet, ca. 2.0 nm apart. The side chain carboxylate group of D₁₉ binds one calcium ion, and the side chain of D₂₅ binds another calcium ion. Each of the ions in turn binds either D₁₉ or D₂₅ residues of two other peptide molecules. This mode of coordination, in conjunction with the symmetry present in the crystal lattice, results in the formation of an assembly containing eight calcium ions and twelve molecules of peptide 1. Even though the arrangement of edges and vertices resembles a cube (Figure 2A and B), the point symmetry is actually tetrahedral, with four of the vertices arising from the coordination of D₁₉ and the other four vertices arising from the coordination of D₂₅.

Figure 2.

Three-dimensional assembly observed in the X-ray crystallographic structure of peptide 1. Individual cube (A), infinite array of cubes (C), and two interpenetrating arrays (E), along with their respective schematic representations (B, D, and F). The cubes are all symmetry-equivalent, and the coloring scheme has been chosen arbitrarily for clarity.

The cubes stack face-to-face in the crystal lattice in all three directions, forming an infinite 3-dimensional array (Figure 2C and D). Consistent with the body-centered lattice type, the complete crystal lattice contains two symmetry-equivalent mutually-interpenetrating arrays, related through a [0.5, 0.5, 0.5] Bravais translation vector (Figure 2E and F).³² Such a polycatenated structure belongs to the bcu-y class in knot theory — a 3D ‘chainmail’of linked cubes.³³

The packing of the cubes, as shown in Figure 2C and D, is stabilized by hydrogen bonding and hydrophobic interactions between molecules of peptide 1 located at the edges of the cubes. The four molecules of peptide 1 that constitute the edges of every four adjacent cubes form a tetramer (Figure 3A). The tetramer can be interpreted as a dimer of dimers, in which each dimer is stabilized by six intermolecular hydrogen bonds involving side chains and main chains of T₃₇, Q₂₃, and I₃₅ (Figure 3B). The dimers stack to form the tetramer. Packing of the hydrophobic side chains of I₃₅, V₃₂, and L₂₂ — a total of twelve residues, three per monomer — help stabilize the tetramer. Eight hydrogen bonds between Q₂₀ and V₃₂ further stabilize the assembly.

Figure 3.

(A) The location of a tetramer in within the cubic assembly of peptide 1. (B) Hydrogen-bonded dimer within the tetramer.

The monomeric β-hairpins that constitute the building blocks of the crystal lattice have an unexpected folding pattern. Peptide 1 was designed to contain two hydrogen-bonded β-strands — DQWLQVD (top strand) and EVTGIIT (bottom strand) — linked by two ^δOrn turn units. In the X-ray crystallographic structure, the top and bottom strands do indeed form a hydrogen-bonded β-sheet, but not precisely as designed (Figure 4). Residues T₃₃ and G₃₄ in the bottom strand form a β-bulge,³⁴ and six of the seven residues of the top strand (Q₂₀ through D₂₅) participate in β-sheet formation. The β-bulge causes a shift in the registration of the strands, which brings D₁₉ into an extended loop with ^δOrn and causes T₃₇ to pair with Q₂₀ instead of D₁₉ (Figure 4B).

Figure 4.

β-Bulge in the X-ray crystallographic structure of peptide 1. (A) X-ray crystallographic structure of the monomeric β-hairpin subunit of the crystal lattice. (B) Hydrogen bonding pattern observed in the X-ray structure.

Discussion

Coordination of Ca²⁺ by carboxylates in the peptide 1 structure is reminiscent of calcium binding in structures ranging from EF-hand proteins to metal-organic frameworks (MOFs). The D₂₅-calcium binding site of peptide 1 contains three aspartate carboxylates and three water molecules in an octahedral arrangement (Figure 5A), while the D₁₉-calcium binding contains three aspartate carboxylates and one water molecule (Figure 5B). EF-hands are a class of calcium-binding proteins featuring a helix-loop-helix motif, containing multiple carboxylates and other ligands that bind Ca²⁺. Calmodulin is a prominent example of this class of calcium-binding proteins, employing three aspartates to bind Ca²⁺, in a fashion similar to the binding mode within the peptide 1 cubes (binding of Figure 5C).³⁵ Only a handful of calcium-based metal-organic frameworks have been reported, including MOFs suitable for removal of heavy metals from water³⁶ and slow release of pesticides,³⁷ far fewer than the number of MOFs containing transition metals.³⁸ Calcium-based materials, such as alginate gels, are of interest for applications in the foodstuffs and medicines, where only non-toxic metals are permissible.^39,40 We anticipate that calcium-based MOFs may also have potential applications in these areas.

Figure 5.

(A and B) Calcium binding geometry observed in the X-ray crystallographic structure of peptide 1. (C) Calcium binding site in the calmodulin protein (PDB 1CLL). Calcium ions are shown as yellow spheres and water molecules as red spheres.

The non-covalent interactions in the crystal structure of peptide 1 are diverse: metal coordination, hydrogen bonding, and hydrophobic packing. Although it is not possible to assess the relative contributions of each of these interactions in stabilizing the assembly, metal coordination is critical. We have not observed growth of any stable crystals in the absence of divalent metal ions (Ca²⁺, Ba²⁺, or Co²⁺). The interpenetrating cubes formed by peptide 1 and Ca²⁺ can thus be considered a calcium-based MOF with peptide ligands, and not merely an artificial calcium receptor.

The interpenetrating cubes cannot form separately, but rather must form simultaneously from tetramers, dimers, single molecules, or other assemblies of peptide 1 during the crystallization process. For this reason, the lattice may also be interpreted as a result of the assembly of tetramers, with the tetramers acting as ligands bearing eight carboxylate groups. Regardless of the process by which the lattice assembles, it can ultimately be viewed as an assembly of cubes with Ca²⁺ at the vertices, because of the crucial role metal cations play in the assembly.

Intrinsic chirality and biocompatibility have allowed peptides to act as ligands in a number of MOFs possessing remarkable properties. These structures include MOFs exhibiting selective binding of chiral molecules,⁴¹ reversible conformational changes upon binding,^42,43 and high mechanical stability.⁴⁴ The relative flexibility of peptide chains, which imparts these interesting properties, is also a major limiting factor. In longer peptide chains, it is almost impossible to predict and control secondary structure. A notable example is a pentapeptide-cadmium complex reported by Yaghi et al. that exhibits no porosity.⁴⁵ In the case of the framework structure formed by peptide 1, the greatly reduced conformational flexibility of the cyclic β-sheet peptide is sufficient to allow the formation of a well-defined assembly. The preorganized spatial arrangement of the two aspartate side chains allows the peptide to behave like a rigid ligand. Thus, a design principle emerges for the construction of other peptide-based MOFs, whereby a preference for a particular secondary structure allows preorganization of metal-binding ligands (e.g., Glu, Asp, or His).

It is important to note that larger peptides with well-defined secondary structures have previously been used with metals to construct various nanostructures. Notable examples include α-helix bundles^46,47,48,49 and collagen triple helices.⁵⁰ Many of these structures rely upon the incorporation of non-natural ligands — most often various pyridine derivatives — into the peptide sequences. The three-dimensional framework formed by peptide 1 is distinctive, because the metal binding involves side chains of unmodified aspartic acid residues. We envision that it will be possible to construct other large calcium-containing MOFs from peptides with well-defined secondary structures that display Glu or Asp side chains. We further envision that this design principle might be extended to other metals and amino acid side chains such as His or Cys.

Conclusion

The assembly formed by peptide 1 in the presence of divalent metal ions constitutes a rare example of a metal-organic framework in which a large polypeptide acts as a ligand. The MOF formed by peptide 1 is among the largest peptide-based MOFs. While the structure exhibits interpenetration, and thus does not contain solvent-accessible pores, it demonstrates important rules for the construction of peptide-based MOFs. The peptide’s cyclic β-sheet design exemplifies one such emergent principle, in which a preference for a particular secondary structure and supramolecular assembly allows the preorganization of metal-binding sites. We envision that different peptides with well-defined secondary structures, bearing multiple metal-binding residues may — in conjunction with metal ions — result in MOFs. Amyloidogenic peptides are particularly suited for this purpose, since they are predisposed toward supramolecular assembly. Amyloidogenic peptides other than medin might also serve as ligands for MOF structures if they contain, or are altered to contain, suitable metal-binding residues.