Living systems are a source of inspiration for man-made robotics1-3 with regard to flexibility, scalability, and resilience. Recent progress in micro-3D printing and synthetic biology4 raises expectations that a bottom-up design of nano- to microscale biorobots5,6 directly from the level of biomolecules may become reality. While the design and fabrication of soft biomimetic robots has become increasingly successful over the past decades, a major remaining challenge for potential use in or on living systems is their biocompatible actuation. Cellular motor proteins7-11, which directly turn metabolic energy into mechanical work, represent promising candidates from nature to execute mechanical operations on soft materials12. In contrast to electromechanical and biohybrid actuators13-17, which require considerable efforts for downscaling to the nano/microscale3, a key challenge for utilizing molecular motors is to upscale force and work to operate devices that are many orders of magnitude larger18-21. Efficient large-scale use of such protein motor systems for actuation of soft robots has so far been mainly realized through cyborg constructions hybridizing living muscle cells and tissue with soft polymer materials such as silicone13,14,22. Recently, a large-scale kinesin motor-based network has been successfully engineered to actuate established soft microstructures23; the actuators were, however, untethered and had to be recruited specifically from the environment. Coupling the protein motors to soft materials has still been complicated by the challenges of producing biomimetic templates that are both sufficiently malleable and physiologically compatible with the large-scale operation of motor protein systems.
We here introduce shape-morphing protein-based robotic structures composed of programmable 3D printed modular units24,25 (Fig. 1a), which can be actuated by a minimal artificial actomyosin exoskeleton. This motorized exoskeleton can be viewed as a scaled-up actomyosin cortical layer26 connecting the different parts of the deformable robotic structure (Fig. 1b). Our exoskeleton design harnesses the contractility generated by myosin molecular motors and transduces the resulting active stresses to perform large-scale mechanical work on the 3D soft device, without the need of hierarchically structured motor assemblies as in sarcomeres. To characterize force production by the contractions of actomyosin layers and the resulting deformation of soft frames, we designed contractile pillar ring modules of different sizes and stiffnesses. The simple geometry of this design allows us to develop a theoretical model to conceptually understand contraction dynamics and demonstrate stiffness-insensitive force generation. We then apply our approach to more complex shape-morphing structures, by combining basic modules, such as pillars, panels, and hinges, thereby mimicking key elements of human hands (Fig. 1c). Finally, we show that the exoskeleton design can be extended and scaled up to realize complex programmable shape transformations of active mechanical devices, and further achieve multi-stage functions of soft robotic elements with (selectively) light-activatable modules (Fig. 1d).
To quantitatively understand the basic performance of our contractile motorized exoskeleton with regard to force generation and transduction in shape-morphing of 3D protein-based modules, we first designed a simple structure consisting of a ring of ten soft pillars (Fig. 2a, Extended Data Fig. 1a-b). The hydrogel pillar units are printed by two-photon polymerization of bio-resin consisting of bovine serum albumin (BSA) and rose bengal as photoinitiator. The Young’s modulus of the protein hydrogel can be tuned from 10 kPa to 250 kPa (Extended Data Fig. 2), corresponding to moduli of biological tissues27,28. Subsequently, the ring structure is biotinylated and decorated by an F-actin meshwork via biotin-actin and neutravidin coupling. The actin filaments are further cross-linked by neutravidin to enhance the network’s mechanical integrity (Fig. 2b, Extended Data Fig. 3-4). To complete the exoskeleton-like minimal actomyosin cortex, pre-assembled myofilaments with skeletal muscle myosin II were included. Upon ATP addition these myosin motors generate contractile forces26,29 that are transduced by the exoskeleton resulting in large-scale active stresses that drive the inward deflection of the pillar ring (Fig. 2, Movie S1).
Next, we investigated how the contractile performance of the 3D structures depends on their dimensions by varying the diameter D of the pillar ring. The diameter can be enlarged by increasing the interpillar distance d and maintaining a constant pillar number N (N=10), or by increasing N and keeping d constant (d=5 µm) (Fig. 2c-d, Extended Data Fig. 5). We noticed that the deflections of pillars towards the centre were determined by the density and integrity of the actin mesh located inside the ring (Fig. 2c-d). Rings with larger diameters develop holes in the actomyosin meshwork, resulting in asymmetric and incoherent contraction of the ring as a whole (Fig. 2d). This effect is exacerbated by larger interpillar distance (Fig. 2d left, Extended Data Fig. 5c-e). In contrast, increasing the pillar numbers at similar diameters (D=26 µm) results in significantly higher network density (Fig. 2c), better preserving the integrity of the networks. However, isotropic contractions are rarely observed for pillar rings with diameters ≥30 µm (Fig. 2d). These insights lead us to a key design feature with regard to the scale over which the exoskeleton scaffold can achieve effective and coherent contractions, suggesting that concatenating smaller force-transducing modules can be a good option to engineer larger device designs.
Another design feature for robotics engineering is to combine modules with different mechanical properties and programmable macroscopic deformations30. To explore the effect of elasticity on the contraction capacity of the actomyosin exoskeleton, we varied the stiffness of the pillars in the pillar-ring test assay, while keeping the density of the meshwork constant (Fig. 2e-f, Extended Data Fig. 6). We found that the pillar deformations can be tuned by varying the material stiffnesses. The largest pillar deflection was observed in rings with a Young’s modulus of 57±12 kPa (Extended Data Fig. 2f). In this case, the contractile exoskeleton is able to reduce its radial dimension by up to a factor 0.5, resulting in large pillar strains up to 0.3 (Fig. 2e). Importantly, the pillar ring design allows us to quantify the forces generated by the active contraction. By comparing the measured pillar deflection profiles with Euler beam theory31 (Fig. 2f, SI-Sec. 1), we measure active forces generated by the actomyosin exoskeleton of 126±22 pN (n=12) per pillar upon full contraction. Individual myosin filaments under these conditions can generate forces close to 20-60 pN26,32,33, suggesting that only a small fraction of the estimated (mean ± s.d.; n=9) myofilaments (Supplementary Fig. S1) in the exoskeleton effectively contribute to the contraction (SI-Sec. 2). However, the force generation in the system can be tuned by varying the density of the actin network coated on to the 3D scaffolds, by either changing the amount of anchor points (Extended Data Fig. 3c-d, Extended Data Fig. 7c-d) or by varying crosslinker concentrations (Extended Data Fig. 4e-g). For a given actin density, however, the force generation is robust over a wide range of pillar stiffnesses (Fig. 2g). The insensitivity of the contractile forces to the frame stiffness facilitates the predictability of force generation in more complex structures, which consist of modules with different stiffnesses.
To understand the build-up of these generated forces, we next investigated the contraction dynamics of actomyosin networks on the microstructures. Upon myosin activation, the pillars start to bend inwards (Fig. 2h). The contraction velocity initially increases markedly over the course of approximately one minute, followed by a decelerating contraction towards a final state (Fig. 2i). The maximal velocity that is reached during the contraction depends sensitively on pillar stiffness (Fig. 2j, Extended Data Fig. 7, Movie S1). We observed a clear increase in peak velocity with decreasing pillar stiffness below a value of =100 pN/μm. Interestingly, the acceleration lasts much longer than the dynamics of single or spatiotemporally coordinated myosin motors in sarcomeres, which operate on the timescale of milliseconds34.
To elucidate the mechanisms that are underlying the observed robust force generation and the dynamics of the contraction, we describe our contractile system by a simple, one-dimensional analytical model. In this viscoelastic model the pillar’s and the network’s elastic response is represented by elastic springs and the viscous response of the network is characterized by a dashpot. The myosin activity is modelled by a time-dependent contractile force, acting on the pillars (Fig. 2a, details in SI). To capture the contraction dynamics, we include a microscopic description of force generation by the molecular motors: myosin filaments transiently bind to the F-actin network where they contribute to contractile force generation. These myosin binding dynamics, and the resulting force generation, thus depend on the number of myosin filaments and the density of actin in the gel35. Finally, we find that it is essential to account for the known load-dependence of the myosin binding dynamics36 (SI-Sec. 3.4). As the force builds up, the load-dependent myosin kinetics results in an increased number of motors being engaged in force generation.
The inherent positive feedback between the slow build-up of viscoelastic network stresses and the active force generation by fast load-dependent myosin binding kinetics in our model gives rise to the intricate contraction dynamics of the pillar-ring, in quantitative agreement with our experiments (Fig. 2i). When possible, the parameters of our dynamic 1D contractility model are chosen based on literature values (see table S1). The remaining parameters are fully constrained by fitting the model to a single contraction curve at one pillar stiffness (Fig. 2i). This model accurately predicts the dynamics and steady-state values of the actively generated forces over broad range of pillar stiffness (steady state: Fig. 2e-g, Extended Data Fig. 6), including the stiffness-dependence of the contraction velocities (Fig. 2j). The stiffness-insensitivity of the steady-state force can be explained by our contractility model (Fig. 2g, solid line), provided that the network’s elastic response is much softer than the pillar stiffness and can thus be neglected (SI-Sec. 3.1). Conceptually, the load-sensitivity of the myosin binding kinetics results in active force generation, which is largely controlled by the internal stress of the actin network and is insensitive to the stiffness of the frame. In contrast, when only accounting for an actin density-dependent contractility as in other models35, the stiffness dependence was predicted incorrectly (Fig. 2g, dash line, SI-Sec. 3.3). Finally, using our 1D contractility model together with the experiments, we can also estimate that generated mechanical power of the exoskeleton in the pillar-ring assay peaks for 43x10-18 W (SI-Sec. 3.5).
Having gained conceptual understanding of the force generation and contraction dynamics of the actomyosin exoskeleton on soft-frames with simple geometries, we next turn to applications involving complex 3D structures with programmable transformations. The central idea is to assemble 3D structures from modules with tunable stiffnesses to perform controllable deformations. A key structural element to advance complexity to the next level is a hinge module. Thus, we designed a V-shaped hinge with two stiff arms (pillars) connected by a soft joint (thickness ratio arm/joint: 4:1) and a stabilizing apex to inhibit overstretching to angles beyond 180o (Fig. 3a, Extended Data Fig. 1c). Upon myosin activation, the hinge is actuated, i.e., the free arm rotates about the soft joint towards the fixed arm, which follows the dynamic behaviours of pillar rings and exhibits with an initial acceleration phase and a deceleration phase (Movie S2, Extended Data Fig. 8). The active closure of V-units can be accomplished for a range of convex angles, e.g. from 45o to 160o (Fig. 3b), and can be tuned by the fabrication parameters (Extended Data Fig. 8).
Further, to explore the possibilities for reversible operation of our protein hydrogel devices, we designed an alternative round joint V-unit with homogeneous elasticity (Supplementary Fig. S2). Since the activity of myosin motors is sensitive to the ATP concentration, the reversible shape transformation can be implemented by switching between high and low ATP conditions (Fig. 3c, Movie S3). Low concentration ATP (0.5 mM) can initiate the active closure of the hinge, as demonstrated above. In contrast, high concentrations of ATP (4 mM) cause myofilaments to detach from the actin exoskeleton, resulting in a relaxation of the elastic hydrogel structures. We successfully performed two repeats of contraction by manually exchanging the ATP concentrations (Fig. 3c). After the second iteration, the structures started to lose their reversibility in response to high ATP concentration. The robust reversibility is partially hampered by the non-reversible breakage and crosslinking of actin networks26,29. We also noticed softer structures showed larger displacements, but exhibit lower reversibility, due to larger energy dissipation37. Thus, structures with different rigidity exhibit an opposite trend between displacement and reversibility (Fig. 3d). To engineer a reversible shape change, we seek an optimal tradeoff between these factors.
Scaling up the mechanical work performed by our actomyosin-actuated protein-based robots can now be achieved by concatenating active modules to engineer larger structures. For example, by concatenating the V-units, large modular architectures can be assembled, as illustrated in Fig. 4a. Specifically, V-units were combined into a zigzag module and the connection sites between V-units were stabilized to permit folding only on the bottom joints. The actomyosin actuators then independently trigger the closure of all V-units, resulting in a rapid curling of the zigzag module (Movie S4). Instead of using one large hinge angle, 90o and 135o angle folding can be accomplished by combining two or three 45o V-units, respectively (Extended Data Fig. 9a). Similarly, large angles that are impossible to achieve with a single unit, such as 180o and 360o, can also be successfully accomplished by concatenating modules. The spring-like zigzag module combined with six V-units could revolve around the first joint on the right, circularly coiling up its long V-chain and finally forming a closed hexagonal star (Fig. 4b). Besides the curling-up, a hydrogel spring-shape structure with additional degrees of freedom can contract along its axis upon myosin-triggered actuation (Extended Data Fig. 9b-e). These examples demonstrate the versatility and potential upscaling of programable mechanical operations that can be achieved in our protein-based active robotics.
Another desired feature of active biomimetic devices is the ability to fold 2D surfaces into complex 3D shapes. To accomplish this, we expanded our approach to generate a more intricate shape-morphing, such as the self-folding of a hydrogel cube from a cruciform precursor consisting of six hinged panels (Fig. 4c-d, Extended Data Fig. 1d). The solid-supported central face was connected to the other five free-standing faces with inner hinges. An extra rigid block was placed between the two hinging faces to serve as a sill for controlling the folding angle. The flat panels and the rigid block formed a 90o angle, which enabled a simple 2D self-folding sheet to yield a 3D cube under active contraction (Movie S5).
To further extend the functional complexity of our protein-based robotics, we created programmable and reversible microscale robots with light-induced spatiotemporal control. First, we employed the bio-actuated 3D protein hydrogel to mimic grasping micro-hands consisting of a panel with five attached fingers (Fig. 4e). The connections between phalanges and the palm were mediated with soft joints. In our designs, the actomyosin exoskeleton functions as an external “muscle” layer to actuate the five fingers with respect to the palm into a grasping shape (Extended Data Fig. 9f, Movie S6). By selectively constraining the finger joints kinematically with triangular-shaped blocker modules (wedges), we programmed anthropomorphic gestures, like “OK” (Extended Data Fig. 9f). Sign language that is even more complicated can be executed successfully within minutes, such as I Love You (ILY) (Fig. 4f).
Finally, we combined the micro-hand and the chain of V-units into a miniature robotic arm - fully made from and actuated by proteins (Fig. 4g-h, Movie S7). Initiating the contraction of the external actomyosin exoskeleton triggered the arm into a raising and grasping motion. As a first proof of concept of reversible operation, the artificial arm was subjected to a high-ATP condition (4mM ATP) and responded with a twitch (Movie S7). We next explored how artificial arm motions could be designed as goal-directed behaviors subject to multistage control. Using photocaged (NPE-caged) ATP as molecular light sensor, the ATP-dependent actuation module, i.e. the actomyosin exoskeleton on the surface of the robotic arm, can be remotely controlled with light (Fig. 4i, Extended Data Fig. 10). Through spatiotemporally targeted release of ATP by illumination with a focused 405 nm diode laser, stepwise arm and hand movements can be guided by light stimuli within one minute (Fig. 4i, Movie S8). Due to the photo damage, the light activation can only be used for single-run applications. Interestingly, as expected based on our model, the complex structure activated by light and preloaded motors can respond more rapidly than by recruiting motors and energy out of solutions, suggesting a new avenue for improving the performance of our system in the future.
To conclude, we demonstrate the ability of 3D printing complex protein-based microrobotics functionalized with a minimal actomyosin exoskeleton as its actuating system. Hereby, the efficient and scalable operation of reconstituted biological motor assemblies as contractile layers enabled the large-scale shape-morphing of complex 3D microstructures by converting chemical energy directly into mechanical work. Furthermore, being designed and assembled entirely from biomolecules in a bottom-up fashion, the self-powered soft robotic system constitutes an excellent starting point as a chassis. The performance in terms of speed, force, and reversibility, may in the future be advanced by operating under automated microfluidics2, by optimizing actomyosin network composition38 with natural crosslinking proteins39, and by integrating other biological or biomimetic modules11,40, such as actin recycling systems41. We consider this an exciting step in engineering and programming arbitrarily shaped bio-motor-based actuators for future soft robotics1, which no longer rely on the swelling and shrinking of materials. In the framework of the bottom-up assembly of life-like systems, as pursued by us and others1,4,13, the technology we developed here opens new vistas for various applications, such as custom-shaped dynamic sensors and bioassays, microrobots for in vivo biomedical tasks, and proto-tissue engineering, when introduced to the life sciences42,43.