An agent-guided peptide hydrogel bio-stabilizer clamps pericellular viscoelastic drift Authors: Xiao Wei, Liqiang Zhang, Zhuo Chang, Fan Ding, Ziyan Qu, Jianghao Chen, Guangkui Xu, Wangxiao He, Wenjia Liu Published: 2026 ABSTRACT Living tissues continually remodel the extracellular matrix, causing pericellular mechanics to drift over time. Chronic glycation amplifies this drift, destabilizing mechanoregulatory cues and compromising regenerative outcomes. Here we introduce Micro-Mechanical Homeostasis Reset, a material-based strategy that targets cell-to-pericellular viscoelastic homeostasis rather than static bulk stiffness. Using an agent-guided, rule-based enumeration-and-scoring workflow, we design ViscoClamp, an extracellular matrix-interpenetrating peptide hydrogel with affinity for calcium ions and advanced glycation end products. We show that ViscoClamp buffers chronic glycation loading and constrains glycation-driven viscoelastic drift, while remaining permissive to mesenchymal stem cell-mediated matrix remodeling and mineralization. We quantify stabilization using a dispersion-transfer ratio that measures how variable glycation inputs are compressed into stable micromechanical outputs. We demonstrate that in glycation-rich hyperglycemic craniofacial defects, ViscoClamp stabilizes cell-scale micromechanics, restores osteogenic activation, and improves bone repair and mineralization in rats, rabbits, beagle dogs, and rhesus macaques. Introduction Cell fate during tissue repair is governed by a niche that integrates soluble signals, cell-cell interactions and the extracellular matrix (ECM), and this niche is neither static nor spatially uniform1–3. Beyond composition, the niche is dynamically remodeled by cells and host factors, meaning that the mechanical cues presented to cells are local, heterogeneous and time‑dependent1–4. Mechanical information is now recognized as a core component of this integration: cells continuously interrogate matrix mechanics, including viscoelasticity, and translate these inputs into signaling and transcriptional programs1–4. Yet biomaterial design still frequently relies on bulk descriptors (for example, modulus or macroscopic rheology), whereas cells primarily experience mechanics at a mesoscopic, cell‑to‑pericellular scale (approximately 1–50 μm), where adhesion, force generation, and nascent ECM deposition remodel local architecture and thereby reshape local mechanics in situ5,6. Emerging micromechanical imaging further highlights pronounced cell-scale heterogeneity and temporal evolution of mechanics, reinforcing that the instructive niche is local and time-dependent rather than reducible to a single bulk parameter7. This scale mismatch becomes consequential in regeneration, where fate decisions depend on what cells experience at the pericellular interface as the niche evolves1–4,8. In bone, this challenge is sharpened by the temporal structure of commitment. Mesenchymal stem cell (MSC) osteogenic initiation exhibits an early “window” in which sustained exposure to an osteo-permissive mechanical milieu, including time‑dependent relaxation and dissipation, facilitates stabilization of osteogenic programs3,9–11. Because these cues are filtered through the pericellular microdomain, the critical variable is not simply bulk stiffness, but whether mesoscopic/pericellular viscoelasticity remains within an osteo‑permissive range over time3,5,6,8,11. When cell-scale (micromechanical) cues at the mesoscopic/pericellular interface drift or fluctuate during this period, osteogenic activation and downstream matrix mineralization become less robust and more variable3,9–11. Chronic disorders frequently impose persistent microenvironmental stressors that progressively modify ECM composition and mechanics4,8. Advanced glycation end products (AGEs) serve as a prototypical example: under hyperglycemic conditions, AGEs accumulation and covalent crosslinking of long-lived matrix proteins remodel local network mechanics and pericellular viscoelasticity, disrupt cell–matrix interactions, and compromise bone material quality and biomineralization12–14. In craniofacial bone, functional loading adds recurrent mechanical disturbance to this chronic chemical drift, creating a stringent setting in which the pericellular niche can repeatedly deviate from an osteogenesis-matched state1,2,10,11. Together, progressive glycation and dynamic loading create a stress test for any strategy that seeks to sustain osteogenic fate control over the full trajectory of healing. Regenerative strategies for metabolic disease-associated bone defects, therefore, commonly focus on augmenting osteogenic factors or modulating inflammation and vascularization12,15. These approaches remain essential, yet their durability can be constrained if the pericellular mechanical context continues to drift, effectively narrowing or shortening the window in which cells can respond productively to osteogenic stimuli3,4,11–14. In this work, we hereby propose Micro-Mechanical Homeostasis Reset (MMHR): a materials paradigm that targets mesoscopic, cell-to-pericellular viscoelastic homeostasis as a complementary control layer for regenerative fate specification1–4,9–11. Implementing MMHR requires a bio‑stabilizer that interpenetrates MSC-deposited ECM to form a hybrid pericellular matrix and clamps viscoelastic drift without obstructing ECM remodeling and mineralization5,6,16–18. We frame bio‑stabilization as an input–output problem, defining a dispersion‑transfer ratio (CVoutput/CVinput) that quantifies how variable pathological inputs are compressed into stable viscoelastic outputs. Here, we implemented an agent-guided and transparent modular biomaterial-design workflow that translates a circuit-inspired “protect-tune-clamp” decomposition into peptide sequence and network architecture, generating the peptide hydrogel ViscoClamp19–22. As a rigorous proof‑of‑concept, we challenged ViscoClamp in AGEs‑rich hyperglycemic craniofacial defects and observed consistent improvements in bone repair and mineralization across rats, rabbits, beagle dogs and rhesus macaques. Results Agent-guided, circuit-inspired design yields a dual-affinity peptide hydrogel bio-stabilizer (ViscoClamp) To implement MMHR, we sought a peptide‑hydrogel bio‑stabilizer that (i) maintains a steady viscoelastic response under variable glycation loading, (ii) can interpenetrate and hybridize with cell‑deposited ECM at the pericellular scale, and (iii) remains permissive to mineral deposition. Guided by a circuit-inspired “protect-tune-clamp” decomposition (the left panel of Fig. 1A), we implemented an agent-guided workflow that converts module-level functions into constrained peptide-sequence candidates and network constraints (the right panel of Fig. 1A and Fig. S1). No new deep-learning model was trained for peptide generation or ranking. Instead, the workflow used deterministic modules for motif enumeration, physicochemical feature calculation, surrogate AGEs-HSA binding scoring, gel-feasibility scoring and predicted viscoelastic-drift scoring, with candidate ranking determined by predefined formulas and objective weights. The three objective weights were assigned a priori before experimental validation (AGEs-HSA adsorption robustness, 0.45; gel feasibility, 0.25; predicted viscoelastic stability, 0.30) and were not adjusted after observing the experimental or biological outcomes. In a four-step workflow, we first adapted a mineralization motif from the Statherin segment DSSEEKF and introduced a phosphomimic substitution (SS → DD) to generate DDDEEKF, which increased the Ca2+-affinity proxy relative to the native motif (Fig. 1A and Fig. S1, Step 1). We next exhaustively enumerated 32 ion‑complementary 4‑aa motifs using a [Pos][Hyd][Neg][Hyd] grammar and ranked them for adsorption to representative AGEs-HSA systems using surrogate binding scores, yielding a Top-5 set (RFDF, RFEF, RMDM, RADA, RMEM) that included the canonical self‑assembling motif RADA (Fig. 1A and Fig. S1, Step 2). We then concatenated DDDEEKF with each (motif)4 block and applied a multi-objective selection balancing AGEs-HSA binding, gel feasibility, and minimal predicted viscoelastic drift across simulated binding states, selecting DDDEEKF‑(RADA)4 as the most stable combined design (Fig. 1A and Fig. S1, Step 3). Finally, an alanine-to-cysteine scan across the (RADA)4 block predicted the strongest mechanical “locking” effect when substituting the terminal alanine, leading to the cysteine-stabilized sequence DDDEEKF‑(RADA)3-RADC, termed ViscoClamp (Fig. 1A and Fig. S1, Step 4).Fig. 1Agent-guided design, synthesis and characterization of ViscoClamp.A Schematic illustrating the conceptual mapping between an electronic voltage stabilizer and the ViscoClamp bio-stabilizer, together with the workflow of agent-guided peptide hydrogel design. B Representative analytical HPLC chromatogram and LC-MS spectrum of synthetic ViscoClamp. C Micro-Fourier transform infrared (micro-FTIR) characterization of ViscoClamp. D Representative macroscopic appearance of ViscoClamp hydrogels at 1% and 6% (w/v), contact-angle measurements, and hydrogel microstructure characterized by scanning electron microscopy (SEM) and transmission electron microscopy (TEM) (scale bars, 10 μm for SEM; 500 nm for TEM). E Isothermal titration calorimetry (ITC) quantifying the binding affinity between ViscoClamp and Ca²⁺. F Fluorescence polarization (FP) assay quantifying the binding affinity between ViscoClamp and AGEs. G TEM images of ViscoClamp after incubation with increasing concentrations of AGEs (scale bar, 500 nm). H Confocal laser scanning microscopy (CLSM) showing colocalization of FITC-labeled ViscoClamp (6%) with Cy3-labeled AGEs (0.1 mg/mL) after binding (scale bar, 50 μm). B, C n = 3 independent replicates; D–H n = 6 independent replicates. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. Agent-guided design, synthesis and characterization of ViscoClamp. A Schematic illustrating the conceptual mapping between an electronic voltage stabilizer and the ViscoClamp bio-stabilizer, together with the workflow of agent-guided peptide hydrogel design. B Representative analytical HPLC chromatogram and LC-MS spectrum of synthetic ViscoClamp. C Micro-Fourier transform infrared (micro-FTIR) characterization of ViscoClamp. D Representative macroscopic appearance of ViscoClamp hydrogels at 1% and 6% (w/v), contact-angle measurements, and hydrogel microstructure characterized by scanning electron microscopy (SEM) and transmission electron microscopy (TEM) (scale bars, 10 μm for SEM; 500 nm for TEM). E Isothermal titration calorimetry (ITC) quantifying the binding affinity between ViscoClamp and Ca²⁺. F Fluorescence polarization (FP) assay quantifying the binding affinity between ViscoClamp and AGEs. G TEM images of ViscoClamp after incubation with increasing concentrations of AGEs (scale bar, 500 nm). H Confocal laser scanning microscopy (CLSM) showing colocalization of FITC-labeled ViscoClamp (6%) with Cy3-labeled AGEs (0.1 mg/mL) after binding (scale bar, 50 μm). B, C n = 3 independent replicates; D–H n = 6 independent replicates. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. To determine whether the final sequence represented a meaningful selected design rather than an arbitrary peptide within the present constrained design space, we benchmarked ViscoClamp (Pep 0) against four scaffold-matched comparator variants (Pep 1-Pep 4; sequences provided in Supplementary Methods). Pep 1-Pep 4 were not selected as literature state-of-the-art peptides. Instead, they were generated from the same DDDEEKF mineralization module and the same ion-complementary [Pos][Hyd][Neg][Hyd] grammar to perturb specific sequence features around the ViscoClamp scaffold while preserving comparable peptide length and terminal cysteine stabilization. Pep 1 tests cation identity by replacing Arg with Lys; Pep 2 tests acidic side-chain length by replacing Asp with Glu; Pep 3 and Pep 4 increase hydrophobic residue content while retaining an ion-complementary repeat. Thus, these peptides serve as bounded design controls for the agent-guided workflow rather than exhaustive best-in-class benchmarks. These comparators were evaluated for AGEs-binding affinity and rheological behavior (Fig. S2A, S2B). Pep 0/ViscoClamp showed an apparent AGEs-binding affinity of 1.0 ± 0.2 μM, whereas Pep 1 and Pep 2 showed weaker binding (5.4 ± 0.5 μM and 8.9 ± 0.7 μM, respectively) and Pep 3/Pep 4 showed no detectable binding under the tested conditions. Representative biological validation further showed that ViscoClamp more effectively preserved Runx2 nuclear localization and Alp staining/activity under glycation stress than Pep 1-Pep 4 (Fig. S2C-S2H). These comparator experiments support the selection of Pep 0/ViscoClamp within the scaffold-matched comparator set, but they do not establish universal superiority over all available mineralization, AGEs-binding or hydrogel-forming peptides. ViscoClamp was synthesized by Fmoc-based solid-phase peptide synthesis and verified by mass spectrometry (Fig. 1B). FTIR spectra showed characteristic amide and thiol absorbance bands, supporting successful cysteine incorporation (Fig. 1C), and RP-HPLC purification yielded 98% purity (Fig. 1B). ViscoClamp formed a gel at ≥1% (w/v) (Figs. 1D), and 6% (w/v) hydrogels exhibited robust integrity and injectability with favorable rheological behavior and fatigue resistance (Fig. 1D and Figs. S2I-S2L). Coarse-grained molecular dynamics suggested a structural basis for Ca²⁺ association (Fig. S2M), which was validated by ITC (Kd = 3.15 ± 1.13 μM) (Fig. 1E). This ITC-derived Kd was used to validate Ca²⁺-association competence of the final hydrogel rather than to establish best-in-class Ca²⁺ affinity. We therefore interpret the Ca²⁺-binding data together with the subsequent mineral-deposition assays as evidence that ViscoClamp remains mineralization-permissive. Fluorescence polarization indicated strong binding to AGEs-HSA with an apparent dissociation constant of 1018 ± 224 nM (Fig. 1F). Consistent with adsorption‑induced network reorganization, TEM showed progressive aggregation and restructuring of ViscoClamp nanofibers upon AGEs addition (Fig. 1G), and confocal imaging of Cy3-AGEs mixed with FITC-ViscoClamp revealed co‑clustering within the gel matrix (Fig. 1H). Finally, cytocompatibility assays indicated that 1% ViscoClamp did not impair MSC proliferation or induce apoptosis (Figs. S2N–S2P). Because ViscoClamp binds Ca2+, we further tested MSC viability under increasing CaCl2 concentrations in the presence of 1% ViscoClamp; CCK-8 analysis showed no overt Ca2+-related cytotoxicity within the tested range (Fig. S2Q), supporting its use as a cell-contacting biomaterial. ViscoClamp compresses dispersed glycation inputs into stable viscoelastic outputs, defined by a low dispersion-transfer ratio To test whether ViscoClamp behaves as a bio‑stabilizer, we framed stabilization as an input-output problem: AGEs concentration serves as a disturbance input, whereas cell‑scale viscoelastic parameters derived from atomic force microscopy (AFM) serve as outputs (Fig. 2A, B). We combined quasi‑static indentation to extract an apparent elastic modulus (EEexp; Sneddon model) with dynamic creep indentation to obtain creep compliance curves (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J\left(t\right)$$\end{document}Jt; Lee-Radok formulation), and then used hierarchical and power‑law fits to decompose multiscale viscoelastic behavior. This pipeline yielded relaxation time (τ), effective viscosity (η), a power‑law exponent (αlong), the stiffness of a local gel unit (E1), and the stiffness of the fiber network (E2) as a compact set of cell‑scale mechanical readouts (Fig. 2A).Fig. 2ViscoClamp buffers glycation fluctuations to stabilize viscoelastic outputs.A Schematic of atomic force microscopy (AFM)-based mechanical characterization and parameter extraction. Static elasticity testing (peak force 300 nN) and dynamic creep testing (peak force 300 nN; hold 10 s) were used to obtain directly measured parameters (E, J and αlong) and model-derived parameters (E₁, E₂, Esum, η and τ); AFM tapping-mode imaging was used for morphological characterization. B Conceptual schematic of bio-stabilization: fluctuating AGEs “inputs” are converted into buffered, low-dispersion mechanical “outputs” by ViscoClamp. C AFM-derived viscoelastic descriptors (Eexp, η, τ, αlong, E₁ and E₂) of single gel units after 24 h incubation with increasing concentrations of AGEs. D Coefficient of variation of the AGEs inputs versus the viscoelastic outputs, where CV = SDpop/mean and SDpop denotes population standard deviation across the predefined AGEs input states; CVoutput/CVinput denotes the dispersion-transfer ratio (lower values indicate stronger stabilization). E Schematic of time-domain bio-stabilization assessment across 6–72 h. (F) AFM-derived viscoelastic descriptors of individual gel units measured at the indicated time points. G Time-resolved CV analysis of input and output and the corresponding CVoutput/CVinput ratio. n = 6 independent replicates for (C, D, F and G). Quantitative data are presented as mean ± SD. For box plots in (C and F), the center line denotes the median, the box bounds denote the 25th and 75th percentiles, and the whiskers extend from the minimum to the maximum values. Source data are provided as a Source Data file. ViscoClamp buffers glycation fluctuations to stabilize viscoelastic outputs. A Schematic of atomic force microscopy (AFM)-based mechanical characterization and parameter extraction. Static elasticity testing (peak force 300 nN) and dynamic creep testing (peak force 300 nN; hold 10 s) were used to obtain directly measured parameters (E, J and αlong) and model-derived parameters (E₁, E₂, Esum, η and τ); AFM tapping-mode imaging was used for morphological characterization. B Conceptual schematic of bio-stabilization: fluctuating AGEs “inputs” are converted into buffered, low-dispersion mechanical “outputs” by ViscoClamp. C AFM-derived viscoelastic descriptors (Eexp, η, τ, αlong, E₁ and E₂) of single gel units after 24 h incubation with increasing concentrations of AGEs. D Coefficient of variation of the AGEs inputs versus the viscoelastic outputs, where CV = SDpop/mean and SDpop denotes population standard deviation across the predefined AGEs input states; CVoutput/CVinput denotes the dispersion-transfer ratio (lower values indicate stronger stabilization). E Schematic of time-domain bio-stabilization assessment across 6–72 h. (F) AFM-derived viscoelastic descriptors of individual gel units measured at the indicated time points. G Time-resolved CV analysis of input and output and the corresponding CVoutput/CVinput ratio. n = 6 independent replicates for (C, D, F and G). Quantitative data are presented as mean ± SD. For box plots in (C and F), the center line denotes the median, the box bounds denote the 25th and 75th percentiles, and the whiskers extend from the minimum to the maximum values. Source data are provided as a Source Data file. We challenged ViscoClamp with a broad range of AGEs concentrations (0 ~ 25.6 mg/mL), lyophilized the gels, and performed AFM measurements. Across increasing AGEs loading, η and τ showed a modest downward trend, whereas EEexp, αlong, E1, and E2 remained largely unchanged (Fig. 2B, C). We therefore describe ViscoClamp as maintaining overall stabilization within the tested AGEs range, with modest dose-dependent changes in viscous/relaxation-related parameters rather than complete invariance of every parameter. Within the current tested range, no abrupt loss of stabilization was evident from the AFM and bulk rheology profiles (Fig. 2B, C; Fig. S3B, S3C). Bulk rheology of AGEs‑exposed gels similarly showed no substantive alteration in elastic behavior or force‑transmission characteristics (Fig. S3B, S3C). Consistent with its preferential modulation of viscous (stress-relaxation) behavior, supplementation with ViscoClamp reduced soluble AGEs levels in MSC culture supernatants, yielding a ≥ 50% decrease in measurable AGEs within 4 h (Fig. S3A). To quantify input-output stabilization, let the AGEs input concentration series be A = {ai}{i=1}N, with mean abar = (1/N)∑{i=1}N ai. We used population standard deviation for the DTR calculation because the AGEs concentrations were predefined input states rather than randomly sampled observations: SDpop(A) = sqrt[(1/N)∑{i=1}N (ai − abar)2], CVinput = SDpop(A)/abar. For each AFM-derived output parameter m ∈ M = {Eexp, αlong, τ, E1, η, E2}, let Ym = {y{m,i}}{i=1}N denote the values measured across the same AGEs input states, with mean ybarm = (1/N)∑{i=1}N y{m,i}. The output coefficient of variation for parameter m was defined as: SDpop(Ym) = sqrt[(1/N)∑{i=1}N (y{m,i} − ybarm)2], CVm = SDpop(Ym)/ybarm. The composite output dispersion was calculated as the arithmetic mean of the six output CV values: CVoutput = (1/|M|)∑{m∈M} CVm = (1/6)∑{m∈M} CVm. Finally, the dispersion-transfer ratio was defined as: DTR = CVoutput/CVinput. A low DTR indicates that a broad pathological input distribution is compressed into a narrow mechanical-output distribution. We used CV-based DTR as the primary stabilization metric because it is dimensionless, directly interpretable as relative input-output attenuation and robust for the current small AGEs dose-series design. Entropy and Jensen-Shannon divergence were not used as primary metrics because they require binning or density estimation, and mutual information was not used because reliable estimation would require a substantially larger paired input-output distribution than the current AFM dose series. A defining property of a stabilizer is that its CVoutput is far less than its CVinput, which means that large fluctuations in the input are compressed into a narrow output range. Indeed, the AGEs input exhibited a CVinput of approximately 200%, whereas the output parameters showed far smaller dispersion (CVoutput values: Eexp 9.1%, αlong 9.7%, τ 15.3%, E1 3.7%, η 14.3%, E2 4.3%) (Fig. 2D). Aggregating these outputs yielded a stabilization index CVoutput/CVinput = 4.70%, corresponding to an approximately 20‑fold attenuation of dispersion from glycation input to viscoelastic output (Fig. 2D). Because these six AFM-derived descriptors are extracted from the same creep-indentation framework, we next examined whether the stabilization index was sensitive to physical redundancy among descriptors. Correlation/PCA-based redundancy assessment and subset recalculation showed that the output-dispersion signature remained in the same order of magnitude when representative parameter subsets were used (Supplementary Table 1). For example, the two-parameter subset Eexp + η yielded values of 11.693% at the single time point and 11.363% across defined time points. The three-parameter subset Eexp + αlong + η yielded values of 11.044% and 11.163%, respectively, and a de-redundant minimal set spanning apparent elasticity, viscosity and network stiffness (Eexp + η + E2) yielded values of 9.217% and 8.997%, respectively. Thus, the low dispersion-transfer signature was not driven by a single choice of six-parameter aggregation, but was retained across representative elastic, viscous and network-level descriptors. To test temporal robustness, we profiled mechanics at 6, 12, 24, 48, and 72 h after AGEs exposure (Fig. 2E). The time‑series outputs recapitulated the single‑timepoint pattern (Fig. 2F), and the time‑dependent stabilization index remained essentially unchanged (CVoutput/CVinput = 4.75%) (Fig. 2G). Together, the full-parameter, subset-based, and time-series analyses identify CVoutput/CVinput as a robust quantitative signature of bio-stabilization and show that ViscoClamp maintains stable viscoelastic outputs despite highly variable glycation loading. Stable viscoelastic output rescues the cell–ECM interface by restoring collagen assembly and pericellular mechanics Having established that ViscoClamp compresses dispersed glycation inputs into a narrowly distributed viscoelastic output, we next asked how such mechanical stabilization becomes biologically actionable. Because MSC fate decisions are filtered through the pericellular ECM microdomain, a bio‑stabilizer must not only resist drift in bulk mechanics but also interpenetrate and hybridize with MSC‑deposited matrix, thereby stabilizing the mechanical microenvironment that cells actually sense (Fig. 3A). We therefore modeled glycation‑driven niche failure at the ECM level. MSCs were first cultured to confluence to promote robust ECM deposition, and then exposed to AGEs (0.1 mg/mL) with or without 1% ViscoClamp. After 24 h, SEM and AFM measurements showed that AGEs induced marked abnormalities in ECM organization and altered cell‑scale stiffness, whereas ViscoClamp largely restored both ECM ultrastructure and mechanical readouts toward control levels (Fig. 3B, C). Because absolute AFM values from isolated gel units and living cell-ECM hybrids are not directly interchangeable, we compared stabilization trends rather than raw parameter equivalence. Undergraded AGEs exposure, cell-surface AFM showed that MSC-surface mechanical readouts remained relatively stable in the presence of ViscoClamp across the tested concentration range, supporting consistency of the anti-drift trend in the cell-hybrid pericellular matrix system (Fig. S3D). In the same setting, quantification of conditioned medium showed that ViscoClamp reduced soluble AGEs by 61%, consistent with direct adsorption and sequestration (Fig. 3D).Fig. 3ViscoClamp restores collagen organization at the cell–ECM interface by adsorbing AGEs.A Schematic: AGEs disrupt collagen organization and weaken the MSC mechanical niche, impairing osteogenesis; ViscoClamp adsorbs AGEs, restores collagen assembly and promotes mineralization. B Pseudo-color SEM images of MSCs treated with AGEs (0.1 mg/mL) with or without 1% ViscoClamp for 24 h (MSC, red; ECM, cyan; scale bar, 1 μm). C AFM measurement of the MSC elastic modulus under the indicated conditions. D ELISA quantification of AGEs concentration in culture supernatants. E Collagen I (Col I) fluorescence staining of MSC-derived matrix under the indicated treatments (scale bar, 10 μm). Representative fluorescence images were acquired by CLSM (left), and the corresponding collagen architecture was reconstructed for structural analysis (right). Scale bar, 10 µm. F Quantification of Col I fibril length, diameter and Col I-positive area. G, K Confocal/3D Imaris analysis of FITC-ViscoClamp (green) and Cy3-AGEs (red) on the MSC surface after 6 h, with phalloidin/DAPI counterstaining: G images/reconstruction (scale bars as indicated); H high-magnification 3D rendering (scale bar, 3 μm); I line-scan profiles; J signal heatmap; K percentage colocalization. L–N Confocal/3D reconstruction of ViscoClamp, AGEs and MSC-secreted Col I after 6 h, with line-scan profiles and collagen-associated ViscoClamp-AGEs heatmap (scale bars as indicated). O, P RNA-seq and gene set enrichment analysis (GSEA) of AGEs-challenged MSCs treated with 1% ViscoClamp for 24 h versus AGEs-alone control, highlighting collagen biosynthesis and ECM remodeling pathways. B–N, n = 6 independent replicates. O, P, n = 3 independent replicates. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. ViscoClamp restores collagen organization at the cell–ECM interface by adsorbing AGEs. A Schematic: AGEs disrupt collagen organization and weaken the MSC mechanical niche, impairing osteogenesis; ViscoClamp adsorbs AGEs, restores collagen assembly and promotes mineralization. B Pseudo-color SEM images of MSCs treated with AGEs (0.1 mg/mL) with or without 1% ViscoClamp for 24 h (MSC, red; ECM, cyan; scale bar, 1 μm). C AFM measurement of the MSC elastic modulus under the indicated conditions. D ELISA quantification of AGEs concentration in culture supernatants. E Collagen I (Col I) fluorescence staining of MSC-derived matrix under the indicated treatments (scale bar, 10 μm). Representative fluorescence images were acquired by CLSM (left), and the corresponding collagen architecture was reconstructed for structural analysis (right). Scale bar, 10 µm. F Quantification of Col I fibril length, diameter and Col I-positive area. G, K Confocal/3D Imaris analysis of FITC-ViscoClamp (green) and Cy3-AGEs (red) on the MSC surface after 6 h, with phalloidin/DAPI counterstaining: G images/reconstruction (scale bars as indicated); H high-magnification 3D rendering (scale bar, 3 μm); I line-scan profiles; J signal heatmap; K percentage colocalization. L–N Confocal/3D reconstruction of ViscoClamp, AGEs and MSC-secreted Col I after 6 h, with line-scan profiles and collagen-associated ViscoClamp-AGEs heatmap (scale bars as indicated). O, P RNA-seq and gene set enrichment analysis (GSEA) of AGEs-challenged MSCs treated with 1% ViscoClamp for 24 h versus AGEs-alone control, highlighting collagen biosynthesis and ECM remodeling pathways. B–N, n = 6 independent replicates. O, P, n = 3 independent replicates. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. Collagen staining further revealed that AGEs caused pronounced collagen disorganization, shortening fiber length, reducing fiber number, and increasing apparent fiber diameter, whereas adding ViscoClamp substantially improved collagen alignment and polymerization features (Fig. 3E, F). To determine whether this rescue reflected physical integration at the cell surface, we labeled ViscoClamp with FITC and AGEs with Cy3 (Fig. 3G). Three‑dimensional confocal reconstruction indicated strong co‑localization of ViscoClamp and AGEs on the surface of MSCs (Fig. 3H). The relative coefficient R was 0.95 (Fig. 3I), and the overlapping area exceeded 81% (Fig. 3J, K). Moreover, imaging showed that the ViscoClamp-AGEs complex reorganized with MSC surface collagen I (Col I), consistent with formation of a hybrid pericellular matrix that counteracts AGEs‑induced suppression of extracellular collagen assembly (Fig. 3L–N). To link structural rescue to molecular programs, we performed transcriptomic profiling of AGEs‑challenged MSCs with and without ViscoClamp. Transcriptomic profiling followed by gene set enrichment analysis (GSEA) showed enrichment of collagen formation/organization and ECM remodeling pathways in ViscoClamp‑treated MSCs compared with the AGEs alone condition (Fig. 3O, P). Collectively, these results indicate that ViscoClamp adsorbs AGEs, integrates with the pericellular matrix, and restores collagen architecture and mechanics at the cell scale, providing a mechanistic bridge from stabilized viscoelastic output to stabilized cellular niche input. ViscoClamp remains permissive to mineralization while rescuing osteogenic differentiation under glycation stress Because craniofacial regeneration requires not only early osteogenic initiation but also sustained matrix maturation, we next evaluated whether ViscoClamp, while stabilizing and hybridizing with the ECM, remains compatible with mineral deposition and osteogenic progression under glycation. We first challenged ViscoClamp with CaCl₂ in the presence or absence of AGEs and tracked mineral deposition over time. Cryo‑EM showed progressive Ca²⁺ accumulation on ViscoClamp with time, and notably, AGEs further enhanced mineral deposition relative to Ca²⁺ alone (Fig. 4A, B). AFM topographic imaging confirmed increased surface mineral build‑up after ViscoClamp treatment (Fig. 4C, D). In agreement, Von Kossa staining across early time points (0 ~ 6 h) indicated that, when AGEs were present, ViscoClamp showed stronger mineral-associated staining (Fig. 4E–G).Fig. 4ViscoClamp preserves mineralization competence and restores Runx2 signaling under AGEs disruption.A, B Cryo-EM of Ca²⁺ deposition on ViscoClamp at 0, 6, 12 and 24 h after CaCl₂ incubation, with or without AGEs; red dots mark deposits (scale bar, 10 nm), and B quantifies Ca²⁺-positive area. C, D AFM 3D topography and Ca²⁺ deposition-intensity heatmap after 24 h CaCl₂ incubation with or without AGEs. E, F Von Kossa staining and quantification of ViscoClamp mineralization at 0, 0.25, 0.5, 1, 3 and 6 h after CaCl₂ incubation, with or without AGEs. G Summary of Ca²⁺ deposition across conditions and time points. H–J Osteogenic differentiation of MSCs under control, 0.1 mg/ml AGEs or 0.1 mg/ml AGEs + 1% ViscoClamp: ARS staining and Col I IF at days 7, 14 and 21 (scale bar, 10 μm), ARS absorbance (I) and ARS-Col I colocalization (J). K Runx2 IF in MSCs treated with 1% ViscoClamp and increasing AGEs, with nuclei highlighted (scale bar, 50 μm). p values versus MSC for 0, 0.05, 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: Runx2 expression, 2.0e-7, 2.5e-8, 1.4e-8, 9.1e-10, 8.5e-10 and 1.8e-10; Runx2 nucleus/cytosol ratio, 1.6e-9, 2.9e-11, 3.0e-10, 6.2e-10, 3.9e-11 and 2.8e-9. L Runx2 luciferase reporter readout; p values versus MSC for 0.05, 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: 3.9e-7, 1.8e-10, 2.8e-8, 3.2e-9 and 4.3e-9. M Alp activity in supernatants; p values versus MSC for 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: 10e-7, 7.1e-6, 5.1e-6 and 3.5e-9. N, O Western blot and heatmap of nuclear/cytosolic Runx2, with Lamin A/C and β-actin controls. (P, Q) Western blot and heatmap of Runx2 and Alp after 7 d osteogenic induction under AGEs with or without ViscoClamp. A–M, n = 6 independent replicates; N–Q were repeated independently three times. Data are mean ± SD. Box plots in (B, F, I and J) show median, 25th/75th percentiles and min–max whiskers. Unpaired two-tailed Student’s t test; p < 0.05. Source data are provided as a Source Data file. ViscoClamp preserves mineralization competence and restores Runx2 signaling under AGEs disruption. A, B Cryo-EM of Ca²⁺ deposition on ViscoClamp at 0, 6, 12 and 24 h after CaCl₂ incubation, with or without AGEs; red dots mark deposits (scale bar, 10 nm), and B quantifies Ca²⁺-positive area. C, D AFM 3D topography and Ca²⁺ deposition-intensity heatmap after 24 h CaCl₂ incubation with or without AGEs. E, F Von Kossa staining and quantification of ViscoClamp mineralization at 0, 0.25, 0.5, 1, 3 and 6 h after CaCl₂ incubation, with or without AGEs. G Summary of Ca²⁺ deposition across conditions and time points. H–J Osteogenic differentiation of MSCs under control, 0.1 mg/ml AGEs or 0.1 mg/ml AGEs + 1% ViscoClamp: ARS staining and Col I IF at days 7, 14 and 21 (scale bar, 10 μm), ARS absorbance (I) and ARS-Col I colocalization (J). K Runx2 IF in MSCs treated with 1% ViscoClamp and increasing AGEs, with nuclei highlighted (scale bar, 50 μm). p values versus MSC for 0, 0.05, 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: Runx2 expression, 2.0e-7, 2.5e-8, 1.4e-8, 9.1e-10, 8.5e-10 and 1.8e-10; Runx2 nucleus/cytosol ratio, 1.6e-9, 2.9e-11, 3.0e-10, 6.2e-10, 3.9e-11 and 2.8e-9. L Runx2 luciferase reporter readout; p values versus MSC for 0.05, 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: 3.9e-7, 1.8e-10, 2.8e-8, 3.2e-9 and 4.3e-9. M Alp activity in supernatants; p values versus MSC for 0.1, 0.2, 0.4 and 0.8 mg/ml AGEs: 10e-7, 7.1e-6, 5.1e-6 and 3.5e-9. N, O Western blot and heatmap of nuclear/cytosolic Runx2, with Lamin A/C and β-actin controls. (P, Q) Western blot and heatmap of Runx2 and Alp after 7 d osteogenic induction under AGEs with or without ViscoClamp. A–M, n = 6 independent replicates; N–Q were repeated independently three times. Data are mean ± SD. Box plots in (B, F, I and J) show median, 25th/75th percentiles and min–max whiskers. Unpaired two-tailed Student’s t test; p < 0.05. Source data are provided as a Source Data file. Next, we examined whether ViscoClamp could rescue osteogenic differentiation of MSCs under glycation. During osteogenic induction, Alizarin Red S (ARS) and collagen staining at days 7, 14, and 21 showed that AGEs markedly suppressed mineralized nodule formation and reduced ARS-collagen I (Col I) co‑localization. Co‑treatment with 1% ViscoClamp largely restored mineral deposition and matrix organization (Fig. 4H-J). Because nuclear translocation of Runx2 is a key step in osteogenic commitment, we assessed Runx2 localization and activity. ViscoClamp alone did not perturb Runx2 nuclear localization in the absence of AGEs, but under AGEs exposure it increased nuclear Runx2 accumulation (Fig. 4K; Fig. S4A). Consistently, a Runx2 reporter assay showed sustained transcriptional activation across a range of AGEs doses in the presence of ViscoClamp, accompanied by increased Alp activity (Fig. 4L, M). To further test whether this osteogenic rescue was retained in a three-dimensional cell model, we cultured MSCs as spheroids in spheroid-forming honeycomb plates. After overnight spheroid formation, spheroids were switched to osteogenic induction medium containing 0.1 mg/mL AGEs and treated with PBS or 1% ViscoClamp. After 7 days of induction, ViscoClamp increased Alp and Sp7 fluorescence in AGEs-challenged MSC spheroids compared with PBS control, indicating that ViscoClamp preserves osteogenic marker induction in a 3D pericellular context (Fig. S4I–S4L). To connect pericellular mechanical stabilization with biochemical osteogenic signaling, we examined mechanotransduction pathways in MSCs under AGEs stress. RNA-seq/GSEA showed enrichment of Integrin complex, Focal adhesion, together with a significant shift in Hippo-associated signaling in ViscoClamp-treated MSCs (Fig. S4B). Functionally, under AGEs stress, inhibition of Fak using PF-573228 attenuated ViscoClamp-mediated Runx2 nuclear localization and reduced Runx2, Alp and Sp7 expression during osteogenic induction (Fig. S4C–S4H). These data support a Fak-associated route by which stabilized pericellular mechanics may be converted into Runx2-centered osteogenic transcriptional activation. Notably, varying AGE concentrations did not appreciably alter the magnitude of ViscoClamp‑mediated Runx2 nuclear translocation (Fig. 4K–O). Immunoblotting further confirmed that AGEs reduced the expression of Alp and Runx2 by more than 2-fold compared to the control group, while ViscoClamp restored the levels of both markers (Figs. 4P, Q). Together, these data indicate that ViscoClamp mitigates glycation stress while preserving a microenvironment that supports mineralization and Runx2‑dependent osteogenic activation. Pericellular micromechanics stabilization is associated with early regenerative programs and improves repair in glycation-challenged rat mandibular defects We next asked whether pericellular viscoelastic clamping translates into robust regeneration in vivo under sustained glycation burden. We evaluated ViscoClamp using an AGEs-injection rat mandibular defect model (Fig. S5A). Repeated AGEs administration did not appreciably alter body weight (Fig. S5B) but increased serum AGEs by 2.9-fold relative to PBS-injected controls (Fig. S5C), confirming a sustained glycation burden. Three days after the final injection, bilateral mandibular defects were created and immediately treated with either saline (Ctrl) or ViscoClamp (Fig. S5A). To capture early regenerative programs before substantial mineral deposition, defect-region tissues were harvested at day 7 for single-cell RNA sequencing (scRNA-seq) (Fig. 5). For each condition, three mandibular-defect tissues were pooled before single-cell library preparation, yielding one pooled Ctrl library and one pooled ViscoClamp library; therefore, single-cell analyses were used to generate descriptive, hypothesis-generating evidence rather than donor-level statistical inference.Fig. 5Single-cell transcriptomics delineates ViscoClamp-associated programs in rat mandibular defect tissues at day 7.A t-distributed stochastic neighbor embedding (t-SNE) of single-cell RNA-sequencing (scRNA-seq) profiles from mandibular defect and peri-defect tissues 7 d after treatment. Three mandibular-defect tissues were pooled per condition before library preparation, yielding one Ctrl and one ViscoClamp library; dots represent cells colored by annotated cell type. B Heatmap of the top three marker genes for each major cell type. C, D Re-clustering of osteogenesis-related cells identifies mesenchymal stem cells (MSCs), osteoprogenitors, osteoblasts, osteocytes and adipocytes, shown by t-SNE with stacked bars and pie charts summarizing Ctrl and ViscoClamp proportions. E Monocle pseudotime trajectory showing osteogenic (branch 1) and adipogenic (branch 2) fates. F Differentiation-potential scores projected on t-SNE and summarized by subtype box plots. G, H Feature plots and expression dynamics of osteogenic/adipogenic lineage markers, including fitted curves and violin plots comparing Ctrl versus ViscoClamp. I, J GO and KEGG analyses of MSCs in the Ctrl and ViscoClamp pooled libraries. K KEGG pathway enrichment highlighting MSC adhesion- and differentiation-associated programs and representative genes. L CellChat-inferred MSC-centered intercellular communication network. M, N CellChat summary of the top significant ligand-receptor interactions among osteogenic-lineage subpopulations. O–Q Visualization of cellular communication in the Integrin-Fak pathway and Wnt pathway. p values shown for gene-expression or enrichment panels reflect the indicated cell-level or gene-set tests. E, H–J, n = 3 per group. Quantitative data are presented as mean ± SD. Box plots in F, H and Q: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using Wilcoxon rank-sum tests for E and H and unpaired two-tailed Student’s t tests for I and J; p < 0.05 was considered significant. Single-cell transcriptomics delineates ViscoClamp-associated programs in rat mandibular defect tissues at day 7. A t-distributed stochastic neighbor embedding (t-SNE) of single-cell RNA-sequencing (scRNA-seq) profiles from mandibular defect and peri-defect tissues 7 d after treatment. Three mandibular-defect tissues were pooled per condition before library preparation, yielding one Ctrl and one ViscoClamp library; dots represent cells colored by annotated cell type. B Heatmap of the top three marker genes for each major cell type. C, D Re-clustering of osteogenesis-related cells identifies mesenchymal stem cells (MSCs), osteoprogenitors, osteoblasts, osteocytes and adipocytes, shown by t-SNE with stacked bars and pie charts summarizing Ctrl and ViscoClamp proportions. E Monocle pseudotime trajectory showing osteogenic (branch 1) and adipogenic (branch 2) fates. F Differentiation-potential scores projected on t-SNE and summarized by subtype box plots. G, H Feature plots and expression dynamics of osteogenic/adipogenic lineage markers, including fitted curves and violin plots comparing Ctrl versus ViscoClamp. I, J GO and KEGG analyses of MSCs in the Ctrl and ViscoClamp pooled libraries. K KEGG pathway enrichment highlighting MSC adhesion- and differentiation-associated programs and representative genes. L CellChat-inferred MSC-centered intercellular communication network. M, N CellChat summary of the top significant ligand-receptor interactions among osteogenic-lineage subpopulations. O–Q Visualization of cellular communication in the Integrin-Fak pathway and Wnt pathway. p values shown for gene-expression or enrichment panels reflect the indicated cell-level or gene-set tests. E, H–J, n = 3 per group. Quantitative data are presented as mean ± SD. Box plots in F, H and Q: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using Wilcoxon rank-sum tests for E and H and unpaired two-tailed Student’s t tests for I and J; p < 0.05 was considered significant. scRNA-seq profiling identified 18 major cell types spanning immune, stromal and skeletal compartments (Fig. 5A, B; Fig. S5D and S5E). Focusing on the osteogenesis-related compartment, we resolved five transcriptionally distinct subpopulations, including MSCs, osteoprogenitors, osteoblasts, osteocytes, and adipocytes (Fig. 5C). In the pooled-library comparison, MSCs represented the predominant osteogenic-lineage state, and the MSC fraction was higher in the ViscoClamp library than in the Ctrl library (Fig. 5D; Fig. S5F). To distinguish broad population enrichment from lineage-state redistribution, we further summarized pooled-library cell-state distributions as delta-proportion effect sizes across 18 major cell types, five osteogenic-lineage states, Monocle-inferred pseudotime bins and osteogenic/adipogenic trajectory branches (Figs. S5D and S5F-S5H). These analyses suggest that the ViscoClamp-associated single-cell pattern reflects both enrichment of osteogenic-lineage states and redistribution of cells along pseudotime and trajectory branches, rather than only a pooled increase in one pre-existing subpopulation. Nevertheless, because the dataset contains one pooled library per condition, these distribution analyses remain descriptive and cannot by themselves distinguish causal lineage-intrinsic normalization from population-level selection. Exploratory trajectory inference reconstructed a continuous differentiation landscape with a prominent bifurcation consistent with osteogenic versus adipogenic fates, and differentiation-potential scoring suggested a pooled-library-level shift toward higher progenitor-like potential within the ViscoClamp-associated compartment (Fig. 5E, F; Fig. S5E–S5I). At the transcriptional level, ViscoClamp was associated with coordinated changes in lineage programs, including altered expression patterns of canonical osteogenic differentiation markers across subpopulations (Fig. 5G, H; Fig. S5J). Pathway analyses further suggested that ViscoClamp-treated MSCs were enriched for signaling modules linked to bone formation and matrix mineralization, including integrin-mediated cell-ECM adhesion, Ca2+-linked signaling, mechanotransduction and positive regulation of ossification (Fig. 5I–K; Fig. S5K and Fig. S5L). We also examined immune-lineage responses in the same pooled day-7 scRNA-seq dataset. Immune-compartment analysis showed ViscoClamp-associated changes in macrophage/monocyte, neutrophil and T-cell programs (Fig. S6A–S6D). Macrophage reclustering further revealed altered macrophage functional states, with the ViscoClamp library showing lower pro-inflammatory/M1-like and inflammatory-response signatures and higher M2-like, phagocytosis/clearance, ECM-remodeling and wound-healing signatures (Fig. S6E–S6G). Because the libraries were generated after pooling three defect tissues per group, these immune analyses are interpreted as descriptive and hypothesis-generating, but they support an association between ViscoClamp treatment and a more reparative early immune ecology. Beyond intrinsic MSC programs, exploratory CellChat-based inference positioned MSCs as a potential central hub of intercellular communication within the early osteogenic niche, showing prominent incoming and outgoing signaling connections in the pooled-library comparison (Fig. 5L). Ligand-receptor analyses highlighted distinct axes associated with lineage commitment, and network interrogation of ECM-associated and developmental signaling circuits identified prominent Integrin-Fak and Wnt communication patterns across osteogenic-lineage populations (Fig. 5M-Q). Together, these pooled single-cell data suggest that ViscoClamp is associated with an early MSC-centered, pro-osteogenic and more reparative immune-stromal niche in glycation-perturbed mandibular defects. Because the scRNA-seq libraries were generated from pooled samples at one early time point, we interpret the trajectory, cell-communication and macrophage-program analyses as hypothesis-generating molecular evidence that complements the later structural repair phenotype. Consistent with these early niche-level shifts, ViscoClamp produced marked improvements in defect repair at later time points. At 2 months, micro-CT analysis revealed a marked enhancement in bone regeneration within the ViscoClamp-treated defects (Fig. 6A). Quantitative assessment indicated increases in bone mineral density (BMD) by a factor of 1.2, bone volume to total volume ratio (BV/TV) by 1.3 times, trabecular number (Tb.N) by 1.9 times, and trabecular thickness (Tb.Th) by 1.5 times, along with decreases in bone surface to bone volume ratio (BS/BV) by approximately half and trabecular separation (Tb.Sp) by 69% compared with saline controls (Fig. 6B). Sequential fluorochrome labeling demonstrated faster mineral apposition: the mineralization rate increased 1.5-fold from weeks 2 to 4 and by 87% from weeks 4 to 8 in the ViscoClamp group relative to controls (Fig. 6C). Histological staining (Masson, VG, H&E) corroborated these findings, revealing substantially greater bone fill and maturation, with approximately doubled new-bone area (Fig. 6D, E). To benchmark the regenerated tissue against native bone, we further compared ViscoClamp-treated new bone with adjacent autologous mandibular bone. Quantification of H&E staining intensity showed a comparable histological-intensity range between newly formed and autologous bone, whereas micro-CT BMD remained significantly different from the native/autologous reference, indicating active remodeling rather than complete restoration of mature native architecture (Fig. S7A, B). Von Kossa staining further demonstrated 1.7-fold higher mineral deposition in the ViscoClamp group (Fig. 6F, G).Fig. 6ViscoClamp restores mineralization and promotes bone regeneration in AGE-injection (AI) rats.A Representative micro-CT images of rat mandibular defects at 2 months after treatment (Ctrl versus ViscoClamp) (scale bar, 1 mm). B Micro-CT quantification of BMD, BV/TV, BS/BV, Tb.N, Tb.Th and Tb.Sp. C Sequential fluorochrome labeling (calcein, alizarin red and tetracycline) and quantification of new bone formation across weeks 2 to 4 and 4 to 8 (scale bar, 30 μm). D, E Representative Masson’s trichrome, Van Gieson (VG) and H&E staining of the mandibular defect regions 2 months after treatment, with histomorphometric quantification (E) of new bone formation (new bone area/defect area) (scale bar, 1 mm). F, G Von Kossa staining and calcium-deposition area fraction (scale bar, 1 mm). H, I CD73/Alp co-staining by CLSM and ImageJ quantification of CD73-Alp signal correlation and CD73⁺Alp⁺ cells (scale bar, 1 mm; 500 μm). J, K AGEs staining with fluorescence intensity and AGEs-positive area fraction quantification (scale bar, 1 mm; 500 μm).Immunofluorescence staining for osteocalcin (Ocn) and Col I in mandibular defects 2 months after treatment (L, N), with quantification of Ocn and Col I intensity and area (M, O) (scale bar, 1 mm; 500 μm). P–S Proteomics, GSEA plots and heatmap comparing ViscoClamp versus Ctrl, highlighting collagen/ECM, bone-development, mineralization and mesenchymal-differentiation programs. NB, new bone; JB, jawbone. A, B, D–O, n = 6 per group; C, P–S, n = 3 per group. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. ViscoClamp restores mineralization and promotes bone regeneration in AGE-injection (AI) rats. A Representative micro-CT images of rat mandibular defects at 2 months after treatment (Ctrl versus ViscoClamp) (scale bar, 1 mm). B Micro-CT quantification of BMD, BV/TV, BS/BV, Tb.N, Tb.Th and Tb.Sp. C Sequential fluorochrome labeling (calcein, alizarin red and tetracycline) and quantification of new bone formation across weeks 2 to 4 and 4 to 8 (scale bar, 30 μm). D, E Representative Masson’s trichrome, Van Gieson (VG) and H&E staining of the mandibular defect regions 2 months after treatment, with histomorphometric quantification (E) of new bone formation (new bone area/defect area) (scale bar, 1 mm). F, G Von Kossa staining and calcium-deposition area fraction (scale bar, 1 mm). H, I CD73/Alp co-staining by CLSM and ImageJ quantification of CD73-Alp signal correlation and CD73⁺Alp⁺ cells (scale bar, 1 mm; 500 μm). J, K AGEs staining with fluorescence intensity and AGEs-positive area fraction quantification (scale bar, 1 mm; 500 μm).Immunofluorescence staining for osteocalcin (Ocn) and Col I in mandibular defects 2 months after treatment (L, N), with quantification of Ocn and Col I intensity and area (M, O) (scale bar, 1 mm; 500 μm). P–S Proteomics, GSEA plots and heatmap comparing ViscoClamp versus Ctrl, highlighting collagen/ECM, bone-development, mineralization and mesenchymal-differentiation programs. NB, new bone; JB, jawbone. A, B, D–O, n = 6 per group; C, P–S, n = 3 per group. Quantitative data are presented as mean ± SD. Statistical significance was assessed using unpaired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. Immunofluorescence analyses provided cellular and matrix correlates for this repair phenotype. ViscoClamp led to a 3.7-fold increase in the abundance of CD73⁺Alp⁺ double-positive osteoblast precursor cells (Fig. 6H, I). Meanwhile, the local AGEs signal within the defect exhibited an increase (4.5-fold in area; 6.9-fold in intensity) (Fig. 6J, K), which was accompanied by elevated levels of osteogenic matrix markers such as Ocn (16.9-fold in area; 3.4-fold in intensity) and Col I (6.3-fold in area; 3.9-fold in intensity) (Fig. 6L–O). This pattern is consistent with the sequestration of AGEs within the gel-ECM hybrid niche, rather than their systemic clearance. Proteomics analysis and GSEA of 1-week tissues showed enrichment of collagen formation/fibril organization and ECM remodeling pathways, as well as bone development, mineralization and mesenchymal differentiation programs in ViscoClamp-treated defects (Fig. 6P–S). Consistent with the immune-compartment scRNA-seq analysis, early tissue immunostaining showed that ViscoClamp-treated defects had a higher M2/M1-like macrophage-polarization ratio (9.2-fold) and increased vascular-marker-positive area or vessel density (3.9-fold), supporting an association between ViscoClamp treatment and a more reparative immune-vascular niche during early repair (Fig. S7C, S7D). Systemic biosafety was corroborated by hematology, serum biochemistry, and organ histology at 2 months, which revealed no discernible toxicity (Fig. S8A, S8B). To further evaluate oral-exposure safety margins, healthy Sprague-Dawley rats were administered a 10-fold dose of ViscoClamp by gavage for 7 consecutive days; hematology, serum chemistry, and H&E histology remained within normal ranges (Fig. S8C, S8D). ViscoClamp improves the repair of rabbit mandibular and calvarial defects while maintaining long-term biosafety To validate efficacy in a larger craniofacial context, we established an AGEs‑injection New Zealand white rabbit model (Fig. 7A). AGEs administration did not significantly impact body weight but increased serum AGEs by 3.8‑fold (Fig. 7B). After creating bilateral mandibular defects, saline (Ctrl) and ViscoClamp were applied to the contralateral sides (left Ctrl; right ViscoClamp; n = 6) (Fig. 7A). At 1 month, micro‑CT revealed markedly enhanced mandibular regeneration in the ViscoClamp group (Fig. 7C). Specifically, BMD increased by 38%, BV/TV increased 2.1‑fold, Tb.N increased 1.2‑fold, and Tb.Th increased 1.1‑fold, whereas BS/BV decreased by 35% and Tb.Sp decreased by 58% (Fig. 7D). Histological analysis (Masson, VG, and H&E) confirmed increased bone fill and maturation (Fig. 7E), and quantitative analysis demonstrated a 1.9‑fold increase in new‑bone area (Fig. 7F). Immunofluorescence further indicated increased CD73⁺Alp⁺ double‑positive osteoblast precursor cells (2.7‑fold) (Fig. 7G, H), increased local AGEs signal (area 3.1‑fold; intensity 2.5‑fold) (Fig. 7I, J), and higher osteogenic matrix markers including Ocn (area 7.1‑fold; intensity 1.6‑fold) and Col I (area 4.4‑fold; intensity 2.4‑fold) (Fig. 7K–N). These findings again support a mechanism in which ViscoClamp locally captures the glycation burden while stabilizing an osteo‑permissive niche.Fig. 7ViscoClamp promotes glycation-resilient craniofacial bone regeneration in AGEs-injection (AI) rabbits.A Study design: 3-month-old male New Zealand white rabbits received AGEs (10 mg/kg) three times weekly for 4 weeks before mandibular and calvarial defects; contralateral sites received saline (Ctrl) or 6% (w/v) ViscoClamp, and AGEs injections continued. B Body weight and serum AGEs levels in rabbits during the AGEs-injection period. C, D Micro-CT images and quantification of mandibular defects 1 month after treatment (scale bar, 2 mm). E, F Representative Masson’s trichrome, VG and H&E staining of the mandibular defect regions 1 month after treatment, with histomorphometric quantification (F) of new bone formation (scale bar, 1 mm). G, H Immunofluorescence co-staining of CD73 and Alp in mandibular defects imaged by CLSM and ImageJ quantification of signal correlation and CD73⁺Alp⁺ cells (scale bar, 1 mm; 500 μm). I, J Immunofluorescence staining for AGEs in mandibular defects (I), with quantification of AGEs fluorescence intensity and AGEs-positive area fraction (J) (scale bar, 1 mm; 500 μm). K–N Immunofluorescence staining for Ocn and Col I in mandibular defects 1 month after treatment (K, M), with quantification of Ocn and Col I intensity and area (L, N) (scale bar, 1 mm; 500 μm). O, P Micro-CT analysis of calvarial defects 3 months after treatment (scale bar, 1 mm). Q, R Histological staining of calvarial defects 3 months after treatment (Masson’s trichrome, H&E and VG) (R), with histomorphometric quantification of new bone formation (new bone area/defect area) (Q) (scale bar, 500 μm). S, T Calvarial AGEs staining with fluorescence intensity (S) and AGEs-positive area fraction (T) (scale bar, 1 mm; 500 μm). AB alveolar bone; NB new bone. A–N, n = 6 per group. O-T, n = 6 per group. Quantitative data are presented as mean ± SD. Box plots: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using paired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. ViscoClamp promotes glycation-resilient craniofacial bone regeneration in AGEs-injection (AI) rabbits. A Study design: 3-month-old male New Zealand white rabbits received AGEs (10 mg/kg) three times weekly for 4 weeks before mandibular and calvarial defects; contralateral sites received saline (Ctrl) or 6% (w/v) ViscoClamp, and AGEs injections continued. B Body weight and serum AGEs levels in rabbits during the AGEs-injection period. C, D Micro-CT images and quantification of mandibular defects 1 month after treatment (scale bar, 2 mm). E, F Representative Masson’s trichrome, VG and H&E staining of the mandibular defect regions 1 month after treatment, with histomorphometric quantification (F) of new bone formation (scale bar, 1 mm). G, H Immunofluorescence co-staining of CD73 and Alp in mandibular defects imaged by CLSM and ImageJ quantification of signal correlation and CD73⁺Alp⁺ cells (scale bar, 1 mm; 500 μm). I, J Immunofluorescence staining for AGEs in mandibular defects (I), with quantification of AGEs fluorescence intensity and AGEs-positive area fraction (J) (scale bar, 1 mm; 500 μm). K–N Immunofluorescence staining for Ocn and Col I in mandibular defects 1 month after treatment (K, M), with quantification of Ocn and Col I intensity and area (L, N) (scale bar, 1 mm; 500 μm). O, P Micro-CT analysis of calvarial defects 3 months after treatment (scale bar, 1 mm). Q, R Histological staining of calvarial defects 3 months after treatment (Masson’s trichrome, H&E and VG) (R), with histomorphometric quantification of new bone formation (new bone area/defect area) (Q) (scale bar, 500 μm). S, T Calvarial AGEs staining with fluorescence intensity (S) and AGEs-positive area fraction (T) (scale bar, 1 mm; 500 μm). AB alveolar bone; NB new bone. A–N, n = 6 per group. O-T, n = 6 per group. Quantitative data are presented as mean ± SD. Box plots: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using paired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. In the same AGEs‑injection rabbit model, we further investigated a calvarial defect model to assess bone formation under a distinct craniofacial geometry and healing environment (Fig. 7A). At 3 months, micro‑CT analysis indicated that ViscoClamp significantly enhanced calvarial regeneration (Fig. 7O). Specifically, BMD increased by 57%, BV/TV increased 1.1‑fold, Tb.N increased 1.2‑fold, and Tb.Th increased by 28%, whereas BS/BV decreased by 21% and Tb.Sp decreased by 59% (Fig. 7P). Histological examination (Masson, H&E, and VG staining) confirmed improved defect bridging and matrix deposition, with a 1.1‑fold increase in new‑bone area (Fig. 7Q, R). AGEs staining was more intense in the ViscoClamp group (area 5.5‑fold; intensity 7.2‑fold) (Fig. 7S, T), consistent with sustained local sequestration over long‑term healing. Biosafety was corroborated by hematology/biochemistry and organ histology, which indicated no adverse alterations at 1 month (Fig. S9A, S9B). Long‑term biosafety was further supported by normal organ histology and hematology/biochemistry findings at 3 months (Fig. S9C, S9D). Large-animal and non-human primate models demonstrate robustness under hyperglycemia and functional loading Small‑animal models can establish mechanistic plausibility. However, craniofacial translation necessitates validation under human‑scale defect geometry, slower remodeling kinetics, and continuous masticatory loading, factors that become even more challenging under hyperglycemia. Therefore, we evaluated ViscoClamp in STZ‑induced hyperglycemic large animals and a non‑human primate, where both the mechanical environment and regenerative timescale more closely resemble clinical craniofacial repair. In 12-month-old male beagle dogs, a single intravenous injection of STZ (40 mg/kg) successfully induced hyperglycemia, as depicted in Fig. 8A. The fasting blood glucose level increased by 1.6-fold, and the serum AGEs level increased by 3.3-fold when compared with solvent controls, as shown in Fig. 8B, C. A mandibular incisor extraction socket model was established using a paired design. Specifically, the left sockets were treated with saline (Ctrl), while the right sockets were immediately filled with 6% ViscoClamp (n = 3 per group), as presented in Fig. 8A. At the 2-month time point, in vivo micro-CT analysis revealed significantly enhanced socket bone regeneration in the ViscoClamp group, as shown in Figs. 8D and 8E. There were notable increases in BMD (36%), BV/TV (1.8-fold), Tb.N (43%), and Tb.Th (30%), along with significant decreases in BS/BV (42%) and Tb.Sp (46%). Histological examinations (Masson, VG, and H&E) confirmed an increased trabecular bone formation and maturation, with a 1.5-fold increase in the new-bone area, as shown in Fig. 8F, G. Systemic hematology/biochemistry and organ histology results showed no detectable toxicity, as presented in Fig. S10A, S10B, which supports the biosafety of the treatment in a large-animal oral environment.Fig. 8ViscoClamp enhances glycation-resilient craniofacial bone regeneration in hyperglycemic large-animal models.A Beagle dog model: male animals (12-month-old) received streptozotocin (STZ; 40 mg/kg, i.v.) to induce hyperglycemia, followed by creation of extraction-socket defects treated with saline (Ctrl) or 6% (w/v) ViscoClamp. (B) Blood glucose levels measured 2 weeks after STZ injection in beagle dogs. C Serum AGEs levels measured 2 weeks after STZ injection by ELISA. D, E In vivo micro-CT analysis and quantification of extraction-socket defects in hyperglycemic beagle dogs 2 months after treatment (Ctrl versus ViscoClamp) (scale bar, 2 mm). F, G Representative Masson’s trichrome, VG and H&E staining of extraction sockets 2 months after treatment, with histomorphometric quantification of new bone formation in G (scale bars, 2 mm). H Rhesus macaque model: male animals (5-year-old) received STZ (45 mg/kg, i.v.) to induce hyperglycemia, followed by creation of mandibular defects treated with saline (Ctrl) or 6% (w/v) ViscoClamp. I Blood glucose levels measured 2 weeks after STZ injection in rhesus macaques. (J) Serum AGEs levels measured 2 weeks after STZ injection in rhesus macaques. K Intraoperative photographs of mandibular defect creation, implantation of 6% ViscoClamp and wound closure (scale bar, 2 mm). L, M Micro-CT analysis and quantification of mandibular defects in hyperglycemic rhesus macaques 1 week after treatment (Ctrl versus ViscoClamp; scale bar, 3 mm). N, O Micro-CT analysis and quantification of mandibular defects in hyperglycemic rhesus macaques 1 month after treatment (Ctrl versus ViscoClamp; scale bar, 3 mm). ES, extraction socket; NB, new bone. A–G, n = 3 per group. H-O n = 3 per group. Quantitative data are presented as mean ± SD. Box plots: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using paired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. ViscoClamp enhances glycation-resilient craniofacial bone regeneration in hyperglycemic large-animal models. A Beagle dog model: male animals (12-month-old) received streptozotocin (STZ; 40 mg/kg, i.v.) to induce hyperglycemia, followed by creation of extraction-socket defects treated with saline (Ctrl) or 6% (w/v) ViscoClamp. (B) Blood glucose levels measured 2 weeks after STZ injection in beagle dogs. C Serum AGEs levels measured 2 weeks after STZ injection by ELISA. D, E In vivo micro-CT analysis and quantification of extraction-socket defects in hyperglycemic beagle dogs 2 months after treatment (Ctrl versus ViscoClamp) (scale bar, 2 mm). F, G Representative Masson’s trichrome, VG and H&E staining of extraction sockets 2 months after treatment, with histomorphometric quantification of new bone formation in G (scale bars, 2 mm). H Rhesus macaque model: male animals (5-year-old) received STZ (45 mg/kg, i.v.) to induce hyperglycemia, followed by creation of mandibular defects treated with saline (Ctrl) or 6% (w/v) ViscoClamp. I Blood glucose levels measured 2 weeks after STZ injection in rhesus macaques. (J) Serum AGEs levels measured 2 weeks after STZ injection in rhesus macaques. K Intraoperative photographs of mandibular defect creation, implantation of 6% ViscoClamp and wound closure (scale bar, 2 mm). L, M Micro-CT analysis and quantification of mandibular defects in hyperglycemic rhesus macaques 1 week after treatment (Ctrl versus ViscoClamp; scale bar, 3 mm). N, O Micro-CT analysis and quantification of mandibular defects in hyperglycemic rhesus macaques 1 month after treatment (Ctrl versus ViscoClamp; scale bar, 3 mm). ES, extraction socket; NB, new bone. A–G, n = 3 per group. H-O n = 3 per group. Quantitative data are presented as mean ± SD. Box plots: center line, median; bounds, 25th/75th percentiles; whiskers, minimum/maximum. Statistical significance was assessed using paired two-tailed Student’s t test; p < 0.05 was considered significant. Source data are provided as a Source Data file. We then transitioned to a rhesus macaque model, which provides the closest approximation to human craniofacial anatomy and loading patterns among preclinical systems (Fig. 8H). In 5‑year‑old male macaques, a single intravenous administration of STZ (45 mg/kg) induced hyperglycemia, increasing fasting glucose by almost 94% and serum AGEs by 2.9‑fold (Fig. 8I, J). A mandibular defect (10 × 5 × 6 mm) was created adjacent to the molar region, and a paired design was used: the left‑sided defect received saline (Ctrl) and the right‑sided defect received 6% ViscoClamp (Fig. 8H, K). Micro-CT examination at 1 week indicated accelerated early bone formation in the ViscoClamp-treated defects (Fig. 8L). There were increases in BMD by 29%, BV/TV by 5.6-fold, Tb.N by 2.4-fold, and Tb.Th by 64%, as well as decreases in BS/BV by 47% and Tb.Sp by 65% when compared to the paired controls (Fig. 8M). At 1 month, micro-CT scans continued to demonstrate greater bone fill and maturation in the ViscoClamp-treated defects (Fig. 8N, O). Given that a posterior mandibular defect with a depth of 6 mm can pose a risk of injury to adjacent tooth roots, the molar within the defect region exfoliated during the follow-up period and was not visible on the 1-month micro-CT scan. Systemic monitoring, including electrocardiography, organ ultrasound, and hematology/biochemistry, revealed no signs of cardiotoxicity or organ abnormalities (Fig. S10C, S10D). Collectively, these data suggest that ViscoClamp maintains functional performance and biosafety under hyperglycemia, chronic glycation burden, and craniofacial mechanical loading—key prerequisites for translating MMHR bio‑stabilizers toward clinically relevant craniofacial repair. Discussion MMHR reframes mechano-regulation as a materials problem: maintaining the mechanics that cells actually integrate—mesoscopic cell‑to‑pericellular viscoelasticity—rather than optimizing static bulk stiffness1–4,8. By treating pericellular viscoelastic homeostasis as a controllable design variable, MMHR complements biochemical osteoinductive strategies and offers a physical rationale for why osteogenic stimuli can become temporally fragile when the mechanotransduction baseline drifts under chronic stress3,4,11,12. We used glycation as a stringent proof‑of‑concept perturbation because AGEs accumulation is persistent, mechanically consequential, and difficult to eliminate; craniofacial sites add dynamic loading that further destabilizes the niche12–14. The cross‑species improvements in bone repair and mineralization support the idea that stabilizing pericellular mechanics can enhance the durability and reproducibility of regenerative trajectories under chronic microenvironmental drift. At the material level, ViscoClamp is best understood as a mesoscopic bio‑stabilizer rather than a conventional scaffold: it interpenetrates MSC‑deposited ECM and maintains a constrained viscoelastic output within the pericellular microdomain while remaining permissive to matrix remodeling and mineralization5,6,16–18. Dynamic networks that relax stress and dissipate energy can shape mechano-sensing and lineage specification in 3D8–10, but persistent pathological loading can drive long‑term drift in such systems. ViscoClamp addresses this challenge by coupling pathological glycation load to adaptive buffering while imposing a self‑limiting clamp that prevents runaway network evolution. We operationalized this function as an input–output stability problem and introduced CVoutput/CVinput as a dispersion‑transfer ratio that quantifies how variable biochemical disturbance is compressed into stable micromechanical outputs. A low dispersion‑transfer ratio provides a quantitative signature of bio‑stabilization, complementing conventional bulk metrics and enabling comparison across perturbation spectra and material designs. Beyond the materials concept, we contribute an interpretable agent-guided design workflow for a constrained peptide-hydrogel problem. Unlike de novo generative peptide-design frameworks that primarily optimize sequence-target interactions, our workflow was designed to couple material-level requirements, including AGEs adsorption, Ca2+-associated mineralization, gel feasibility and suppression of viscoelastic drift. The agent-guided component organized the protect-tune-clamp specification into deterministic enumeration, scoring and multi-objective ranking steps, but did not constitute a newly trained universal peptide generator. We therefore use the term “agent-guided” rather than “AI-designed” and view this workflow as complementary to sequence-based and generative peptide-design methods, including RFpeptides, PepMLM, PepPrCLIP, EvoBind2, PepTune and SaLT&PepPr19–22. This agentic approach is particularly relevant for MMHR materials, which must co‑encode recoverability, disease‑input buffering, and drift suppression while integrating with a living ECM that continually remodels5,6. More broadly, modularity enables adaptation: by substituting the sensing/tuning module, MMHR materials could be reconfigured to buffer other chronic perturbations, including inflammatory mediators, oxidative stress, or fibrosis‑associated crosslinking, without changing the output objective of pericellular mechanical homeostasis1–4,8. A limitation of the single-cell analysis is that tissues were pooled before library preparation, yielding one Ctrl library and one ViscoClamp library at a single early time point. Accordingly, trajectory inference, differentiation-potential analysis and CellChat predictions should be interpreted as descriptive and hypothesis-generating rather than as donor-level mechanistic proof. Future studies with independent single-cell libraries across multiple time points, lineage tracing and perturbation experiments will be required to define causal cell-state transitions and distinguish population-level enrichment from lineage-intrinsic normalization. Although the pooled scRNA-seq analysis and the early immune-vascular staining panels suggest an association between ViscoClamp treatment and a more reparative immune-stromal program, more extensive spatial and time-course validation of macrophage polarization and angiogenesis will be required to establish the immune-vascular remodeling process. The beagle extraction-socket experiment was used to evaluate alveolar socket bone regeneration under hyperglycemic loading; the present study did not directly assess regeneration of a fully organized periodontal ligament/cementum-bone interface, which should be addressed in future dedicated periodontal-regeneration studies. Although ViscoClamp showed micromolar Ca²⁺ affinity by ITC, the present study did not perform exhaustive head-to-head ITC benchmarking against optimized mineral-binding peptides under identical buffer, concentration and peptide-state conditions. The Ca²⁺-binding data therefore support mineral-association competence and compatibility with mineral deposition, but do not establish universal superiority in Ca²⁺ affinity. The present study focuses on cell-ECM and pericellular mechanics, but the same homeostatic-control principle may be extendable to cell-cell interface mechanics. Adherens junctions, cadherin-mediated force transmission and intercellular mechanical coupling can influence lineage specification and tissue organization. Future MMHR designs could incorporate modules that tune cell-cell junctional mechanics or spatially coordinate cell-ECM and cell-cell force transmission, thereby expanding agent-guided bio-stabilizer design beyond the pericellular ECM interface. Although demonstrated here in glycation‑challenged craniofacial regeneration, MMHR is intended as a generalizable therapeutic axis for conditions in which chronic microenvironmental drift destabilizes fate programs1–4,8. A mesoscale stabilizer such as ViscoClamp can be deployed as a mechanical “background” therapy that extends the effective duration of biochemical cues and reduces response variability, and can be combined with growth‑factor delivery, cell therapy, or implantable scaffolds12,15,18. By showing that pericellular viscoelastic homeostasis can be maintained without compromising mineralized matrix maturation, this work establishes mesoscopic mechanical homeostasis as an actionable target and provides a blueprint for designing bio‑stabilizers that operate within living, disease‑modified ECM. Methods ViscoClamp sequence selection, synthesis, and material characterization Agent-assisted, rule-based sequence-selection workflow Sequence selection followed an auditable rule-based pipeline designed to balance four properties: adsorption of AGEs-modified HSA, association with Ca²⁺, self-assembly into a gel, and resistance to adsorption-induced viscoelastic drift. The agent translated the protect-tune-clamp design brief into a fixed series of computational operations; it was not used to train or deploy a new deep-learning model. Candidate generation, physicochemical descriptor calculation, surrogate scoring, and multi-criterion ranking were performed with predefined deterministic equations. Step 1: Construction of a calcium-binding phosphomimetic segment The calcium-interacting module was derived from the seven-residue N-terminal statherin fragment DSSEEKF. Both serines were replaced by aspartate so that the additional negative charge normally supplied by phosphorylation could be represented with canonical residues, producing DDDEEKF. Expected ionization at pH 7.4 was calculated with the Henderson-Hasselbalch relation using fixed pKa values of 9.6 for the N terminus, 2.4 for the C terminus, 3.9 for Asp, 4.1 for Glu, 10.5 for Lys, and 12.5 for Arg. The calcium-association proxy was Ca_score = 1.0× |negative net charge|+ 0.5 × (acidic-site count) + 0.7× (length of the longest continuous acidic cluster). This step returned the selected mineralization segment and its charge- and cluster-based descriptors. Step 2: Complete enumeration of candidate AGEs-HSA-binding repeats A four-residue repeat was specified as [Pos][Hyd][Neg][Hyd], with Pos selected from {R, K}, Hyd from {A, V, I, L, F, Y, W, M}, and Neg from {D, E}. The Cartesian product generated 32 unique motifs. Mean Kyte-Doolittle hydropathy, average beta-sheet propensity, and the presence or absence of an aromatic residue were tabulated for every motif. Separate surrogate scores were then calculated for CML-HSA and MG-HSA adsorption using Score = w_pos × pos_strength + w_hyd × exp[-(H − μ)²/(2σ²)] + w_arom × I(aromatic) - neg_penalty × neg_strength. Parameters for CML-HSA were w_pos = 1.00, w_hyd = 0.30, μ = -1.0, σ = 0.90, w_arom = 0.02, and neg_penalty = 0.08; the corresponding MG-HSA values were 0.72, 0.40, −0.8, 1.10, 0.04, and 0.05. Each score was rescaled to the interval [0,1]. Cross-model performance was summarized as Robust = min(CML_norm, MG_norm) + 0.05 × mean(CML_norm, MG_norm). Motifs were ordered by this value, and the highest-ranked four-residue units were expanded to 16-residue (motif)₄ blocks. Step 3: Joint ranking of fused mineralizing and adsorption modules Each Step-2 repeat block was appended to DDDEEKF, yielding candidates of the general form DDDEEKF-(motif)₄. Predicted rheological behavior was represented by two descriptors: assembly_strength, based on proximity of hydropathy and beta-sheet propensity to the ion-complementary assembly regime, and sensitivity, representing the expected change in bundle organization after adsorption-driven charge screening. The unbound storage-modulus proxy was G′₀ = 3.2 + 2.7 × assembly_strength + 1.2 × Ca_norm, where Ca_norm is the normalized Step-1 Ca_score. Binding fractions were sampled at f = {0, 0.2, 0.4, 0.6} for CML-HSA and f = {0, 0.15, 0.30, 0.45} for MG-HSA. The corresponding responses were G′(CML,f) = G′₀ × (1 + 0.90 × sensitivity × f) and G′(MG,f) = G′₀ × (1 + 0.65 × sensitivity × f). Viscoelastic drift was expressed as the coefficient of variation across all simulated states. A gel-feasibility prior was independently derived from hydropathy and beta-propensity proximity to the self-assembly window. Final ranking used Score_final = 0.45 × Robust + 0.25 × GelPrior_norm + 0.30 × (1 - CV_norm). These coefficients were fixed before experimental testing and were not retuned in response to experimental outcomes. The outputs were an ordered list of fused sequences and their predicted G′ trajectories for both AGEs-HSA models. Step 4: Single-site cysteine optimization To identify a position capable of mechanically reinforcing the leading Step-3 sequence, every alanine in its 16-residue repeat block was evaluated as a single A-to-C substitution. Terminal exposure was represented by p = exp(-d/3), where d is the residue distance from the substituted site to the nearer end of the block. For each variant, CV_adj = CV_base/(1 + 1.2 × p) and G′₀_adj = G′₀_base × (1 + 0.22 × p). Variants were ranked by CV_adj, and the top cysteine-containing sequence was advanced for experimental assessment. Fmoc solid-phase peptide synthesis Peptide assembly was performed in a solid-phase reactor on 10 g dichlorotrityl chloride resin with a nominal loading of 1.0 mmol/g. Before chain elongation, the resin was conditioned for 10 min in 10 mL dichloromethane (DCM) and then for 10 min in 10 mL N, N-dimethylformamide (DMF); solvent was removed by filtration after each step. Every coupling cycle contained 33 mmol of the required Fmoc-protected amino acid and 100 mmol N, N-diisopropylethylamine in DMF and proceeded for 120 min at 30 °C with agitation. The resin was washed five times with DCM and five times with DMF. Fmoc groups were removed in two consecutive 10-min treatments with 20% (v/v) piperidine/DMF at 30 °C, followed by the same DCM and DMF wash sequence. The development of a blue ninhydrin reaction was used to verify deprotection. Peptide purification and LC-MS confirmation Crude monomer was separated by preparative reversed-phase HPLC on a Gemini NX-C18 column (Phenomenex; 5 μm, 110 Å, 150 × 30 mm). The aqueous phase was water containing 0.1% (v/v) trifluoroacetic acid (TFA), and the organic phase was acetonitrile containing 0.1% (v/v) TFA. A 5-60% organic gradient was run for 50 min at 30.0 mL/min. Fractions containing ViscoClamp were pooled and lyophilized. Product identity and purity were then checked on a Waters binary LC-MS platform at 50 μL/min with water/acetonitrile mobile phases containing 0.1% (v/v) formic acid. Electrospray spectra were collected in positive full-scan mode and used to verify the expected molecular mass. Micro-FTIR spectroscopy Lyophilized ViscoClamp monomer was analyzed with a VERTEX 70 micro-Fourier transform infrared spectrometer (Bruker, Germany). Interpretation focused on the peptide amide regions and other sequence-associated absorption features. Hydrogel assembly and macroscopic handling The monomer was dissolved directly in phosphate-buffered saline to 1% or 6% (w/v). Self-supporting gel formation was judged by whether the material remained in place after tube inversion. Macroscopic shapeability was illustrated by extruding the 6% formulation to form letters. Wettability measurement A DSA100 goniometer (KRÜSS, Germany) was used to record equilibrium contact angles. For each test, a 50-μL aliquot of either the 1% or 6% (w/v) formulation was placed on the measurement substrate, and the angle was read after the droplet had stabilized. Transmission electron microscopy For dose-dependent ultrastructural imaging, 6% (w/v) ViscoClamp was mixed with AGEs (BIOSS, bs-1158P) to final concentrations of 0.05, 0.1, 0.2, 0.4, 0.8, or 1.6 mg/mL. Each mixture was applied to a carbon-coated copper grid and examined on a Talos L 120 C G2 transmission electron microscope (Thermo Fisher Scientific, USA). Scanning electron microscopy The 1% and 6% (w/v) gels were lyophilized, secured to specimen stubs, and coated with an approximately 8-nm layer of platinum. Surface architecture was recorded with a GeminiSEM 500 instrument (Zeiss, Germany). Isothermal titration calorimetry Association of ViscoClamp with CaCl₂ was quantified at 25 °C on an ITC200 calorimeter (Switzerland). Ultrapure water was used to prepare 20 μM ViscoClamp and 200 μM CaCl₂. Instrument cleanliness and baseline behavior were first checked by titrating water into water. The peptide solution was loaded into the sample cell without introducing bubbles, and the salt solution was delivered from the syringe in 20 injections of 2.0 μL, each lasting 4 s and separated by 150 s. Reference power was 5 μcal/s, stirring was set to 60×g, and the reference cell contained water. The resulting thermograms were fitted in the Origin package supplied with the calorimeter. Fluorescence-polarization binding assay ViscoClamp was labeled at its N terminus with FITC and subsequently dialyzed, purified, and freeze-dried. AGEs binding was assessed in black 96-well plates on a SpectraMax iD5 reader (Molecular Devices, USA). A dilution series of AGEs was combined with FITC-ViscoClamp at 6% (w/v) in a total volume of 100 μL per well and equilibrated for 2 h at room temperature. Polarization was then measured with 470-nm excitation and 530-nm emission settings. Apparent dissociation constants were estimated by nonlinear fitting to the following one-site mass-balance equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{F}}}}={F}_{0}+\left(\frac{{F}_{c}-{F}_{0}}{2\left[{Pep}\right]}\right)\left(\left[{Pep}\right]+\left[{AGEs}\right]+{K}_{d}-\sqrt{{\left(\left[{Pep}\right]+\left[{AGEs}\right]+{K}_{d}\right)}^{2}-4\left[{Pep}\right]\left[{AGEs}\right]}\right)$$\end{document}F=F0+Fc−F02PepPep+AGEs+Kd−Pep+AGEs+Kd2−4PepAGEs In Eq. (1), F is the observed polarization, F₀ is the signal from FITC-ViscoClamp without AGEs, Fc is the plateau value, [Pep] is the final peptide concentration, and [AGEs] is the total concentration of the glycation product preparation. Scaffold-matched comparator peptides Four related sequences were included as internal design controls rather than as claims of best-in-class literature benchmarks. Every comparator retained the DDDEEKF mineralization segment, a similar overall length, and terminal cysteine stabilization. Pep 1, DDDEEKF-(KADA)₃-KADC, replaced Arg with Lys to isolate the effect of cation identity. Pep 2, DDDEEKF-(RAEA)₃-RAEC, replaced Asp with Glu to alter acidic side-chain length. Pep 3, DDDEEKF-(RVDV)₃-RVDC, and Pep 4, DDDEEKF-(RIDI)₃-RIDC, increased repeat-block hydrophobicity while preserving alternating charge complementarity. These variants, therefore, tested local sequence perturbations around the selected ViscoClamp scaffold. Pep 0–Pep 4 were produced and formulated under matched conditions whenever the chemistry allowed. Their apparent affinity for AGEs was determined by the fluorescence-polarization protocol and fitting model described above. Gel formation and bulk mechanics were compared by frequency-sweep rheology. Selected variants were also examined during AGEs challenge for Runx2 nuclear localization, nuclear-to-cytoplasmic Runx2 distribution, and Alp staining or enzymatic activity.Peptide sequencePep 0DDDEEKF-(RADA)3-RADCPep 1DDDEEKF-(KADA)3-KADCPep 2DDDEEKF-(RAEA)3-RAECPep 3DDDEEKF-(RVDV)3-RVDCPep 4DDDEEKF-(RIDI)3-RIDC Peptide sequencePep 0DDDEEKF-(RADA)3-RADCPep 1DDDEEKF-(KADA)3-KADCPep 2DDDEEKF-(RAEA)3-RAECPep 3DDDEEKF-(RVDV)3-RVDCPep 4DDDEEKF-(RIDI)3-RIDC Confocal imaging of peptide-AGEs organization FITC-labeled ViscoClamp and Cy3-labeled AGEs were used for dual-channel visualization. The 6% (w/v) gel, either untreated or supplemented with 0.1 mg/mL AGEs, was mounted on a glass slide and imaged with a Leica TCS SP8 STED 3X system (Germany) to determine whether AGEs altered mesoscale gel organization. Oscillatory and steady-shear rheology Mechanical spectra were collected for 6% (w/v) ViscoClamp (Pep 0) with no added AGEs or with 0.05, 0.1, 0.2, 0.4, 0.8, or 1.6 mg/mL AGEs (BIOSS, bs-1158P). At 25 °C, angular-frequency sweeps covered 0.1–40 rad/s at 1% strain, and steady-shear measurements covered 0.1–100 s⁻¹. Storage modulus (G′), loss modulus (G″), loss tangent (G″/G′), and viscosity were recorded. Repeated frequency scans were additionally used to follow changes in G′ and G″ as the gels matured. For Pep 1-Pep 4, viscosity was likewise measured from 0.1 to 100 s⁻¹ at 25 °C. Their evolving viscoelastic spectra were compared by repeated 0.1–40 rad/s frequency sweeps performed at 1% strain. Coarse-grained molecular dynamics Simulations were run in GROMACS 2021.4 with the MARTINI 2.2 coarse-grained force field. ViscoClamp molecules were initially distributed at random within a 15 × 15 × 15 nm³ periodic cube containing coarse-grained water and 30 Ca²⁺ ions; Na⁺ and Cl⁻ were introduced as needed for charge neutrality. Both electrostatic and van der Waals interactions used a 1.05-nm cutoff. Peptide and solvent/ion groups were coupled to separate temperature baths, and pressure was maintained with the Parrinello-Rahman scheme. A Verlet neighbor list with a 1.05-nm search radius was rebuilt every 10 steps. Each trajectory extended to 2000 ns. Standard GROMACS utilities were used for analysis, with a component contact assigned when the minimum separation was below 0.8 nm. Molecular configurations were inspected in PyMOL and VMD. Cryogenic electron microscopy The 1% (w/v) peptide preparation was assigned to three conditions: no additive, 0.1 mg/mL CaCl₂, or 0.1 mg/mL CaCl₂ together with 0.1 mg/mL AGEs. Samples were held at 4 °C, and aliquots were taken at 0, 6, 12, and 24 h. Thin aqueous layers were formed on copper grids and vitrified by rapid immersion in liquid ethane. Images were acquired on a Talos F200C cryo-electron microscope (FEI, USA). Von Kossa mineral staining in gels A 1% (w/v) ViscoClamp preparation was incubated without additives, with 0.1 mg/mL CaCl₂, or with 0.1 mg/mL each of CaCl₂ and AGEs. Material in 24-well plates was sampled at 0, 15, and 30 min and at 1, 3, and 6 h. Using the Solarbio Von Kossa kit (G3282, China), samples were exposed to silver nitrate under intense illumination for 15–60 min, washed in distilled water for 1 min, and treated with sodium thiosulfate for 2 min to stop development. ImageJ was used for quantitative image analysis. AFM-based mapping of ViscoClamp ultrastructure and multiscale mechanics Freeze-dried 6% (w/v) gels were examined with a CSI Nano-Observer atomic force microscope (France). Measurements used an ACTA conical probe (APPNANO, USA; nominal resonance frequency, 300 kHz; spring constant, 37 N/m; half-opening angle, 18°). Before data acquisition, the cantilever spring constant was obtained by thermal-noise calibration, and optical-lever sensitivity was established on glass. Tapping-mode topographs were collected at 1 line/s with 512 scan lines. For the AGEs dose series shown in Fig. 2C, D, gels were exposed to the specified concentrations before lyophilization. For the temporal series in Fig. 2F, G, samples were collected 6, 12, 24, 48, or 72 h after AGEs addition and processed in the same manner. Quasi-static force curves were acquired by approaching the sample at 200 nm/s until the load reached 300 nN. Indentation depth, δ, was defined as:2δ=z-dδ=z−d Here, z is the cantilever z-displacement and d is its measured deflection. To limit influence from the underlying substrate, indentation was restricted to less than 2 μm, corresponding to under 10% of specimen thickness. Force was obtained from Hooke’s relation:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F=k\cdot \delta$$\end{document}F=k⋅δ The apparent elastic response was fitted with the conical Sneddon expression:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F=\frac{2}{\pi }{E}^{*}{\delta }^{2}\tan \varphi$$\end{document}F=2πE*δ2tanφ In Eq. (4), F is force, δ is indentation, and φ is the probe half-angle. The sample modulus was recovered from the reduced-modulus relation:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,{E}^{*}=\frac{E}{1-{v}^{2}}$$\end{document}E*=E1−v2 E denotes the elastic modulus. A Poisson ratio of 0.5 was assigned because the gel was treated as incompressible. For creep testing, the approach velocity was increased to approximately 20 μm/s. When the load reached 300 nN, it was maintained for 10 s while the change in indentation was recorded. Creep compliance J(t) = ε(t)/σ(t) was calculated with the Lee-Radok formulation:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F\left(t\right)={F}_{0}H\left(t\right)$$\end{document}Ft=F0Ht7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J\left(t\right)=\frac{4\tan \varphi }{\pi \left(1-v\right){F}_{0}}{\delta (t)}^{2}$$\end{document}Jt=4tanφπ1−vF0δ(t)2 In these expressions, H(t) is the Heaviside step function, F(t) describes the imposed loading history, and F₀ is the target force. Together, Eqs. (6) and (7) provide the viscoelastic extension of conical Sneddon indentation used to recover time-dependent compliance from the creep trace. Short-time creep data (0.01–0.1 s) were interpreted with a two-level self-similar hierarchical model. The two component relations were:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${J}_{1}=\frac{-1+\sqrt{1+4t/\tau }}{2{E}_{1}}$$\end{document}J1=−1+1+4t/τ2E19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${J}_{2}=\frac{-1+\sqrt{1+4{E}_{2}\;{J}_{1}}}{2{E}_{2}}$$\end{document}J2=−1+1+4E2J12E2 Combining the two levels gives:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J(t)=\frac{-1+\sqrt{1+4{E}_{2}\left[\left(-1+\sqrt{1+4t/\tau }\right)/2{E}_{1}\right]}}{2{E}_{2}}$$\end{document}J(t)=−1+1+4E2−1+1+4t/τ/2E12E2 Within this representation, E₁ is assigned to the local gel-unit stiffness, E₂ to the fibrous-network stiffness, and τ to the relaxation time. Fitting J(t) between 0.01 and 0.1 s yielded E₁, E₂, and τ. Effective local viscosity was then calculated as η = E₁ × τ. The later portion of the creep response (0.1–10 s) was described by a power law:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J\left(t\right)\propto {t}^{\alpha }$$\end{document}Jt∝tα The exponent αlong served as the index of time-dependent deformation. Values approaching 0 indicate an elastic-solid response, whereas values approaching 1 indicate a viscous-fluid response. αlong was obtained as the slope of a linear fit after log transformation of the J(t) data. Dispersion-transfer ratio. Let A = {aᵢ} denote the predefined AGEs concentration states. Input dispersion was CVinput = SD(A)/mean(A). For each AFM output m in {Eexp, αlong, τ, E₁, η, E₂}, CVm = SD(Ym)/mean (Ym), where Ym is the set of measurements across the same input states. CVoutput was the arithmetic mean of the six CVm values, and the dispersion-transfer ratio was DTR = CVoutput/CVinput. DTR was used as an engineering summary of input-to-output attenuation and does not assume independence among the AFM descriptors. Population standard deviation with denominator N was used for DTR and subset-DTR because the AGEs levels were fixed experimental states; replicate-level biological summaries elsewhere are reported as mean ± SD. For cell-surface AFM, MSCs were exposed for 24 h to 1% ViscoClamp together with 0.025, 0.05, 0.1, 0.2, or 0.4 mg/mL AGEs. Eexp, η, τ, αlong, E₁, and E₂ were extracted from the resulting force curves with the same conical-contact and creep-fitting framework used for isolated gels. Redundancy and subset analysis. Correlation and principal-component analyses were used to assess covariation among Eexp, αlong, τ, E₁, η, and E₂. DTR was then recalculated for every two-, three-, four-, and five-parameter combination. Within a given subset, CVoutput was the mean of the CVm values for the included parameters. Examples reported in Supplementary Table 1 include Eexp + η, Eexp + αlong + η, and the minimally redundant set Eexp + η + E₂. Calculations were performed for both the single-time-point and the specified time-series datasets. Cell culture and functional assays Isolation and expansion of jawbone-derived MSCs Mandibles were removed aseptically from euthanized male rats and briefly surface-decontaminated. Until processing, the specimens were held on ice in PBS containing 3% fetal bovine serum (FBS) and antibiotics. Adherent soft tissues were stripped away, teeth were removed when required, and the marrow spaces were opened by trimming the alveolar and ramus ends. The cavities were flushed with complete α-MEM containing 10% FBS, 2 mM L-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin. The recovered material was gently dispersed, passed through a cell strainer, and centrifuged at 800 × g for 3-5 min. To recover stromal cells retained in compact mandibular bone, the remaining bone was diced into 1-3 mm³ fragments and agitated for 60 min at 37 °C in 0.1% collagenase II. After washing, these fragments were plated together with the marrow-derived cell fraction. Cultures were maintained in 100-mm dishes at 37 °C and 5% CO₂. The first medium change, performed after 48–72 h, removed nonadherent cells; subsequent changes were made every 2–3 days. At 80–90% confluence, cells were split approximately 1:2. Passages 2 and 3 were used for all in vitro experiments. Cell-ViscoClamp co-culture. Immediately before treatment, ViscoClamp was diluted to 1% (w/v) in complete medium or in osteogenic induction medium. The peptide and AGEs were added directly to monolayer MSC cultures and remained present for the specified exposure interval. During prolonged osteogenic experiments, freshly prepared treatment medium was supplied every 2–3 days. Three-dimensional spheroids were generated by seeding MSCs in honeycomb spheroid plates and allowing them to assemble overnight. The following day, 50% of the culture medium was exchanged for osteogenic induction medium containing 0.1 mg/mL AGEs plus either PBS or 1% (w/v) ViscoClamp. The treatment medium was renewed every 2–3 days. On day 7, spheroids were fixed and permeabilized, labeled for Alp (HUABIO, ET1601, 1:200) or Sp7 (Proteintech, 28694-1-AP, 1:200), and counterstained with DAPI (1:1000). ImageJ was used to quantify fluorescence. Six independent spheroids were analyzed per condition, and the two groups were evaluated with an unpaired Student’s t test. SEM of cell-associated matrix MSCs were grown to confluence on coverslips in 24-well plates and then challenged for 24 h with 0.1 mg/mL AGEs in the presence or absence of 1% (w/v) ViscoClamp. After fixation, samples were dehydrated and dried, coated with approximately 8 nm platinum, and visualized on a GeminiSEM 500 microscope (Zeiss, Germany). AFM measurement of MSC surface mechanics Cells were treated for 24 h with 0.025-0.4 mg/mL AGEs, either alone or together with 1% ViscoClamp. Force-indentation data were collected on an MFP-3D Infinity system (Asylum Research/Oxford Instruments, USA) using a HYDRA6V-200NG conical cantilever. Thermal-noise calibration gave a spring constant of 107.63 pN/nm; the nominal tip radius and half-angle were 20 nm and 18°, respectively. Cellular force curves were fitted with the conical Sneddon contact relation described above. Alp enzymatic activity Alp activity in culture supernatants was determined with the NJJCBIO assay (A059-2-2, China) using the supplier-specified procedure. Activity (U·L⁻¹) was calculated as [(AT − A0)/(AS − A0)] × 100 × Cs × 7.14. AT, A0, and AS are the sample, blank, and standard absorbances at 520 nm, respectively, and Cs is the concentration of the standard. Immunofluorescence microscopy MSCs were established in confocal dishes at 1 × 10⁴ cells per dish and allocated to control, AGEs, or AGEs plus 1% (w/v) ViscoClamp conditions. A separate Runx2 dose series used 1% ViscoClamp with 0, 0.05, 0.1, 0.2, 0.4, or 0.8 mg/mL AGEs for 24 h. Once cultures reached approximately 80% confluence, osteogenesis was initiated with medium containing 10 nM dexamethasone, 100 μg/mL ascorbic acid, and 1 mM beta-glycerophosphate. Samples collected on days 7, 14, and 21 underwent fixation for 10 min in 4% paraformaldehyde at ambient temperature, followed by 30 min of permeabilization with 0.3% Triton X-100 and serum blocking. Primary antibodies to collagen I (Proteintech, 66761, 1:100), AGEs (MCE, HY-P81087, 1:100), and Runx2 (CST, 12556S, 1:100) were applied at 4 °C overnight. Fluorophore-coupled secondary antibodies (YeasenBio, China, 1:200) were added for 1 h at room temperature. Alizarin Red S (Beyotime, C0148S) was used, where mineral deposition was also visualized. After a 10-min DAPI (1:1000) counterstain, fields were acquired on an Olympus FV3000 confocal microscope (Japan), and signal intensity was measured in ImageJ. For co-localization imaging, MSCs were plated at 1 × 10⁴ cells per dish and exposed for 6 h to FITC-ViscoClamp and Cy3-AGEs. After fixation, permeabilization, and serum blocking, collagen I was labeled overnight at 4 °C with antibody 66761 (Proteintech, 1:100). YSFluor 647 secondary antibody (YeasenBio, 34413ES60, China, 1:200) was applied for 1 h, followed by a 10-min DAPI (1:1000) stain. Multichannel images were recorded with a Leica TCS SP8 STED 3× microscope (Germany). Immunoblotting MSCs were plated in 6-well format at 1 × 10⁵ cells per well and treated according to the relevant experiment. Protein extraction was carried out for 30 min on ice in RIPA buffer supplemented with PMSF; insoluble material was then cleared by centrifugation for 15 min at 4 °C (13,800 × g). Where subcellular localization was required, cytoplasmic and nuclear fractions were generated with the Minute Plasma Membrane Protein Isolation and Cell Fractionation Kit (Invent, SM-005-4, USA). Samples were mixed with SDS loading buffer containing 50 mM DTT and heated to 100 °C for 10 min. Total protein was quantified by BCA assay (Beyotime, P0010S). Equivalent protein amounts were resolved on 10% SDS-PAGE and transferred to 0.45-μm PVDF membranes (Millipore, IPFL00010). Membranes were blocked for 1 h in 5% bovine serum albumin and incubated at 4 °C overnight with antibodies to Runx2 (CST, 12556S, 1:1000), Alp (HUABIO, ET1601, 1:1000), Sp7 (Proteintech, 28694-1-AP, 1:1000), Lamin A/C (Beyotime, AG2517, 1:1000), or beta-actin (PTM BIO, PTM-5018, 1:1000). After TBST washing, HRP-linked secondary antibodies (YeasenBio, China, 1:5000) were added for 1 h at room temperature, and chemiluminescent signal was developed with ECL reagent (Millipore). Runx2 dual-luciferase assay For the reporter assay, each well of a 24-well plate received 1 × 10⁵ MSCs before co-transfection with pRunx2-TA-Luc (D4313) and the pRL-TK-Renilla normalization plasmid (D2760; Beyotime, China). Transfected cultures were exposed for 24 h to 1% (w/v) ViscoClamp and to AGEs concentrations of 0, 0.05, 0.1, 0.2, 0.4, or 0.8 mg/mL. Firefly and Renilla signals were read on a SpectraMax iD5 instrument using the Beyotime dual-luciferase reagent set (RG088M). Runx2 reporter activity was expressed as the firefly-to-Renilla luminescence ratio. Bulk RNA sequencing of MSCs MSCs in 100-mm dishes were treated for 24 h with 0.1 mg/mL AGEs alone or with 0.1 mg/mL AGEs plus 1% (w/v) ViscoClamp. TRIzol was used to recover total RNA. Libraries were constructed with the NEBNext Ultra RNA Library Prep Kit, catalog E7530L (NEB, USA), for Illumina sequencing and run as 2 × 150-bp paired-end reads on a NovaSeq 6000. Qlucore Omics Explorer 3.2 was used for heat maps and gene-set enrichment, and Ingenuity Pathway Analysis was used for pathway-level interpretation. Alizarin Red S mineralization assay Each well of a 24-well plate was inoculated with 1 × 10⁵ MSCs. At approximately 80% confluence, growth medium was replaced by osteogenic medium containing 10 nM dexamethasone, 100 μg/mL ascorbic acid, and 1 mM beta-glycerophosphate. Mineralized matrix was stained on days 7, 14, and 21 with Alizarin Red S (Beyotime, C0148S). Bound dye was released with 2% cetylpyridinium chloride, and optical density was read at 570 nm on a SpectraMax 190 instrument (Molecular Devices, USA). ALP staining Cell-associated Alp was visualized after 7 days of osteogenic induction with the Beyotime staining kit C3250S, following the procedure supplied with the kit. Fak inhibition During challenge with 0.1 mg/mL AGEs, MSCs received 1% (w/v) ViscoClamp with or without 10 μM PF-573228. Runx2 immunofluorescence and nuclear/cytoplasmic fractionation were evaluated after 24 h of osteogenic induction. Immunoblotting for Runx2 (CST, 12556S, 1:1000), Alp (HUABIO, ET1601, 1:1000), and Sp7 (Proteintech, 28694-1-AP, 1:1000) was performed after 7 days. CaCl₂-dependent viability MSCs were cultured with 1% ViscoClamp in media containing the indicated CaCl₂ concentrations. At the specified intervals, viability was quantified by CCK-8 using the same 450-nm absorbance procedure described below. Imaging-based cytocompatibility MSCs were seeded at 1 × 10⁴ cells per well and exposed to 1% (w/v) ViscoClamp for 24 h. Fixed cells were permeabilized in 0.3% Triton X-100 for 30 min and blocked with goat serum. Proliferation and apoptosis were assessed by Ki-67 immunolabeling (CST, 9129S, 1:100) and the Beyotime TUNEL kit (C1089, China), respectively. FITC-conjugated secondary antibodies (YeasenBio, China, 1:200) were used where required, and nuclei received a 1-min DAPI (1:1000) counterstain. Images were captured with an Olympus FV3000 confocal microscope. CCK-8 viability and proliferation assay Cells were dispensed into 96-well plates at 1 × 10³ per well. Two exposure schemes were examined: continuous treatment with 1% (w/v) ViscoClamp for 7 days, and 1% (w/v) ViscoClamp followed by 0, 2, 4, or 6 mM CaCl₂ for 3 days. CCK-8 reagent (Biosharp, BS350C, China) was applied daily, and absorbance at 450 nm was measured with a SpectraMax 190 plate reader (Molecular Devices, USA). Animal studies Animal sources, husbandry, and ethical oversight Male Sprague-Dawley rats (8 weeks, 220–240 g), New Zealand white rabbits (3 months, 1.5–2.0 kg), and beagle dogs (12 months, 8–10 kg) were supplied by the Laboratory Animal Center of Xi’an Jiaotong University (SYXK 2020-001). Male rhesus macaques (5 years, 10.0–10.5 kg) were obtained from Shaanxi Longteng Macaque Breeding Co., Ltd. (SYXK2023-004). Animals were maintained under specific-pathogen-free conditions on standard chow with a conventional light-dark schedule. The Xi’an Jiaotong University Animal Care and Use Committee supervised all procedures. Rat, rabbit, and dog protocols were authorized as No. 20211732; the macaque protocol was authorized as No. XJTUAE202608. Model induction and surgical procedures AGEs-injection model AGEs (BIOSS, bs-1158P, China) were prepared in sterile PBS and delivered intraperitoneally three times each week. The dose was 20 mg/kg for rats and 10 mg/kg for New Zealand white rabbits. Twenty-four hours after the final injection in week 4, peripheral blood was obtained and serum AGEs were measured with the Abcam kit ab238539. Defective surgery was undertaken on the next day. The same AGEs dosing schedule was resumed after surgery and continued until tissue collection. STZ-induced hyperglycemia Streptozotocin (Solarbio, S8050, China) was dissolved immediately before use in 0.05 M sodium citrate, pH 4.5. Animals had no food for 8 h before dosing but retained access to water. Under anesthesia, beagle dogs received 40 mg/kg STZ intravenously and macaques received 45 mg/kg intravenously, each as a single injection. Fasting glucose was followed longitudinally, and subsequent interventions began only after hyperglycemia had stabilized. Rat mandibular defect Four weeks after starting AGEs administration, rats were anesthetized with inhaled isoflurane. A 4-mm circular outline was created in the periapical region of the first and second molars. Bone was removed to approximately 0.8 mm with a 4 × 4 mm trephine, and a ball burr was then used to complete and regularize the defect under continuous saline cooling. The left defect received normal saline and the right defect received 6% (w/v) ViscoClamp. Animals were killed with pentobarbital sodium 2 months later. Blood, serum, mandibles, and major organs were collected at necropsy. Rabbit mandibular defect Rabbits underwent surgery after 4 weeks of AGEs exposure. Under isoflurane, the lateral mandibular field was shaved and opened inferior to the molars. With continuous Hank’s balanced salt solution irrigation, an 8-mm hollow trephine was used to establish the defect, and a 2-mm ball burr was used to level its floor. The target geometry was 8 mm wide and 2 mm deep. Saline was placed on the left and 6% (w/v) ViscoClamp on the right. At the 1-month endpoint, animals were euthanized and blood, serum, mandibles, and organs were recovered. Rabbit calvarial defect After 4 weeks of AGEs administration, rabbits were anesthetized with isoflurane and the scalp was prepared. A midline incision of approximately 50 mm exposed the calvarial periosteum. An 8-mm trephine was used in each cranial quadrant under saline cooling, followed by a 2-mm ball burr to produce an 8-mm-wide, 2-mm-deep defect with a level base. The left side was filled with saline and the right with 6% (w/v) ViscoClamp. Blood, calvariae, and visceral organs were harvested after 3 months. Beagle extraction-socket model Once STZ-induced hyperglycemia was stable, beagles were anesthetized with pentobarbital sodium and both mandibular lateral incisors were removed. After residual blood was cleared, saline was placed in the left socket and 6% (w/v) ViscoClamp in the right. The 2-month collection included blood for biochemical testing and serum preparation, the mandibular specimens, and visceral organs. Rhesus macaque mandibular defect Hyperglycemic macaques were deprived of food for 24 h and water for 12 h before surgery. Propofol was used for intravenous induction and isoflurane for maintenance. A 10 × 5 × 6 mm defect was prepared in the posterior molar region. In the paired design, the left side received saline, whereas the right side was gradually filled with 6% (w/v) ViscoClamp until the cavity was occupied. Nonabsorbable sutures were used for gingival closure while avoiding displacement of the gel. Ceftiofur sodium was given intramuscularly at 3–5 mg/kg once daily for 7 days, and the operative site was irrigated daily for the same period with metronidazole in saline. Blood glucose monitoring Glucose was read with a OneTouch UltraEasy meter (Johnson & Johnson, USA), which uses the glucose oxidase principle. Measurements followed the instrument procedure and are expressed as mmol/L. Quantification of AGEs by ELISA Serum and conditioned-medium AGEs were assayed with the Abcam kit ab238539. Standards and specimens (50 μL) were first placed in AGEs-conjugate-coated wells for 10 min, followed by 50 μL diluted anti-AGEs antibody for 1 h. After a 250-μL wash with 1× buffer, 100 μL diluted HRP-secondary antibody was added for 1 h. Wells were washed again, developed with 100 μL TMB for 2–20 min, and stopped with 100 μL stop reagent. Absorbance was recorded immediately at 450 nm. Concentrations were interpolated from the blank-corrected standard curve. CT acquisition and bone morphometry Rat, rabbit, and dog specimens were scanned ex vivo on an Xradia 610 Versa system (Carl Zeiss, Germany) at 100 kV, 110 μA, and 500 ms per projection. Macaques underwent in vivo imaging at 1 week and 1 month with an AX3000-D scanner (Always Imaging, China) using 160 or 180 kV, 0.1–0.5 mA, and a 1-s integration. After three-dimensional reconstruction, the analysis volume was restricted to the defect or extraction socket. Manufacturer software was used to obtain bone mineral density, BV/TV, BS/BV, trabecular number, trabecular separation, and trabecular thickness. Histological processing Decalcified bone sections were processed with hematoxylin and eosin (Baso, BA4097/BA4098, China), Masson’s trichrome (Solarbio, G1340, China), or Van Gieson stain (Baso, BA4084A, China), using the respective kit procedures. Whole-slide images were generated on a Pannoramic DESK scanner (3D HISTECH, Hungary). Immunofluorescence of bone tissue Following heat-mediated antigen retrieval, sections were serum-blocked for 1 h at room temperature. Primary antibodies against CD86 (Abcam, ab238468, 1:200), CD206 (Abcam, ab64693, 1:200), CD31 (Abcam, ab182981, 1:200), CD73 (Invitrogen, PA5-85958, 1:200), Alp (HUABIO, ET1601, 1:200), osteocalcin (Proteintech, 23418, 1:200), AGEs (Sigma, MABN1837, 1:200), and collagen I (Proteintech, 66761, 1:200) were incubated with the tissue at 4 °C overnight. Fluorescent secondary antibodies (YeasenBio, China, 1:200) were then applied for 1 h, and nuclei were labeled with DAPI (1:1000) for 10 min. An Axio Scan Z platform (Zeiss, Germany) was used for image acquisition. Sequential fluorochrome administration Rats received calcein green at 20 mg/kg in week 2, alizarin red at 30 mg/kg in week 4, and tetracycline at 25 mg/kg in week 8. Three days after the tetracycline dose, animals were euthanized (n = 3). Mandibles underwent 48 h of fixation in 4% formaldehyde before resin embedding, sectioning at approximately 150 μm, and grinding to approximately 40 μm. Fluorochrome bands were imaged on a Leica SP8 DIVE confocal microscope (Germany) at excitation/emission settings of 488/517 nm for calcein, 543/617 nm for alizarin, and 405/580 nm for tetracycline. Von Kossa staining of undecalcified tissue Undecalcified mandibular sections were processed with the Solarbio Von Kossa kit G3282. Silver nitrate development was carried out for 15–60 min under strong illumination, followed by a 1-min distilled-water rinse and a 2-min sodium-thiosulfate stop. Stained slides were digitized with the Pannoramic DESK system (3D HISTECH, Hungary). Single-cell RNA sequencing Mandibles were collected 7 days after local saline or 6% (w/v) ViscoClamp treatment, rinsed in ice-cold PBS, and processed within 24 h. Tissue dissociation combined collagenase/DNase digestion, gentleMACS mechanical processing, and passage through a 70-μm filter. The resulting cells were partitioned on a 10× Chromium Controller (10× Genomics), and libraries were constructed with the corresponding droplet workflow. cDNA and final library quality were checked with an Agilent Bioanalyzer high-sensitivity DNA assay. Sequencing was performed on a NovaSeq 6000 with paired 150-bp reads. For each treatment, mandibles from three rats were pooled before library construction, producing one control library and one ViscoClamp library. Consequently, inference was restricted to descriptive comparisons at the pooled-library level. Following quality control, clustering, and cell annotation, cell-state fractions were computed within each library relative either to all annotated cells or, for lineage-specific questions, to all osteogenic-lineage cells. Changes were summarized as absolute proportion differences and log₂ ratios after addition of a small pseudocount. A symmetric-prior Dirichlet posterior model was used to address the compositional closed-sum constraint and to derive posterior medians, 95% credible intervals, and probabilities of directional change. Robustness was also examined by a reference log-ratio analysis. Trajectory reconstruction and cell-cell communication analyses were treated as exploratory visualizations of treatment-associated states and candidate signaling links. Four-dimensional label-free proteomics Mandibular defect tissue was obtained from AGEs-injected rats 7 days after application of 6% (w/v) ViscoClamp. Samples were pulverized and sonicated for protein extraction, and concentration and integrity were checked by Bradford assay. Proteins underwent magnetic-bead-assisted tryptic digestion, followed by peptide desalting. Nano-flow HPLC was used for separation before analysis on a timsTOF HT mass spectrometer. Data-dependent acquisition was combined with parallel accumulation-serial fragmentation to obtain the four-dimensional datasets. FragPipe searches were conducted against the Rattus norvegicus UniProt reference proteome UP000002494. Peptide- and protein-level false-discovery rates were each maintained below 1%. Quantitative intensities were median-normalized, and missing entries were imputed before testing. Proteins with an absolute fold change of at least 1.2 and p < 0.05 were classified as differentially expressed. Gene Ontology and KEGG analyses were used for functional enrichment, and protein-interaction networks were generated for interpretation. Safety assessments For acute oral testing, healthy male Sprague-Dawley rats weighing 220–240 g received ViscoClamp by gavage once daily for 7 days at ten times the experimental dose. At necropsy, major organs were fixed in 4% paraformaldehyde, whole blood was used for complete blood counts and serum chemistry, and tissues were subjected to routine H&E histology. Healthy rabbits and beagle dogs were evaluated after ViscoClamp exposure by complete blood count (RBC, WBC, LYMPH, NEUT, PLT, and HGB), serum ALT, AST, BUN, and creatinine, and H&E examination of major organs fixed in 4% paraformaldehyde. In macaques, complete blood counts and serum chemistry were compared before implantation and 1 month afterward. Electrocardiography and ultrasonography of the heart, liver, spleen, and kidneys provided additional systemic safety readouts. Statistical procedures Analyses were performed in SPSS 22.0 and GraphPad Prism 10. Normality was assessed by Shapiro-Wilk testing. Independent comparisons used two-sided unpaired Student’s t tests; paired comparisons used paired t tests. Data are mean ± SD, with SD error bars. The symbols *, **, ***, and **** indicate p values of <0.05, <0.01, <0.001, and <0.0001, respectively. Unless otherwise specified, experiments were repeated at least three times independently, and n values are given in the figure legends. Sample size was not statistically predetermined, and no observations were excluded. Randomization and blinding were used only when stated. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Source Data