Nanopillar templating of biopolymer hydrogels augments stiffness and strength

In nature, structural biopolymers are highly organized to allow for the development of complex tissues within a living entity, including the human body. To match the properties found in these brous structural tissues, synthetic biomimetic hydrogels must have an optimal combination of stiffness, strength, and toughness; though an ideal combination remains challenging to achieve. Here, we report a general strategy to design stiff, strong, and tough hydrogels by conning biopolymers with a balance of rigid and weak domains into nanopillar topography. The connement within nanopillars templates the ber assembly process throughout the bulk of the lm. Compared to a at control, the application of the nanopillar topography increases the bulk stiffness ~ 160% to 20 MPa, strength ~ 350% to 36 MPa, and toughness ~ 450% to 8,500 kJ m − 3 . This simple templating strategy is suitable for a vast range of hydrogels, opening up the potential applications for a diverse array of materials.


Introduction
Naturally occurring tissues, such as skin and cartilage, have robust mechanical properties which are obtained via the assembly of brous building blocks into larger structures. These structures have outstanding and unusual mechanical properties due to the hierarchical variation of orientation and concentration [1][2][3] . The brillar structures are rich in both strongly and weakly bonded domains that allow for re-organization and adaptation with an applied load 4 . Synthetic hydrogels can closely mimic biological tissues due to the copious amounts of water within the matrix. However, historically synthetic hydrogels have been considered weak, and not suitable for structural applications such as in tissues 5 . There has been signi cant interest in the development of stiff, strong and tough biomimetic hydrogels for applications in arti cial tissues such as cartilage and skin 6,7 .
Strategies to obtain strong and tough hydrogels include the use of double-network hydrogels, noncovalent interactions, and applying orientation to the network. A few of the rst strong and tough structural hydrogels were double-network hydrogels consisting of a tightly crosslinked rigid and brittle network and a loosely crosslinked soft and ductile network 8 . The rigid and brittle hydrogel network acts as sacri cial bonds to dissipate energy by breaking into small clusters upon crack propagation. The sacri cial bonds are critical for dissipating energy and achieving the high toughness characteristic of the double-network gels 9 . Noncovalent interactions, such as ionic 10 and hydrogen bonding 11,12 (H-bonding) can lead to increased strength and stiffness. When multiple bonds are clustered together, the bonds act cooperatively and break concurrently, therefore stabilizing and strengthening the material 13,14 . Imparting orientation through the physical con nement of constituents during fabrication can control tensile stiffness, strength, and toughness [15][16][17] .
Embossing and imprinting techniques have been used for millennia, an early demonstration of this is Egyptians molding with clay. Thousands of years later, this technique has been developed into a quick and straightforward method to replicate features from the macro into the nanoscale onto a polymer surface 18,19 . When the dimension of the replicated pattern approaches the scale of the polymer chain, the polymer molecule is con ned within the pattern 20 . The ow of the polymer chain into the cavity causes viscous ow and chain orientation due to the nanoscale dimension, and dimensional reduction during shear 21,22 . Polymers tend to align along ow lines during deformation, subsequently the lithography process induces long range polymer alignment 23,24 .
Herein, we demonstrate a new approach to modulating the mechanical response of brous linear biopolymer hydrogels and lms through the application of a con ning physical surface topography. The free-standing lms with nanopillars are generated by dropcasting the biopolymer solutions onto a mold with nanohole topography, subsequently the polymer solution in ltrates the nanoholes and forms nanopillars. We demonstrate that the polymer chain alignment induced by nanohole in ltration increases the clustered, rigid bonds and ordered domains. The domain formation causes a cascading effect that increases the inter-and intra-chain interactions in regions far from the nanostructure interface, subsequently templating the brillar assembly throughout the material. The strong, clustered bonds serve to strengthen the matrix of weak, reversible bonds, analogous to the toughening mechanism in doublenetwork hydrogels 25 . Through the design of our materials, we introduce the use of surface topography for con ning nano brils into knots that enhance stiffness, strength, and toughness. Nanopillar Topography Templates The Nano ber Assembly Chitosan, alginate, cellulose, and chitin are all hydrogen bonding, ber-forming biopolymers. Chitin is an abundant linear polysaccharide found in the structural skeleton in crustacean and fungal walls 26 . The transformation of the C-2 acetamide groups in chitin into primary amino groups results in chitosan, although the deacetylation is rarely complete 27 . A high degree of deacetylation in chitosan renders it soluble in weak acids 28 . Here, we dissolved chitosan powder in 0.1 M acetic acid. SEM imaging reveals the aggregation of chitosan building blocks into nano brils approximately 58 nm in diameter and up to the micrometer range in length ( Fig. 1-b, Supplementary Fig. 1). The nano brils bundle to form bers ~ 1 µm in diameter ( Fig. 1-c).
The free-standing lms were fabricated by drop-casting the polymer solutions onto molds (additional details in Supplementary information). The solvent was evaporated overnight, creating water-soluble chitosan lms. The lms are rendered water insoluble by neutralizing them in a 10 wt.% sodium hydroxide solution 29 . The neutralization prevents ionic repulsion between the chains, allowing the formation of weak reversible H-bonds, rigid H-bond clusters, and crystallites between the brous chitosan chains 30 ( Fig. 1-d). The lms are dried, then peeled from the mold.
Films with nanopillar topography were made in a similar fashion ( Fig. 1-a). We used molds with nanohole periodicities 200 nm, 300 nm and 500 nm, referenced here as P200, P300 and P500 (Supplementary Table 1). The nanohole dimensions are comparable to but larger than the characteristic nano bril size of chitosan 20,31 . These length scales ensured that the brils were con ned within the nanopillar topography if they were able to in ltrate the nanoholes during the drop casting process. The bril polymer chains are aligned during in ltration via extensional ow at the nanohole entrance and high shear stress at the nanohole wall 32 . The brils consequently bundle into bers. The aligned brils in the bers form rigid bond clusters and ordered domains that enhance the ber alignment and interactions throughout the bulk of the lm. As the solvent is evaporated, the ber chain alignment is preserved 33 . The brous network throughout the bulk lm is therefore anchored by the orientation and alignment at the surface layer ( Fig. 1-e).
SEM images of the lms ( Fig. 1-f) con rm good replication of the nanopillar arrays across the entire sample surface. The P200 lm nanopillars, however, collapsed into hierarchical clusters ( Fig. 1-f), while the P300 and P500 lm show stable nanopillars without clustering or bunching. We con rmed the nanoscale morphology of the lms formed by this process by freeze-drying them and examining them with AFM ( Supplementary Fig. 2). The root mean square roughness (rms) of the at lm is 12.4 nm, increasing to 72.5 nm in the P300 lm.

Mechanical Response Of Templated Nanopillar Films
We characterized the mechanical properties of the brous lms in uniaxial tension. The lms were trimmed into rectangular lm specimens with a gauge length (between grips) of 10 mm and width of 2 mm, such that the nanopillars were perpendicular to the plane of the lm ( Fig. 2-a, Supplementary Fig. 3).
The tensile tests were performed at a strain rate of 0.3 mm min − 1 . A at hydrogel lm at equilibrium swelling was used as a control. The typical stress-strain curve of the control is shown by the blue curve in  Table 2).
The size of the nanopillars signi cantly enhanced the macroscopic mechanical response, following the order: P500 > P300 > P200 > Flat ( Fig. 2-c, Supplementary Fig. 4). The slope in the initial region of the stress-strain curve, which represents the Young's modulus E of the lm, is strongly affected by increasing the nanopillar dimensions ( Fig. 2 Table 3). The greatest increase in E was observed in the P500 lm, with E ~ 20.7 MPa, a ~ 160% increase relative to the control. The enhanced modulus is driven by a high total number of crosslinks (both weak and rigid). Thus, simply by adding the surface topography, the modulus of the material became comparable to that of skin ( Fig. 2 Table 4).

-j, Supplementary
The slope of the stress-strain curve after the proportionality limit, the tangent modulus E tangent , increased to 89.7 MPa in the P500 lm, a ~ 170% increase relative to the control (Fig. 2-e). Additionally, the toughness of the P500 lm, de ned here as the work at break, increased signi cantly to 8.5 MJ m − 3 , a ~ 450% increase (Fig. 2-f). The high toughness observed is consistent with the existence of large, rigid sacri cial domains within the nanopillar topography lms (Fig. 2-h). The stress at break , increased to 36.8 MPa, a ~ 350% increase in strength compared to the control ( Fig. 2-g). The high strength of the lm is attributed to the large number of rigid H-bond clusters. The strength surpasses other high-strength hydrogels recently developed ( Fig. 2 Table 5).
While the stiffness, strength, and toughness increased the most with the P500 lm, the maximum strain, , of the lms was unique in that the P300 lm had the greatest fracture strain of ~ 80%, compared to 50% in the at lm ( Supplementary Fig. 4). The fracture strains decreased in the order P300 > P200 ~ P500 > at. The smaller P300 size results in a less highly aligned brous network, and fewer sacri cial, rigid H-bond clusters. The bers within the P500 topography are more highly aligned within the cavityperhaps due to the greater hole depth. Consequently, the converging brous network is more highly aligned within the P500 lm, leading to a denser network of rigid bonds, and the lm is unable to stretch as much as the P300 lm. The situation is analogous to ow convergence just before the entrance region in a spinneret 34 . Origin Of Process-structure-property Relationship The periodicity Ps of the nanopillars in our strained samples was determined by application of twodimensional fast Fourier Transform (2D FFT) to the AFM images ( Fig. 3-b). The periodicity of the unstrained sample, Po = Ps (ε=0), was calculated to be 460 nm and 205 nm for the P500 and P300 samples, respectively. Po slightly decreased in comparison to the corresponding freeze-dried sample, probably due to the loss of some water from the lms. In both the P300 and P500 samples, the periodicity increased exponentially with the application of strain, Ps(ε)=Poexp(ε r-1), similar to the nonlinear stress-strain behavior observed in the tensile experiments (Fig. 3-c).
The aspect ratio (η=hd^(-1)) of the P500 lm decreased from 1.7 to 1.2 ( Fig. 3-d, Supplementary Fig. 5, Supplementary Table 6). On the other hand, that of the P300 lm was largely independent of the applied tensile strain. The difference may indicate that there is a threshold that must be met for the nanopillars to deform during tensile deformation. This may be related to the diameter of the cavity, which must be wide enough to accommodate multiple nano brils to form bundles. Our results suggest that the ber bundles pulled out during loading in the P500 lm, but not in the P300 lm.

Inter-chain And Inter-layer Microstructural Organization
We sought to understand the in uence of nanopillars on the bulk lm composition and microstructure.
The water content of the hydrated lms remained at 60-70%, similar to that in naturally occurring cartilage (Supplementary Table 3) 35 . Even before hydration, in the as-prepared state, the modulus E increased to 2.2 GPa in the P500 lm, a ~ 57% increase compared to the control ( Supplementary Fig. 6, Supplementary Table 7). This result indicates that rigid, sacri cial bonds had formed during the fabrication procedure, prior to the introduction of water.
We con rmed that the as-prepared P500 and at lms contained different quantities of rigid bonds with loading-unloading tests. Both the at and P500 lms demonstrated mechanical hysteresis, indicating the existence of rigid, sacri cial bonds and weak, reversible bonds within the lms (Fig. 4-a). The P500 lm exhibited a large hysteresis, which is consistent with the fracture of rigid, sacri cial clusters and bonds 36 (Fig. 4-b). After the rst loading, some weaker bonds are able to recover and reform within both the at ϵ m a x and P500 lms. The greater portion of fractured sacri cial bonds within the P500 lm are unable to completely recover and reform, consistent with the greater hysteresis area in the P500 lm compared to the at lm ( Fig. 4-b). The results show that the at lm contains weak, reversible bonds.
In the as-prepared dry state the lm thicknesses are ~ 20 µm (Fig. 4-c). At equilibrium swelling, the thickness of the at lms increased 144%, while the thickness of the lms with P500 topography increased by only 45%. The restriction of polymer chains during fabrication causes a denser, more ordered matrix 37 . X-ray diffraction results con rm that both the as-prepared nanopillar and control lms contain crystalline domains between random amorphous coils ( Fig. 4-f, Supplementary Table 8). The more prominent diffraction peaks in the P500 lms indicate that they have a larger number of crystalline domains than the control lm. The strong attractive effect of the H-bonding results in the large number of crystalline domains. This is consistent with the existence of rigid, sacri cial bonds.

Generality Of Nanopillar Templating Strategy
Our templating strategy con nes polymer chains within nanopillar topography enhancing the formation of rigid, sacri cial bonds within a weakly bonded network. We chose to contrast the rigid, sacri cial bonds against weak, reversible bonds by crosslinking chitosan lms with either the chemical crosslinker genipin, or the physical crosslinker tripolyphosphate (TPP). The chemical crosslinks should act as rigid bonds within the network and enhance the mechanical properties ( Fig. 5-a). The physical crosslinks should act as weak, reversible bonds within the network and should not enhance the mechanical properties ( Fig. 5-b). We chose to utilize only the P300 and P500 surfaces, as we expect that the most signi cant mechanical improvement will occur with these larger surface structures.
We chemically crosslinked the chitosan lms with genipin, a common crosslinker for amine-containing biomaterials 38,39 . The genipin crosslinked chitosan system displays hierarchical ber assembly ( Supplementary Fig. 7). Uniaxial tensile testing veri ed that the nanopillar topography enhanced the mechanical properties of the material (Fig. 5-c). The control lm has a Young's modulus E of 3.8 MPa.
The E increased to 13.1 MPa for the P500 lm, a ~ 245% increase compared to the control. The enhanced E in the nanopillared lms con rms that a greater quantity of rigid crosslinks and interactions formed in comparison to the control. However, these permanent, rigid crosslinks are unable to reform after fracture, resulting in a low . These results illustrate that permanent crosslinks networks con ned into nanopillar topography form rigid bond clusters that tune the mechanical properties.
Next, we examined chitosan lms crosslinked with the physical crosslinker TPP, which forms weak, reversible bonds. The ionically TPP crosslinked network lacks rigid, sacri cial components. The TPP crosslinked control lm had a E of ~ 1.3 MPa, much less than the chemically crosslinked lm, an expected result due to the dynamic nature of ionic bonds 30 . The TPP crosslinked lms with nanopillar topography had a similar E to the control lm (Fig. 5-d). This is due to the dynamic, reversible bonds in the ionic network which can quickly reform after deformation. These results illustrate that a balance of weak and strong bonds are necessary to enhance the mechanical properties via nanopillar templating. We demonstrate the generality of our templating strategy by applying nanopillar surface topography onto lms cast from the biopolymer alginate. The alginate lm is ionically crosslinked with the divalent cation of calcium (Ca 2+ ) to form a Ca-alginate hydrogel. Unlike the chitosan lms which contain rigid H-bond clusters and weak, reversible H-bonds, the Ca-alginate hydrogel contains both hydrogen and weak, sacri cial ionic bonds 40 . The H-bonds should only be rigid prior to hydration, when there is no water to compete with them.
We con rmed that nanopillars were replicated onto the Ca-alginate surface (Supplementary Fig. 8). We found an increase in the number of crystalline domains in the as-prepared Ca-alginate lm when imprinted with a nanopillar surface ( Supplementary Fig. 9), consistent with the hypothesis that the interchain interaction in the nanopillared lms was enhanced. We began to characterize the mechanical properties of the dry, as-prepared Ca-alginate lms. The mechanical tests show that E increased ~ 38% from 2.6 GPa in the at lm to 3.6 GPa in the P500 lm, respectively ( Supplementary Fig. 10). This con rms the nanopillared ca-alginate lm contains a greater quantity of crosslinks. After 24 hours immersed in water, the nanopillar surface relaxed into the bulk as the ionic crosslinks dynamically broke and reformed 41 , therefore disrupting the balance of strong and weak bonds. As a result, the mechanical properties of the hydrated lm are independent of surface topography.

Discussion
In summary, we demonstrate that con nement within nanopillar topography templates the ber assembly process of brous biopolymers and represents a general strategy to design stiff, strong and tough hydrogels. The biopolymer ow into the nanohole during fabrication facilitates the con nement and close packing of its chains which causes a cascading effect of enhanced rigid, sacri cial domains throughout the bulk lm. To the authors' knowledge, the previous free standing polymer lms that feature nanostructure topography have demonstrated a decrease in the mechanical properties compared to a at control 42,43 . These nanopillared hydrogels are the rst example to demonstrate that surface topography can make a hydrogel stiff and strong. As revealed in the present study, a requirement for this templating strategy is the existence of both rigid, sacri cial domains and weak bonds. In short, this nanopillar templating method presents a unique strategy to achieve high stiffness, strength, and toughness in biopolymer lms.
Considering that strong interactions are present in many forms, such as hydrogen bonds, covalent bonds, and crystalline domains, we foresee this templating method will not be restricted to the systems presented here. We envision that this relationship could be extended to other length scales in which strengthening and stiffening behavior is desirable. If our conclusion is correct, then the structural surface pattern only needs to be larger than the typical density uctuation in a material. This generalized templating strategy is suitable for a vast range of originally weak hydrogels, opening up the range of possible materials for a diverse range of applications. Figure 1 Integration of the nanopillar surface structure that templates the ber assembly in the hydrogel lms (a)

Figures
An illustration of the nanopillar fabrication process. The polymer solution is dropcast onto the nanohole mold, and the polymer chains in ltrate the mold nanocavities. As the solvent is evaporated the chitosan chains bundle together into bers with strong and weak bonds that stabilize the structure. The nanopillars enhance the formation of strong bonds. A free-standing polymer lm is peeled from the mold.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.