Introduction

Formation of dental enamel, the hardest and most mineralised tissue of vertebrates1, relies on the 3D assembly and organisation of the protein amelogenin2. During development, this protein transforms from a disordered conformation to ordered β-rich fibrillar structures3, nucleating and epitaxially growing hierarchically organised apatite nanocrystals4,5. This process results in a highly structured enamel tissue that protects teeth throughout life by providing strength and anti-abrasive properties against physical, chemical, and thermal insults6. However, irreversible enamel loss due to mechanical wear, trauma, or erosion from acidic foods or fermentation by-products from bacteria can lead to carious lesions and ultimately tooth loss[7](https://www.nature.com/articles/s41467-025-64982-y#ref-CR7 “Johansson, A.-K., Omar, R., Carlsson, G. E. & Johansson, A. Dental erosion and its growing importance in clinical practice: from past to present. Int. J. Dent. https://doi.org/10.1155/2012/632907

(2012).“). These oral problems affect nearly half of the global population at an annual cost of ~ US$544 billion8.

The properties of enamel result from its composition and structure1. Carbonated hydroxyapatite (CHAp) nanocrystals of ~ 50 nm in diameter and micrometres in length are the building blocks of dental enamel. They are aligned, densely packed, and organised in specific configurations in different anatomical regions. The outermost aprismatic enamel layer in permanent teeth is ∼ 16–45 μm thick and composed of aligned apatite nanocrystals with their c-axis perpendicular to the tooth surface9. Subjacent to this, prismatic enamel comprises nanocrystals organised into ~ 4–8 μm thick prisms10, which are delineated by inter-prismatic crystallites aligned at a ~ 60° angle to the prism axis1. Here, undulating trajectories of prisms create decussation patterns known as Hunter-Schreger bands, forming longitudinal (parazone) and transverse (diazone) bundles11,12. The stiffness of enamel, expressed as Young’s modulus (E), allows small deformations under high forces (e.g., during mastication) while its hardness (H) enables it to resist localised plastic deformation due to indentation or scratching (e.g., during teeth grinding)13. The wear strength of enamel protects teeth from chemical erosion (e.g., from extrinsic or intrinsic acids), abrasion (e.g., toothbrushing), or attrition (e.g., tooth-on-tooth contact), while its toughness allows it to tolerate fracture without falling apart. The combination of all these properties allows enamel to withstand thousands of chewing cycles per hour[14](https://www.nature.com/articles/s41467-025-64982-y#ref-CR14 “Fuentes, R. et al. A new tridimensional insight into geometric and kinematic characteristics of masticatory cycles in participants with normal occlusion. Biomed. Res. Int. 527463 https://doi.org/10.1155/2018/2527463

(2018).“), biting forces of up to 770 N, and decades of exposure to harsh oral environments1.

The recreation of dental enamel has been a major goal in materials science15. Methods based on acidic16, hydrothermal17, and high-power laser18 treatments have enabled the ambient deposition of mineral, but without the capacity to recreate the architecture and function of healthy enamel. Recently, several enamel analogues comprising organic-inorganic nanocomposites such as polyacrylic acid-zinc oxide nanowires19 and polyvinyl alcohol-HAp nanowires20 have been reported to generate functional hierarchical architectures. However, these technologies rely on non-physiological fabrication conditions, which restrict their applicability in clinical settings21.

To address the clinical need, bioinspired approaches offer promising alternatives to regrow different anatomical regions of enamel22. For instance, Moradian-Oldak and colleagues have developed amelogenin-based scaffolds capable of growing layers of aprismatic enamel-like structures ex vivo9. Other approaches based on phase-transited lysozyme films with amelogenin-like peptides23, amelogenin-derived shortened ADP5 (shADP5) oligopeptide24, modified leucine-rich amelogenin peptide (mLRAP)25, phosphorylated and phosphonated peptides26, or triethylamine stabilised calcium phosphate ions27 have reported remineralisation of prismatic enamel. However, these strategies suffer from drawbacks that restrict their clinical translation such as having toxic and noxious components (e.g., triethylamine27), time-consuming application processes (e.g., 30 min23, 15 min27, 12 h26, 10 min/day28), partial recovery of architecture and functional properties (e.g., aprismatic enamel9, prismatic enamel23,25,26,27), and limited control of the mineralisation process. Thus, remineralising technologies that recreate the diverse architecture and spectrum of functional properties of dental enamel in a patient and clinically-friendly manner remain an unmet challenge.

In this study, we report on the design and performance of a clinically-friendly supramolecular protein matrix using disordered elastin-like recombinamer (ELR) molecules to emulate key structural and functional features of the β-rich fibrillar amelogenin matrix driving epitaxial and hierarchical mineralisation in dental enamel. We use both computational and experimental work to describe the matrix design and demonstrate the epitaxial growth of mineralised layers up to 10 μm thick from the surface of teeth exhibiting different levels of erosion down to bare exposed dentine ex vivo. The matrix is highly stable and able to regrow all the anatomical features of enamel, including prismatic, interprismatic, and aprismatic regions, while restoring tissue stiffness, hardness, toughness, coefficient of friction, wear strength, and stability to extensive abrasion from toothbrushing, chewing and grinding, and exposure to acidic solutions.

Results and discussion

Rationale for the design of the mineralising ELR matrix to emulate enamel development

Recent studies have shown that intrinsically disordered amelogenin molecules assemble into functional β-rich fibrillar structures via intermolecular interactions in the presence of Ca2+ ions29,30. These structures nucleate and grow apatite nanocrystals in vitro and during enamel development4,5. We have demonstrated the capacity to grow organised mineralised structures by modulating the levels of order and disorder of an intrinsically disordered elastin-like recombinamer (ELR) consisting of hydrophobic (VPGIG), hydrophilic (VPGKG), and acidic statherin (DDDEEKFLRRIGRFG) motifs31. We hypothesised that the disordered nature of the ELR molecules can be harnessed to generate supramolecular ELR ensembles that imitate structural and functional characteristics of the amelogenin matrix2,3,4. By generating these amelogenin-like β-rich supramolecular assemblies over enamel or dentine, it may be possible to facilitate epitaxial, integrated, and hierarchical enamel-like mineralisation from dental tissues. Thus, we combined Ca2+ ions with ELR molecules, which are reported to promote β-sheet conformation in proteins32 and peptides33 via ion bridge formation. We also used drying as a mechanism to induce molecular order through molecular crowding, as previously reported34,35. We reasoned that the incorporation of Ca2+ ions and drying of the ELR solution can together be used to promote β-rich fibrillar ELR structures, which we refer to as ‘ELR fibrils’. Furthermore, the incorporated Ca2+ ions would also serve as nucleating points to facilitate mineralisation. In addition, given the capacity of the ELR matrix to create a confined environment for mineral nucleation and growth31, we envisage that the thickness of the ELR layer applied to dental tissue can spatially regulate the resulting mineral layer thickness in situ. To further support clinical translation, the choice of components in the formulation has been guided by considerations of rapid application, ease of use, and overall practicality in a clinical setting.

Supramolecular assembly of the mineralising ELR matrix

We placed a drop of 1% w/v ELR solution without or with 1.5 mM Ca2+ ions on either scanning electron microscope (SEM) stubs or transmission electron microscope (TEM) grids and allowed them to dry at room temperature (25 °C). SEM and TEM imaging revealed that the ELR molecules assembled into ELR fibrils only in the presence of Ca2+ ions (Fig. 1a, Supplementary Fig. 1 and Supplementary Discussion 1). These fibrils ranged between 15–40 nm in width and several micrometres in length, which corresponds to the classical fibrillar geometry36. Presence of Ca2+ ions within these fibrils was confirmed by X-ray photoelectron spectroscopy (XPS) while weak electrostatic interactions between Ca2+ ions and ELR molecules were confirmed by isothermal titration calorimetry (ITC) and UV-Vis (Supplementary Fig. 2). Furthermore, we demonstrate the possibility to use Ca2+ ions and crosslinking during matrix fabrication to engineer ELR matrices with tuneable levels of ordered β-conformations, fibril structures (Supplementary Discussion 2, 3; Supplementary Figs. 3–6), and number of potential nucleation points for mineralisation (Supplementary Fig. 7). These results demonstrate the synergistic role of Ca2+ ions, crosslinking, and solvent evaporation in forming ELR fibrils, while generating a tuneable matrix with Ca-rich nucleating sites for mineralisation.

Fig. 1: ELR fibril formation and mineralisation capacity.

a SEM micrograph of ELR fibrils formed after drying a drop of 1% w/v ELR solution (in DMF/DMSO at 9/1 ratio) containing 1.5 mM Ca2+ ions. Representative micrograph from *n *= 5 independent experiments. b WAXS data of ELR fibrils show reflections at Q = 1.34 and 0.6 Å−1 which correspond to 4.7 and 10 Å, respectively. The inset shows a 1030 pixel wide WAXS detector image with the beam position slightly off-centred due to its setup. The inset shows a WAXS detector image of dried ELR fibrils. c Deconvoluted FTIR spectra (1600 to 1700 cm−1) showing the antiparallel arrangement of the β-sheet. d (i) Computational model depicting self-assembly of ELR molecules in the presence of Ca2+ ions, 30 equilibrated ELR molecules (β-sheet secondary structure with coiled secondary structure for the statherin motif) were inserted into a 16 x 16 x 30 nm aqueous rectangular cuboid box and equilibrated with a 0.2 M concentration of Ca2+ ions. (ii) RDF plot of statherin – Ca2+ and VPGKG – Cl1- interactions with statherin shown in red, Ca+2 ions shown in green within and at the surface of the protein aggregate, other protein residues are shown in transparent yellow. (iii) Aggregation of 6 individually coloured protofilaments, each containing 30 ELR proteins. The system starts at an evenly distributed collection of filaments that, over time, aggregate along a line. e Schematic illustration of the proposed model showing the supramolecular organisation of the ELR molecules within ELR fibrils. Each ELR fibril (blue layer) is ~300 Å in thickness and comprises of 6 individual filaments (grey sticks) measuring 48.7 Å thick each. f Schematic illustration and their corresponding TEM images showing the formation and mineralisation of the ELR matrix. (i) a 2 μL drop of 5% w/v ELR solution containing 1.5 mM Ca2+ ions and hexamethylene diisocyanate (HDI, 0.56% v/v) crosslinker was casted on a TEM grid and dried to form a crosslinked matrix of ELR fibrils. (ii) Formation of amorphous calcium phosphate (ACP, blue arrows) phase at 2 h of incubation in mineralisation solution containing 10 mM Ca2+, 6 mM PO43-, and 2 mM F- ions. (iii) Transformation of ACP into apatite nanocrystals (white arrows) after 24 h of mineralisation. Schematic illustrations in (e, f) were prepared by scientific illustrator Leonora Martínez Nuñez.

Structural characterisation of the ELR fibrils

Wide-angle X-ray scattering (WAXS) analysis on the ELR fibrils revealed a sharp peak at Q = 1.3 Å−1 corresponding to a periodic β-strand separation dβ = 4.7 Å perpendicular to the fibril axis and a broader hump at Q = 0.6 Å−1 corresponding to a characteristic inter-sheet packing distance of 10 Å in a two-dimensional (2D) crystalline lattice (Fig. 1b)37. This is a typical cross-β diffraction pattern observed in several peptides and proteins that exhibit fibrillar structures37. Furthermore, the 4.7 Å β-strand separation corresponds to stacks of hydrogen-bonded β-sheets in folded protein structures37,38 and is similar to the structural motifs found in both the protein matrix of developing human enamel39 and amelogenin ensembles formed in vitro[40](https://www.nature.com/articles/s41467-025-64982-y#ref-CR40 “Zhang, J. et al. Calcium interactions in amelogenin-derived peptide assembly. Front. Physiol. 13, https://doi.org/10.3389/fphys.2022.1063970

(2022).“). Peak at Q = 1.04 Å⁻¹ can be attributed to interactions between α-helices41 while peaks at Q = 1.07, 1.49, 1.72, and 2.0 Å⁻¹ (Fig. 1b) align with the (001), (111), (112), and (211) Miller indices of CaCl2 suggesting peaks arising from CaCl2 crystallisation (Supplementary Fig. 8c). In addition, FTIR analysis revealed the presence of antiparallel β-sheets (Fig. 1c)42, which are also comparable to those in the enamel-developing amelogenin matrix3,5. On the other hand, small-angle X-ray scattering (SAXS) analysis showed a low Q-decay of 4.1, characteristic of a large surface fractal nanostructure with smooth surfaces (Supplementary Fig. 8a), which correlates with SEM observations (Fig. 1a) depicting a network of smooth fibrils. A small broad peak at Q = 0.129 Å−1 (Supplementary Fig. 8a) corresponding to a d-spacing of 48.7 Å was attributed to the width of an individual filament43, which undergoes periodic stacking to form ELR fibrils similar to those reported for β-rich fibrillar structures43. Interestingly, 48 Å wide filaments arranged parallel to each other have been observed in embryonic bovine enamel matrix44, which are structurally similar to the 48.7 Å wide filaments of our ELR fibrils. These results demonstrate the capacity of the ELR matrix to form cross-β fibrillar structures exhibiting similar supramolecular organisation and dimensions as those found in the amelogenin-rich enamel matrix.

Coarse-grained simulation of the formation of ELR fibrils

To better understand the underlying matrix assembly mechanism, we performed simulations of the ELR molecules using the MARTINI (v3) forcefield. Thirty ELR molecules were equilibrated in water with Ca2+ ions and self-assembled into an elongated filament with VPGIG forming the core of the structure and VPGKG and statherin largely localised at the surface (Fig. 1d (i)). We only observed filament formation in the presence Ca2+ ions (Supplementary Fig. 9), which is consistent with the experimental SEM results (Fig. 1a). Furthermore, measurement of radial distribution function (RDF) confirmed interactions between Ca2+ ions and statherin regions of single filaments as well as ELR aggregates of 6 filaments, where Ca2+ ions interacted with their surface and absorbed into their internal structure (Fig. 1d (ii)). On the other hand, interactions between the counter ion (Cl1-) and VPGKG motifs were not observed (Fig. 1d (ii)). In addition, we investigated the aggregation of these filaments to form the fibril structure by inserting 6 filaments into a box with water between them and equilibrated. These filaments aggregated parallel to each other (Fig. 1d (iii)) and in concordance with experimental observations, displayed a width of 51.6 Å/filament. Interestingly, increasing the number of filaments to 8 did not lead to assembly along a single plane. These results confirmed the role of Ca2+ ions in facilitating ELR assembly into filaments, which are in turn assembled into the fibrillar structure.

ELR fibril formation model

Based on experimental and simulation results, we propose a model for the molecular packing of the ELR fibrils (Fig. 1e). From WAXS data, we assigned the 10 Å to the inter-sheet packing distance and 4.7 Å to the periodic β-strand separation along the fibril axis37,38, while SAXS data was used to assign the 48.7 Å to the individual filament width43. Furthermore, from the simulations, the hydrophobic VPGIG motifs would aggregate at the core of the filament, while charged VPGKG and statherin motifs would localise in the outer layers of each filament. In addition, V and G residues would promote ordered β-conformation45 and the negatively charged regions (statherin) would interact electrostatically with Ca2+ ions during the formation of the ELR fibrils (Fig. 1e). This arrangement of hydrophobic and charged motifs is in agreement with peptide-based fibrillar models3,46. Furthermore, given the average fibril width of ~ 300 Å (Fig. 1a) and individual filament width of 48.7 Å, we conclude that an average ELR fibril consists of ~ 6 filaments. These results demonstrate the potential to use our mineralising technology to recreate structural features of the natural amelogenin matrix with high precision and reproducibility.

Mineralisation of the ELR fibrils

Given the key role of β-rich fibrillar structures in enamel development4, we then investigated the mineralisation capacity of our ELR matrix. First, a drop of 5% w/v ELR solution containing 1.5 mM Ca2+ ions and 0.56% v/v hexamethylene diisocyanate (HDI) crosslinker was casted and dried on a TEM grid to form a 50 –100 nm thick ELR coating. TEM imaging confirmed the presence of ELR fibrils within the coating (Fig. 1f, (i)). Upon exposure to a mineralisation solution supersaturated with respect to fluorapatite31, ELR fibrils templated the formation of ~ 20–40 nm thick needle-like amorphous mineral platelets were observed within the ELR matrix at 2 h (Fig. 1f, (ii)), which evolved into highly crystalline fluorapatite nanocrystals after 24 h (Fig. 1f, (iii), Supplementary Fig. 10). These nanocrystals were ~ 50 nm in diameter, ~ 1 μm in length, and aligned along the direction of the ELR fibrils. These geometrical characteristics are similar to those growing within the amelogenin-rich matrix during enamel development4,5, as well as in systems involving amelogenin peptide47 and other protein-templated mineralisation31.

ELR matrix-guided preferential crystal growth along the c-axis

During early enamel development, the amelogenin matrix stabilises and fuses prenucleation clusters via a non-classical crystallisation process, guiding the growth of apatite nanocrystals preferentially along the c-axis4,47. Thus, we focussed on assessing the capacity of the ELR matrix to preferentially control the direction of crystal growth. To do this, we embedded commercially available HAp nanocrystals within the ELR matrix deposited on the TEM grid. After 24 h in the mineralisation solution, precursor amorphous mineral was observed, which may arise by the fusion of prenucleation clusters along the c-axis of the existing nanocrystals (Fig. 2a, blue square), as previously reported47,48. This amorphous mineral undergoes transformation into crystalline structures with identical orientation (Fig. 2a, yellow square), as seen in classical amorphous-to-crystalline transitions during apatite formation27,49. These results demonstrate the capacity of our supramolecular ELR matrix to facilitate the oriented, integrated, and anisotropic growth of the existing crystals, reminiscent of the mineralisation mechanism observed during amelogenesis4,5. In addition, these results were complemented with umbrella sampling simulations, which revealed that more energy is required to remove ELR fragments bound on the a-axis than the c-axis of the apatite nanocrystal (Supplementary Fig. 11), as reported previously for a small 12-mer amelogenin sequence50, thus energetically favouring crystal growth along the c-axis. In contrast, nanocrystals mineralised without an ELR matrix exhibited irregular growth due to the fusion of amorphous mineral along all the crystal axes (Supplementary Fig. 12). Overall, these results confirmed the role of the ELR matrix in directing epitaxial crystal growth preferentially along the c-axis.

Fig. 2: Epitaxial remineralization from different anatomical regions of the tooth.

a TEM image showing the preferential growth of an apatite nanocrystal along the c-axis while embedded within the ELR matrix (representative image from *n *= 3 independent experiments). High-resolution TEM and FFT images are presented in the Supplementary Fig. 13. The lattice lines in the apatite nanocrystal grown within the ELR coating appear blurry due to imaging interference and potential beam-induced damages caused by ionisation and electrostatic charging of the beam-sensitive ELR matrix, a common issue for poorly conducting organic samples. b Illustration of different anatomical parts of a tooth. c Illustration of ELR coating on a section of aprismatic enamel and (d) SEM images of a ~10 μm thick apatite layer grown on aprismatic enamel (representative images from n = 3 independent experiments). e Illustration of ELR coating on a section of prismatic enamel and SEM images of prismatic enamel (f) before and (g) after remineralisation (representative images from n = 8 independent experiments for each group). h Illustration of ELR coating on dentine surface and (i) SEM images of a ~5 μm thick apatite layer grown on dentine surface (representative images from n = 5 independent experiments). Schematic illustrations in (b, c, e, h) were prepared by scientific illustrator Leonora Martínez Nuñez.

Remineralization of enamel and dentine

An ideal enamel regenerative technology would enable remineralisation of the different architectures of the natural tissue (Fig. 2b) and the recovery of its functional properties. Towards this goal, we first assessed the capacity of the ELR matrix to regrow aprismatic enamel. We drop cast 10 μL of 5% w/v ELR solution containing 1.5 mM Ca2+ ions and 0.56% v/v HDI on a 2 mm x 4 mm area of acid-etched aprismatic enamel to form a ~ 10 μm thick ELR coating as observed under SEM (Fig. 2c and Supplementary Fig. 14c) and exposed it to the mineralisation solution supersaturated with respect to fluorapatite31 for 10 days. The ELR matrix induced epitaxial growth of apatite nanocrystals from the enamel-ELR interface through the ELR matrix, forming an integrated and organised mineral layer (Fig. 2d and Supplementary Discussion 4). This layer comprised ~ 50 nm diameter nanocrystals that were several microns in length and exhibited the typical hexagonal apatite morphology (Fig. 2d inset) of native enamel. Then, to assess remineralization of prismatic enamel (Fig. 2e), we similarly deposited a ~ 2 μm thick ELR coating on acid-etched prismatic enamel (Fig. 2f and Supplementary Fig. 14a) and repeated the 10-day mineralisation process, which triggered the epitaxial growth of similar nanocrystals but now maintaining the distinctive prismatic and inter-prismatic architectures (Fig. 2g). Finally, to test the remineralizing potential of the ELR matrix on teeth with completely eroded enamel (Fig. 2h), we deposited a ~ 5 μm thick ELR coating on acid-etched dentine, repeated the mineralisation process, and again observed the epitaxial growth of apatite nanocrystals. In this case, the remineralized layer resembled the architecture of aprismatic enamel but now growing from the dentine surface (Fig. 2i). In all three cases, similar results were obtained using artificial saliva (1.2 mM Ca2+, 0.72 mM PO43−) containing 1 ppm F1− ions to imitate physiological conditions (Supplementary Fig. 15). We used 1 ppm of F1− concentration in artificial saliva, which has been reported to be the average amount present in natural saliva ranging between 0.02 ppm to 1.93 ppm51. We have conducted experiments using three different thicknesses of the ELR coating to demonstrate (i) the tuneability of our process to deposit coatings of specific thicknesses (Supplementary Fig. 14) and (ii) that nanocrystal growth is defined by the ELR coating, generating a mineralised layer of equal thickness to the ELR coating (Supplementary Figs. 16, 17). In contrast, tissue sections without ELR coating developed misoriented crystals (Supplementary Fig. 18) commonly found in mineralisation processes without organic matrices52.

These results suggest that our technology could potentially provide a one-pot solution for the regeneration of dental enamel independently of the level of tooth erosion. To the best of our knowledge, this capacity to regenerate enamel from different anatomical regions of teeth expanding from aprismatic enamel, prismatic enamel, and exposed dentine has not been achieved before. Here, we define regeneration as the capacity to recreate the architecture of the different anatomical regions of enamel and regain its functional properties. Thus, given the potential clinical implications of such a versatile regenerative technology, we then performed in depth investigations of the remineralisation from each of these anatomical regions.

Regeneration of prismatic enamel

We fabricated a uniform ~2 μm thick ELR coating on an acid-etched prismatic enamel (Fig. 3a, b and Supplementary Fig. 19). Upon mineralisation, the ELR coating induced the epitaxial growth of apatite nanocrystals from the enamel-ELR interface through the ELR matrix (Fig. 3c and Supplementary Fig. 20), recovering the native architecture of both diazone (Fig. 3d and Supplementary Fig. 21) and parazone (Fig. 3e and Supplementary Figs. 22, 23) prisms as well as inter-prismatic regions (Fig. 3d, e). Similarly, a 5 μm thick ELR coating on prismatic and aprismatic regions of enamel triggered the uniform growth of a ~ 5 µm thick mineral layer over large and uneven areas, recreating the microstructure of both enamel anatomies (Supplementary Fig. 15). While the ELR matrix slowly degrades during the mineralisation process, it remains stable and functional during the enamel remineralization process (Supplementary Fig. 24 and Supplementary Discussion 5). To expedite this ELR degradation process and expose the underlying remineralised layer, we treated the samples with elastase after the remineralisation process was completed (Fig. 3d, e and Supplementary Figs. 21–23). Nanoindentation and microtribological tests were then conducted to assess the capacity of this remineralised enamel to restore key mechanical properties of the native tissue, including Young’s modulus (E), hardness (H), coefficient of friction (CoF), and specific wear rate (SWR). As expected, acid-etched enamel displayed dramatically lower E (36.9 ± 14.3 GPa) and H (1.1 ± 0.6 GPa) compared to E (80.7 ± 18.