Sequence-encoded hexagonal lattices in multichannel peptide nanofibrils

Biological materials derive functional complexity from molecular information that encodes their three-dimensional structure and hierarchical organization across length scales. Polypeptides offer a rich chemical design space^13,14,15, and β-sheet-rich amyloid fibrils represent one of their most prevalent supramolecular states, with roles ranging from hormone storage to pathological aggregation^16,17,18. Recent advances in cryo-electron microscopy (cryo-EM) have revealed these assemblies at high resolution, resolving their atomic structures in the hydrated state^17,18, and disclosing the packing and polymorphism of cross-β-motifs.

Most peptide assemblies elongate as one-dimensional fibrils whose lateral interfaces are polymorphic and therefore difficult to design, preventing controlled tiling into higher-dimensional lattices (Extended Data Fig. 1a). Programmable hierarchical growth has been achieved in coiled-coil proteins^{19,20,21,22,23,24,25,26} and collagen-mimetic systems through large pre-organized interfaces^27,28, whereas peptide nanochannels formed from cyclic scaffolds rely on covalent backbone pre-organization^{29,30,31,32,33,34}. By contrast, controlled two- or three-dimensional lattice formation from minimal linear β-sheet peptides is rare^{5,6,7,8,9,10,11,12}. Consequently, a geometry-guided framework for programmable multichannel architectures from short linear β-sheet sequences has remained elusive.

Here we show that amphiphilic nonapeptides encode discrete interaction motifs that direct the formation of hexagonal pores, which hierarchically tile into laterally expandable multichannel nanofibrils: a cross-β-dimer that mediates lateral and axial growth, a trimeric junction that encodes lateral connectivity and symmetry, and a central inversion residue that orients the two motifs (Fig. 1a) to create complementary interfaces that couple lateral organization to axial stacking. Cryo-EM, sequence perturbation, molecular dynamics simulations and vibrational spectroscopy reveal how these motifs couple axial stacking, lateral tiling and sequence-dependent channel hydration. Together, these results show that minimal sequence-encoded interaction motifs can programme supramolecular symmetry and hierarchical lattice growth, enabling multichannel peptide nanofibrils from short linear sequences.

Cryo-EM resolves the encoded interaction geometry as a hexagonal β-sheet lattice containing a continuous central nanochannel. DILT1 forms highly ordered fibrils that adopt two principal structural isomorphs: a single-channel hexagonal architecture (1-1-1; 12% relative abundance) and a laterally expanded honeycomb lattice (3-3-3; 73% relative abundance) (Fig. 2a and Extended Data Fig. 2a,b). The three-digit notation denotes the number of hexagonal pores in each of the three lattice sectors visible in the cross-section. Helical reconstruction resolved the 1-1-1 and 3-3-3 structures to 4.0 Å and 2.9 Å, respectively, revealing a repeating unit composed of 12 peptide chains arranged as 6 face-to-face packed cross-β-dimers around a central cavity (Fig. 2b–e and Extended Data Fig. 2c–g). In the plane perpendicular to the fibril axis, this unit exhibits C₆ symmetry and encloses an approximately 5-nm central cavity that propagates continuously along the fibril. Successive layers therefore preserve pore geometry along the fibril axis, defining a continuous hexagonal channel. In the single-channel morphology (1-1-1), the peripheral density connects adjacent dimers through partially resolved triangular features consistent with predicted trimeric junctions. In the predominant honeycomb architecture, these junctions are fully resolved and mediate symmetry and lateral growth at the lattice perimeter (chains I–VI, Fig. 2b). This lateral tiling yields 7 hexagons per layer and a total of 60 peptide chains in a single cross-section.

The clear axial peptide stacking at 4.7–4.8 Å and well-resolved electron density in the higher-resolution reconstruction enable atomic modelling of the complete lattice and assignment of the principal dimer and junction interfaces. The atomic model of the 3-3-3 fibril yields a diameter of 210 Å, a helical rise of 4.77 Å and a twist of −0.64°. Application of C₆ symmetry reduces the asymmetric unit to ten unique chains per layer, three of which were built directly into the density and propagated by symmetry to generate the full architecture (Extended Data Fig. 2g). The peripheral regions show weaker electron density, indicating local dynamics or incompletely formed junctions, which account for the semi-triangular features observed in the 1-1-1 morphology. The absence of the third junction-forming chain reflects insufficiently defined cryo-EM density to model its sequence, although the residual density is consistent with its presence. Thus, the 1-1-1 and 3-3-3 morphologies represent isomorphs of the same underlying lattice geometry that differ in the extent of lateral expansion, rather than distinct assembly modes. Helical torsion decreases with fibril diameter: fibrils containing more hexagons show larger crossover distances and smaller twist angles (Extended Data Fig. 2b).

To define the interaction network associated with the hexagonal lattice, we performed molecular dynamics simulations based on the cryo-EM structure of the DILT1 3-3-3 architecture. The trajectories reveal a cooperative and spatially constrained set of hydrophobic and hydrogen-bonding interactions that couple axial propagation to lateral connectivity and thereby maintain the lattice geometry over the simulated timescale in aqueous buffer solution (Fig. 3a). Within the cross-β-dimer, face-to-face packed chains are stabilized by hydrophobic packing of the V2 and V4 side chains (d-interface; Fig. 3b and Extended Data Fig. 3a), which defines the lateral growth interface. This packing is reinforced by recurrent inter-chain hydrogen bonds. In the fibril plane, hydrogen bonds between Q6 and the N-terminal amine (d(N)-lock) and polar interactions between N8 and K1 (d(K1) lock) provide additional directional stabilization of the lateral registry. At the junction, three peptide chains form a compact triangular hub in which interdigitated I7 and M9 residues create a densely packed hydrophobic core (j-interface; Fig. 3b and Extended Data Fig. 3a), topologically equivalent to a threefold steric-zipper interaction. Persistent attractive interactions between the M9 C-terminus and the side chains of K3 and S5 of a neighbouring chain (j-lock; Fig. 3b and Extended Data Fig. 3b), including hydrogen bonding and electrostatic contacts, further stabilize this connectivity. Along the fibril axis, backbone hydrogen bonding within the cross-β-architecture drives periodic stacking of the hexagonal layers into an extended nanochannel fibril (Fig. 3c and Extended Data Fig. 3c).

Simulations of a 94-layer fibril (44.8 nm in length; total twist 60.16°) retained the experimental lattice geometry over 100 ns (Extended Data Fig. 3d). Root-mean-square fluctuation (RMSF) analysis showed mean backbone fluctuations below 2.5 Å, supporting structural integrity of the model over this timescale (Extended Data Fig. 3e,f). By contrast, isolated cross-β-dimers and their hexagonal assemblies lacking junctions rapidly destabilize, whereas isolated junction motifs and the hexagonal nanochannels remain intact over the simulated timescale (Extended Data Fig. 3g,h), supporting a model in which the junction provides the principal lateral cohesion required for hexagonal channel formation. The enhanced persistence of the junction is consistent with contributions from both the compact hydrophobic j-interface and the three j-locking interactions, which together reinforce lateral cohesion through combined packing and hydrogen-bonding contacts. Mapping these interactions onto the fibril cross-section reveals pronounced amphiphilic partitioning, with buried hydrophobic cores and hydrophilic surfaces lining the pore. This organization is consistent with a solvent-accessible nanochannel in which solvent-exposed polar residues define the channel environment, whereas hydrophobic interfaces stabilize the lattice. Thus, axial propagation, lateral junction formation, locking interactions and backbone hydrogen bonding form a coupled interaction hierarchy that supports an axially and laterally expandable six-fold-symmetric architecture in aqueous solution.

To test whether lattice topology is encoded in the peptide sequence, we determined the structures of the arginine variants DILT2 (RVKVSQINM) and DILT3 (KVRVSQINM) using cryo-EM. Both assemble into hexagonal nanochannel architectures with conserved pore geometry and junction connectivity, showing that the lattice organization is preserved across these sequence variants (Fig. 4a). DILT2 predominantly forms a 4-4-4 isomorphology (approximately 80%), with minor 2-2-2 and 1-1-1 populations (approximately 10% each). The first two isomorphologies (4-4-4 and 2-2-2) were reconstructed at 1.9 Å and 3.4 Å resolution, respectively (Fig. 4b and Extended Data Fig. 4a,b). The helical rise and twist were determined to be 4.781 Å and −0.409°, and 4.812 Å and −8.2°, respectively. The symmetry centre depends on the parity of the hexagon arrangement. For odd-numbered hexagon arrangements (for example, 3-3-3), the symmetry axis is located in the centre of the central pore, yielding C₆ symmetry, whereas for even-numbered morphologies (for example, 4-4-4), the symmetry centre is located in the central junction, yielding C₃ symmetry. DILT3 adopts mainly the 3-3-3 morphology (approximately 70%) and was resolved at 1.78 Å resolution, with a helical rise of 4.791 Å and a twist of −0.6115°. At this resolution, side-chain conformations within both the cross-β-dimers and the junction interfaces are directly visualized.

Across all variants, the hexagonal lattice, the trimeric junction and the approximately 5-nm nanochannel remain invariant (Fig. 4b and Extended Data Fig. 4c), showing that global topology emerges from the encoded interaction geometry. Thus, sequence perturbations in these cases modulate local packing without altering the underlying symmetry or pore architecture.

The cryo-EM structures and the molecular-dynamics-derived interaction hierarchy are consistent with lateral growth into additional hexagonal units while maintaining the central pore. Cryo-EM reconstructions support this model by visualizing a family of geometrically related isomorphs that differ in the number of laterally connected hexagons (Fig. 4a,b and Extended Data Fig. 4d–f). These range from single-channel fibrils (1-1-1) to multichannel architectures containing up to 14 pores (4-4-5), comprising several hundred peptide chains in a single cross-section. In all reconstructed cases, the cross-β-registry and pore diameter remain constant, supporting lateral growth by integer tiling of a conserved hexagonal unit rather than by extensive polymorphic reorganization. Simulated projections derived from the atomic model reproduce the intermittent stripe pattern observed in the cryo-EM micrographs. This supports assignment of the fibril pattern to the hexagonal nanochannel architecture of the DILT family. The number of connected hexagons correlates with the characteristic intermittent striped pattern in the cryo-EM images (Fig. 4a,c and Extended Data Fig. 4d–f), providing a morphological fingerprint of the lattice organization. Because this relationship is conserved across DILT1–3, the stripe pattern provides a diagnostic morphological signature that can be used to assess whether mutations preserve hexagonal symmetry.

The related peptides DILT4 (GVKVSQINM) and DILT5 (YVKVSQINM) show the same intermittent stripe pattern, indicating conservation of the hexagonal lattice despite non-conservative substitution at the N-terminal position (Fig. 4d). The sequences of DILT2, DILT4 and DILT5 differ from DILT1 and DILT3 only at residue X1, where K1 is replaced by R, G or Y, suggesting this site as symmetry tolerant and involved in modulating surface chemistry rather than lattice formation. Despite the large variation in fibril width among the detected isomorphs of DILT1–DILT3 (approximately 60–250 Å; Extended Data Fig. 2b), β-strand spacing, pore diameter and junction geometry remain invariant, showing that the encoded interaction geometry constrains the supramolecular topology. The hexagonal nanochannel, therefore, acts as a laterally expandable structural module that supports symmetry-defined multichannel peptide fibrils.

To define the sequence constraints required for hexagonal lattice formation under the standard DILT assembly conditions, we systematically perturbed residues within the dimer, junction and inversion motifs. All variants were screened under matched concentration and buffer conditions, allowing direct comparison of their assembly behaviour within this defined regime. We evaluated the structural outcome by cryo-EM, analysis of the characteristic stripe pattern and molecular dynamics simulations. This perturbation analysis yields experimentally validated sequence–structure rules for hexagonal nanochannel assembly (Fig. 5a). Conservative substitution at the N-terminal X₁ residue (DILT2) and variants carrying non-conservative substitutions at the N-terminus (DILT4 and DILT5) preserved hexagonal symmetry and stripe pattern organization, implying that this position is able to tolerate sequence variation without detectable loss of lattice organization. In molecular dynamics trajectories, the d-locks involving N8 and K1 are transient in DILT1 and can be replaced by the outwards projection of the K1 side-chain into the nanochannel, indicating that individual d-lock interactions contribute to stabilization but are not required for maintenance of global topology (Figs. 3 and 5a and Extended Data Fig. 3d). Consistent with this interpretation, non-conservative substitutions at position 1 (G in DILT4; Y in DILT5) abolish this d-lock interaction yet retain honeycomb architectures, identifying the N-terminus as a symmetry-tolerant site that modulates surface chemistry. Atomic force microscopy further reveals the grooved surface topology expected for multichannel fibrils and well-defined fibril boundaries, consistent with the presence of laterally expanded nanochannel architectures (Fig. 5b and Extended Data Fig. 5b).

In contrast, perturbation of the hydrophobic dimer interface did not yield detectable hexagonal assembly under these matched screening conditions. Alanine or leucine substitution (P1, A2/A4; P2, L2/L4, Supplementary Fig. 2) alters V2/V4-mediated packing and cross-β pairing, and results in loss of the stripe pattern under these conditions, indicating that the hydrophobic dimer interface is required for the formation of the ordered hexagonal lattice in the DILT assembly regime examined here. Moreover, extension of the amphiphilic dimer motif (P3, KVKVKVSQINM) yields fibrils lacking the stripe pattern, consistent with the loss of hexagonal lattice formation under the tested conditions. Charge inversion at the X₃ position (P4, E1/E3) is expected to perturb junction stabilization and suppress lateral fusion, resulting in fibrils that likewise lack the characteristic intermittent stripe pattern. The inversion residue is positioned at the junction between the dimer-forming and trimer-forming segments and disrupts the otherwise regular amphiphilic alternation, consistent with a role in coupling the two interfaces geometrically, and participates in j-lock formation. Accordingly, mutations that alter this residue (P5, S5G; P6, S5T) produce fibrils with crossover but lacking stripe patterns, indicating loss of hexagonal organization. Similarly, perturbation of the trimeric QINM junction motif, including geometric reversal of hydrophobic (P7, M7/I9) or hydrophilic side chains (P8, N6/Q8), does not produce detectable honeycomb formation under the matched screening conditions and yields predominantly non-striped fibrils, underscoring the requirement for precise spatial positioning of residues within the trimeric interface (Fig. 5c and Extended Data Fig. 6a,b).

Together, these results define a hierarchical set of sequence constraints for hexagonal lattice formation under the matched assembly conditions used here (Extended Data Fig. 6c). The N-terminal position is symmetry tolerant and primarily modulates surface chemistry, whereas the hydrophobic dimer interface, the inversion residue and the trimeric junction are required to preserve the cooperative interactions that support cross-β-registry, lateral tiling and honeycomb topology in this assembly regime. These comparative data do not constitute an exhaustive phase diagram for each variant, and future studies may reveal additional morphologies under different buffer, ionic-strength or dilution conditions.

Under the tested aqueous buffer conditions, the self-assembly of these elements to form micrometre-long honeycomb fibrils is initiated within minutes. After overnight incubation, the formed honeycomb fibrils remain structurally intact in solution and these assemblies tolerate dilution, heating to 95 °C and sonication, indicating cooperative stabilization by the same cross-β- and junction interfaces that define the hexagonal nanochannel topology at the molecular scale (Extended Data Fig. 7). Fibrils are observed at concentrations above approximately 48 µM when incubated overnight, as shown by a concentration-dependent analysis of DILT1 by transmission electron microscopy (TEM) together with a fluorescence-based aggregation assay (Extended Data Fig. 8a,b). The DILT1 fibrils form self-supporting hydrogels, as demonstrated by vial inversion and mechanical property characterization, through reversible physical cross-links most likely between solvent-exposed peripheral interfaces (Extended Data Fig. 8c–e). Although the detailed pathway linking nanoscale integer tiling in DILT1 and interfibril interactions to macroscopic network formation remains to be resolved, these data link concentration-dependent fibril formation and network connectivity to macroscopic gelation.

The fibrillar lattice contains a continuous solvent-accessible nanochannel whose physicochemical properties are modulated by the amino acid exposed to the pore interior. Molecular dynamics simulations reveal a water-filled channel with elevated water density relative to bulk solvent and reduced diffusional mobility along the radial axis (Fig. 6a–d, Extended Data Fig. 9a and Supplementary Fig. 3). To probe pore hydration experimentally, Fourier-transform infrared (FTIR) spectroscopy was performed during controlled dehydration at defined water contents (Fig. 6e and Extended Data Fig. 9b–g). Normalized FTIR spectra after subtraction of the corresponding fully dry peptide fibril spectrum highlight the water-related O–H stretching response during dehydration. In DILT1, progressive dehydration leads to a broadened and redshifted normalized O–H stretching envelope, where a low-wavenumber component distinct from bulk liquid water becomes increasingly pronounced (Fig. 6e), consistent with an increasingly heterogeneous and more strongly hydrogen-bonded water population during drying. Because the O–H stretching region overlaps with peptide N–H contributions and contains multiple water populations, we analysed the dehydration series by deconvoluting the 2,700–3,800 cm⁻¹ region into one narrow N–H component near 3,300 cm⁻¹ and three broad O–H components centred at approximately 3,000 cm⁻¹, 3,400 cm⁻¹ and 3,600 cm⁻¹ (refs. ^39,40; Extended Data Fig. 9e,f).

During DILT1 drying, the total integrated O–H response decreases continuously, whereas the higher-wavenumber O–H components (approximately 3,400 cm⁻¹ and approximately 3,600 cm⁻¹) assigned to more weakly hydrogen-bonded or mobile (bulk-like) water decay more rapidly than the low-wavenumber component near 3,000 cm⁻¹ assigned to strongly hydrogen-bonded and more confined water. Consequently, the relative contribution of the strongly hydrogen-bonded water population increases during drying and dominates at later time points before being progressively depleted. These data support preferential removal of more weakly associated external or mobile water molecules, followed by retention and gradual loss of more strongly hydrogen-bonded confined water, rather than abrupt channel emptying. By contrast, the cross-β-control-peptide CKFKFQF^41,42, which forms β-sheets but lacks a nanochannel topology, shows a progressive reduction in the overall O–H stretching response under dry nitrogen while the bandwidth remains essentially unchanged, consistent with uniform water loss rather than confinement-dependent dehydration.

For DILT1, a semi-quantitative estimate based on deconvolution-derived N–H/O–H integrated band areas and literature infrared absorption cross-sections, as described in Methods, yields approximately 25 confined water molecules per peptide at around 120 min in the drying stage when the relative fraction of strongly bound water reaches a plateau (Fig. 6e and Extended Data Fig. 9f), consistent with a fully filled nanochannel. This estimate is inherently approximate and model dependent as it relies on environment-sensitive absorption coefficients, separation of overlapping N–H and O–H bands, and assumptions about the number of peptide N–H oscillators contributing to the spectrum.

For DILT2, DILT4 (Extended Data Fig. 9c,d) and DILT5 (Fig. 6e), the presence of confined water is likewise inferred based on the observations made for DILT1. Importantly, the O–H stretching envelopes (Fig. 6e and Extended Data Fig. 9e,f) indicate that sequence variation at the pore-lining position can tune the local hydrogen-bonding state of the confined water in both directions: K1 and R1 substitutions shift the band to a lower wavenumber, consistent with enhanced hydrogen bonding, whereas G1 shows no discernible shift, suggesting hydrogen bonding comparable to that of bulk water. For DILT5 containing the more lipophilic Y1 residue, we do not observe the characteristic dehydration profile in the O–H stretching region, suggesting that the water environment in DILT5 differs from that in the other samples, which may reflect differences in peptide–water interactions arising from variations in local electrostatic environments. DILT1 (K1), DILT2 (R1), DILT4 (G1) and DILT5 (Y1), therefore, illustrate sequence-controlled modulation of nanochannel polarity while preserving the hexagonal lattice. In particular, replacing K1 with R1 (DILT2; Extended Data Fig. 9c) preserves a highly polar channel environment, whereas replacing it with G1 (DILT4; Extended Data Fig. 9d) or Y1 (DILT5; Fig. 6e) reduces the effective polarity of the pore surface. We next examined whether dehydration detectably alters the peptide architecture. Raman spectra recorded during drying show no detectable frequency shifts of the peptide C–H or N–H modes between hydrated and powder-like states, although the overall intensities change as water is removed (Extended Data Fig. 9g). The FTIR spectra likewise show that the peptide N–H stretch near 3,300 cm⁻¹ remains essentially unchanged in frequency during dehydration. In addition, electron microscopy of samples dried for around 120 min under the same conditions used for FTIR measurements retains fibrillar structures with features consistent with the characteristic hexagonal fibril fingerprint (Extended Data Fig. 9h). These observations indicate that dehydration does not cause detectable collapse of the peptide architecture within the sensitivity of the vibrational and electron-microscopy measurements, although a full diffraction-based comparison of hydrated and dehydrated states will be required to quantify changes in crystallinity. Together, molecular dynamics simulations and vibrational spectroscopy show that the pores are not empty geometric cavities, but solvent-accessible nanochannels whose confined water can be tuned between more strongly and more weakly hydrogen-bonded states by sequence variation at a single pore-lining residue.

We demonstrate that a minimal linear peptide sequence can encode a symmetry-defined supramolecular architecture through a hierarchy of complementary interaction motifs. By integrating a cross-β-dimer interface, a trimeric junction, an inversion element and locking interactions within a nine-residue scaffold, amphiphilic peptides assemble into laterally expandable honeycomb lattices containing continuous hexagonal nanochannels. This architecture emerges directly from sequence-encoded geometry rather than from pre-organized cyclic scaffolds, folded protein domains or metal coordination, showing that programmable higher-order organization can arise from minimal β-sheet building blocks. Systematic sequence perturbation identifies symmetry-critical and symmetry-tolerant positions, establishing design rules for laterally expandable β-sheet multichannel nanofibrils.

Unlike conventional amyloid fibrils, which predominantly elongate through repetitive stacking and achieve macroscopic cohesion through polymorphic entanglement, the assemblies reported here expand laterally by integer tiling of a conserved hexagonal unit. This mechanism generates multichannel fibrils with conserved pore geometry across sequence variants within the set experimental conditions and links molecular sequence to mesoscale architecture. Although the present data do not resolve the exact growth pathway, the hexagonal honeycomb nanochannels probably form through a nucleation-and-growth process in which initial assembly of a hexagonal unit is coupled to subsequent lateral tiling and axial propagation. Future studies will be needed to define the precise sequence and relative contributions of these growth steps, involving the role of buffer ions, concentrations, kinetic and thermodynamic factors, as well as environmental parameters. The resulting pores are solvent accessible and show sequence-dependent hydration, indicating that molecular sequence can tune not only the supramolecular topology but also the physicochemical environment within confined spaces. The same interfaces that define nanoscale symmetry also support reversible interfibrillar association and hydrogel formation, although the detailed pathway from integer tiling to macroscopic network formation remains unresolved. More broadly, these results establish short linear β-sheet peptides as modular units for programmable, symmetry-defined lateral growth, opening opportunities for multichannel biomolecular materials with tunable confined solvation and transport-related properties.

Materials

Chemicals were purchased from commercial suppliers (Thermo Scientific, Merck KGaA, VWR and Honeywell). Solvents and reagents used for peptide synthesis were obtained in peptide grade. Solvents used for peptide purification were of high-performance liquid chromatography (HPLC) grade; dimethyl sulfoxide (DMSO) was BioUltra grade for molecular biology (≥99.5%, Sigma Aldrich); and water was obtained from a Millipore purification system. Dulbecco’s phosphate buffered saline (DPBS; 1x, without CaCl₂ and MgCl₂, composition: KCl 0.2 g l ⁻¹, KH₂PO₄ 0.2 g l ⁻¹, NaCl 8.0 g l ⁻¹, Na₂HPO₄ (anhydrous) 1.15 g l ⁻¹) was purchased from Sigma Aldrich (DPBS) or from Thermo Fisher Scientific (Dulbecco’s balanced salt solution) with minor variations in inorganic salt water content. DILT1 was in part purchased from GenScript Biotech with a purity of ≥98%. DILT2-5 and peptides P1 and P3–P8 were purchased from GenScript Biotech with a purity of ≥98%.

Solid-phase peptide synthesis of DILT1 (KVKVSQINM) and P2 (KLKLSQINM)

For peptide synthesis, the fluorenylmethyloxycarbonyl (Fmoc) solid-phase peptide synthesis strategy was applied from the C-terminus to the N-terminus using N,N′-diisopropylcarbodiimide (DIC) and ethyl cyano(hydroxyimino)acetate (Oxyma) on an automated microwave-assisted peptide synthesizer (CEM, Liberty Blue). The peptides were synthesized onto pre-swollen Fmoc-Met-Wang resin (0.1 mmol, 0.147 g, substitution 0.68 mmol g⁻¹, 100–200 mesh, pre-swollen in dimethylformamide (DMF; 2–3 ml) on a shaker at room temperature) by sequential coupling of activated Nα-Fmoc-amino acid in DMF (2.5 ml, 0.2 M) in the presence of DIC (1 ml, 0.5 M) and Oxyma (0.5 ml, 1.0 M) in DMF via microwave-assisted reaction at 75 °C (170 W) for 15 s and 90 °C (30 W) for 110 s followed by multiple washing of the resin (DMF, 2 × 3 ml, 1 × 4 ml). Sequentially, for DILT1 Fmoc-Asn(Trt)-OH, Fmoc-Ile-OH, Fmoc-Gln(Trt)-OH, Fmoc-Ser(tBu)-OH, Fmoc-Val-OH, Fmoc-Lys(Boc)-OH, Fmoc-Val-OH and Fmoc-Lys(Boc)-OH were coupled to the resin via cycling coupling with each amino acid introduced in one sequential coupling cycle, washing, and Nα-Fmoc-deprotection steps. For P2, Fmoc-Leu-OH was used instead of Fmoc-Val-OH. Nα-Fmoc deprotection was conducted by application of piperidine (3 ml, 20% v/v in DMF) under heating (75 °C, 155 W, 15 s, followed by 90 °C, 30 W, 50 s), followed by washing with DMF (2 × 2 ml, 1 × 3 ml). The synthesized peptides on resin were kept in DMF at 4 °C until further use. Subsequently, the peptides were cleaved from the resin using a cleavage cocktail (10 ml) comprising trifluoroacetic acid (TFA) (95% v/v), Milli-Q water (2.5% v/v) and triisopropylsilane (TIPS) (2.5% v/v). After 2 h incubation on a shaker at room temperature, the peptides were precipitated with diethyl ether pre-cooled to 4 °C (40 ml) and centrifuged at 4,000 rpm for 16 min (×2, 0 °C). The resultant crude products were dried overnight at room temperature. The crude products were purified by preparative reverse-phase HPLC using a 0.1% TFA water–acetonitrile mixture as eluant and lyophilized to yield the peptide as a white amorphous powder. The purified peptides were characterized by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS). The lyophilized solids were stored at −20 °C until use.

MALDI-TOF MS (DILT1): m/z_calc = 1,045.60, m/z_found: 1,046 [M + H]⁺, 1,062 [M + O + H]⁺, 1,068 [M+Na]⁺, 1,084 [M + K]⁺, 1,100, 1,106.

MALDI-TOF MS (P2): m/z_calc = 1,073.63, m/z_found: 1,074 [M + H]⁺, 1,090 [M + O + H]⁺, 1,096 [M+Na]⁺, 1,112 [M + K]⁺, 1,128, 1,134, 1,150, 568.

High-performance liquid chromatography

The synthesized crude peptide was purified via preparative HPLC of the dissolved and filtered (0.22 µm) precipitate (Shimadzu system). Multiple manual injections (2× or 3 × 8 mL;) of the peptide crude solution (c = 1.3 mg ml⁻¹, water with 0.1% TFA (v/v) and 5% acetonitrile (ACN) (v/v)) were performed. Separation was performed by a preparative reverse-phase HPLC column (Phenomenex 5 µm C18, 150 × 30 mm) using a 0.1% TFA water–ACN mixture as eluant (flow 25 ml min⁻¹), the applied gradient was as follows: (1) 0.01 min: 5% ACN, 0.1% TFA (A) with 95% 0.1% TFA in water (B); (2) 5 min: 5% A with 95% B; (3) 17 min: 70% A with 30% B; (4) 19 min: 100% A with 0% B; (5) 24 min: 100% A with 0% B; (6) 28 min: 5% A with 95% B; (7) 30 min: 5% A with 95% B. Chromatography was monitored with an ultraviolet absorption detector at 214 nm. Alternatively, the gradient can be applied as follows (for example, P2); (1) 0.01 min: 0% ACN, 0.1% TFA (A) with 100% 0.1% TFA in water (B); (2) 5 min: 0% A with 100% B; (3) 15 min: 40% A with 60% B; (4) 17 min: 100% A with 0% B; (5) 19 min: 100% A with 0% B; (6) 21 min: 5% A with 95% B; (7) 24 min: 5% A with 95% B. Data were evaluated in LabSolutions software.

MALDI mass spectrometry

Peptide mass spectra were recorded in positive ion reflector mode on a Bruker rapifleX MALDI-TOF/TOF mass spectrometer (Bruker Daltonik, scanning smartbeam 3D 10-kHz neodymium-doped yttrium aluminium garnet (Nd:YAG) laser, wavelength of 355 nm, and 10-bit 5-GHz digitizer, acceleration voltage 20 kV) using α-cyano-4-hydroxycinnamic acid (HCCA) as the matrix. Samples were measured with random walk ionization across the sample spot, and usually 8,000 shots were averaged per spectrum. The sample was prepared by mixing an acidic aqueous analyte solution (0.1% TFA) in equal amounts with the supernatant of a saturated HCCA solution acetonitrile/water (30:70 + 0.1% TFA). Subsequently, this mixture (1 µl) was applied to a stainless steel target and allowed to dry before measurement. Calibration was conducted with sodium adducts of polyethylene glycol in a mass range of 500–3,000 Da. Data were evaluated with mMass software.

MSMS sequencing

The sequence analysis was carried out on a rapifleX MALDI-TOF/TOF mass spectrometer from Bruker Daltonik. The instrument is equipped with a scanning smartbeam 3D 10-kHz Nd:YAG laser at a wavelength of 355 nm and a 10-bit 5-GHz digitizer. The acceleration voltage was set to 20 kV, and the mass spectra were recorded in positive ion MSMS mode. The protonated molecular ion of the peptide at 1,046 Da was selected as precursor with an isolation window of +6 Da and −6 Da. Basic instrument calibration was done with the Bruker peptide mix and the Bruker protein calibration standard I and II in a mass range up to 70 kDa. Before fragment ion analysis, the instrument was manually recalibrated with the exact mass of the precursor ion at 1,046.5418 m/z. Samples were measured with random walk ionization across the sample spot. The laser power was adjusted to provide significant signal intensity for b- and y-series fragments, and 2,000 shots were averaged per spectrum. Sample preparation was done by mixing a 50 mg ml⁻¹ solution of super-DHB matrix in acetonitrile/water (1:1 + 0.1% TFA) with equal amounts of the peptide solution (10 mg ml⁻¹) in acetonitrile/water (1:1 + 0.1% TFA). Finally, 1 µl of the mixture was applied to a stainless steel target and allowed to dry before measurement. The mass spectrometer was controlled by the software Bruker flexControl (Version 4.0) and data analysis was done by Bruker flexAnalysis (Version 4.0).

Peptide nanofibril formation

Nanofibril formation was induced by introducing the pre-dissolved peptide stock solution (10 mg ml⁻¹, DMSO) into DPBS (pH 7.4), yielding a 1 mg ml⁻¹ solution. After brief vortexing, the samples were incubated on a shaker (500 rpm, room temperature) overnight. Further incubation was conducted as indicated without shaking at room temperature or at 37 °C for the indicated periods.

Peptide nanofibril electron microscopy screening

Cryo-EM

For cryo-EM micrograph screening of DILT2–DILT5 and peptides P1 and P3–P8, samples were incubated as described above (1 mg ml⁻¹) overnight (500 rpm, 25 °C), subsequently kept at room temperature without shaking and plunge-frozen after 4 days. Holey-carbon-coated grids (400 mesh C-flat 1.2/1.3) were glow-discharged (PELCO easiGlow glow discharge cleaning system; TED PELLA), 3 µl sample was applied, blotted for 8 s and plunge-frozen in liquid ethane using an Automatic Plunge Freezer EM GP2 (Leica Microsystems) for each sample. Grids were screened on a JEM-2100 transmission electron microscope (Jeol) at 200 kV using a TVIPS F416 camera.

Transmission electron microscopy

P2 (L2/L4) was imaged after overnight incubation via TEM as described previously^42,43. The pre-incubated sample (5 μl, 1 mg ml⁻¹) was placed on a copper grid coated with formvar layer (etched with oxygen plasma). Following a short incubation, excess liquid was removed with filter paper, and the grids were stained with uranyl acetate solution (4% w/v, 2 min) and washed with water. Measurements were performed on a Jeol 1400 electron microscope with 120 kV acceleration voltage.

Cryo-EM sample preparation and data acquisition

For DILT1 analyses, samples were incubated overnight (500 rpm, 22 °C), and further sample incubation was conducted without shaking at 37 °C for the indicated periods. For DILT2 (2-2-2) and DILT3 (2-2-2) cryo-EM reconstruction analyses, samples were incubated overnight (500 rpm, 25 °C), subsequently kept at 37 °C without shaking, and plunge-frozen after 7 days and 1 day, respectively. In case of DILT1 and DILT3, a 4 µl aliquot of the DILT1 fibril sample (from the indicated incubation time points) was applied on a Quantifoil grid (Q R2/2, 400 mesh) that was glow-discharged in a 6:1 oxygen/hydrogen plasma (Diener Nano, Diener Electronic) for 30 s shortly before. The excess solution was blotted for 4 s at 4 °C and >80 % humidity. The specimen was cryo-plunged in liquid ethane using a Vitrobot Mark V (Thermo Fisher Scientific). Subsequent imaging was done using a Titan Krios G4 transmission electron microscope (Thermo Fisher Scientific) at 300 kV, equipped with a Gatan GIF continuum spectrometer. Micrographs and videos were acquired on a 6,912 × 6,912-pixel K3 Gatan direct electron-detection camera with the spectrometer operated in imaging mode using a slit width of 20 eV. For data acquisition, the K3 camera was operated in super resolution and correlated double sampling mode. Extended Data Table 1 lists the data acquisition parameters.

For DILT2 (4-4-4) and DILT3 (3-3-3) cryo-EM reconstruction analyses, samples were incubated overnight (500 rpm, 25 °C), subsequently kept at room temperature without shaking, and plunge-frozen after 15 days for DILT2 and after 8 days for DILT3. For each, 3 µl of sample was applied to a glow-discharged (PELCO easiGlow glow discharge cleaning system; TED PELLA) holey-carbon-coated grid (400 mesh C-flat 1.2/1.3). After blotting with filter paper for 8 s and plunge-freezing in liquid ethane using an Automatic Plunge Freezer EM GP2 (Leica Microsystems), the grids were screened on a JEM-2100 transmission electron microscope (Jeol) at 200 kV, equipped with a TVIPS F416 camera. The dataset for reconstruction was recorded on a Krios G4 transmission electron microscope (Thermo Fisher Scientific) with a Falcon4i (Thermo Fisher Scientific) direct electron detector and a Selectris X energy filter (Thermo Fisher Scientific) operated with a 10-eV slit width. Extended Data Table 1 lists the data acquisition parameters.

Helical reconstruction

In the case of DILT1 fibrils, MotionCor2⁴⁴ was used for movie correction. Contrast transfer function (CTF) for all micrographs was performed using CTFFIND⁴⁵. Further image processing and 3D helical reconstructions were performed with RELION-4.0^46,47 on the 1-1-1 (hexagon fibril) and 3-3-3 (honeycomb fibril) morphologies. For the 3-3-3 morphology, micrographs containing ice or lacking the 4.8-Å-resolution information were excluded. The 3-3-3 morphology filaments were manually picked. As an initial model for the refinement, a featureless cylinder was used with a 60-Å low-pass filter and C₁ symmetry. From the 3D auto-refinement, a C₆ symmetry was observed and imposed, reaching a 4.3-Å-resolution volume with clearly visible β-sheets. To further improve resolution, the full dataset was re-processed via 3D classification and further 3D auto-refine to reach the final 3 Å volume. The volume was then post-processed with a soft-edge mask and sharpened with a B-factor of −50. All 3D classifications were carried out on the central 10% of the reconstruction. For the 1-1-1 morphology, a similar approach was performed. Details of particle extraction and helical reconstruction are reported in Extended Data Table 1.

In the case of the 2-2-2 DILT2 fibrils, CTF for all single images, acquired with drift correction in SerialEM, was performed using CTFFIND⁴⁵. Further image processing and 3D helical reconstructions were performed with RELION 5.0⁴⁶. The 2-2-2 morphology filaments were manually picked. As an initial model for the refinement, a simulated 2-2-2 hexagonal morphology was used. From the 3D auto-refinement, a C₃ symmetry was imposed. The volume was then post-processed with a soft-edge mask and sharpened with a B-factor of −50. All 3D analyses were carried out on the central 10% of the reconstruction.

In the case of DILT2 (4-4-4), helical reconstruction was performed in RELION 5.0⁴⁶. Raw video frames were aligned and corrected for beam-induced motion using RELION’s own implementation of MotionCorr2⁴⁴, followed by CTF estimation using CTFFIND 4.1⁴⁵. Fibrils displaying the 4-4-4 morphology were manually selected from the micrographs. Binned segments were extracted and subjected to reference-free two-dimensional classification until a homogeneous set of particles was obtained. Selected segments were extracted for 3D refinement without binning for 3D reconstruction. Helical refinement was performed using a low-passed cylinder and helical twist estimated from the raw micrographs by manual measurement in Fiji⁴⁸, and a helical rise of 4.75 Å without the imposition of rotational symmetry. For the purpose of map improvement, rotational symmetries C₂, C₃ and C₆ were imposed in the subsequent refinement, of which C₃ symmetry yielded the best quality map as validated by visual inspection. Furthermore, the local search for helical parameters was turned on to optimize the twist and rise of the fibril. After the initial reconstructions, 3D classification and selection of the refined particle stack with progressive increase of the Tau fudge factor were carried out to remedy residual heterogeneity. The resulting particles were subjected to Bayesian polishing. The polished, final stack underwent two cycles of CTF refinement and reconstruction until no further improvement of the reconstructed map was noticeable. The final map was masked with a soft-edged mask and post-processed, yielding a map with a nominal gold-standard resolution of 1.9 Å.

In the case of DILT3 (3-3-3), image processing was performed in cryoSPARC 4.7.1⁴⁹. Raw video frames were aligned and corrected for beam-induced motion using cryoSPARC’s internal patch motion correction. CTF parameter estimation was performed by patch CTF estimation. Exposures were manually curated to exclude micrographs with thick ice and fibril overcrowding. Filament segments were automatically picked using the filament tracer, and segments were extracted with binning for initial classification. Reference-free two-dimensional classification was used to separate segments belonging to different fibril morphologies. Particles corresponding to the 3-3-3 morphology were re-extracted without binning. The resulting stack was subjected to multiple rounds of two-dimensional classification to remove low-quality particles. An initial 3D helical reconstruction was performed using a featureless cylinder without the imposition of helical symmetry. The obtained map was used to bootstrap a 3D helical refinement with enforcement of helical symmetry parameters as obtained by measurement in Fiji⁴⁸. Three-dimensional classification without particle alignment was used to resolve structural heterogeneity. Classes displaying clear separation of fibril layers along the helical axis were selected, and helical symmetry search was enabled during subsequent reconstructions. The homogeneous particle stack was subjected to reference beam motion correction in cryoSPARC and subsequent local and global CTF refinements. Several rounds of local CTF refinement and helical refinement were performed until no further improvement of the 3D map was observable. The final particle stack was symmetry expanded, and a final local refinement with recentring was performed. The resulting half-maps and reconstruction mask were imported into RELION 5.0⁴⁶ where masking and post-processing were performed, yielding a final map with a gold-standard resolution of 1.78 Å. Details of particle extraction and helical reconstruction are reported in Extended Data Table 1.

Atomic model building and refinement

In the case of DILT1, the atomic model for morphology 3-3-3 was built de novo using the program Coot⁵⁰. Given the large number of chains present in the structure, model building was performed stepwise. The fundamental chains of the fibril are termed I.II.III–IV.V.VI–VII.VIII–IX.X (Extended Data Fig. 2g). Initially three chains (I-II-III) were constructed in Chimera⁵¹ by applying the standard β-parallel parameters ((Φ, Ψ) = (–135°, 135°)). These were manually aligned to the junction motif in the 3D volume. Next, to limit atomic clashes in the fibril axis, three layers of such chains were constructed. The three-layer system was refined further in Coot⁵⁰ and Chimera⁵¹. An iterative optimization process was performed to achieve optimal fitting. Next, the structure was further extended to take into consideration the cross-β-motif. Once the I-II-III chains were optimized, they were imposed on the remaining outer electron density. MolProbity⁵², the comprehensive validation tool in Phenix⁵³, was used to produce a validation of the atomic model to assess atomic clashes, rotamer and Ramachandran outliers, and model geometry. The structural statistics for refinement and model building are listed in Extended Data Table 1 and refer to the deposited atomic model (18 chains and 3 layers).

In the case of 2-2-2 DILT2 fibrils, the starting model for refinement was the atomic model of DILT1, and the applied rotational symmetry was C₃. The structural statistics for refinement and model building are listed in Extended Data Table 1 and refer to the deposited atomic models.

In the case of DILT3 fibrils, an initial atomic model was generated de novo using ModelAngelo⁵⁴. An iterative model refinement process was used in which an asymmetric unit of a single layer of the C₆-symmetrical DILT3 fibril cross-sections was refined in WinCoot 1.1.18⁴⁹. Phenix 2.0.5936⁵³ was used to evaluate model geometry and clash score. ChimeraX 1.10⁵⁵ was used to obtain a helical assembly of three layers, which was again validated. The three-layer model was further manually refined in WinCoot 1.1.18⁴⁹ using regularization with Ramachandran, torsion and planar peptide restraints, followed by a subsequent real-space refinement. ISOLDE⁵⁶ was used to improve the clash score. After each model refinement cycle, the contact sites between asymmetric units were checked for steric clashes by symmetry expansion to the original fibril symmetry and renewed validation in Phenix. Final images were produced using Fiji⁴⁸ and ChimeraX⁵⁵.

In the case of DILT2 fibrils, the same procedure was used as for DILT3 fibrils, with the exception of the starting model and the applied symmetry. The starting model for refinement of DILT2 was the atomic model of DILT3, and the applied rotational symmetry was C₃. The structural statistics for refinement and model building are listed in Extended Data Table 1 and refer to the deposited atomic models.

Polymorphism analysis

The CTF-corrected micrographs were used for the polymorphism analysis. A subset of 100 micrographs was randomly generated with the RELION ‘Subset Selection’ function. Each of these micrographs was analysed, and the different morphologies were manually picked and extracted using a box size of 350 pixels, nanochannel diameter of 220 Å, number of asymmetric units 7 and rise of 4.7 Å. The number of extracted particles was used to quantify the different morphologies. The polymorphism prediction was performed using Chimera⁵¹, RELION-4.0⁴⁷ and Fiji⁴⁸. Three layers of the atomic model of the central hexagon obtained from the 3-3-3 morphology were copied and shifted manually to create plausible hexagonal patterns. In Chimera⁵¹, the atomic model was converted into a volume with a 3-Å resolution (molmap #0 3) and saved as an .mrc file. The file was then processed with RELION-4.0⁴⁷ to create a prolonged volume. Specifically, the file was converted into a file with a box size of 1,750 pixels, a nanochannel diameter of 250 Å, a rise of 4.77 Å and a twist of −0.64°. The following commands were used:

relion_image_handler–i 2Hex_221.mrc–new_box 1750–o 2Hex_221_BS1750.mrc relion_helix_toolbox –impose–i 2Hex_221_BS1750.mrc–o 2Hex_221_BS1750_4.77_t0.64.mrc–cyl_outer_diameter 230–angpix 0.8875–rise 4.77–twist -0.64–z_percentage 0.1

The volume was then processed in Fiji⁴⁸. The central 1,511 slices of the volume (which correspond to the number of slices that generate a single crossover distance) were analysed to visualize the projection. To visualize the full 360° rotation pattern, the images were rotated by 180°, and the 2 projection images were combined.

Simulation details

Simulations were performed using the GROMACS 2021.7⁵⁷ molecular dynamics package in the isothermal–isobaric NPT ensemble. The temperature was imposed using a velocity rescale thermostat⁵⁸ at 298.15 K with a time constant of 1.0 ps⁻¹. The pressure was controlled at 1.0 bar using the Parrinello–Rahman barostat⁵⁹ with a time constant of 2.0 ps⁻¹. The equations of motion were integrated using the leap-frog algorithm with a time step of 1.0 fs. Peptide chains and sodium, potassium and chloride ions were modelled using CHARMM36m force-field parameters⁶⁰, and water was described by the TIP3P model^61,62. The H₂PO₄⁻ and HPO₄²⁻ ions parameters were taken from CHARMM FF on CHARMM-GUI^63,64. Electrostatic interactions were treated using the smooth particle mesh Ewald method⁶⁵. The solvent molecules and ions were equilibrated by the steepest descent energy minimization with a tolerance of 100.0 kJ mol⁻¹, with position restraints in all peptide atoms, followed by 25 ns of molecular dynamics simulation with position restraints only in the backbone atoms of the peptides. After the equilibration of the solvent, the systems were simulated for 100 ns without restraints. All simulations were performed on RAVEN HPC at Max Planck Computing and Data Facility, Garching, Germany.

Molecular dynamics analysis

All analyses were performed with the GROMACS 2021.7 molecular dynamics package⁵³. The RMSF was calculated by residue and averaged over time and layer. The size-independent comparison of the 3D structures was done using the scaled dissimilarity index proposed by Maiorov and Crippen⁶⁶. The trajectories were least-squared fitted to the first snapshot of the simulation, using its backbone atoms as reference. The visualization and simulation snapshots were generated with visual molecular dynamics⁶⁷.

DILT1 fibril integrity and solvent interaction

A DILT1 fibril in explicit solvent, consisting of 94 layers with each layer rotated by −0.64° (total twist approximately 60°) was simulated to investigate its integrity. The fibril extends across the 22.0 × 22.0 × 44.8 nm³ simulation box along the z direction, forming an infinitely periodic fibril owing to periodic boundary conditions. The system composition was determined using the systematic equilibrium treatment described previously. The final composition includes: 1,692 peptide chains, 4,739 Cl⁻, 18 H₂PO₄⁻, 96 HPO₄²⁻, 44 K⁺, 1,521 Na⁺, 14,637 DMSO and 509,721 H₂O.

Composition of the system for molecular dynamics

The integrity of the DILT1 junction and cross-β-motif was studied by simulating 1-, 3-, 5-, 10- and 15-layer models for each motif. The same number of layers was simulated to verify the conformational persistence of the hexagon fibril. The layer structures were generated by rotating and translating the fundamental junction unit along the z direction. The geometric centre of the peptides was placed in a 22.0 × 22.0 × 22.0 nm³ simulation box and solvated with a 9:1 (% v/v) mixture of PBS and DMSO.

The system composition was determined from equilibrium calculations at 298.15 K based on the procedure described in ref. ⁶⁸. The dissociation constants of ionizable amino acids (for example, lysine, pK_a = 10.82) were assumed to be independent of each other. Activity coefficients were computed using the temperature-dependent extended Debye–Hückel equation⁶⁸. The experimental value of the dielectric constant of the water/DMSO mixture, needed to obtain the system composition, was obtained from ref. ⁶⁹, corresponding to 78.08 at 298.15 K.

The equilibrium composition of each system was computed numerically using the Newton–Raphson method, with the change in pH, ΔpH <10⁻⁷, as the convergence criterion.

Structure–property relationship simulations

As we have observed that 15 layers are a persistent fibril fragment for DILT1, we have generated samples for the other sequences, DILT2–DILT5, and P1 and P4, with this number of layers. Once the peptide sequence has been mutated in PyMOL, systems were built as before, starting from a junction and applying the corresponding symmetry operations, and simulated following the protocol described above.

DILT 1 nanofibril characterization

The incubation times and temperatures were varied as indicated below for kinetics and stability (solvent, temperature, sonication) characterizations as follows. (1) Preformed fibrils (1 mg ml⁻¹, 10% v/v DMSO, 90% v/v PBS) were obtained by incubating at 22 °C for 24 h with 500 rpm shaking (Extended Data Fig. 7a). (2) Nanofibril formation without pre-dissolved DMSO peptide stock was determined by dissolving the lyophilized peptide directly in DPBS (pH 7.4) yielding a 1 mg ml⁻¹ solution (Extended Data Fig. 7b). (3) A dilution stability test was performed on the preformed fibrils as described in (1), which were incubated at 37 °C for 1 week. The fibrils were then diluted in water. Specifically, 10 μl of the preformed fibrils (1 mg ml⁻¹) was added to 90 μl Milli-Q H₂O (0.1 mg ml⁻¹) with a final composition of 1% v/v DMSO, 9% v/v PBS and 90% v/v Milli-Q H₂O. The sample was incubated for 3 days at 37 °C (Extended Data Fig. 7c). This sample was used for the reconstruction of the 1-1-1 morphology (#2; Extended Data Table 1). (4) Early kinetics experiments were conducted by inducing nanofibril formation via the general method with incubation (22 °C, 500 rpm) for either 30 min, 45 min, 1 h, 3 h, 6 h, 9 h, 15 h, 21 h or 24 h (Extended Data Fig. 7k–s). (5) Late kinetics experiments were conducted using the preformed fibrils as described in (1) and transferring them into a 37 °C oven without shaking. The samples were then imaged after 4 days, 1 week, 2 weeks and 53 days (Extended Data Fig. 7t–w). The 53-day sample was used for reconstruction of the 3-3-3 morphology and for the polymorphism analysis (Extended Data Fig. 2). (6) Temperature stability tests were performed by using the preformed fibrils as described in (1) and transferring them for 15 min or 60 min in an oven at 60 °C, 85 °C or 95 °C (Extended Data Fig. 7d–i). (7) Sonication stability tests were performed by exposing preformed fibrils as described in (1) to sonication (15 min, sonication bath) (Extended Data Fig. 7j).

To minimize contamination during long-term incubation, peptide assemblies were prepared from fresh stock solutions using sterile-filtered buffer and clean consumables; full-grid cryo-EM mapping of long-incubation samples did not reveal contaminating micrometre-scale objects.

Atomic force microscopy

For atomic force microscopy (AFM) screening of DILT1–DILT5 and P3 samples were prepared by introducing the pre-dissolved peptide stock solution (10 mg ml ⁻¹, DMSO) to DPBS (pH 7.4), yielding a 1 mg ml⁻¹ solution, and analysed after overnight incubation (500 rpm, 25 °C, DPBS, pH 7.4). For imaging the DILT and P3 architectures, a Bruker Dimension FastScan BioTM atomic force microscope was used in the liquid state, which was operated in PeakForce mode. FastScan-D tips from Burker with a nominal spring constant of 0.25 Nm⁻¹ were used.

For AFM sample preparation, the pre-incubated peptide solution (100 µl, 1 mg ml⁻¹) was added to a circular mica substrate (20 mm) and incubated for 15 min. The excess liquid was removed, washed with 200 µl buffer (DPBS, pH 7.4), excess liquid was removed and 300 µl buffer was added to the mica to measure in liquid. Images were analysed with NanoScope Analysis 1.9.

In situ humidity-controlled FTIR experiments

Nanofibril formation was induced either by dissolving the DILT1 solid powder directly in the DPBS to yield a 1 mg ml⁻¹ solution (DILT1, Fig. 6e and Extended Data Fig. 9b,e–h) or, alternatively, peptide nanofibril formation was induced as described above by introducing the pre-dissolved peptide stock solution (10 mg ml⁻¹, DMSO) to DPBS (pH 7.4), yielding a 1 mg ml⁻¹ solution (DILT 5, Fig. 6e; DILT 2, Extended Data Fig. 9c; DILT 4, Extended Data Fig. 9d). The same protocol was applied for the control-peptide CKFKFQF. After overnight incubation (500 rpm, 25 °C), DILT1, DILT2, DILT4, DILT5 and CKFKFQF samples, were submitted to in situ humidity-controlled FTIR experiments, respectively. FTIR spectra were recorded in transmission mode using a Bruker VERTEX 70 spectrometer. The spectrometer was purged with nitrogen and all measurements were conducted at room temperature. The pre-incubated peptide samples were loaded into a Teflon flow cell fitted with CaF₂ windows (1 mm thickness) and a 2 mm path length; the sample (approximately 8 µl) was applied with a thickness ≤0.1 mm. The flow cell was mounted in the spectrometer sample compartment and connected to a dry-nitrogen supply to initiate and control dehydration. Spectra were recorded at regular intervals throughout the drying process, which proceeded until no further spectral change indicated complete desiccation.

Estimation of the number of water molecules per peptide

The number of water molecules per peptide was estimated using the known infrared absorption cross-sections of the respective functional groups. Molar extinction coefficients of ϵ_N–H = 100 M⁻¹ cm⁻¹ for the peptide N–H (ref. ⁷⁰) and ϵ_O–H = 400 M⁻¹ cm⁻¹ for the O–H stretch of water molecules⁷¹ were applied. Considering four N–H groups per peptide and two O–H bonds per water molecule, the integrated intensity ratio satisfies \(\frac{{I}_{{\rm{N}}-{\rm{H}}}}{{I}_{{\rm{O}}-{\rm{H}}}}\approx \frac{{N}_{{\rm{N}}-{\rm{H}}}\times {{\epsilon }}_{{\rm{N}}-{\rm{H}}}}{2{N}_{\mathrm{water}}\times {{\epsilon }}_{{\rm{O}}-{\rm{H}}}}\), where N_N–H is the number of peptide N–H groups and N_water is the number of water molecules.

For DILT1, after about 110 min in dry nitrogen, the N–H peak at 3,277 cm⁻¹ shows an integrated intensity of 0.15, and the broader water O–H band centred at 3,300 cm⁻¹ shows an integrated intensity of 7.5, corresponding to approximately 25 water molecules per peptide, assuming negligible contribution from backbone N–H groups (Extended Data Fig. 9b). Considering the channel geometry (0.48 nm peptide-layer height, 6 peptides per layer, 5 nm channel diameter) and a water density of 1.05 g cm⁻³ from simulations, a fully filled channel would contain about 55 water molecules per peptide. The FTIR-derived estimate is inherently approximate, as it relies on extinction coefficients from related systems and does not account for peptide side-chain O–H or N–H contributions. Nevertheless, the number inferred from the infrared spectral intensities is consistent with fully filled tubes at about 100 min drying time inferred from the convergence of the spectral shape at that time.

In situ dehydration Raman spectroscopy and Cryo-EM experiments

In situ Raman spectra were recorded using a WITec alpha300 Raman microscope equipped with a 532-nm excitation laser. The laser power at the sample was set to 5 mW. The pre-incubated peptide samples (1 mg ml⁻¹, directly dissolved in DPBS, 500 rpm overnight shaking, 25 °C) were loaded into a Teflon flow cell fitted with CaF₂ windows (1 mm thickness) and a 2 mm path length; the sample, approximately 8 µl, was applied with a thickness ≤0.1 mm. The flow cell was fixed on the microscope sample stage and connected to a dry-nitrogen supply, allowing dehydration to be initiated and controlled during Raman acquisition. Spectra were recorded at regular intervals throughout the drying process until no further spectral changes were observed, indicating complete desiccation.

For cryo-EM analysis after controlled drying, a similarly pre-incubated peptide sample was dehydrated using the same Teflon flow-cell and dry-nitrogen protocol as used for the Raman and FTIR experiments. After approximatey 120 min of drying, the flow cell was opened and a TEM grid was dripped into the residual sample liquid on the CaF₂ window and the TEM grid was then immediately plunge-frozen using a Vitrobot Mark V for subsequent cryo-EM examination.

Characterization of concentration-dependent self-assembly of DILT1

Sample preparation and incubation

DILT1 solutions of various concentrations were prepared. For this, a DILT1 10 mg ml⁻¹ DMSO stock solution, prepared in filtered DMSO, was diluted to yield DILT1 DMSO stock solutions at varying concentrations. Subsequently, these pre-dissolved DILT1 DMSO stock solutions were introduced to DPBS (pH 7.4) in a 1:9 ratio, respectively, yielding solutions with varied concentrations (1,000 µg ml⁻¹, 500 µg ml⁻¹, 100 µg ml⁻¹, 50 µg ml⁻¹, 35 µg ml⁻¹, 17.5 µg ml⁻¹, 8.8 µg ml⁻¹, 4.4 µg ml⁻¹, 2.2 µg ml⁻¹, 500 µl, 10% v/v DMSO in DPBS). The respective samples were mixed by brief vortexing and incubated on a shaker (500 rpm, room temperature) overnight. The given values were accurately rounded up to the full number in Extended Data Fig. 8.

Vial flip test

After overnight incubation of the DILT1 samples (2–1,000 µg ml⁻¹, 10% v/v DMSO in DPBS, 500 µl), the macroscopic gelation behaviour of DILT1 at varied concentrations was examined by vial flip test.

Proteostat assay

The critical aggregation concentration of the DILT1 peptide was examined via the commercial Proteostat protein aggregation assay kit by Enzo Life Sciences. Following the manufacturer’s recommendation, adapting a previously published protocol⁷², the Proteostat working solution was prepared by diluting Proteostat stock solution (0.50 µl) and assay buffer (1 µl 10x assay buffer) in 98.5 µl Milli-Q water. To conduct the assay, 36 μl pre-incubated DILT1 peptide solutions at various concentrations (2–1,000 µg ml⁻¹, 10% v/v DMSO in DPBS) were mixed with 4 μl Proteostat working solution. The samples were transferred into a Greiner 384 flat black well plate (3 wells per sample (technical replicates n = 3), that is, per peptide concentration, 9 µl sample per well), incubated in the dark (15 min) while shaking, and the fluorescence intensity of the Proteostat dye was determined λ_excitation = 550 nm, λ_emission = 600 nm; bandwidths 20 nm, multiple reads per well, respectively. At high DILT1 concentration (956 µM), the aggregation affected sample loading to a maximum of 2 wells, each containing 9 µl of the mix (n = 2).

Transmission electron microscopy

To analyse the aggregate and assembly morphologies at various concentrations, the incubated DILT1 sample solutions (2–1,000 µg ml⁻¹, 10% v/v DMSO in DPBS) were submitted to TEM analysis. Grid preparation was carried out as follows. Five microlitres of the sample solution was deposited onto a TEM grid and allowed to stand for 5 min to enable adequate adsorption. Excess solution was then gently removed using filter paper. Subsequently, staining was performed with 5 µl of 4% w/v aqueous uranyl acetate for 2 min. After staining, the grids were rinsed three times with Milli-Q water to remove excess stain. Finally, the water was gently removed from the grid with filter paper, and the grids were left to dry overnight before TEM measurement. TEM was measured at 120 kV.

Hydrogel and honeycomb fibril preparation

Hydrogels were prepared by either pre-dissolving the DILT1 peptide in DMSO (10 mg ml⁻¹) and adding this DMSO peptide stock solution into DPBS to yield a 1 mg ml⁻¹ solution (0.1 wt%) or by dissolving solid DILT1 powder directly in the respective solvent to yield 1–4 wt% hydrogels. For example, 0.3 mg peptide was dissolved in 30 µl buffer (for example, DPBS, pH 7) and mixed for approximately 5 s to yield a 1 wt% (10 mg ml⁻¹) hydrogel. Very soft hydrogels were obtained at 0.1 wt%, for example, by the nanofibril-formation procedure described in ‘Peptide nanofibril formation’.

Rheology

Rheological characterization was conducted using a DHR3 rheometer (TA Instruments) equipped with a temperature controller and a solvent reservoir to prevent hydrogel drying. Experiments were performed using an 8-mm parallel-plate geometry with hydrogels of approximately 30 μl volume (gap size of about 0.5 mm). Characterization of the hydrogel mechanical properties was conducted at 25 °C. Gels were prepared directly on the plate by the procedure described in ‘Hydrogel and honeycomb fibril preparation’. Oscillatory time-sweep measurements monitored the gelation at a fixed strain of 0.1% and a fixed frequency of 1 Hz. Extended Data Fig. 8e presents representative sections after complete gelation (approximately last 300 s of gelation curves), which summarizes individual measurements recorded with time stamps at approximately 6-s intervals (time stamps are indicative). Oscillatory strain sweeps (0.01–200% or 0.01–1000%, as indicated) were conducted at a fixed frequency of 1 Hz. Oscillatory frequency sweeps (0.05–100 Hz) were performed at a fixed strain of 0.1%.

DILT1 gel stability

Photographs were taken with a Canon EOS 250D with a Canon EF 24–105 mm f/4L IS lens. Manual mode, exposure of 1/50 with aperture of f/8 and focal length of 105 mm. ISO 100. Saved in jpg format with a resolution of 5,284 dpi.

1. Rothemund, P. W. K. Folding DNA to create nanoscale shapes and patterns. Nature 440, 297–302 (2006).

   Article  ADS  CAS  PubMed  Google Scholar 

2. Douglas, S. M. et al. Self-assembly of DNA into nanoscale three-dimensional shapes. Nature 459, 414–418 (2009).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

3. Woo, S. & Rothemund, P. W. K. Self-assembly of two-dimensional DNA origami lattices using cation-controlled surface diffusion. Nat. Commun. 5, 4889 (2014).

   Article  ADS  CAS  PubMed  Google Scholar 

4. Cheng, D., Jia, F., Jiang, Y., Conticello, V. P. & Jiang, T. Assembly of peptide nanostructures with controllable sizes. Nano Res. 17, 151–161 (2024).

   Article  ADS  Google Scholar 

5. Bianco, S. et al. Modular salt-induced nanostructures formed by a functionalized dipeptide system. Matter 8, 102334 (2025).

   Article  CAS  Google Scholar 

6. Vila, Á, González, S. & Insua, I. Emergent properties of supramolecular peptide assemblies. Chem. Commun. 62, 2097–2109 (2026).

   Article  CAS  Google Scholar 

7. Fang, Y., Shi, J., Liang, J., Ma, D. & Wang, H. Water-regulated viscosity-plasticity phase transitions in a peptide self-assembled muscle-like hydrogel. Nat. Commun. 16, 1–11 (2025).

   Article  ADS  CAS  Google Scholar 

8. Sinha, N. J., Langenstein, M. G., Pochan, D. J., Kloxin, C. J. & Saven, J. G. Peptide design and self-assembly into targeted nanostructure and functional materials. Chem. Rev. 121, 13915–13935 (2021).

   Article  CAS  PubMed  Google Scholar 

9. Wang, F. et al. Deterministic chaos in the self-assembly of β sheet nanotubes from an amphipathic oligopeptide. Matter 4, 3217–3231 (2021).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

10. Kralj, S. et al. Heterochirality and halogenation control Phe-Phe hierarchical assembly. ACS Nano 14, 16951–16961 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

11. Abb, S., Harnau, L., Gutzler, R., Rauschenbach, S. & Kern, K. Two-dimensional honeycomb network through sequence-controlled self-assembly of oligopeptides. Nat. Commun. 7, 10335 (2016).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

12. Athiyarath, V. et al. Water-mediated reconfigurable topology and mechanics in porous peptide materials. Matter 9, 102669 (2026).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

13. De Santis, E. & Ryadnov, M. G. Peptide self-assembly for nanomaterials: the old new kid on the block. Chem. Soc. Rev. 44, 8288–8300 (2015).

    Article  PubMed  Google Scholar 

14. Acar, H. et al. Self-assembling peptide-based building blocks in medical applications. Adv. Drug Deliv. Rev. 110–111, 65–79 (2017).

    Article  PubMed  Google Scholar 

15. Cui, H., Webber, M. J. & Stupp, S. I. Self-assembly of peptide amphiphiles: from molecules to nanostructures to biomaterials. Pept. Sci. 94, 1–18 (2010).

    Article  CAS  Google Scholar 

16. Maji, S. K. et al. Functional amyloids as natural storage of peptide hormones in pituitary secretory granules. Science 325, 328–332 (2009).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

17. Fitzpatrick, A. W. P. et al. Cryo-EM structures of tau filaments from Alzheimer’s disease. Nature 547, 185–190 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

18. Gremer, L. et al. Fibril structure of amyloid-β(1–42) by cryo-electron microscopy. Science 358, 116–119 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

19. Vidmar, S. et al. Beyond dimerization: harnessing tetrameric coiled-coils for nanostructure assembly. Angew. Chem. Int. Ed. 64, e202422075 (2025).

20. Gupta, A. K., Esih, H., Gradišar, H. & Jerala, R. Self-assembling protein cages: from coiled-coil module to machine learning-driven de novo design of next-generation biomaterials. Mater. Adv. 7, 1378–1396 (2026).

    Article  CAS  PubMed  Google Scholar 

21. Tang, Y. et al. Patchy peptide particles for pH-responsive assembly into liquid crystals or lattices. Science 391, 604–610 (2026).

    Article  ADS  CAS  PubMed  Google Scholar 

22. Zhou, W., Strmšek, Ž, Snoj, J., Škarabot, M. & Jerala, R. Programmable protein–DNA composite nanostructures: from nanostructure construction to protein-induced micro-scale material self-assembly and functionalization. Small 21, 1–12 (2025).

    Google Scholar 

23. Zhang, T. et al. Terminal-directed supramolecular liquid crystal formation by designed coiled-coil interparticle stacking. ACS Nano 20, 5818–5829 (2026).

    Article  CAS  PubMed  Google Scholar 

24. Grosvirt-Dramen, A. et al. Hierarchical assembly of conductive fibers from coiled-coil peptide building blocks. ACS Nano 19, 10162–10172 (2025).

    Article  CAS  PubMed  Google Scholar 

25. McCahill, A. L. et al. Designed coiled-coil peptide nanoparticles with tunable self-assembly: distinct ordered nanostructures via nonnatural side chain modification and electrostatic screening. Soft Matter 22, 428–437 (2026).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

26. Huddy, T. F. et al. Blueprinting extendable nanomaterials with standardized protein blocks. Nature 627, 898–904 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

27. Pandya, M. J. et al. Sticky-end assembly of a designed peptide fiber provides insight into protein fibrillogenesis. Biochemistry 39, 8728–8734 (2000).

    Article  CAS  PubMed  Google Scholar 

28. Sarkar, B., O’Leary, L. E. R. & Hartgerink, J. D. Self-assembly of fiber-forming collagen mimetic peptides controlled by triple-helical nucleation. J. Am. Chem. Soc. 136, 14417–14424 (2014).

    Article  ADS  CAS  PubMed  Google Scholar 

29. Ghadiri, M. R., Granja, J. R., Milligan, R. A., McRee, D. E. & Khazanovich, N. Self-assembling organic nanotubes based on a cyclic peptide architecture. Nature 366, 324–327 (1993).

    Article  ADS  CAS  PubMed  Google Scholar 

30. Insua, I. et al. Self-assembly of cyclic peptide monolayers by hydrophobic supramolecular hinges. Chem. Sci. 14, 14074–14081 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

31. Díaz, S., Sanchez-Fernandez, A., Granja, J. R., Insua, I. & Montenegro, J. 3D self-assembly of cyclic peptides into multilayered nanosheets. Chem. Sci. 17, 6727–6733 (2026).

    Article  PubMed  PubMed Central  Google Scholar 

32. Chen, K. H., Corro, K. A., Le, S. P. & Nowick, J. S. X-ray crystallographic structure of a giant double-walled peptide nanotube formed by a macrocyclic β-sheet containing Aβ 16–22. J. Am. Chem. Soc. 139, 8102–8105 (2017).

    Article  ADS  CAS  PubMed  Google Scholar 

33. Zeng, M., Parsons, W., Chen, Y., Chalmers, D. K. & Perrier, S. Hydrophobicity-controlled self-assembly of supramolecular peptide nanotubes in water. Angew. Chem. Int. Ed. 64, e202423828 (2025).

    Article  CAS  Google Scholar 

34. Song, Q. et al. Molecular self-assembly and supramolecular chemistry of cyclic peptides. Chem. Rev. 121, 13936–13995 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

35. Nagy-Smith, K., Moore, E., Schneider, J. & Tycko, R. Molecular structure of monomorphic peptide fibrils within a kinetically trapped hydrogel network. Proc. Natl Acad. Sci. USA 112, 9816–9821 (2015).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

36. Yolamanova, M. et al. Peptide nanofibrils boost retroviral gene transfer and provide a rapid means for concentrating viruses. Nat. Nanotechnol. 8, 130–136 (2013).

    Article  ADS  CAS  PubMed  Google Scholar 

37. Hervas, R. et al. Cryo-EM structure of a neuronal functional amyloid implicated in memory persistence in Drosophila. Science 367, 1230–1234 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

38. Bücker, R. et al. The cryo-EM structures of two amphibian antimicrobial cross-β amyloid fibrils. Nat. Commun. 13, 4356 (2022).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

39. Baumgartner, B. et al. Pore size-dependent structure of confined water in mesoporous silica films from water adsorption/desorption using ATR–FTIR spectroscopy. Langmuir 35, 11986–11994 (2019).

    Article  CAS  PubMed  Google Scholar 

40. Podbevšek, D. et al. The role of water mobility on water-responsive actuation of silk. Nat. Commun. 15, 8287 (2024).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

41. Heerde, T. et al. Cryo-EM structure and polymorphic maturation of a viral transduction enhancing amyloid fibril. Nat. Commun. 14, 4293 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

42. Sieste, S. et al. Supramolecular peptide nanofibrils with optimized sequences and molecular structures for efficient retroviral transduction. Adv. Funct. Mater. 31, 2009382 (2021).

43. Kaygisiz, K. et al. Peptide amphiphiles as biodegradable adjuvants for efficient retroviral gene delivery. Adv. Healthc. Mater. 13, 1–14 (2024).

    Google Scholar 

44. Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat. Methods 14, 331–332 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

45. Rohou, A. & Grigorieff, N. CTFFIND4: fast and accurate defocus estimation from electron micrographs. J. Struct. Biol. 192, 216–221 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

46. Scheres, S. H. W. RELION: implementation of a Bayesian approach to cryo-EM structure determination. J. Struct. Biol. 180, 519–530 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

47. Zivanov, J. et al. A Bayesian approach to single-particle electron cryo-tomography in RELION-4.0. eLife 11, e83724 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

48. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

49. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017).

    Article  CAS  PubMed  Google Scholar 

50. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. Sect. D 66, 486–501 (2010).

    Article  ADS  CAS  Google Scholar 

51. Pettersen, E. F. et al. UCSF Chimera—a visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605–1612 (2004).

    Article  ADS  CAS  PubMed  Google Scholar 

52. Williams, C. J. et al. MolProbity: more and better reference data for improved all-atom structure validation. Protein Sci. 27, 293–315 (2018).

    Article  CAS  PubMed  Google Scholar 

53. Liebschner, D. et al. Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. Acta Crystallogr. Sect. D 75, 861–877 (2019).

    Article  ADS  CAS  Google Scholar 

54. Jamali, K. et al. Automated model building and protein identification in cryo-EM maps. Nature 628, 450–457 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

55. Meng, E. C. et al. UCSF ChimeraX: tools for structure building and analysis. Protein Sci. 32, e4792 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

56. Croll, T. I. ISOLDE: a physically realistic environment for model building into low-resolution electron-density maps. Acta Crystallogr. Sect. D 74, 519–530 (2018).

    Article  ADS  CAS  Google Scholar 

57. Abraham, M. J. et al. GROMACS: high performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX 1–2, 19–25 (2015).

    Article  ADS  Google Scholar 

58. Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J. Chem. Phys. 126, 14101 (2007).

    Article  Google Scholar 

59. Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: a new molecular dynamics method. J. Appl. Phys. 52, 7182–7190 (1981).

    Article  ADS  CAS  Google Scholar 

60. Huang, J. et al. CHARMM36m: an improved force field for folded and intrinsically disordered proteins. Nat. Methods 14, 71–73 (2017).

    Article  CAS  PubMed  Google Scholar 

61. Mahoney, M. W. & Jorgensen, W. L. A five-site model for liquid water and the reproduction of the density anomaly by rigid, nonpolarizable potential functions. J. Chem. Phys. 112, 8910–8922 (2000).

    Article  ADS  CAS  Google Scholar 

62. Jorgensen, W. L., Chandrasekhar, J., Madura, J. D., Impey, R. W. & Klein, M. L. Comparison of simple potential functions for simulating liquid water. J. Chem. Phys. 79, 926–935 (1983).

    Article  ADS  CAS  Google Scholar 

63. Kim, S. et al. CHARMM-GUI ligand reader and modeler for CHARMM force field generation of small molecules. J. Comput. Chem. 38, 1879–1886 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

64. Jo, S., Kim, T., Iyer, V. G. & Im, W. CHARMM-GUI: a web-based graphical user interface for CHARMM. J. Comput. Chem. 29, 1859–1865 (2008).

    Article  ADS  CAS  PubMed  Google Scholar 

65. Essmann, U. et al. A smooth particle mesh Ewald method. J. Chem. Phys. 103, 8577–8593 (1995).

    Article  ADS  CAS  Google Scholar 

66. Maiorov, V. N. & Crippen, G. M. Size-independent comparison of protein three-dimensional structures. Proteins Struct. Funct. Bioinform. 22, 273–283 (1995).

    Article  CAS  Google Scholar 

67. Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. J. Mol. Graph. 14, 33–38 (1996).

    Article  CAS  PubMed  Google Scholar 

68. Harris, D. C. Quantitative Chemical Analysis 8th edn (W. H. Freeman and Company, 2010).

69. Płowaś, I., Świergiel, J. & Jadżyn, J. Relative static permittivity of dimethyl sulfoxide + water mixtures. J. Chem. Eng. Data 58, 1741–1746 (2013).

    Article  Google Scholar 

70. Park, J. & Hochstrasser, R. M. Multidimensional infrared spectroscopy of a peptide intramolecular hydrogen bond. Chem. Phys. 323, 78–86 (2006).

    Article  CAS  Google Scholar 

71. Nienhuys, H.-K. Femtosecond Mid-Infrared Spectroscopy of Water (Technische Universiteit Eindhoven, 2002).

72. Chagri, S. et al. Intracellular formation of synthetic peptide nanostructures causes mitochondrial disruption and cell death in tumor spheroids. Adv. Sci. 12, 1–13 (2025).

    Article  Google Scholar 

73. Tuttle, M. D. et al. Solid-state NMR structure of a pathogenic fibril of full-length human α-synuclein. Nat. Struct. Mol. Biol. 23, 409–415 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

74. Kyte, J. & Doolittle, R. F. A simple method for displaying the hydropathic character of a protein. J. Mol. Biol. 157, 105–132 (1982).

    Article  CAS  PubMed  Google Scholar 

75. Sunde, M. et al. Common core structure of amyloid fibrils by synchrotron X-ray diffraction 1 1Edited by F. E. Cohen. J. Mol. Biol. 273, 729–739 (1997).

    Article  CAS  PubMed  Google Scholar 

Download references

Author notes

1. These authors contributed equally: Jasmina Gačanin, Francesca Mazzotta

Authors and Affiliations

1. Max Planck Institute for Polymer Research, Mainz, Germany

   Jasmina Gačanin, Francesca Mazzotta, Luis Andre Baptista, Nico Alleva, Thunchanok Thummaraj, Fanny Bonnicel, Cong Zhou, Lei Gao, Mischa Bonn, Ingo Lieberwirth, Robinson Cortes-Huerto, Katharina Landfester & Tanja Weil

2. Institute of Inorganic Chemistry I, Ulm University, Ulm, Germany

   Jasmina Gačanin

3. Institute of Protein Biochemistry, Ulm University, Ulm, Germany

   Nikolay Stoyanov, Matthias Schmidt & Marcus Fändrich

4. Institute of Molecular Virology, Ulm University Medical Center, Ulm, Germany

   Jan Münch

Contributions

J.G. designed the amino acid sequences of the peptides, performed or supervised peptide synthesis, purification and analysis, identified the self-assembly, fibrillating and gelating behaviour of the amino acid sequence of the DILT peptides and mutants, designed experimental procedures and conducted or supervised sample preparation for DILT peptide fibril assemblies and mutants, prepared and characterized the hydrogels and interpreted, under the supervision of T.W. J.G. and L.A.B. acquired photographs of the DILT1 hydrogel. F.M. performed the cryo-EM DILT1, DILT2 and DILT3 preparation, image acquisition, image analysis and atomic modelling under the supervision of M.S., I.L. and K.L. F.B. performed cryo-EM imaging and image analysis of DILT2, DILT3, P1, P2, P3, P4 and P5 under the supervision of F.M., I.L. and K.L. L.A.B. assisted in the atomic modelling of the DILT1 cryo-EM data and building and refining the model. N.S. performed the cryo-EM preparation, image acquisition, image analysis and atomic modelling of DILT2 (4-4-4) and DILT3 (3-3-3) fibrils. M.S. performed cryo-EM DILT and mutated peptides imaging. L.A.B. and R.C.-H. designed the simulation protocols. L.A.B. prepared the input files and performed the simulations, which data both L.A.B. and R.C.-H. analysed and interpreted. N.A. conducted DILT and mutated PNF characterization via atomic force microscopy, and T.T. helped with DILT and mutated peptide sample preparation or incubation, and conducted TEM imaging and analysis under the supervision of J.G. and T.W. C.Z. and L.G. conducted humidity-controlled FTIR experiments and calculations, under the supervision of M.B. J.G., F.M., L.A.B. and T.W. created images and all figures; N.A., C.Z., L.G., N.S. and M.S. created images. J.G., F.M., L.A.B., N.S., M.S., I.L., M.S., M.F., R.C.-H., T.W. and K.L. discussed and interpreted the experimental data. J.G., F.M., L.A.B., I.L. R.C.-H., M.F., J.M., T.W. and K.L. contributed to the writing of the paper. J.G., T.W. and K.L. raised the funds. T.W. and K.L. supervised the research.

Corresponding authors

Correspondence to Katharina Landfester or Tanja Weil.

Competing interests

J.G., F.M., I.L., K.L. and T.W. declare potential financial competing interest from a patent application filed by their institution. Sequences as well as related applications are covered in the patent application (patent application number EP26197462.0). The other authors declare no competing interests.

Peer review information

Nature thanks Hye-soo Lee, Yong-beom Lim and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions