Serine-16 phosphorylation, c-terminal truncation, and ion-specific interactions coordinate amelogenin nanoribbon formation

Abstract

Amelogenin self-assembly is a critical step in enamel matrix organization, supporting the spatial confinement of transient ACP and orchestrating hydroxyapatite nucleation and ordered growth. Studies have demonstrated that amelogenin self-assembles into nanoribbons by forming an amyloid-like structure that matches the cross-β pattern observed in the developing enamel matrix. However, molecular determinants remain unclear. In this work, we demonstrate that Ser16 phosphorylation (pS16) and Mmp20-mediated C-terminal truncation exert complementary control over amelogenin self-assembly and templated mineralization. Using advanced microscopy and spectroscopic methods, we show that C-terminal truncation lowers the energetic threshold for the disordered-to-ordered β-sheet phase transformation, producing rapid nanoribbon growth, whereas pS16 prolongs this transformation barrier and channels assembly into a gradual, more orderly templating process. Remarkably, pS16 alone is sufficient to drive nanoribbon formation under ion-free conditions, facilitating ion-mediated charge compensation at the self-assembly N-terminal domain that otherwise requires calcium and phosphate ions. These ions play complementary roles: phosphate appears to promote longitudinal elongation through interactions with protonated His-rich motifs, whereas calcium strengthens lateral cohesion and bundling by binding acidic C-terminal residues and engaging with the pS16 site, which could also foster ACP accumulation along the nanoribbon central zone. In a revised structural model, this central zone likely represents a longitudinal interface between two strands of beta-sheets, forming a steric-zipper by interactions of isoleucine and phenylalanine. Together, these results establish a framework in which pS16, C-terminal truncation, and ion-specific interactions cooperate to fine-tune amelogenin self-assembly and direct ribbon-like mineral formation in enamel.

Introduction

Dental enamel is a remarkable mineralized tissue known for its exceptional mechanical properties and intricate architecture [1, 2]. Its formation is a highly orchestrated process reliant on the precise interplay between matrix proteins, enzymatic activity, and mineral precursors, unfolding in two main stages: secretory and maturation [3, 4]. The secretory stage involves ameloblast secretion of an amelogenin-rich matrix that self-assembles and interacts with other enamel proteins and proteases like matrix metalloproteinase 20 (Mmp20) to create a structured system capable of influencing mineral phase transformation, mineral particle shape, and their self-alignment. Maturation involves ameloblast secretion of proteolytic enzymes, mainly kallikrein 4 (Klk4), to degrade enamel proteins and allow the growth in width and thickness of thin enamel crystallites initially deposited along the protein scaffold to achieve its final hardness [4-6].

Amelogenin, accounting for approximately 90–95% of the enamel matrix during the secretory stage [7], is widely recognized as a critical regulator of enamel matrix organization and biomineralization. In AMELX knockout mice, the absence of amelogenin drastically alters enamel architecture, leading to a reduction in enamel thickness by nearly 90%. Moreover, the typical rod and interrod organization is lost and replaced by disorganized octacalcium phosphate-like structures[8-11]. Similarly, in humans, X-linked forms of amelogenesis imperfecta (MIM #301200), resulting from mutations in the AMELX gene, exhibit milder phenotypes characterized by hypoplastic enamel and structural disorganization [12, 13].

Full-length amelogenin is an amphiphilic molecule comprising three likely functionally distinct regions [14, 15] - Fig. 1. The N-terminal domain, commonly referred to as the tyrosine-rich amelogenin peptide (TRAP), contains the only known phosphorylation site at serine-16, a posttranslational modification that has been implicated in self-assembly events and stabilization of amorphous calcium phosphate (ACP) [16-20]. The central domain is predominantly hydrophobic, characterized by a high content of proline (P=32%), glutamine (Q=21%), and histidine (H=12%) residues, which contribute to the protein’s intrinsic disorder and monomer flexibility [21, 22]. Finally, a C-terminal domain that is hydrophilic and comprises charged cationic and anionic residues, indicating a key role in stabilizing interactions with mineral ions and mineral surfaces [23-25]. Domain-specific functions of amelogenin become increasingly prominent in developing enamel, particularly those of the N-terminal and C-terminal regions, as Mmp20-mediated cleavage progressively isolates these structural motifs [6, 11, 26], allowing them to perform distinct yet complementary roles within the enamel matrix.

Fig. 1.

Sequence map of recombinant human full-length amelogenin (rH174) highlighting the self-assembly domain P2 (box) and major MMP20 cleavage product. Proteolytic processing by Mmp20 during the secretory stage releases the hydrophilic C-terminal domain (AA147–174), yielding a predominant N-terminal fragment, rH146 (AA1–146). Recombinant human forms used in this study lack the initial methionine residue (M) present in the native sequence (UniProt ID: Q99217-1).

In vitro studies have demonstrated that amelogenin forms nanospheres [22, 27-31] and nanoribbon-like [32-38] assemblies in aqueous environments, with the resulting morphology depending on factors such as protein concentration, pH, and the composition and strength of the ionic milieu. As initially proposed by Fincham et al. [39], the dominant belief for many years was that amelogenin forms nanospheres with the N-terminus oriented inward and the C-terminal tail outward. These structures were thought to shape newly formed HAp crystals, restricting lateral crystal growth until their enzymatic removal during maturation [40]. While providing a basic framework, the classical theory of enamel biomineralization presents significant inconsistencies with in vivo observations [41]. The substantial lateral crystal growth during the transition and early maturation stage in wild-type mice and the extensive mineralization observed in Klk4-deficient mice challenge the hypothesis of sphere-mediated inhibition of lateral crystal growth [42, 43]. More importantly, the theory also lacks a satisfactory explanation for how enamel mineral ribbons' unique shape and spatial organization are established before they become crystalline [44]. The oddity lies in how such spheres could consistently mold the initial ACP into the precisely organized mineral ribbon, demanding a yet-unidentified mechanism for this highly ordered process.

Previous research from our group showed that, when incubated in calcium- and phosphate-rich environments, recombinant human amelogenin redirects the assembly pathway toward the formation of tens of micrometer-long nanoribbons with amyloid-like characteristics [32, 36]. These nanoribbons, typically measuring ~17 nm in width and ~3-4 nm in height [34, 35, 45]. Evidence suggests that although the secondary structure of amelogenin is predominantly disordered under physiological conditions [46], a localized β-sheet motif within the N-terminal region, particularly in the P2 domain containing the pS16, can initiate a disorder-to-order transition during self-assembly [32, 36]. This conformational shift promotes the formation of β-sheet-rich architectures and facilitates aggregation into amyloid-like intermediates, stabilized through Ca²⁺ bridging. The morphological and dimensional resemblance between these in vitro structures and enamel ribbons found in vivo[47], combined with growing evidence for their role in templated mineralization [33, 48], has prompted new perspectives on organic matrix-templated enamel biomineralization.

Studies on amelogenin nanoribbons have significantly advanced our understanding of enamel matrix organization, yet key mechanistic questions remain unresolved. Structural and biochemical factors governing the nanoribbons formation pathway, particularly domain-specific interactions and conformational changes, remain poorly investigated. A particularly intriguing factor is the unique phosphorylation at Ser16 residue, a post-translational modification commonly found in bone and dentin [49, 50]. In vivo studies using a knock-in mouse model in which Ser16 was substituted with alanine revealed hypoplastic and mechanically weak enamel characterized by irregular surfaces and ectopic, spherically shaped mineral deposits [19, 20, 51]. Ultrastructural analysis further demonstrated that disruptions in matrix assemblage profoundly affect enamel formation, leading to the absence of a prismatic architecture [51], alterations in mineral composition and crystal orientation, and a marked reduction in enamel hardness [19]. These findings underscore pS16 as a central regulatory element that influences the mineralization and emergence of enamel’s hierarchical structure. In this context, in vitro studies using pS16 recombinant porcine amelogenin demonstrated a strong inhibitory effect on ACP stabilization and control of HAp crystallization, in contrast to the non-phosphorylated counterpart [52-54]. Importantly, while phosphorylation exerts a pronounced regulatory influence on mineralization, self-assembly studies from the same group revealed that pS16 does not fundamentally alter the intrinsic assembly pathway of amelogenin; rather, it acts as a context-dependent modifier that fine-tunes assembly behavior [52, 55]. Despite its potential significance, the mechanistic investigation of amelogenin phosphorylation has been constrained by the widespread use of recombinant amelogenin expressed in E. coli, which lacks the enzymatic machinery for phosphorylation. While phosphorylation at Ser16 has been explored in porcine systems, comparable investigations in human amelogenin, particularly across full-length and physiologically relevant MMP20 cleavage products, have been scarce. Addressing this gap, our research group has taken a pioneering step by successfully synthesizing a phosphorylated version of full-length human amelogenin (rH174) and its major MMP20 cleavage products during the secretory stage of amelogenesis (rH146) using a third-generation genetic code expansion strategy [56] validated earlier for recombinant murine amelogenin [57]. This technological advancement enables, for the first time, direct investigation of how phosphorylation influences the assembly behavior of human amelogenin. Therefore, our study was designed to investigate the regulatory effects of pS16 and enzymatic C-terminal truncation by MMP20 on amelogenin self-assembly, focusing on their roles in directing the formation of nanoribbon-like structures and their interaction with calcium and phosphate ions.

Experimental Section

Study Design

To gain a comprehensive understanding of the effects of pS16 on amelogenin self-assembly, we performed in vitro assays using phosphorylated and non-phosphorylated recombinant human forms of rH174 and rH146 at a protein concentration of 1 mg mL⁻¹. Experiments were conducted at 37 °C and pH 5.0 to 5.5 over four weeks to capture both early and late stages of assembly. The pH was adjusted using dilute NaOH prior to incubation using a calibrated microelectrode. To minimize water evaporation during long-term incubation, samples were sealed in screwcap microtubes, and tube opening was minimized to 1 – 2 times/week. Bacterial contamination was prevented by sterile filtration of all solutions and by the addition of sodium azide (NaN₃, 0.62 mM).

Samples were incubated under four distinct ionic conditions to dissect the roles of calcium and phosphate in amelogenin assembly: (1) deionized water, (2) phosphate (20.4 mM KH₂PO₄/K₂HPO₄), (3) calcium chloride (33.3 mM CaCl₂), and combined calcium and phosphate (20.4 mM KH₂PO₄/K₂HPO₄ + 33.3 mM CaCl₂). Elevated concentrations of calcium and phosphate were selected to promote measurable assembly within experimentally accessible timescales, given the reduced kinetics observed under low-salt conditions (data not shown), and to facilitate systematic assessment of ion-mediated modulation of amelogenin self-assembly. Under these conditions, at pH 5.0 to 5.5, combined CaP condition is strongly supersaturated with respect to multiple calcium phosphate phases (HAp, OCP, DCPD, TCP) and would therefore be expected to undergo spontaneous precipitation in the absence of regulatory factors. However, no macroscopic precipitation was observed during incubation, consistent with stabilization of calcium and phosphate ions in a dispersed, noncrystalline state through interactions with amelogenin (Fig. S1). These ions concentrations fall within the range commonly used in other in vitro amelogenin assembly experiments, where similar trends have been observed [32, 34, 35, 38, 58, 59].

Nanostructural organization was monitored using transmission electron microscopy (TEM) along with atomic force microscopy (AFM), including imaging in fluid and air modes (tapping mode and nanomechanical mapping). In addition, Attenuated total reflection (ATR) – Fourier-transform infrared spectroscopy (FTIR) was employed to characterize secondary structure and potential conformational changes upon phosphorylation and ionic environment.

Recombinant human amelogenin synthesis and pS16 phosphorylation

Non-phosphorylated recombinant human amelogenins (rH174 and rH146) were expressed in Escherichia coli BL21(DE3) cells and purified under denaturing conditions, as described elsewhere [60]. Site-specific phosphorylation at serine-16 was achieved using a modified third-generation genetic code expansion strategy, as defined by Buchko et al. [57, 61] Bacterial strain and vectors for phosphoprotein expression were gifts from Richard Cooley, Ryan Mehl, and Jason Chin, respectively (Add gene: #197655, #174075, and #173897). Protein purity exceeded 95%, as confirmed by high-performance liquid chromatography (HPLC). Phosphorylation was validated by matrix-assisted laser desorption/ionization mass spectrometry (MALDI–MS) and Phos-tag gel electrophoresis (Fig. S2).

ATR-FTIR spectroscopy

An aliquot (30 μL) of each sample was deposited onto prepared substrates consisting of metal discs covered with the dull side of aluminum foil as described previously [62]. Samples were air-dried at room temperature for 2 hours and analyzed by ATR-FTIR using a Spotlight 400 FT-IR Imaging System (PerkinElmer, MA, USA) equipped with a room-temperature triglycine sulfate (TGS) detector. Samples were placed in direct contact with a diamond crystal. The crystal was cleaned with 70% ethanol between each sample to prevent cross-contamination, and a new background spectrum was acquired after every ten measurements to ensure signal consistency.

Spectra were collected in the 900 – 1800 cm⁻¹ range at a spectral resolution of 4 cm⁻¹, with four accumulations per run. Raw spectra were processed using the PerkinElmer FT-IR software. Spectra were baseline-corrected using a polynomial baseline subtraction and normalized over the entire spectral range to account for intensity variations between measurements. The Amide I region (1600–1700 cm⁻¹) was isolated for analysis, and peak positions were determined from the minima of the transmittance spectra. These values were used for comparative analysis of conformational changes among different supramolecular assemblies.

Quantitative deconvolution of the Amide I region was performed using Gaussian peak fitting in OriginPro 2025. Peak positions were initially constrained based on established assignments and allowed to vary within ±5 cm⁻¹ during fitting: 1625 cm⁻¹ for β-sheets, 1650 cm⁻¹ for random coils and α-helices, 1670 cm⁻¹ for β-turns, and 1690 cm⁻¹ for secondary β-sheet structures. Peak assignments were based on established references for protein secondary structure IR signatures [63-66]. The relative contribution of each secondary structure was determined by calculating the area under the fitted peaks and is reported as a percentage [46]. Fit quality was assessed using the reduced χ² value and the coefficient of determination (R²). Only fits with R² > 0.98 and randomly distributed residuals were accepted.

TEM imaging

Samples (10 μL) were blotted onto carbon-coated (5–10 nm), formvar-stabilized copper grids and allowed to absorb for 10 min. Grids were then gently rinsed with deionized water and stained with 2% NanoW^™ (organotungsten compound) for 1 min. Imaging was performed using a Tecnai T12 transmission electron microscope (JEM-1400, JEOL, Tokyo, Japan) operating at 120 kV and equipped with a 4k Gatan Ultra Scan CCD camera (Gatan, USA). Nanostructure morphology, dimensions, and bundling were analyzed using Fiji software, v2.14.0, 2023. [67]

AFM imaging

Sample aliquots (15 μL) of each sample were deposited onto freshly cleaved mica substrate and incubated for 15 minutes in a wet cell to facilitate surface immobilization and avoid evaporation. After incubation, excess liquid was carefully removed by touching the edges of the mica with filter paper, avoiding direct contact with the sample surface to prevent disturbance of the adsorbed material. The surface was then rinsed with 100 μL of deionized water, carefully wicked away at the edges using filter paper, and immediately dried using dust-free canned air.

AFM imaging was conducted in air using a Bruker Multimode 8 system equipped with a Nanoscope V controller, operating in either Tapping mode or PeakForceTM Quantitative Nanomechanical Mapping mode to characterize surface morphology and nanomechanical properties. Scans were acquired using a standard tapping mode AFM probe (160AC-NA, OPUS, OPUS Instruments) featuring a nominal tip radius of <7 nm, a spring constant of approximately 26 N/m, and a resonance frequency of 300 kHz. For QNM maps, a calibrated RTESPA-300 probe (Bruker), featuring a nominal tip radius of ~8 nm, a spring constant of ~40 N/m, and a resonance frequency of 300 kHz, was used. The system was calibrated using a polystyrene reference sample (elastic modulus = 2.7 GPa) to extract quantitative values for modulus, adhesion, and deformation based on the Derjaguin–Müller–Toporov (DMT) model[67]. Images were collected with a resolution of 512 × 512 pixels and a scan rate of 0.5–1.0 Hz, and subsequently analyzed using Nanoscope Analysis software, version 1.8 (Bruker), and Gwyddion software, version 3.3 (2025).

Evaluation of kinetics and maturation behavior of nanoribbon assembly

Nanoribbon assembly kinetics were quantified from 2 independent experiments using longitudinal AFM/TEM analysis designed to capture the morphological transition from oligomeric assemblies to nanoribbon-dominated states with time. AFM/TEM images were acquired at predefined incubation times under identical imaging conditions, and three non-overlapping images acquired at the same magnification were selected from each sample at each time point over a 4 to 6-week period. Image selection was based solely on imaging quality and surface accessibility, without regard to the apparent assembly state. Each image was classified using an ordinal scoring scheme applied consistently across all datasets to reflect the relative prevalence of nanoribbon structures within the field of view as follows: 0, exclusively oligomeric or globular features; 1, fewer than five isolated ribbon-like structures or minimal surface coverage; 2, moderate nanoribbon presence with ribbon-like structures distributed across a substantial portion of the surface; and 3, nanoribbon-dominated fields in which ribbon-like morphologies constituted the majority of surface features.

For each time point and condition, nanoribbon scores from the three images were averaged and normalized to the theoretical maximum score to yield a dimensionless estimate of nanoribbon yield ranging from 0 to 1. These normalized values were plotted as a function of incubation time and fitted using a variable-slope logistic model (four-parameter sigmoidal fit) implemented in GraphPad Prism version 5.0 for Windows, GraphPad Software, www.graphpad.com. The lower asymptote was constrained to zero, reflecting the absence of nanoribbons at early incubation times, while the upper asymptote was either constrained to unity or allowed to vary freely depending on whether complete conversion to nanoribbons was experimentally observed. From these fits, the transition midpoint (t50), Hill slope (describing the sharpness of the oligomer-to-nanoribbon transition), and goodness of fit (R²) were extracted for each recombinant amelogenin variant under each ionic condition.

Lag-phase and plateau times were defined independently of the sigmoidal fitting based on direct morphological observations. The lag phase was identified as the earliest time point at which nanoribbons were consistently detectable across the three images, whereas the plateau corresponded to the time point at which nanoribbon morphology became stable and no longer increased appreciably with further incubation. These landmark-based descriptors were used to corroborate trends observed in the fitted kinetic parameters.

Morphometric measurements

Morphometric analysis of amelogenin assemblies was performed using TEM and AFM images acquired from three to four independent experiments. TEM images were used to quantify nanostructure width and length, and measurements were conducted using ImageJ/Fiji [68]. Manual object selection and measurement were performed using the ROI Manager in combination with the Point Tool to ensure consistent and unbiased sampling. For width analysis, high-magnification TEM images were acquired to resolve individual structures. Measurements were performed on non-rounded morphologies (protofibrils and nanoribbons) by drawing line or point-to-point measurements perpendicular to the long axis of each structure. Multiple measurements were collected along the length of each object to account for local width variations, and the resulting values were pooled for statistical analysis. For length analysis, lower-magnification TEM images were acquired to enable visualization of extended nanostructures over micrometer length scales. This approach allowed accurate measurement of nanoribbons with lengths of up to approximately 4 μm within a single field of view. The full end-to-end contour of isolated nanostructures was measured directly from TEM micrographs, excluding overlapping objects, aggregates, or structures truncated at image boundaries. Only well-resolved, isolated assemblies were included in the analysis. For each recombinant construct, mean values and standard deviations were calculated from measurements of 100 individual nanostructures. All measurements were calibrated using the corresponding image scale bars, and datasets were pooled from multiple independent images to ensure statistical robustness.

Because the geometry of the AFM tip generally causes lateral broadening in measured features, AFM images were used to quantify only height distributions. Vertical (z-axis) measurements remain largely unaffected by tip shape, allowing reliable determination of fibril heights. The resolution along the z-axis is primarily constrained by instrumental noise, which typically remains within a few tenths of a nanometer under standard imaging settings [69]. Height measurements were obtained from flattened AFM topography images acquired over scan areas of 3.3 × 3.3 μm or smaller. Line profiles were drawn perpendicular to the long axis of individual nanostructures, avoiding regions of overlap or aggregation. Multiple profiles were collected along each object to account for local height variations, and the mean peak height from these profiles was used to define the height of each individual nanoribbon. For each recombinant construct, mean values and standard deviations were calculated from measurements of 50 individual nanostructures. Only well-resolved, isolated structures were included in the analysis. All measurements were calibrated using the corresponding image scale bars, and datasets were pooled from multiple independent images to ensure statistical robustness.

All quantitative data are presented as mean ± standard deviation. Statistical comparisons among multiple groups were performed using one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparisons test. Statistical significance was defined as p < 0.05. All statistical analyses were performed using GraphPad Prism version 5.0 for Windows, GraphPad Software, www.graphpad.com.

Computational method for in silico electrostatic surface potential landscape prediction

The H175 and H147 amelogenin structure was obtained from the AlphaFold Protein Structure Database, and protonation states were assigned for pH 5.5 using the PDB2PQR software suite with the PROPKA algorithm. Atomic charges and radii were assigned using the AMBER force field.

Electrostatic surface potentials were calculated using the Adaptive Poisson-Boltzmann Solver (APBS v3.4.1). The finite-difference Poisson–Boltzmann equation (PBE) was solved on a high-resolution grid centered at the protein’s center of mass (0.4645, −9.0470, 1.8250 Å). A grid dimension of 385 × 385 × 385 points and a physical box size of 38.5 Å per axis were used for both the coarse and fine grids, providing sufficient resolution and spatial coverage for modeling the electrostatic environment around the ~3 nm-diameter protein.

Electrostatic potentials were computed under four ionic environments: DI water (control), 33 mM CaCl₂: Ca²⁺ = 33 mM, Cl⁻ = 66 mM, and 20 mM KH₂PO₄: H₂PO₄⁻ = 20 mM, K⁺ = 20 mM. All simulations were performed with a solvent dielectric constant of 78.5 and a solute dielectric constant of 2.0 at a temperature of 298.15 K. An ion exclusion radius of 2.4 Â was applied to account for the effective hydration shell surrounding each ion. Surface electrostatic potential maps were generated in.dx format for all ionic conditions. Visualization and qualitative comparison of surface potentials were performed using ChimeraX and custom Python scripts, which extracted and averaged the electrostatic potential per residue.

Results

Amelogenin nanoribbons assemble through a conserved hierarchical pathway

Using AFM and TEM analysis, all four recombinant human variants (pS16rH174, rH174, pS16rH146, rH146) initially assemble into nanospherical oligomers and follow a conserved hierarchical transition to nanoribbons (Fig. S3), as illustrated for rH146 as a representative example (Fig. 2a). Nanospheres aggregate over days to form short strings of 3 to 10 nanospheres growing into a worm-like structure, which become the precursors for nanoribbons, and are therefore termed “protoribbons”. The width of nanospheres, measured by TEM analysis, decreased slightly when transitioning into protoribbons while their height, determined by AFM, remained stable (Fig. 2b). With continued incubation of the amelogenin suspensions a transition of the protoribbons into nanoribbons was observed starting as early as day 3 as shown in the AFM time-lapse analysis (Fig. S4 - S6) and became the predominant morphology after 2 to 4 weeks under all conditions studied. The transformation from the worm-like short protoribbons to nanoribbons is linked to a significant change in both width and thickness of the self-assembled structures. Upon transformation into nanoribbons, protoribbons increased in width while their thickness was reduced (Fig. 2b), indicating rearrangement of amelogenin molecules associated with a critical conformational transition from disordered oligomeric assemblies to more ordered nanoribbon precursors. This tendency may be explained by the conformational noise emanating from intrinsically disordered proteins [70-72], which generates a rugged energy landscape that favors the accumulation of heterogeneous oligomeric states before their reorganization into more ordered assemblies [73, 74].

Fig. 2.

AFM imaging and morphometric quantification of hierarchical amelogenin self-assembly. [a] AFM height images of rH146 after showing the structural evolution of amelogenin assemblies, progressing from spherical oligomers to protoribbons and mature nanoribbons. This assembly pathway appears to be conserved among recombinant human amelogenins, as illustrated in Fig. S2. The samples were incubated with 33 mM CaCl₂ and 20 mM KH₂PO₄ for 30 days at pH 5.5 and 37 °C. Scale bars: 20 nm. [b] TEM and AFM-based morphometric analysis of spheres, worm-like proto-ribbons, and mature nanoribbons formed by pS16rH174/rH174 and pS16rH146/rH146, illustrating progressive conformational transitions during assembly. Data are mean ± SD; one-way ANOVA with post-hoc testing. Significance levels: ** p < 0.01, *** p < 0.001.

The conformational reorganization observed in Fig. 2a is directly reflected in the Amide I FTIR spectra (1600–1700 cm⁻¹), where assemblies grouped by dominant supramolecular organization exhibit a progressive shift in peak position from ~1650–1660 cm⁻¹ to ~1630 cm⁻¹ (Fig. S7). Peak deconvolution further reveals systematic changes in secondary-structure composition (Table 1; Fig. S8 - S9), characterized by an increasing proportion of β-sheet–associated bands and a corresponding decrease in random coil/α-helical contributions as assemblies evolve from spheres to protoribbons and ultimately mature nanoribbons. These changes indicate enhanced molecular ordering and stabilization of intermolecular β-sheet interactions during nanoribbon formation.

Table 1.

Amide I maximum peak positions and quantitative FTIR peak deconvolution of amelogenin assemblies at different hierarchical morphological stages.

	pS16rH174			rH174
	Sphere	Protoribbons	Nanoribbons	Sphere	Protoribbons	Nanoribbons
Peak-Max (cm−1)	1,654	1,643	1,637	1,646	1,642	1,638
Assignment (%)
ß-sheet	39	69	67	45	46	57
Random coil/α-helix	37	31	33	42	41	33
β-turn	17	*	*	13	12	10
Secondary ß-sheet	7	*	*	*	*	*
	pS16rH146			rH146
	Sphere	Protoribbons	Nanoribbons	Sphere	Protoribbons	Nanoribbons
Peak-Max (cm−1)	1,655	1,641	1,632	1,666	1,642	1,633
Assignment (%)
ß-sheet	46	49	56	14	49	56
Random coil/α-helix	40	40	33	46	39	33
β-turn	13	11	11	31	12	11
Secondary ß-sheet	*	*	*	9	*	*

Samples were incubated with 33 mM CaCl₂ and 20 mM KH₂PO₄ at pH 5.5 and 37 °C and sampled at three hierarchical stages: spheres, protoribbons, and mature nanoribbons. Amide I maximum peak positions (cm⁻¹) were determined directly from the original spectra (Fig. S4), whereas relative band areas (%) were obtained from Gaussian deconvolution of the Amide I region (1600–1700 cm⁻¹) using fixed peak positions corresponding to β-sheet, random coil/α-helix, β-turn, and high-frequency β-sheet components. Representative deconvoluted spectrum and fitted components are shown in Fig. S8 and S9. Asterisks (*) indicate peak maxima falling outside the spectral window analyzed or values that could not be reliably assigned.

Amelogenin pSer16 and C-terminal truncation modulate assembly kinetics

Analysis of assembly kinetics reveals sigmoidal nanoribbon growth with variant-specific lag phases, demonstrating that pSer16 delays, while C-terminal truncation accelerates nanoribbon formation (Fig. 3a and Fig. S4 - S6). Full-length pS16rH174 exhibited the slowest assembly among the four variants, remaining largely in a lag phase through Day 17, with only occasional short nanoribbons observed, and extensive assembly becoming evident only after Day 21. In contrast, non-phosphorylated rH174 showed a delayed onset initially, but detectable nanoribbons formed already around Day 14 and became preeminent by Day 17. Interestingly, both full-length proteins exhibited a burst-like behavior, where the extended lag phase is followed by a rapid formation of short nanoribbons that agglomerate laterally, forming thick bundled structures (Fig. 3b and Fig. S4 - S6). C-terminal truncation significantly enhanced assembly rates. rH146 displayed the most rapid and extensive assembly behavior, with abundant elongated structures already present at Day 7 and a dense fibrillar network by Day 21. pS16-rH146 produced well-defined elongated ribbons as early as Day 14 and reached high nanoribbon density by Day 21. Notably, truncated variants predominantly formed long nanoribbons with minimal bundling relative to full-length constructs.

Fig. 3.

TEM imaging and time-dependent assembly kinetics of phosphorylated and non-phosphorylated amelogenin variants in the presence of CaCl₂/KH₂PO₄. [a] Sigmoidal nanoribbon assembly kinetics reveal delayed transition in Ser-16-phosphorylated variants and accelerated nanoribbon formation upon C-terminal truncation. Estimated nanoribbon yield (normalized, 0–1) was plotted as a function of incubation time and fitted using a variable-slope logistic model (four-parameter fit). For each recombinant amelogenin variant, fitting parameters are: pS16rH174 (t50 = 18.9 days, Hill slope = 0.18, R² = 0.99), rH174 (t50 = 14.2 days, Hill slope = 0.20, R² = 0.99), pS16rH146 (t50 = 15.2 days, Hill slope = 0.09, R² = 0.95), and rH146 (t50 = 6.3 days, Hill slope = 0.19, R² = 0.99). Experiments were conducted over 30 days of incubation in the presence of 33 mM CaCl₂ and 20 mM KH₂PO₄ at pH 5.5 and 37 °C. [b] Representative negatively stained TEM images acquired at the plateau phase illustrate variant-dependent agglomeration behaviors, with extensive lateral bundling in full-length variants and comparatively reduced bundling in truncated constructs. Scale bar: 500 nm.

Amelogenin pS16 overrides ionic requirements in nanoribbon assembly

Under calcium- and phosphate-free conditions, representative TEM images illustrate that phosphorylated amelogenins transition from protoribbons to nanoribbons with enhanced parallel alignment and bundling; however, nanoribbon conversion remained incomplete, as disordered oligomers persist even after 40–50 days of incubation (Fig. S10 - S11, Fig. 4a; Tables 2 and 3). Nanoribbons formed in deionized water and CaCl₂/KH₂PO₄ exhibit a similar morphology (Fig. S12), including a central dark line corresponding to the mature nanoribbon core, indicating conservation of the underlying packing architecture. Notably, nanoribbons assembled in deionized water show reduced dimensions relative to CaCl₂/KH₂PO₄, suggesting tighter packing when Ca²⁺ and PO₄⁻³ stabilization are absent (Fig. S13).

Fig. 4.

TEM imaging and time-dependent assembly kinetics of phosphorylated and non-phosphorylated amelogenin variants in the absence of CaCl₂/KH₂PO₄. [a] Representative negatively stained TEM images acquired at the plateau phase in [b] (pS16rH174=45 days, rH174=28 days, pS16rH146=42 days, and rH146=28 days) illustrate that phosphorylated amelogenin forms nanoribbons that progressively bundle into dense aggregates, accompanied by persistent disordered oligomers, reflecting incomplete conversion into nanoribbons. Scale bar: 200 nm. [b] Sigmoidal nanoribbon assembly kinetics reveal that nanoribbon formation can occur without Ca²⁺ and PO₄³⁻ in phosphorylated amelogenin variants, albeit with a pronounced rightward shift reflecting an extended lag phase. Estimated nanoribbon yield (normalized, 0–1) was plotted as a function of incubation time and fitted using a variable-slope logistic model (four-parameter fit). For each recombinant amelogenin variant, fitting parameters under CaP conditions are reported in Fig. 3 caption, whereas fits obtained under DI water are: pS16rH174 (t50 = 24.3 days, Hill slope = 0.10, R² = 0.97) and pS16rH146 (t50 = 22.3 days, Hill slope = 0.06, R² = 0.96). Experiments were conducted over 50 days of incubation in with DI water at pH 5.5 and 37 °C.

Table 2:

Comparative effects of calcium and phosphate ions on structural organization, nanoribbon formation, and bundling behavior of full-length amelogenin.

	p16rH174				rH174
Feature	No ions	Ca2+ only	PO43− only	Ca2+/ PO43−	No ions	Ca2+ only	PO43− only	Ca2+/ PO43−
Assembly pathway	NCC*	NCC*	Unusual**	NCC*	Arrested Nucleation	NCC*	NCC*	NCC*
Lag Phase	4th week	3rd week	2nd week	3rd week	-	3rd week	4th week	2nd week
B-sheet/random coil ratio ¥	1.93	2.33	1.78	2.13	1.07	1.77	2.10	2.03
Predominant structure type	Nanoribbons/Oligomers	Nanoribbons/Bundles	Filaments/Bundles	Nanoribbons/Bundles	Oligomers	Nanoribbons/Bundles	Nanoribbons	Nanoribbons/Bundles
Nanoribbon width nm (sd) #	15.3 (1.5)	17.6 (1.8) b	9.65 (1.5) c	19.35 (2.0) a	-	19.9 (1.5) a	15.3 (1.7) b	19.1 (2.7) a
Nanoribbon length nm (sd) #	92.8 (54.3)	173.9 (126.1) b	277.61 (152.8) a	102.89 (27.5) c	-	129.9 (98.9) a	93.6 (46.1) b	86.1 (29.9) b
Nanoribbon curvature	Curved, flexible	Straight, rigid	Curved, flexible	Straight, rigid	-	Straight, rigid	Curved, flexible	Straight, slightly flexible
Degree of bundling	High	High	Moderate	High	-	High	Low	High
Parallel alignment	High	High	High	High	-	High	Low	High
Bundle density	Dense	Dense	Dense	Dense	-	Dense	Loose	Dense

^*Assembly mechanism as described in Fig. 1.

^**Alternative pathway: slender, well-defined filamentous precursors emerging from a diffuse region.

^#Values are presented as mean (standard deviation) of nanoribbon widths and lengths in nanometers (nm) based on N = 100 measurements per condition, measured at 4 weeks of incubation. Superscript letters denote statistically significant differences within each recombinant human form as determined by two-way ANOVA followed by Tukey’s HSD post hoc test (p < 0.05).

^¥β-sheet/random coil ratio at week 4.

Table 3:

Comparative effects of calcium and phosphate ions on structural organization, nanoribbon formation, and bundling behavior of C-terminal truncated amelogenin.

	p16rH146				rH146
Feature	No ions	Ca2+ only	PO43− only	Ca2+/ PO43−	No ions	Ca2+ only	PO43− only	Ca2+/ PO43−
Assembly pathway	NCC*	NCC*	NCC*	NCC*	Arrested Nucleation	NCC*	NCC*	NCC*
Lag Phase	3rd week	2nd week	3rd week	2nd week	-	1st week	1st week	1st week
B-sheet/random coil ratio	2.57	2.84	3.35	2.23	0.84	3.16	4.61	2.22
Predominant structure type	Nanoribbons/Bundles	Nanoribbons/Bundles	Nanoribbons	Nanoribbons	Oligomers	Nanoribbons	Nanoribbons	Nanoribbons
Nanoribbon width nm (sd) #	17.1 (1.5)	17.9 (1.8) b	19.2 (1.6) a	18.2 (1.6) b	-	18.9 (1.7) a	17.4 (2.3) c	18.0 (1.6) b
Nanoribbon length nm (sd) #	804.9 (507.3)	328.8 (247.1) c	1467 (772.7) a	1084.0 (1092.2) b	-	265.5 (153.7) c	2295 (1171.2) a	1495.4 (1302.9) b
Nanoribbon curvature	Curved, flexible	Straight, rigid	Curved, flexible	Straight, rigid	-	Straight, rigid	Curved, flexible	Curved, flexible
Degree of bundling	High	Moderate, mesh-like	Low	Low	-	Moderate	None	Low
Parallel alignment quality	High	Moderate	Absent	Moderate	-	Low	Absent	Low
Bundle density	Dense	Dense	Loose	Loose	-	Dense	-	Loose

^*Mechanism as described in Fig. 2. Upward and downward arrows denote statistically significant differences compared with the CaP-incubated condition (positive control; physiologically relevant). The equals sign indicates no statistically significant difference.

^#Values are presented as mean (standard deviation) of nanoribbon widths and lengths in nanometers (nm) based on N = 100 measurements per condition, measured at 4 weeks of incubation. Superscript letters denote statistically significant differences within each recombinant human form as determined by two-way ANOVA followed by Tukey’s HSD post hoc test (p < 0.05).

^¥β-sheet/random coil ratio at week 4.

Nanoribbon assembly followed a time-dependent progression that was strongly influenced by both pSer16 and C-terminal truncation (Fig. 4b; Tables 2 and 3). pS16rH174 exhibited a prolonged lag phase, with nanoribbon formation initiating only after ~30 days of incubation and reaching a plateau near Day 45. C-terminal truncation markedly accelerated nanoribbon emergence, with pS16rH146 displaying shorter lag phases and reaching the plateau phase by ~40 days. FTIR analysis further demonstrated a progressive increase in β-sheet content among pSer16 variants during incubation (Table S1; Fig. S14 - S15), rising progressively throughout the incubation time and with a concomitant reduction in random coil/α-helix contributions. This shift indicates a gradual disorder-to-order transition that stabilizes after approximately 2 weeks and coincides with the emergence of nanoribbon assemblies, supporting a pS16-dependent conformational reorganization underlying higher-order self-assembly.

Synergistic role of calcium and phosphate on the amelogenin assembly mechanism

Under both CaCl₂ and KH₂PO₄ conditions, nanoribbon assembly proceeded for all constructs, but the response to the ionic environment differed markedly between full-length and C-terminal truncated variants. Sigmoidal analysis revealed a strong dependence of full-length amelogenin assembly on ionic identity (Fig. 5b; Fig. S16). Under phosphate conditions, rH174 exhibited a pronounced kinetic delay with a rightward shift in the transition midpoint relative to Ca²⁺, whereas pS16 partially alleviated this delay. In contrast, Ca²⁺ supported earlier and more gradual assembly for both phosphorylated and non-phosphorylated rH174. Across all conditions, nanoribbon yields converged to similar plateau values, indicating that ionic effects primarily modulate assembly timing rather than final outcome. For the truncated rH146 constructs, nanoribbon exhibited similar assembly kinetics under Ca²⁺ and PO₄³⁻ conditions, indicating reduced sensitivity to ionic identity (Fig. 5d; Fig. S17). Both Ca²⁺- and PO₄³⁻-containing conditions yielded comparable transition midpoints and final nanoribbon yields for rH146 and pS16rH146, with only minor shifts in the timing and sharpness of the transition.

Fig. 5.

TEM imaging and time-dependent assembly kinetics of phosphorylated and non-phosphorylated amelogenin variants in CaCl₂ or KH₂PO₄. [a, c] Representative negatively stained TEM images acquired at the plateau phase illustrate that calcium yields straighter, more rigid ribbons with pronounced bundling, particularly in full-length and pS16 variants. Phosphate ions promote extended, flexible nanoribbons in truncated variants, but induce alternative nucleation pathways in full-length constructs, as explored in Fig. S17. Scale bar: 50 nm. [b] Sigmoidal nanoribbon growth curves reveal that full-length forms exhibit greater sensitivity to phosphate, with rH174 showing an extended lag phase, whereas pS16rH174 overcomes it earlier. Estimated nanoribbon yield (normalized, 0–1) was plotted as a function of incubation time and fitted using a variable-slope logistic model (four-parameter fit). For pS16rH174, fitting parameters are: Ca²⁺ (t50 = 25.3 days, Hill slope = 0.08, R² = 0.99) and PO₄⁻³ (t50 = 16.6 days, Hill slope = 0.07, R² = 0.96). For rH174, fitting parameters are: Ca²⁺ (t50 = 22.3 days, Hill slope = 0.08, R² = 0.99) and PO₄⁻³ (t50 = 35.1 days, Hill slope = 0.08, R² = 0.99). [d] Sigmoidal nanoribbon growth curves reveal that C-terminal truncation strongly accelerates nanoribbon assembly compared to full length in [b], with pS16rH146 exhibiting a pronounced rightward shift relative to rH146. Ca²⁺ and PO₄³⁻ produce only minor differences in assembly outcome, with both conditions converging to a comparable transition profile. Estimated nanoribbon yield (normalized, 0–1) was plotted as a function of incubation time and fitted using a variable-slope logistic model (four-parameter fit). For pS16rH146, fitting parameters are: Ca²⁺ (t50 = 16.4 days, Hill slope = 0.08, R² = 0.99) and PO₄⁻³ (t50 = 22.6 days, Hill slope = 0.12, R² = 0.99). For rH146, fitting parameters are: Ca²⁺ (t50 = 11.1 days, Hill slope = 0.11, R² = 0.99) and PO₄⁻³ (t50 = 9.2 days, Hill slope = 0.12, R² = 0.99). Experiments were conducted over 50 days of incubation in 33 mM CaCl₂ or 20 mM KH₂PO₄

Beyond the kinetics, construct-dependent differences emerged in nanoribbon rigidity, alignment, and higher-order organization (Fig. 5a and c; Fig. S16 - S17; Tables 2 and 3). In the presence of calcium, nanoribbons were consistently straighter and more rigid, with enhanced parallel alignment and lateral association, culminating in dense bundle formation, particularly for the full-length and pS16 variants. In contrast, phosphate exerted a strong truncation-dependent effect, as in pS16rH146 and rH146, this ionic species promoted longitudinal elongation, yielding long, flexible nanoribbons that remained largely unbundled. By comparison, in pS16rH174 and rH174, phosphate did not drive extensive elongation but instead favored alternative nucleation pathways, characterized by thin filament-like intermediates that matured into nanoribbons over time, suggesting polymorphism in the assembly landscape (Fig. S18).

Notably, combined Ca²⁺/PO₄³⁻ conditions revealed calcium-dominated behavior, with calcium effectively suppressing phosphate-driven elongation across all constructs (Tables 2 and 3; Fig. S5 - S6 compared with Fig. S16 - S17). Full-length variants formed straight, rigid nanoribbons with high parallel alignment and dense bundling. In contrast, truncated variants, although similarly straightened under Ca²⁺/PO₄³⁻ conditions, exhibited reduced lateral association, lower bundle density, and only moderate alignment. Together, these observations indicate that while calcium governs overall assembly behavior in the presence of phosphate, robust nanoribbon bundling under physiologically relevant Ca²⁺/PO₄³⁻ conditions require the C-terminal domain.

ATR-FTIR analysis (Table S2; Fig. S19) shows that calcium and phosphate ions elicit comparable overall trends in amelogenin secondary structure, but with distinct magnitudes depending on construct length. Full-length variants display moderate β-sheet enrichment (55–70%) accompanied by measurable β-turn contributions, reflecting a broader Amide I envelope. In contrast, C-terminal truncation results in substantially higher β-sheet content (74–82%) and loss of detectable β-turn components under both ionic conditions, yielding elevated β-sheet/random coil ratios and a more homogeneous secondary-structure profile. In the mixed Ca²⁺/PO₄³⁻ condition, all four constructs converge to a very similar two-component secondary-structure distribution, with β-sheet around 67–69% and random coil/α-helix around 31–33% (Table S3; Fig. S20).

Calcium and phosphate promote similar electrostatic charge redistribution in amelogenin

In silico Electrostatic surface mapping revealed that both ions, calcium and phosphate, induce broadly similar charge redistributions in full-length H175 and truncated H147 (Fig. 5a -f; Fig. S21). In DI water, both H175 and H147 exhibited a pronounced and uniformly distributed positive electrostatic potential localized at the N-terminal region. Upon addition of either calcium (33 mM CaCl₂) or phosphate (20 mM KH₂PO₄), both constructs showed a reduction in the magnitude of surface positivity and a pronounced rebalancing of charge across the molecular contour of the N-terminal region. This ion-driven reorganization resulted in a more heterogeneous but overall neutral electrostatic landscape, consistent with enhanced electrostatic screening and stabilization of intermolecular interactions. Within this state, regions that retain positive potential (6H, 9H, 30R, 46H, 47H, 54S, 57H, 66H, 67H, 68H, 91H, 98H, 99H) likely constitute phosphate interaction sites, whereas persistently negatively charged residues (157D, 161E, 168D, 172E, 173E, 175D) remain accessible for calcium association.

pS16-dependent stiffening and stabilization of the amelogenin nanoribbon midline.

PeakForce QNM measurements showed that pS16rH146 nanoribbons exhibited significantly higher stiffness compared to rH146 under identical assembly conditions, indicating a more mechanically rigid supramolecular architecture (Fig. 7a –c). Correspondingly, pS16rH146 nanoribbons displayed greater tip-sample adhesion and reduced deformation under applied force, whereas rH146 nanoribbons were softer, showed lower adhesion, and underwent higher indentation. Line profile analyses confirmed that these differences were localized along the nanoribbon backbone, particularly within the central axis, rather than arising from substrate effects (Fig. 7c). The pronounced stabilization of the central zone in pS16rH146 nanoribbons is characterized by consistently elevated stiffness and adhesion coupled with reduced deformation (Fig. 7c; Fig. S22), highlighting pS16-dependent reinforcement of the nanoribbon core.

Fig. 7.

PeakForce QNM mechanical mapping of pS16rH146 and rH146 nanoribbons. [a – b] Stiffness and adhesion maps revealed that pS16rH146 nanoribbons exhibit consistently higher mechanical resistance compared to rH146. Deformation maps further reveal that pS16rH146 deformed more evenly across the nanoribbon, suggesting a cohesive and resilient structure, likely reflecting phosphoserine/carboxylates-calcium bridges. Height, stiffness, adhesion, and deformation maps were acquired under ambient conditions on a mica substrate. A lower PeakForce setpoint of 0.5 nN was used to minimize structural deformation during scanning. Variations in the background signal may reflect differential blotting or residual unassembled protein adsorption. [c] Cross-sectional profiles corresponding to the yellow lines in [a] and [b] highlight differences in nanoribbon topography and mechanical properties. Notably, the nanoribbon midline appeared stiffer and less flexible than the periphery, suggesting a more tightly packed core architecture. Samples were incubated for 21 days in 33 mM CaCl₂ and 20 mM KH₂PO₄ at pH 5.5 and 37 °C.

Discussion

Previous models of amelogenin assembly into nanoribbons were derived from oil-water emulsion studies [34, 38] and SAXS studies [35], which both indicated that the N-termini of amelogenin molecules overlap and most likely form a dimer that could act as a building block for assembly. Subsequently, XRD studies [75, 76] have shown that amelogenin nanoribbons comprise a cross-beta motif similar to that observed in developing enamel [32, 76]. An amyloid-forming region at the amelogenin N-terminus, the P2 domain, was identified as a key contributor to this architecture. Fourteen–amino acid peptides derived from the P2 domain (residues #7 to #20) can adopt a steric-zipper dimer configuration, providing a stable β-sheet–based template for supramolecular growth. In the current model, self-assembly into nanoribbons occurs when P2-derived steric-zipper dimers align through hydrogen bonding and calcium-mediated bridging to form extended β-sheet arrays that are parallel and in register within strands, but antiparallel out of register between the adjacent strand (see Fig. 8a). Consistent with this model, aromatic residues (tyrosine and phenylalanine) within the N-terminal region are essential for amelogenin self-assembly, as their substitution abolishes oligomerization and protein-protein interactions [77].

Fig. 8.

Structural model for steric-zipper–mediated coalescence of amelogenin nanoribbons. [a] Current atomic model of the p14P2 peptide assembly modeled after the S. cerevisiae Sup35 cross-β amyloid spine. Antiparallel dimers initially nucleate via hydrophobic contact between isoleucine (ILE) and phenylalanine (PHE), generating a dry interface that stabilizes early assemblies. These dimers propagate through in-register inter-ß-strand hydrogen bonds, in a process facilitated by Ca²⁺-mediated interactions with glutamate and pS16. [b] Schematic model illustrating β-sheet alignment and supramolecular nanoribbon assembly driven by amelogenin interactions. Formation of a cross-β architecture in which β-strands from adjacent protein segments pack face-to-face, generating a paired, zipper-like interface that propagates longitudinally to produce a nanoribbon. The P2-domain-enrichment core region represents the inter-strand interface, identified as the central dark line in negatively stained TEM images and as central topographic depression resolved in AFM images. [c] Schematic illustration of the role of Ca²⁺ on acidic C-terminal cluster and pS16 binding sites in lateral cohesion, and His-rich domains in elongation driven by PO₄³⁻.

The original model suggests a classical nucleation-dependent polymerization pathway where monomers assemble directly onto a growing β-sheet with little structural reorganization [32, 75]. This view implies an energetically permissive assembly mechanism well aligned with the behavior of short peptides known for their high amyloidogenic propensity [78, 79]. In contrast, our results demonstrate that the full-length amelogenin and its C-terminal truncated form assemble via a distinct mechanism involving nucleated conformational conversion (Fig. 9), also commonly observed in Aβ [80] fibrillogenesis and other amyloid-forming systems [81, 82]. Rather than direct monomer addition, this process proceeds through oligomeric intermediates, after which β-sheet–rich nanoribbons emerge. The disordered C-terminal domain appears to act as a kinetic barrier to nucleation, likely due to its high conformational entropy, as supported by PONDR-VSL2 predictions (Fig. S23). Proteolytic cleavage by MMP20 reduces this barrier, enabling faster assembly in vitro. Meanwhile, pS16 introduces a layer of regulation by imposing an energetic cost on ordering, thereby increasing the nucleation barrier and promoting a more gradual transition through oligomeric intermediates. Notably, despite these kinetic differences, the mature nanoribbons formed by phosphorylated and non-phosphorylated amelogenins display no discernible morphological differences within the resolution limits of the microscopy techniques used in this study, indicating that pS16 primarily modulates assembly dynamics rather than the final structural outcome.

Fig. 9.

Schematic representation of the proposed amelogenin assembly pathway. Monomers initially associate into prenucleation clusters that accumulate as disordered oligomers. These oligomers can reorganize into protoribbons through the emergence of β-sheet nuclei, which then mature into nanoribbons.

A key and unexpected finding is that pS16 at pH 5.5 is sufficient to induce nanoribbon assembly in the absence of calcium or phosphate, overturning the prevailing assumption that divalent ions are strictly required for amelogenin nanoribbon formation [32-35, 47, 76]. While phosphorylation of amelogenin facilitates assembly in the absence of calcium and phosphate, the protein interaction kinetics are substantially reduced with delayed nanoribbon formation and precursor structures (oligomers and protoribbons) still detectable after 50 days in assembly solutions. Our electrostatic potential calculations suggest that non-phosphorylated amelogenin exhibits strong positive charge density at the N-terminal region, generating repulsive interactions that possibly prevent β-sheet stacking under ion-free conditions. Interestingly, the addition of either calcium or phosphate tends to neutralize these changes, which experimentally favor the assembly of nanoribbons. In this context, one explanation for nanoribbon formation in CaP-free conditions is that pS16 introduces a stable, localized negative charge at Ser16 (−OPO₃²⁻, pKa=5.6) [83] within the amyloidogenic P2 region (GYINFSpYE) [32]. This modification reshapes the electrostatic landscape of the N-terminal domain, enabling attractive interactions with nearby positively charged residues (e.g., His6⁺, His9⁺, Lys24⁺, Arg30⁺). These interactions likely involve a cooperative network of non-covalent forces, including hydrogen bonding, aromatic stacking, and cation-π interactions. Because these interactions are weaker and more dynamic [84] than ion-mediated electrostatic bridging, they likely produce fewer stable oligomers, which, in turn, explains the slower ß-sheet nucleation kinetics and reduced elongation potential compared to ion-assisted conditions.

Our findings indicate that calcium and phosphate ions modulate the amelogenin assembly pathway in distinct yet complementary ways, offering new insight into how amelogenin folds to support nucleation, elongation, and supramolecular bundling. Fig. 8 provides a revised model that integrates these ion-dependent effects with sequence-encoded structural motifs. Some key advancements are emerging from our work.

TEM imaging of filament-filament junctions, open ribbon ends, single-filament tips, and characteristic ~17 nm axial contraction zones (Fig. S24), together with the observation that single-stranded filaments remain stable on their own under calcium-free conditions (Fig. S16a), indicates that nanoribbon growth can proceed through a coalescence mechanism and does not require the obligatory formation of a two opposing monomers as proposed by Martinez-Avila et al. [35]. This coalescence integrates elements of both prior models but introduces a key mechanistic refinement: parallel monomer–monomer units nucleate single-stranded ribbons that later coalesce through a steric zipper, whose stability is enhanced by Ca²⁺ bridging. Consistent with this interpretation, the nanoribbon midline visualized in negatively stained TEM images and AFM images (Fig. S25) may be a demarcation between two opposing strands of about 8-9 nm width that combine to form a double-stranded nanoribbon of ~18 nm.

In the refined model, steric zipper interactions between the P2 domain of two strands act as local stabilizing elements that promote lateral association and contribute to the formation of a persistent nanoribbon midline (Fig. 8b). This zipper-like motif, predicted by a computational model and observed as a cross-beta amyloid spine in previous X-ray diffraction studies of P2 peptides [32, 75], arises from interdigitated isoleucine-phenylalanine interactions that form a dry steric zipper interface [32]. While this dry interface stabilizes the double-strand filament pairing and axial assembly, the opposite face of the β-sheet remains solvent-exposed and chemically active, forming a wet interface enriched in phosphorylated Ser16 and other charged residues (Fig. 8c). AFM PeakForce QNM measurements support this interpretation by showing that the central axis of the ribbon is mechanically and chemically distinct, exhibiting higher stiffness, reduced deformation, and elevated adhesion (Fig. 7), a nanomechanical signature consistent with a charged, hydrated, hydrogen-bond-rich interface. This hydrated surface is well-positioned to engage in electrostatic and coordination interactions with calcium and phosphate ions, providing a mechanistic basis for ion-mediated lateral association, nanoribbon formation, and mineral-directing functionality. This architecture provides a coherent explanation for the nanoribbon midline observed in negatively stained TEM images and the corresponding central topographic depression resolved by AFM, both of which indicate an interface that excludes protein density and enhances solvent accessibility. These observations support a model in which steric zipper interactions nucleate and stabilize ribbon formation, while the nanoribbon midline emerges as a less densely packed mesoscale feature of the assembled architecture.

Our study suggests that calcium acts on the pS16-containing N-terminal region, where the phosphate group and nearby acidic residues (Glu18 and maybe Glu39) form an electrostatic pocket that selectively recruits Ca²⁺. Nanomechanical mapping shows that pS16 stabilizes the nanoribbon midline, producing uniform stiffness and increased adhesion, features consistent with a hydrated, ion-responsive interfacial layer. We propose that this central region functions as a mineral nucleation platform, where the phosphate group of pS16, in coordination with nearby acidic residues (mainly Glu18), forms a localized electrostatic pocket capable of recruiting Ca²⁺. This site-specific enrichment of Ca²⁺ could initiate the formation of ACP precursors directly on the nanoribbon scaffold, linking early amelogenin self-assembly to the spatially restricted onset of mineral stabilization during enamel formation as also suggested by others [19, 51]. Furthermore, phosphate’s pronounced effect on nanoribbon elongation and surface electrostatics suggests that it interacts preferentially with His-rich regions of amelogenin, where protonated His can engage in electrostatic attraction and potentially form bidentate hydrogen bonds through their Nδ1 and Nε2 positions at pH below 6.0 [85, 86]. Although direct structural evidence for these contacts in amelogenin is not yet available, the strong phosphate-dependent increase in nanoribbon length implies that these residues help stabilize longitudinal associations during growth. In this context, although the P2 domain is recognized as the principal amyloidogenic nucleation element in amelogenin, it is unlikely that the nanoribbons formed by the longer polypeptide variants (rH146, rH174) are stabilized by a single, well-defined steric zipper motif. Instead, the extended sequence length and intrinsically disordered regions suggest that amelogenin nanoribbons are stabilized through various distributed, locally formed β-strands where short, transient steric-zipper-like interactions can form in restricted segments. In a revised model, predictions of amyloid-prone regions (Fig. S26) converge to indicate that within amelogenin N-terminal region, short sequences segments, many adjacent to or overlapping His-rich motifs, possess packing propensities compatible with β-sheet stacking. We do not interpret these segments as independent amyloid cores, but rather as potential auxiliary stabilizing interfaces that may become engaged during elongation. This idea is consistent with the behavior of amyloid fibrils comprised by longer polypeptides [78], where elongation often involves the progressive ordering of secondary structural motifs beyond the initial nucleating region.

Conclusion

Taken together, our findings establish a comprehensive framework in which pS16, Mmp20-mediated C-terminal truncation, and the complementary actions of calcium and phosphate function as coordinated regulators of amelogenin nanoribbon self-assembly. By delineating the nucleation barrier, restructuring oligomeric intermediates, and directing both longitudinal elongation and lateral cohesion, our study fills a critical gap in enamel biology and provides a conceptual platform to move toward a molecularly grounded understanding of how amelogenin shapes the creation of one of the most mechanically resilient materials in nature.

Fig. 6.

Electrostatic surface potential of full-length and C-terminal truncated amelogenin under varying ionic conditions. APBS-calculated electrostatic potentials (kT/e) are mapped onto representative conformers of (a–c) H175 and (d–f) H147 at pH 5.5 in four conditions: no ions, 33 mM CaCl₂, and 20 mM KH₂PO₄. The color gradient represents the electrostatic potential, ranging from negative (red) to positive (blue), and is assigned values from −1 to 1 kT/e, highlighting shifts in charge distribution, particularly around the His-rich domain and the C-terminal region. Key acidic and basic residues predicted to interact with calcium and phosphate are labeled based on the predictions of Martinez-Avila et al.[35]. Notably, the combined ion condition exhibits more neutralized and evenly distributed surface potential, suggesting electrostatic screening and potential stabilization of intermolecular interactions. Regions that remain positively (strong: 6H, 9H, 30R, 46H, 47H, 54S, 57H, 66H, 67H, 68H; mild: 91H, 98H, 99H) or negatively (157D, 161E, 168D, 172E, 173E, and 175D) charged are likely to interact with phosphate or calcium, respectively.

Data Availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request.