Mechanical cues of an interpenetrating polysaccharide matrix regulate self-assembly of collagen fibers Authors: Fatemeh Tavakoli Joorabi, Nicholas J. Derr, Zecheng Li, Bryan A. Nerger, Chris H. Rycroft, David J. Mooney, Kyle H. Vining Published: 2026 ABSTRACT Collagen molecules self-assemble into supramolecular fibers within a molecularly crowded, polysaccharide-rich extracellular matrix (ECM) that has fluid-like, viscoelastic properties. Here, we determine that the viscoelasticity of alginate networks regulates the assembly of type I collagen fibers. The viscoelasticity and shear moduli of the alginate network were tuned by the polymer weight percentage and degree of cooperative ionic and covalent norbornene-tetrazine cross-linking. Stepwise shear strain applied to covalently cross-linked hydrogels generated higher stress than in ionic hydrogels. Hydrogels with reduced viscoelasticity also showed reduced water permeability. Second-harmonic generation confocal imaging revealed that decreasing viscoelasticity significantly suppressed collagen fiber self-assembly. Simulations demonstrated mechanical coupling between the hydrogel network and the aggregate size of collagen molecules, which was consistent with experimental results showing impaired rate and magnitude of self-assembly in covalently cross-linked networks. These results provide a framework for understanding how ECM mechanical properties can influence the assembly and organization of fibrillar macromolecules. INTRODUCTION Type I collagen is the most abundant structural protein in the human body, representing nearly 30% of the total protein mass and forming the primary fibrous scaffold of the extracellular matrix (ECM) (comprising up to 90% of the total ECM protein content) (1, 2). Abnormal regulation of type I collagen deposition and organization is implicated in a wide range of pathologies, including fibrosis, cardiovascular disease, and cancer (3). Excessive deposition and cross-linking of collagen type I (Col I) stiffen tissues in fibrotic disease (4), while tumor-associated collagen remodeling promotes invasion and metastasis (5). These observations highlight that understanding how type I collagen assembles within the ECM is not only fundamental to tissue biology but also crucial to uncovering disease mechanisms and identifying therapeutic targets. The biosynthesis of Col I assembly begins with the synthesis of individual pro-α peptide chains, which are the fundamental building blocks of procollagen. These pro-α chains assemble into procollagen molecules that fold into a triple-helical conformation with the assistance of C-terminal propeptides by zipping-up mechanism (6). Once secreted into the ECM, the loose terminal propeptides are cleaved by metalloproteinases, producing tropocollagen monomers of ≈300 nm length that are competent for fibril formation (7). Tropocollagen molecules align in a quarter-staggered configuration to form microfibrils, the basic repeating units of fibrils, characterized by the 67-nm D-period. This arrangement is stabilized through intermolecular cross-links at the N- and C-terminal telopeptides (8–11). Under physiological conditions (including appropriate temperature, pH, and molecular crowding) tropocollagen spontaneously self-assembles into collagen fibrils, which further organize into mature collagen fibers (12, 13). Although the biochemical pathways that regulate collagen self-assembly have been extensively studied (7, 14–16), it is less understood how the physical properties of the surrounding ECM regulate fibrillogenesis. The ECM can be conceptualized as a fibrous polymer network whose stiffness is determined by fiber rigidity, cross-link density, network connectivity, alignment, and compaction (17, 18). However, tissues are not purely elastic: They are viscoelastic, which means that they dissipate stress over time, and they can accumulate residual stress under sustained deformation (19, 20). Meanwhile, the ECM also exhibits finite permeability as a porous network, with pore size, tortuosity, connectivity, compaction, and cross-link density controlling how freely water and mobile molecules move through the matrix (18, 21, 22). In polymer networks, modest changes in cross-linking or compressive deformation can change permeability by orders of magnitude, thus altering the solute diffusivity (23). Accumulated stress, resulting from constrained matrix remodeling, can distort network geometry and densify local pores, further restricting both the translational and rotational mobility of tropocollagen monomers (17, 20). From a physical standpoint, collagen fibrillogenesis is a hierarchical self-assembly of biopolymers characterized by thermally activated nucleation and growth (16, 24, 25). Collagen networks fall within the class of semiflexible polymer system, whose emergent mechanical behavior arises from collective network connectivity (26–28). These fibrous matrices exhibit nonlinear elasticity, strain-stiffening behavior, and architecture-dependent stress transmission characteristic of fibrous polymer gels (26, 28–31). In hydrated environments, poroelastic coupling between the polymer network and interstitial fluid further links structural reorganization to stress redistribution (32, 33). Although many studies have examined individual mechanical properties of extracellular matrices, such as stiffness or viscoelasticity (21, 34), it remains unclear how the coupled influence of permeability, stress relaxation, and stiffness regulates the growth of collagen fibrils within a confined 3D environment. To address this, we developed an artificial ECM hydrogel that incorporates both ionic and covalent cross-linking. Alginate is a polysaccharide that forms ionic bonds with calcium ions (Ca2+), which mimics the ionic interactions found in native ECM (Fig. 1, A and B). Alginate was modified with norbornene (Nb) and tetrazine (Tz) functional groups to facilitate covalent cross-linking using click chemistry (Fig. 1, A and C). This dual cross-linking strategy enables the precise tuning of the mechanical properties of the matrix. To isolate the influence of matrix mechanics on collagen fibrillogenesis, we controlled the sequence of network formation. Alginate hydrogels were first fully cross-linked at 4°C, a temperature at which collagen remains soluble and does not assemble. This step establishes a preexisting polymer network with defined mechanical properties. After complete gelation (3.5 hours), the temperature was raised to physiological conditions (37°C) to initiate collagen self-assembly within the already formed polymer matrix (Fig. 1D). Because fibrillogenesis occurs after the polymer network is established, collagen fibers assemble under the geometric and mechanical constraints imposed by the surrounding hydrogel. Dual-cross-linked alginate hydrogels with tunable ionic and covalent cross-linking for sequential collagen assembly in a preformed matrix. (A) Schematic illustration of interpenetrating collagen-alginate (C-Alg) hydrogels (blue) and interpenetrating collagen-alginate-modified-with-NbTz (C-NbTz) hydrogels (red). Native alginate chains form ionic cross-links with Ca2+ ions to generate C-Alg hydrogels. To introduce covalent cross-linking, alginate was functionalized with norbornene (Nb) and tetrazine (Tz) groups via EDC/NHS coupling, enabling the Nb-Tz click reaction that forms stable covalent junctions in addition to ionic cross-links. (B) Molecular schematic of ionic cross-linking in C-Alg hydrogels mediated by Ca2+ ions. (C) Reaction scheme showing covalent functionalization and Nb-Tz click chemistry for dual-cross-linked C-NbTz hydrogel formation. (D) Schematic of collagen assembly within the alginate matrix. At 4°C, tropocollagen molecules remain unassembled; upon warming to 37°C, hierarchical fibril and fiber formation occurs, resulting in supramolecular collagen networks stabilized by both ionic and covalent cross-links. This sequential assembly strategy mimics key aspects of in vivo collagen deposition, where newly secreted tropocollagen assembles within an existing ECM that already has defined mechanical properties. In contrast to many collagen-alginate systems where both networks form simultaneously (35–39), our approach decouples matrix formation from collagen fibrillogenesis. This design allows us to systematically interrogate how independently controlled mechanical parameters (including stiffness, viscoelastic stress relaxation, and permeability) regulate the hierarchical assembly of collagen fibers within interpenetrating polysaccharide matrices. Combined experimental measurements and computational modeling provide a framework for quantifying how these mechanical properties influence fibril growth, organization, and overall ECM architecture. RESULTS Fabrication and mechanical properties of the hydrogel system Alginate G-blocks (glucuronic acid-rich segments) cooperatively coordinate Ca2+ ions through their carboxyl groups (Fig. 1B), forming ionically cross-linked networks with viscoelastic, time-dependent behavior (40, 41). To further tune the viscoelastic response, alginate chains were covalently functionalized with norbornene (Nb) and tetrazine (Tz) groups via carbodiimide coupling [N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride (EDC)/N-hydroxysuccinimide (NHS)], enabling subsequent Nb-Tz inverse-electron-demand Diels-Alder click reactions that introduce permanent covalent cross-links within the polysaccharide backbone (Fig. 1C) (35, 42). This dual-cross-linking strategy reinforces the ionic network without reacting with proteins or cells, allowing orthogonal control over stiffness and relaxation behavior. C-Alg [collagen-alginate interpenetrating network (IPN)] and C-NbTz (collagen-alginate modified with NbTz IPN) hydrogels enable tunable control over the mechanical microenvironment in which collagen self-assembly occurs. Because collagen remains soluble and unassembled at 4°C but undergoes supramolecular fibrillogenesis upon warming to 37°C (Fig. 1D), rheological measurements performed at 4°C characterize the mechanical properties of the polymer network before collagen fiber formation. This approach isolates the viscoelastic properties of the alginate-based matrix that defines the mechanical environment experienced by collagen monomers during subsequent fibrillogenesis. To quantify these preassembly mechanical properties, we performed oscillatory shear rheology to characterize gelation kinetics and the frequency-dependent viscoelastic behavior of collagen-alginate hydrogels (C-Alg and C-NbTz). Time sweep measurements were conducted at 4°C (0.1 Hz, 2% strain) to monitor the evolution of the storage modulus (G′) and loss modulus (G′′) as the hydrogel networks formed. For consistency across conditions, the time-dependent evolution of G′ and G′′ is reported over the first 4 hours of gelation (14,400 s), enabling comparison of the mechanical properties of ionically cross-linked (C-Alg) and covalently reinforced (C-NbTz) networks before collagen assembly. For ionically cross-linked alginate hydrogels (C-Alg), increasing Ca2+ concentration resulted in progressive stiffening of the network. The plateau storage modulus increased from approximately ∼10 Pa at 0.1% Ca2+ to ∼703 Pa at 0.2% Ca2+ and ∼2642 Pa at 0.3% Ca2+, while the loss modulus increased from ∼1.4 to ∼54 Pa and ∼209 Pa across the same Ca2+ concentrations. Therefore, the loss tangent (tanδ) decreased from ∼0.143 at 0.1% Ca2+ to ∼0.077 at 0.2% Ca2+ and ∼0.079 at 0.3% Ca2+ (Fig. 2A). Because tanδ=G′′/G′, this reduction reflects the increasing dominance of elastic energy storage over viscous energy dissipation as the density of ionic cross-links increases. A similar trend was observed for covalently reinforced hydrogels (C-NbTz) prepared with constant Ca2+ concentration (0.3% w/v) but varying Nb:Tz cross-link ratios. Increasing covalent cross-link density led to progressively higher stiffness. The plateau G′ increased from ∼2442 Pa for Nb:Tz = 2:1 to ∼3091 Pa for 1:2 and ∼3974 Pa for 1:1, while G′′ increased from ∼92 to ∼116 Pa and ∼131 Pa, respectively. In contrast, tanδ remained low (∼0.033 to 0.038), indicating highly elastic network behavior (Fig. 2B). Compared with C-Alg gels, C-NbTz hydrogels reached substantially higher storage moduli and exhibited lower loss tangents, consistent with the introduction of permanent covalent cross-links that stabilize the network and suppress viscous relaxation. Frequency sweep measurements (0.01 to 25 Hz, 2% strain) further revealed that C-Alg hydrogels exhibited modest increases in both G′ and G′′ with frequency, along with relatively higher and weakly frequency-dependent tanδ, indicative of a softer, more dissipative network. In contrast, C-NbTz hydrogels maintained substantially higher G′ with minimal frequency dependence and consistently low tanδ, confirming a more elastic, covalently reinforced network behavior (fig. S1). Rheological characterization of alginate-based hydrogels with ionic and covalent cross-linking. Oscillatory time sweep rheology (0.1 Hz, 2% strain) was used to monitor gelation kinetics and the transition from a more fluid-like state (loss angle >45° or tanδ > 1) to a solid-like gel state (loss angle <45° or tanδ < 1). Storage modulus G′ (solid line), loss modulus G′′ (dashed line), and loss tangent (tanδ) (dotted line) are shown. (A) ionically cross-linked C-Alg hydrogels [blue; 1.5% (w/v) alginate with collagen I (4 mg/ml)] at increasing calcium concentrations [0.1, 0.2, and 0.3% (w/v) Ca2+]. (B) Dual-cross-linked C-NbTz hydrogels formed by sequential ionic and covalent cross-linking [red; 1.5% (w/v) total alginate, collagen I (4 mg/ml)] with varying Nb:Tz ratios (2:1, 1:2, and 1:1) at 0.3% (w/v) Ca2+. Rheological measurements were performed on a TA Instruments HR-30 rheometer at 4°C using a 20-mm parallel-plate geometry with a 350-μm gap and a solvent trap to prevent dehydration. Representative rheological data are shown. The storage modulus (G′) and loss tangent (tanδ) values for all hydrogel formulations are summarized in fig. S2 (A and B). These formulations include 1 and 1.5% (w/v) alginate matrices, comprising ionically cross-linked C-Alg hydrogels and dual-cross-linked C-NbTz hydrogels prepared with Ca2+ concentrations of 0.1, 0.2, and 0.3% (w/v) and Nb:Tz ratios of 2:1, 1:2, and 1:1. Consistent with the trends observed in the gelation and frequency sweep experiments, increasing Ca2+ concentration increased hydrogel stiffness and reduced viscoelasticity, while the incorporation of Nb-Tz covalent cross-links further increased G′ and decreased tanδ. To assess the reproducibility of these trends across independent hydrogel preparations, four representative formulations spanning the range of mechanical properties investigated in this study were selected for additional analysis. The selected hydrogels were independently prepared in triplicate using a separate batch of alginate and Nb-Tz–functionalized alginate, and the resulting mechanical properties are shown in fig. S2 (C and D). The independently prepared samples reproduced the same trends observed in the full dataset, with C-NbTz hydrogels exhibiting higher G′ and lower tanδ than C-Alg hydrogels. These results demonstrate that the observed differences in hydrogel mechanics are robust across independent hydrogel preparations and material batches. Shear strain–dependent rheology and permeability of the alginate network We hypothesized that hierarchical collagen fibrillogenesis generates local volumetric reorganization and internal stress within the surrounding hydrogel network as aggregate size increases several orders of magnitude during assembly. In our precross-linked alginate matrix, individual tropocollagen molecules (length ∼300 nm, D ∼1.5 nm) have a cross-sectional dimension smaller than the characteristic alginate pore size (5 to 10 nm). Their elongated molecular length greatly exceeds the mesh size, meaning that transport requires threading through multiple pores and is therefore sterically constrained. As monomers laterally assemble into fibrils (D ∼100 nm) and subsequently bundle into micrometer-scale fibers (D ∼10 μm), the growing collagen structures rapidly exceed the native mesh size of the hydrogel. This geometric mismatch necessitates deformation of the surrounding alginate chains (D ∼1 nm), leading to localized volumetric expansion, chain stretching, and the development of tensile and compressive stress fields within the network. Thus, collagen assembly in this system does not occur in a mechanically neutral environment; rather, fibril growth progressively perturbs the preexisting polymer network, creating dynamic mechanical feedback that can regulate fibril thickening, bundle formation, and overall matrix architecture (Fig. 3A) (43, 44). We propose that in NbTz hydrogels, covalent cross-links restrict polymer chain mobility and diminish stress relaxation, leading to higher accumulated stress within the network (Fig. 3A). We measured the nonlinear shear response of Alg and NbTz hydrogels up to 100% strain (fig. S3) to evaluate the rheology of the gels under large shear deformations (Fig. 3, B and C). Pregeled hydrogel discs were preloaded with 5% axial compression to ensure full plate contact and to allow compressive stress to relax before testing (fig. S4). Both continuous and stepwise shear tests were performed to evaluate stress accumulation and relaxation behavior (Fig. 3D). Shear stress fully relaxed in Alg hydrogels after each shear step, whereas NbTz hydrogels exhibited more linear stress-strain behavior and accumulated stress over successive cycles (Fig. 3E). During stress relaxation, NbTz hydrogels retained approximately 80% of the applied stress, compared to 60% in Alg hydrogels (Fig. 3, F and G, 5% strain). This trend was consistent across a broader strain range (5 to 40%), where NbTz hydrogels continued to exhibit slower stress relaxation and greater stress retention than Alg hydrogels (fig. S5). We hypothesized that these differences in stress relaxation will affect collagen fibril alignment and fiber growth. Stepwise shear rheology reveals accumulated stress under large deformation in Alg and NbTz hydrogels. (A) Collagen self-assembly progresses over multiple length scales, generating large local deformations within the alginate network that involve both compressive and tensile components. (B) Shear deformation in the hydrogel corresponds to alternating compressive and tensile strain regions. (C) A 5% compressive preload was applied to pregeled 1.5% (w/v) alginate discs with 0.3% (w/v) Ca2+ before shear testing. (D) Stepwise shear relaxation protocol consisting of 10 increments of 5% strain applied at a rate of 1% s−1, each followed by 1500 s of stress relaxation. (E) Comparison between continuous shear and stepwise shear tests for Alg (left) and NbTz (right) hydrogels. (F) Representative normalized stress relaxation profiles for ionic (Alg) and dual-cross-linked (NbTz) hydrogels at the initial 5% strain step. (G) Relaxation percent as a function of strain and violin plot summarizing aggregate relaxation values (n = 8). Statistical significance was determined by an unpaired two-tailed Student’s t test (P < 0.05). The observed differences in storage modulus (Fig. 2) and stress relaxation (Fig. 3, D and F) indicate underlying changes in network architecture and cross-link density. Because these structural features govern both effective diffusivity and mechanical resistance within the hydrogel mesh, we next evaluated permeability to determine whether intrinsic nanoscale transport constraints differ between gel systems. Given that collagen fibrillogenesis occurs within a precross-linked porous network under static conditions, assembly kinetics are expected to depend on diffusion and confinement within the matrix rather than on bulk convective mixing (45). To quantify permeability, we measured fluid transport through the hydrogel network during controlled axial compression using a rheometer equipped with optical imaging. The volumetric flow rate, Q, of fluid exiting the gel was recorded during compression (Fig. 4A). Flux, q, was calculated using q=Q/A, where A denotes the effective cross-sectional area for fluid outflow. Permeability, k, was determined using Darcy’s law, q=(−kΔP)/(μl), where μ represents the dynamic viscosity of the fluid, ΔP is the pressure difference across the sample, and L is the effective distance for fluid flow corresponding to the hydrogel radius (Fig. 4B). These measurements revealed that Alg hydrogels exhibit significantly higher permeability than NbTz hydrogels (Fig. 4C). Permeability of Alg and NbTz hydrogels under axial compression. (A) Rheometer was equipped with a camera to detect cross-sectional area and volume change throughout the compression, enabling real-time volumetric flux measurement. [(B) and (C)] Volumetric flux, q, and normal stress, ΔP, are linearly fitted using Darcy’s law to calculate permeability of Alg and NbTz hydrogels. (B) Linear fitting flux versus normal stress profile and (C) calculated permeability for ionic and NbTz alginate. (D) A schematic of stress gradient and permeability affecting rate of collagen monomer and fibril diffusion in ionic and NbTz alginate. P value < 0.05 shows statistically significant difference by unpaired, two-tailed Student’s t test (n = 3 or 4). These permeability differences have important implications for collagen fibrillogenesis within the network. Continued fibril growth requires deformation of alginate chains and local volumetric reorganization of the matrix, a process that necessarily involves displacement of interstitial solvent and stretching of the surrounding polymer network. The ability of the matrix to accommodate this reorganization is governed in part by its hydraulic permeability, which determines how easily solvent can flow through the porous polymer network. In more highly cross-linked matrices, permeability is reduced, limiting the rate at which interstitial fluid can redistribute during fibril growth. Because fibril thickening and bundle formation increase the local collagen volume fraction, solvent must be displaced from the surrounding matrix to accommodate this structural expansion. When permeability is low and stress relaxation is limited, the network resists this solvent redistribution and polymer deformation, effectively increasing the energetic barrier to continued fibril thickening and bundle formation. Consistent with this model, the reduced permeability of NbTz hydrogels likely promotes the accumulation of mechanical stress on alginate chains surrounding assembling collagen fibrils, thereby restricting fibril elongation and limiting the formation of larger fibrillar structures (Fig. 4D). Collagen fiber self-assembly in collagen-alginate hydrogels We next investigated the impact of hydrogel composition and mechanical properties on collagen organization within the hydrogel matrix. We used second-harmonic generation (SHG) imaging to quantify collagen self-assembly in 1.5% (w/v) hydrogels with varying cross-linking chemistries and calcium ion (Ca2+) concentrations (Fig. 5, A and B) to achieve a specific range of stiffness (G′), energy dissipation (tanδ), and permeability. Results revealed that collagen assembly was impaired in C-NbTz hydrogels compared to C-Alg hydrogels, as evidenced by decreased SHG signal intensity and increased fibril dispersion (Fig. 5, A and B). A similar trend was observed in 1% (w/v) hydrogels, indicating that reducing the overall polymer content (Alg/NbTz chains) did not alter the general trend from C-Alg to C-NbTz (fig. S6). Cross-linking of interpenetrating alginate and collagen networks regulates collagen fiber self-assembly and morphology. (A) SHG images of ionically cross-linked C-Alg hydrogels [1.5% (w/v)] with increasing Ca2+ concentrations [0.1, 0.2, and 0.3% (w/v)]. Scale bar, 20 μm. (B) SHG images of dual-cross-linked C-NbTz hydrogels [1.5% (w/v) alginate and 0.3% (w/v) Ca2+] with varying Nb:Tz ratios (2:1, 1:2, and 1:1). Scale bar, 20 μm. (C) Quantification of collagen network morphology showing fiber area fraction (left), length (middle), and width (right). Statistical comparisons were performed by one-way ANOVA with Tukey’s multiple-comparisons test (n = 5). Pixel-to-micrometer conversion factor: 3.704 < pixel/μm < 5.553. (D) Multiple linear regression analysis correlating hydrogel mechanical parameters (G′, tanδ, permeability) with collagen fiber metrics (area%, length, and width). Dashed red lines indicate y = x; coefficients of determination (R2) show model fit. (E) PCA biplot showing loading vectors (blue arrows) for all input variables, including hydrogel mechanical parameters (G′, tanδ, permeability, stiffness, and shear stress) and collagen fiber metrics (area%, length, width, and straightness). PC1 (73.2%) and PC2 (14.03%) together explain 87.23% of total variance. Blue arrows represent all variable loadings, showing that increased stiffness and decreased viscoelasticity are associated with reduced collagen fiber length, width, and assembly. (F) PCA scores plot showing separation of samples by cross-linking type and composition. Ionically cross-linked C-Alg hydrogels (blue) cluster distinctly from dual-cross-linked C-NbTz hydrogels (red), reflecting how covalent cross-linking alters the mechanical-morphological relationship governing collagen self-assembly. Quantitative image analysis using ImageJ and CT-FIRE revealed pronounced differences in collagen fiber organization between C-Alg and C-NbTz hydrogels. The overall collagen area fraction was significantly lower in C-NbTz hydrogels compared to C-Alg hydrogels, indicating that fewer or smaller collagen networks formed under the more restrictive mechanical environment. Collagen fibers in C-NbTz hydrogels were markedly shorter, narrower, and straighter than those in C-Alg hydrogels, reflecting the influence of matrix mechanics on fiber morphology (Fig. 5C and figs. S12 to S14). At 1.5% (w/v) alginate, the matrix is denser and stiffer compared to 1% (w/v) (fig. S6), which led to shorter fibers and a reduced area fraction of collagen with an increase in fiber width. These findings highlight how matrix density and stiffness modulate collagen fiber morphology and organization, providing critical insights into the mechanics of collagen assembly. Collagen gels formed in alginate matrices lacking Ca2+ cross-linking (C-Alg, C-Nb, and C-Tz) exhibit comparable fiber area fraction and mean fiber thickness despite backbone modification with norbornene or tetrazine. In contrast, only the dual-modified C-NbTz hydrogel suppresses fibrillogenesis, indicating that baseline changes in alginate chemistry or hydrophobicity alone do not account for altered collagen nanoscale transport or assembly in the absence of network cross-linking (fig. S16). Furthermore, in control experiments where alginate and collagen were allowed to gel simultaneously at 37°C, collagen fiber width, length, and overall morphology remained comparable across gel formulations (fig. S17, A and B). Together, these results demonstrate that backbone modification or baseline hydration differences are insufficient to disrupt collagen assembly, and that sequential network formation, where collagen assembles within a preformed matrix, is required to reveal mechanics-dependent modulation of fibrillogenesis. Confocal and SHG imaging further confirmed differences in collagen assembly across gel conditions (fig. S18). Representative images show labeled collagen (detecting both soluble tropocollagen and fibrillar collagen) and SHG signal (detecting only structured collagen fibrils) for C-Alg and C-NbTz hydrogels at varying Nb:Tz ratios (fig. S18A). Quantification of labeled collagen revealed similar total collagen area fractions across all conditions (fig. S18B), consistent with the constant collagen concentration (4 mg/ml). In contrast, SHG imaging showed a significant reduction in fibrillar collagen area fraction in C-NbTz hydrogels compared to C-Alg gels, with the lowest fibrillar signal at Nb:Tz = 1:1 (fig. S18B), indicating reduced fibrillogenesis. Colocalization analysis further supported this observation: Both the fraction of labeled collagen monomers colocalized with fibrillar collagen and the fraction of structured fibrils colocalized with labeled collagen decreased with increasing Nb:Tz cross-link density (fig. S18, C and D). Notably, the percentage of fibrils colocalized with labeled collagen was consistently higher than the fraction of labeled collagen colocalized with fibrils because SHG detects only fibrils, whereas the fluorescence signal includes both fibrillar and soluble collagen pools. Together, these results indicate that while total collagen distribution remains constant, increasing matrix cross-link density suppresses collagen fibrillogenesis rather than simply altering collagen spatial distribution. We performed multiple linear regression analyses on 24 hydrogel formulations [1 and 1.5% (w/v) alginate matrices, including C-Alg and C-NbTz gels prepared with Ca2+ concentrations of 0.1, 0.2, and 0.3% (w/v) and Nb:Tz ratios of 2:1, 1:2, and 1:1] to identify which mechanical and transport parameters most strongly regulate collagen fiber architecture, using storage modulus (G′), viscoelasticity (tanδ), and permeability as predictors (Fig. 5D). The models revealed distinct but complementary relationships across collagen structural features, explaining most of the variance in assembly behavior. For collagen area fraction, the overall model was highly significant (R2 = 0.72 and P < 0.0001), with permeability emerging as the dominant positive predictor (β = +1.45 × 1016 and P = 0.0048), showing that higher matrix permeability facilitates fibril diffusion and lateral aggregation, which results in more extensive network coverage. In contrast, G′ (β = −1.028 and P = 0.1518) and tanδ (β = −37.78 and P = 0.5557) showed negative but nonsignificant trends, suggesting that increased stiffness and reduced stress relaxation or viscoelasticity limit overall collagen fiber formation. For fiber length, the regression model (R2 = 0.73 and P < 0.0001) revealed that higher G′ was associated with shorter fibers (β = −0.9063 and P = 0.0606), while greater tanδ correlated positively with fiber elongation (β = 76.50 and P = 0.0812). Thus, both matrix stiffness and viscoelastic relaxation jointly modulate fibril elongation. Permeability also displayed a positive but nonsignificant effect (β = +2.089 × 1015 and P = 0.4972). Fiber width was similarly governed by mechanical effects (R2 = 0.59 and P = 0.0004), where stiffness showed a near-significant negative correlation (β = −0.1298 and P = 0.0581), and tanδ a weaker positive trend (β = 8.344 and P = 0.1729). Permeability exhibited a negligible and nonsignificant effect (β = 1.019 × 1014 and P = 0.8143). Last, fiber straightness (R2 = 0.33 and P = 0.0429) displayed a significant dependence on tanδ (β = −0.2614 and P = 0.0427), with lower viscoelasticity yielding straighter, more aligned fibers and higher viscoelasticity producing wavier morphologies. Stiffness (G′) showed a weak positive but nonsignificant trend (β = +0.001577 and P = 0.2465), suggesting that increased rigidity may modestly enhance fiber alignment, while permeability also exhibited a nonsignificant positive effect (β = +1.3324 × 1013 and P = 0.1437) (fig. S15). Because multiple matrix properties (e.g., storage modulus, stress relaxation, and permeability) change simultaneously across gel formulations, we next applied principal components analysis (PCA) to identify which physical parameters most strongly correlate with collagen fiber architecture. We performed PCA on the 24 hydrogel formulations to identify the dominant variables governing collagen fiber organization and to visualize how mechanical and compositional features contribute to these differences (Fig. 5, E and F). The first two components accounted for 87.23% of the total variance (PC1 = 73.2% and PC2 = 14.03%), satisfying the Kaiser criterion (λ > 1). PC1 represented the primary mechanical axis, capturing the contrast between stiffness and stress accumulation versus viscoelastic relaxation and permeability. The loading plot (Fig. 5E) showed that G′ (0.611), shear stress (0.928), and collagen fibers’ straightness (0.583) loaded positively along PC1, whereas permeability (−0.928), relaxation percentage (−0.928), tanδ (−0.918), and collagen morphological parameters, such as area% (−0.901), length (−0.939), and width (−0.868) loaded negatively. Thus, stiffer, more elastic hydrogels promoted straighter but shorter and thinner collagen fibers, while more permeable and viscoelastic matrices supported larger, longer, and thicker networks. PC2 (14.03%) captured secondary variations related to transport and local relaxation. Fiber straightness (0.716), permeability (0.361), relaxation percentage (0.361), and stiffness (G′, 0.339) contributed positively along PC2, while fiber width (−0.39), length (−0.25), and shear stress (−0.361) loaded negatively, and tanδ (0.175) and collagen area% (0.029) showed weak contributions. These results indicate that PC2 defines a relaxation-permeability-viscoelasticity axis. Matrices with greater permeability and faster stress relaxation, along with moderately higher tanδ, promote more extensive collagen assembly. This relationship is reflected in higher fiber area% and the formation of straighter, more uniformly distributed collagen networks. The PCA score plot (Fig. 5F) showed a clear separation by cross-linking chemistry along PC1: C-NbTz hydrogels clustered at positive values, reflecting higher stiffness and stress retention, while C-Alg hydrogels clustered at negative values, corresponding to greater relaxation, permeability, collagen area%, width, and length. Overall, PCA revealed that permeability and viscoelastic relaxation cluster most strongly with collagen fiber morphology metrics, indicating that transport and stress-relaxation behavior are dominant predictors of fibril formation, whereas storage modulus alone explains less of the observed variation. Modeling and time-course analysis of collagen assembly in alginate hydrogels We hypothesized that the differences in collagen self-assembly observed between C-Alg and C-NbTz hydrogels arise from changes in the effective pore size of the surrounding hydrogel network. Increased cross-linking in C-NbTz is expected to reduce pore size, which restricts the nanoscale transport of tropocollagen monomers and small assembling clusters required for fibrillogenesis. In the model, these assembling clusters are referred to as collagen aggregates, representing groups of collagen molecules that progressively associate during fibril formation. To describe this process, we developed a simplified kinetic model that tracks how the population of aggregates evolves over time. Rather than treating each discrete aggregate size separately, aggregate size is approximated as a continuous variable, which allows the assembly process to be described by a smooth distribution of aggregate size or, equivalently, mass. This approximation simplifies the mathematical description while still capturing the gradual growth of aggregates during fibrillogenesis. In addition, we neglect spatial dependence, instead focusing on time evolution of the distribution in mass space. This reduction focuses the model on the key process controlling fibril formation (the transport-limited growth of collagen aggregates) while avoiding unnecessary spatial complexity. Although simplified, the model captures the essential physics governing collagen assembly under confinement. In particular, it reproduces the qualitative trends observed experimentally, where increasing covalent cross-link density reduces effective pore size and suppresses fibrillogenesis. The model further predicts that when covalent cross-links strongly restrict nanoscale transport, assembly can drop below measurable levels, while intermediate cross-linking produces a threshold regime in which reduced but still detectable fibril formation may occur. A model for collagen self-assembly in confining hydrogels We consider the hierarchical self-assembly of N collagen particles in a surrounding porous medium and denote the concentration of aggregates of size n as cn(t), where t is time. We write the evolution equations for the number concentration of aggregates with n > 1 asdcndt=kn−1↑c1cn−1+kn+1↓cn+1−(kn↑c1+kn↓)cn(1)and of monomers asdc1dt=−2k1↑c12+∑n=2Nkn↓cn−∑n=2N−1kn↑c1cn−1(2) At this point, we assume N ≫ 1 and treat the aggregate size n as a continuous variable following the approach of van Lith et al. (46). We introduce a normalized size variable ξ(t) given byξ=n−2N−2≈nN(3) Considering the family of functions ψ(ξ,t) such that ψ(ξ,t)=cn(t)/N, and denoting the monomer concentration ϕ(t)=c1(t)/N, we write the family of rate equations as a single conservative advection-diffusion equation in dimensionless form∂ψ∂t+∂j∂ξ=0, j(ξ,t)=q(ϕ,ξ)ψ−D(ϕ,ξ)∂ψ∂ξ(4)whereq(ϕ,ξ)=1N[Aˆa(ξ)ϕ−b(ξ)], D(ϕ,ξ)=12N2[Aˆa(ξ)ϕ+b(ξ)](5)are an effective velocity and diffusion through mass space. The rate constants have been scaled with dimensional A and B such thatk↑=Aa(ξ), k↓=Bb(ξ), Aˆ=NAB(6)for dimensionless functions a and b, which satisfy 0≤a,b≤1. We nondimensionalize time such that t∼1/B, leaving us with the following dimensionless parameters: N (the amount of collagen in units of the effective distance between neighboring aggregates in mass space), Aˆ=NA/B, and the functions a and b. Note the above conserves the total number of aggregates, so that their creation or destruction is determined by fluxes at the boundariesj(0,t)=k↑ϕ2−k↓ψ(0,t), j(1,t)=0(7)allowing for nucleation of monomers at ξ=0 and no flux at ξ=1 (the case where all monomers exist within a single size-N aggregate.) The monomer concentration varies asdϕdt=−2j(0,t)−∫01j N dξ(8)which conserves the number of monomers such thatϕ+∫ξψ N dξ=1(9)at all times. Because the experimental results indicate the size of assembled aggregates is effectively unchanged across the C-Alg and C-NbTz cases, we introduce an equilibrium size L and adopt a simple set of rate relationsa(ξ)=12[1+tanh(−ξN−LΔL)], b(ξ)=1(10)so that a transitions from 1 to 0 as εN (aggregate size) becomes larger than L. The parameter ΔL controls the rate of this change. The interpretation of this simplified model is that aggregates of size less than L undergo assembly before becoming so constrained by the effective pore size of the C-Alg system that assembly stops. To account for the presence of covalent cross-links, we let Aˆ=Aˆ0e−β so that the overall rate is slowed by an increase in the energy required for the forward reaction, ΔE=βkT, associated with the covalent C-NbTz bonds, which must be broken for assembly to continue. Ultimately, even in this case assembly (if it occurs) continues up to size L. Before presenting numerical solutions to the above equations for a selection of parameters, we describe the expected behavior. If Aˆ<1, then the effective velocity q will never be positive, so that while aggregates will form due to the influx associated with nucleation, no continuing assembly will propagate the small-size aggregates forward through mass space. Conversely, if Aˆ≫1, q will be large when assembly begins and this propagation will occur up to size L. Thus, we ascribe the cessation and onset of assembly to the transition between these two regimes. It is useful to consider the expected shape of any steady-state distribution. When ΔL≪L, we can approximate a as a step function, and solving Eq. 4 for the steady case yieldsψ∗∝exp[∫0ξAˆa(ξ′)ϕ∗−1Aˆa(ξ′)ϕ∗+1dξ′](11)where the asterisk subscript indicates the steady equilibrium value. Again, for Aˆ<1 the exponential argument will always be negative, and we expect an aggregate distribution with a maximum at ξ=0 which decays exponentially with ξ. If Aˆ≫1, the argument will be given byAˆa(ξ′)ϕ∗−1Aˆa(ξ′)ϕ∗+1≈Aˆϕ∗−1Aˆϕ∗+1,ξ