Light-triggered topological programmability in a dynamic covalent polymer network

See allHide authors and affiliations

Science Advances  27 Mar 2020:
Vol. 6, no. 13, eaaz2362
DOI: 10.1126/sciadv.aaz2362


Dynamic covalent polymer networks exhibit unusual adaptability while maintaining the robustness of conventional covalent networks. Typically, their network topology is statistically nonchangeable, and their material properties are therefore nonprogrammable. By introducing topological heterogeneity, we demonstrate a concept of topology isomerizable network (TIN) that can be programmed into many topological states. Using a photo-latent catalyst that controls the isomerization reaction, spatiotemporal manipulation of the topology is realized. The overall result is that the network polymer can be programmed into numerous polymers with distinctive and spatially definable (thermo-) mechanical properties. Among many opportunities for practical applications, the unique attributes of TIN can be explored for use as shape-shifting structures, adaptive robotic arms, and fracture-resistant stretchable devices, showing a high degree of design versatility. The TIN concept enriches the design of polymers, with potential expansion into other materials with variations in dynamic covalent chemistries, isomerizable topologies, and programmable macroscopic properties.


Isomerization of small molecules is a foundational concept in organic chemistry. For covalent polymers, the substantially more bond connections in a single macromolecule lead to many more opportunities. Non–cross-linked polymers with identical compositions can exist in various forms, ranging from linear to blocky, grafted, star-shaped, or even bottle-brushed architectures (15). This richness in macromolecular topology plays an essential role in polymer science, as it allows versatile design of materials with markedly different properties using the same set of monomer(s). Most commonly, such topological control is achieved at the polymerization stage, and alteration of topology is not permitted afterward. The same limitation also applies to covalent polymer networks, although the impact of network topology on physical properties is also well recognized (613). Dynamic covalent bonds, on the other hand, open up opportunities in a different way for the design of cross-linked polymer networks. Via reversible covalent bond breaking/reforming, recyclable/reconfigurable/self-healable network polymers can be realized (1427). For these purposes, it is essential that the network topology remains statistically identical throughout the bond exchange process. With an opposite thought in mind, we envision that if a dynamic covalent network is designed to isomerize to different topologies, then new opportunities should arise. We show hereafter a molecular design that allows network isomerization among multiple topological states (>2). We further illustrate that this unique attribute leads to a high degree of versatility for programming polymer properties.


The heart of our topology isomerizable network (TIN) concept is the network topological heterogeneity. For typical dynamic covalent polymer networks, their chain segmental distribution is statistically homogeneous, corresponding to the highest entropic and thermodynamically stable topological state. Dynamic bond rearrangement therefore cannot alter their network topology (1427), although some exceptions exist such as change in network side-chain loops (13). Here, we define topological heterogeneity as the nonuniform chain segmental distribution in a polymer network. It can be purposely designed into a network at the synthesis step. Topological heterogeneity corresponds to a thermodynamically unstable state owing to its low entropy. However, the robustness of covalent bonds allows trapping this kinetically stable state. When the dynamic covalent bonds are activated, bond exchange leads to topological homogenization. Any intermediate topological states before reaching the full topological homogeneity can be kinetically trapped by deactivating the bond exchange. Within the general TIN concept, many design variations are possible in terms of both the choice of dynamic bonds (1427) and the topological heterogeneous states. To ensure the general applicability, we choose transesterification in a polyester network for the following demonstration because ester bonds are the most common polymer building units. For ease of demonstration, we pick a topological heterogeneous polymer network with a permanent network mainframe, long grafted dynamic chains, and exchangeable pendant groups. With the former ensuring the network integrity, the chemical exchange between the latter two allows topological change in terms of the length, density, and distribution of the side chains.

The network is synthesized via two-step thiol-ene click reactions (Fig. 1A). The first step between two diallyl ether monomers and a dithiol yields a thiol-terminated polymer with pendant hydroxyl groups and grafted polycaprolactone chains (PCLs; molecular weight, 5500). The molecular weight of this thiol-terminated precursor polymer as determined via 1H nuclear magnetic resonance (NMR) end group analysis (fig. S1) is about 7500. In the second step, this precursor polymer is cross-linked via reaction with a tetra-allyl ether cross-linker. A photo-latent catalyst that can release an organic base, 1,5,7-triazabicyclo[4.4.0]dec-5-ene (TBD), is subsequently introduced. The gel fraction of the resulting polymer network is 95%. From the network design standpoint, two features are prominent (Fig. 1B): The network mainframe is composed of all permanent covalent linkages, and the ester bonds on the PCL graft chains are dynamic; namely, they can undergo transesterification with the pendant hydroxyl groups in the presence of TBD catalyst. The first feature distinguishes our network from other known dynamic covalent networks for which the mainframes are generally dynamic (1427). This permanent mainframe ensures network integrity while allowing topological change via side-chain bond exchange, a characteristic that is key to the material versatility demonstrated later in the context. We emphasize, however, that as a design variation, the network mainframe can also be dynamic, provided that proper control of the bond exchange equilibrium is in place to avoid liquefying the whole network. Nevertheless, the permanent mainframe simplifies the concept demonstration. Specifically, we envision that, as the grafted PCL chains undergo transesterification with the pendant hydroxyl groups, the PCL chains get shorter and redistribute onto the mainframe. Accordingly, the network should isomerize from a grafted topology (long chain and low grafting density) into a brushy topology with a high density of short PCL chains (Fig. 1C). We note that, for simplicity, Fig. 1C shows only two isomeric states, the starting state and the ultimate equilibrium state. In reality, there are many intermediate isomeric states distributed on a continuum between these two topologies. Except for the equilibrium topology, all others, including the starting topology, represent kinetically trapped states that are stable at temperatures the transesterification is deactivated.

Fig. 1 Synthesis of the polymer network and its topological isomerization mechanism.

(A) Network synthesis. (B) Transesterification between the pendant hydroxyl group and long grafted PCL chain. (C) Network topological isomerization via transesterification. UV, ultraviolet.

Because polymer networks are insoluble, direct verification of the above hypothesis is difficult. We therefore resort to model experiments. In these experiments, with the addition of TBD, a non–cross-linked polymer (Fig. 2A) similar to the network precursor polymer in Fig. 1A is heated to 80°C for 1.5 hours, and its compositional change is investigated using proton NMR (1H NMR) analysis. For more accurate 1H NMR end group analysis, the molecular weight of PCL (3000) on the non–cross-linked polymer for the model experiments is intentionally chosen to be lower than that of the network precursor polymer (5500). Specifically, two types of hydrogen atoms (labeled Ha and Hb in Fig. 2A) on the alpha-carbons (relative to the terminal hydroxyl groups) are monitored. Figure 2B shows that, as the reaction proceeds, chemical shifts for both hydrogen atoms change slightly. The signal intensity of Ha increases, while that of Hb decreases, and such a change is more and more pronounced as the TBD concentration increases. This suggests the occurrence of the transesterification between pendant hydroxyl groups and the PCL chains. The net effect is less and less hydroxyl groups and progressive redistribution and shortening of the grafted PCL chains. The conversion of the pendant hydrogel group and the molecular weights of the grafted PCL chains can be calculated from the 1H NMR spectra using the equations shown in fig. S2. The results are captured in Fig. 2C. The overall results point to the occurrence of topological change outlined in Fig. 2D, and a higher TBD concentration results in a progressively more brushed topology. We should note that kinetics of transesterification in the polymer solution is expected to be different from that in a solid-state polymer network. However, it verifies the molecular mechanism on a qualitative basis.

Fig. 2 Modeling network topological isomerization via non–cross-linked precursor polymer.

(A) Chemical structure of precursor polymer. (B) 1H NMR spectra of the precursor polymer upon isomerization with different amounts of TBD catalyst (full-range spectra provided in fig. S2). (C) Quantitative evolution of the pendant hydroxyl group and the molecular weight of the grafted PCL chain during isomerization. (D) Schematic illustration of the topology isomerization for the precursor polymer.

This strong dependence of the isomeric states on the TBD motivates us to explore potential approaches to manipulate the topology at the polymer network level via spatiotemporal control of the catalyst. This is achievable with the photo-latent catalyst that can release TBD upon light irradiation (Fig. 3A). Accordingly, Fig. 3B shows an approximate linear release profile of TBD triggered by light, reaching a plateau yield of 65% at 11 min. In accordance with Fig. 2C, longer irradiation followed by thermal isomerization (80°C, 2 hours) should lead to progressively more brushed topology with denser and shorter PCL chains. The direct consequence is that the PCL melting temperature decreases with the light irradiation (Fig. 3C). The degree of crystallinity calculated from Fig. 3C simultaneously decreases with the irradiation time (Fig. 3D), resulting in reduction in room temperature modulus until reaching a plateau at 5 min. This is seemingly inconsistent with Fig. 3B, which suggests that the catalyst continues to release upon further irradiation. We believe that this discrepancy is due to the fact that 5-min irradiation releases enough catalyst for the network topology to reach its equilibrium state under the thermal isomerization conditions; thus, releasing more catalyst with long irradiation does not change the modulus. Overall, as the isomerization proceeds, the side-chain length is reduced with a corresponding increase in the grafting density. The combined effect is reduction in crystallinity. The room temperature mechanical properties of these networks are dominated by the crystallinity. As a result, isomerization leads to reduction in room temperature modulus and increase in maximum strain as shown in figs. S4 and S5. The light-triggered mechanism further allows spatio-programming of the topology, thus the corresponding thermomechanical properties within one sample. This is demonstrated in Fig. 3E. Three regions of the sample are light-irradiated for 0, 2, and 5 min, respectively. After the thermal isomerization, the regions become distinctively different materials. Their difference in elongation under the same load reflects their different room temperature moduli. Upon heating to various temperatures, the sample undergoes further elongation in different regions as they melt in a stepwise fashion. We emphasize that, although the transesterification reaction is reversible, the isomerization process is irreversible because the entropy always drives isomerization toward more topologically homogeneous states. This ensures that the topological isomers are stable at temperatures sufficiently below the thermal isomerization temperature (80°C). This is an important attribute for many potential applications.

Fig. 3 Network topological patterning via a photo-latent catalyst.

(A) Photo-latent catalyst and its light-triggering mechanism. (B) Release of the TBD catalyst upon light irradiation (more details in fig. S3). (C) Differential scanning calorimetry curves of the isomerized networks after light irradiation and thermal isomerization. (D) Room temperature moduli (20°C) and PCL crystallinities of the isomerized networks. (E) A topologically patterned sample (with light irradiation time labeled) and its elongation at different temperatures under a constant load of 10 g. Scale bar, 1 cm. Photo credit: Binjie Jin, Zhejiang University.

During the thermal isomerization process, the network integrity is always maintained owing to the permanent mainframe. The solid-state isomerization differs from other topological shifting systems: a supramolecular gel–based system (28) and a dynamic covalent-based system that undergoes topological change only in its liquid state (i.e., with full network disintegration) (29). The supramolecular gel is particularly relevant for further comparison. It relies on the reorganization of metal organic cages for light-triggered topological switching (28). Mechanistically, it has two equilibrium topological states under two different lights. In principle, proper control of light irradiation allows it to access a continuum of states that are mixtures of the two equilibrium states. By comparison, our system is a covalent network with only one equilibrium state, the one that corresponds to topological homogeneity. All the other topological states correspond to different average side-chain length, instead of a mixture of starting long side chains and equilibrium short side chains. Although the topological isomerization in our covalent system is irreversible, the different topological isomers are kinetically stable as long as the temperature is sufficiently below the transesterification temperature. A solid dry polymer (our system) has mechanical properties that are markedly different from supramolecular gels. Consequently, they have their own appeal toward different potential applications. Nevertheless, the solid-state isomerization in our system, along with the spatio-selectivity provided by the photo-latent catalyst, allows control of multiple topological isomeric states (>2) in a pixelated manner (30). These unique advantages are explored in fabricating various geometrically architecture materials/devices in Fig. 4.

Fig. 4 Demonstration of the device potential for the topologically patterned networks.

(A) Shape memory hinge. Credit: Binjie Jin and Huijie Song, Zhejiang University. (B) Adaptive robotic arm (sample length: 8 cm after stretching). (C) Topologically patterned kirigami structure and simulated stress distribution upon stretching by 50%. (D) Comparison of the fracture behaviors upon stretching between non-patterned and patterned kirigami structures. (E) Stress-strain curves of the kirigami structures. Photo credits: Binjie Jin, Zhejiang University. Scale bars, 1 cm.

For simplicity, we first focus on two topological states defined by light irradiation of 0 and 6 min, corresponding to the white and yellow regions in Fig. 4A. A shape memory device is obtained via laser cutting and spatio-selective topological isomerization. The device is uniaxially stretched by 30% at 80°C and cooled down to 20°C under the load. Upon removing the load, a triple-hinged structure (temporary shape 1) is obtained. This triple-hinge geometry is noteworthy. For existing shape memory polymers (3133), their temporary shape correlates directly to the form of the deformation force. Therefore, uniaxial stretching would only lead to an elongated temporary shape, and more complexed shapes would require correspondingly more complex deformation forces. By comparison, a simple in-plane stretching deformation force here results in a more complex out-of-plane temporary shape, namely, the triple hinge. For our system, the different topological states for the white and yellow regions result in difference in shape fixing capability. Specifically, their shape fixities are 98 and 5% (fig. S6), respectively. This nonuniform shape fixing leads to a mechanical imbalance that determines the temporary shape in addition to the contribution of the deformation force. Intriguingly, this triple-hinged shape represents a mechanically bistable state (34). It can readily interconvert with another mechanically equivalent triple-hinged structure (temporary shape 2 in Fig. 3A). The bistable nature does not negatively affect the shape recovery, as both temporary shapes can recover to the original shape. We emphasize that each hinge brings two bistable states; thus, the total number of bistable configurations for one sample can be increased to 2n, with n being the hinge number. The two temporary shapes in Fig. 4A are only representatives of eight temporary shapes associated with the triple hinge (fig. S7).

The simplicity in hinge fabrication, along with the convenient light irradiation, allows easy access to a hextuple-hinged structure involving four different topological states (Fig. 4B). Following the same procedure in Fig. 4A, a single sample can be programmed into three energetically identical temporary shapes. Upon heating, they show distinctively different pathways toward shape recovery, mimicking the function of an adaptive robotic arm (see movie S1 and mechanical simulation in fig. S8). Without the cutting in Fig. 4 (A and B), topological patterning alone can lead to a diverse set of three-dimensional temporary shapes (fig. S9). This exceptional simplicity in converting a two-dimensional film into a three-dimensional one should benefit many technological areas that are typically constrained to two dimensions such as electronics and microfabrication (35, 36). The network topological patterning is also useful in engineering material elasticity. Previous work suggests that geometric architecturing via kirigami cuts represents an effective approach (37); our topological patterning introduces another mechanism that further extends the upper limit. Figure 4C shows that strategic topological patterning substantially alters the stress distribution upon stretching. Consequently, the topologically patterned structure shows much enhanced elasticity over its nonpatterned counterpart with minimal impact on the maximum stress (Fig. 4, D and E, and movie S2).

We emphasize that there are reported methods to spatio-selectively regulate material stiffness, notably network gels that can undergo photodegradation (89) or photo-growth (10). However, systems of those typically require/involve mass change via, for instance, additional monomer addition or water swelling change. Spatio-selective material change via the nonisomerization mechanism unavoidably leads to uneven stress, which would alter the macroscopic geometric shape. In contrast, our isomerization mechanism allows tuning the material properties within a fully enclosed system. This mechanism is particularly suited for a dry solid system as ours since it forgoes the need for external mass transport. The permanent mainframe of our network carries any stress the material may encounter, and the isomerization exclusively occurs through the dynamic exchange of the non–load-bearing side chains. This would allow spatial patterning of material properties without affecting the permanent geometric shape, regardless of the external stress condition. This unique feature provides new opportunities for advanced manufacturing. Stretchable electronics, for instance, typically requires a substrate stiffness pattern to protect the strain-sensitive electronic components while maintaining its overall stretchability (38). With our material system, the stiffness pattern can be introduced during the manufacturing process in which an external stress is difficult to avoid. This is a direction we will pursue in the near future.


In summary, we illustrate a molecular design principle for dynamic covalent polymer network that allows topological isomerization, leading to the capability to program the (thermo-)mechanical properties. This stands in sharp contrast to typical dynamic covalent polymer networks for which the topology is nonprogrammable. The solid-state isomerization, the light-triggering mechanism, and the multiple topological states further allow spatiotemporal topological patterning. Consequently, a single network polymer can be programmed into an unlimited number of polymers. A wide variety of materials with similar characteristics can be designed by exploring the rich library of dynamic covalent chemistry (1427). Beyond the grafting to brush topological shifting, many other opportunities exist. With proper network design, isomerization among many topologies (e.g., block, gradient, and random) is possible. Alternative design may also include dynamic network mainframe, extending beyond the permanent mainframe for the current system. The wide design space allows access to different programmable materials to meet various demands. The conceptual device applications in Fig. 4 represent the tip of an iceberg, as the practical benefit has yet to be fully explored in many other technological areas such as artificial muscle, flexible electronics, and three-dimensional printing.



ε-Caprolactone (ε-CL; TCI), trimethylolpropane diallyl ether (90%; Sigma-Aldrich), tin (II) 2-ethylhexanoate (TCI), 2,2-dimethoxy-2-phenylacetophenone (photoinitiator; TCI), 3,6-dioxa-1,8-octanedithiol (TCI), 1,6-hexanedithiol (TCI), 1,5,7-triazabicyclo [4.4.0] dec-5-ene (TBD; TCI), ketoprofen (TCI), and 1,2-ethanedithiol (TCI) were used in this study. ε-CL was purified by distillation before use. All other chemicals were used as received.

Synthesis of diallyl-functionalized polycaprolactone

Diallyl-functionalized polycaprolactone was synthesized by ring-opening polymerization. ε-CL (40 g), trimethylolpropane diallyl ether (1.712 g), and tin (II) 2-ethylhexanoate (0.2 g) were added into a Hiddink flask. The mixture was kept at 120°C for 12 hours under argon. The obtained polymer was dissolved in toluene and precipitated in cold hexane, followed by overnight vacuum-drying. The white powdery product was dissolved in toluene and reacted with 1,2-ethanedithiol (molar ratio of 1:2). After reaction for 15 min at room temperature under stirring, the precipitant was removed by filtration. A solid product was obtained by precipitation in cold hexane and filtration. The polymer was washed by cold hexane and methanol twice. The final polymer product was vacuum-dried overnight. The molecular weight of the polymer was calculated by NMR end group analysis (fig. S10).

Synthesis of the precursor polymer

Diallyl-functionalized polycaprolactone (0.35 g), trimethylolpropane diallyl ether (0.214 g), 1,6-hexanedithiol (0.18 g), and photoinitiator (0.004 g) were added into a glass vessel. The mixture melted at 80°C was exposed to ultraviolet (UV) light for 180 s. The structure of the polymer precursor was confirmed by 1H NMR (see fig. S1).

Synthesis of the polymer network

The polymer network was synthesized by thiol-allyl addition in two steps. First, diallyl-functionalized polycaprolactone (0.35 g), trimethylolpropane diallyl ether (0.214 g), 3,6-dioxa-1,8-octanedithiol (0.23 g), and photoinitiator (0.004 g) were added into a glass vessel. The melted mixture was exposed to UV light for 180 s. 1,1,2,2-Tetrakis(allyloxy)ethane (0.023 g) and photoinitiator (0.0038 g) were added to the sticky precursor. After degassing under vacuum, the homogeneous mixture was quickly sandwiched into two glass slides with a 0.4-mm spacer and exposed to the UV light for 180 s. The film was kept at an 80°C oven overnight to remove the residual solvent.

Synthesis of the photo-latent catalyst and characterization of its UV-triggered release

TBD (1.39 g) and ketoprofen (2.59 g; molar ratio of 1:1.02) were dissolved in toluene (15.9 g). The transparent solution was stored away from light. For the release reaction, the solid photo-latent catalyst was obtained by drying at 80°C and redissolved in deuterated chloroform (25%). The solution was exposed to the UV light (365 nm; TaoYuan light emitting diode; intensity, 30 mW/cm2) at 80°C at a distance of 16 cm. The release kinetics was monitored by 1H NMR (fig. S5).

Topological patterning

The film was swollen in the 1.5 weight % photo-latent catalyst solution for 30 min, and the solvent was removed at 80°C oven for 30 min. The sample was exposed to masked UV light at 80°C at a distance of 16 cm, followed by heating at 80°C for 2 hours with no UV irradiation to complete the topology isomerization.

Fabrication of bistable hinge structures

Topologically patterned samples were laser cut. The samples were then uniaxially stretched at 80°C and cooled down to 20°C (maintained for 15 min) to fix the temporary shapes.

Materials characterization

Moduli were measured at 20°C with samples in a dog bone shape (11.07 mm by 1.44 mm by 0.35 mm) using a dynamic mechanical analyzer (TA Q800). Melting temperatures and crystallinities of samples were measured using a differential scanning calorimeter (TA Q200). Shape memory characterization was conducted by monitoring the sample length change in uniaxial stretching experiments. The deformation temperature was 80°C, and the fixing temperature was 20°C (15 min).

Finite element simulation

A commercial finite element package ABAQUS was used to simulate the relaxation behavior of the sheets upon stretching, cured with different topological patterns. Since the stiff phase formed in the curing process is stress-free, the main driving force of relaxation comes from the mechanical strain stored in the soft phase, which is spatially patterned. In the simulations, the effects of stretched-induced tension in the soft phase were represented by an equivalent eigenstrain that tends to shrink in the stretching direction. The magnitude of the shrinking eigenstrain was selected to be the same as the preimposed strain. All the mechanical properties of the polymers (soft and stiff phases) used in the numerical simulations were obtained from experimental measurements (table S1). The detailed procedure of the finite element simulations is as follows. A shell structure was created in ABAQUS to reproduce the actual geometry of the polymer sheets after stretching. The actual patterning of soft and stiff phases was implemented in the geometry by partitioning the shell into different regions. Different material properties, obtained from experiments (table S1), were assigned to the corresponding regions. The standard 4-node shell element (S4) was used in the simulations, and mesh refinements were carried out to ensure the convergence of the computational results. A linear perturbation analysis was conducted on the reference state with a superposition of the first 10 buckling modes, yielding the combination of possible configurations. The Riks method was then used to realize the cooperative deformation that occurred in different phases of the patterned composite sheets, which are in agreement with the experimental counterparts.


Supplementary material for this article is available at

Fig. S1. The 1H NMR spectrum of the thiol-terminated precursor polymer.

Fig. S2. More complete 1H NMR spectra of the model compound study corresponding to Fig. 2B in the main text.

Fig. S3. 1H NMR spectra of the photo-latent catalyst after UV irradiation for different time length.

Fig. S4. Dynamic mechanical analysis curves of topologically isomerized sample with different irradiation time.

Fig. S5. Stress-strain curves of topologically isomerized samples with different irradiation time.

Fig. S6. Characterization of the shape memory behaviors of the topologically isomerized network.

Fig. S7. Interconvertible bistable-hinged structures.

Fig. S8. Mechanical simulation of the adaptive robotic arm.

Fig. S9. Stretching-induced three-dimensional structures.

Fig. S10. 1H NMR spectrum of diallyl-functionalized polycaprolactone.

Table S1. The mechanical properties used in simulation.

Movie S1. Adaptive robotic arm (accelerated by 72).

Movie S2. Fracture resistant device (accelerated by 8).

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank L. Xu (State Key Laboratory of Chemical Engineering, Zhejiang University) for assistance in performing differential scanning calorimetry analyses. Funding: We thank the National Natural Science Foundation (21625402, 51822307, and 51673169) for financial support. Author contributions: T.X. conceived the concept and wrote the paper. W.Z. and B.J. designed and conducted the experiments. Y.W. and J.Q. performed the finite element stimulations. T.X. and Q.Z. supervised the project. All authors participated in the discussion of the results. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate to the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article