A New Drive System for Microagent Control in Targeted Therapy Based on Rotating Gradient Magnetic Fields

Using magnetically powered microagents as precise delivery carriers is a promising technology for targeted therapy. It is a significant challenge to converge microagents to the desired position without sufficient support of real‐time imaging feedback. Herein, a new drive solution is proposed, which generates a rotating gradient‐based magnetic field by sequentially energizing each coil of the electromagnetic coil system. With this method, a high magnetic field gradient concentrated onto a specific target site is generated, which will attract microagent swarms to automatically converge to the target site from different directions. Based on the established model of rotating gradient magnetic field actuation, the relationship between the current inputs of coils and the location of the aggregation center is characterized, so that the target site can be adjusted in the entire workspace by changing the coil current inputs. The method drives microagents to rotate while moving forward, thereby reducing the viscous resistance and friction on microagents. It does not rely on specific trajectory planning and real‐time visual guidance to navigate microagents, and can drive magnetic microagents with different characteristics such as size, shape, and materials. This research provides a basis for using microagent delivery in clinical applications and precision‐targeted therapy.

DOI: 10.1002/aisy.202100214 Using magnetically powered microagents as precise delivery carriers is a promising technology for targeted therapy. It is a significant challenge to converge microagents to the desired position without sufficient support of realtime imaging feedback. Herein, a new drive solution is proposed, which generates a rotating gradient-based magnetic field by sequentially energizing each coil of the electromagnetic coil system. With this method, a high magnetic field gradient concentrated onto a specific target site is generated, which will attract microagent swarms to automatically converge to the target site from different directions. Based on the established model of rotating gradient magnetic field actuation, the relationship between the current inputs of coils and the location of the aggregation center is characterized, so that the target site can be adjusted in the entire workspace by changing the coil current inputs. The method drives microagents to rotate while moving forward, thereby reducing the viscous resistance and friction on microagents. It does not rely on specific trajectory planning and real-time visual guidance to navigate microagents, and can drive magnetic microagents with different characteristics such as size, shape, and materials. This research provides a basis for using microagent delivery in clinical applications and precision-targeted therapy. delivery in the in vivo environment, [36] especially in small and complex regions, such as tiny cavities or tortuous ducts across the blood circulation system. [37,38] This paper presents a new method of using a rotating gradientbased magnetic field to transport a microagent swarm to the target site precisely. The rotating gradient magnetic field can be generated by sequentially energizing each magnetic coil of a gradient-based electromagnetic coil system. [28] In comparison with traditional gradient magnetic field drivers, the rotation of microagents will reduce the viscous resistance and friction of the surrounding environment with microagents, thereby greatly enhancing the mobility of each microagent. [39] The shape design of the microagent is unrestricted, considering that the gradient magnetic field remains as the main driving force. Importantly, the rotating gradient field will produce a high magnetic field gradient concentrated onto the target site, which will attract microagent swarms to automatically converge to the target site from different directions under the action of this equivalent centripetal force. Thus, a specific trajectory plan and real-time visual guidance for each microagent navigation are not required under this paradigm.
This research is performed in the following three perspectives. First, by adjusting the rotating frequency of the magnetic field, different types of microagents are driven to converge to the produced aggregation center or scatter at different rates. Considering that the aggregation does not depend on the mutual attraction of the microagents, there is no specific requirement on the distribution density of the microagents. A numerical model of rotating gradient magnetic field actuation is established, based on which the location of the microagent aggregation can be determined by adjusting the input current of each coil. The relationship between the current input and the location of the aggregation center is characterized to ensure that the microagents are controllable in the entire workspace. Second, the driving ability of the rotating gradient magnetic field to transport the microagents to the simulated blood flow environment is investigated. Given that the proposed driving mechanism does not depend on the agent-agent interaction, the aggregation does not depend on the specific characteristics of the microagents, such as size, shape, and materials. Experiments performed in the microfluidic channel have confirmed that when the aggregation area is created, the majority of microagents passing through the area are firmly attracted by the magnetic field. Third, an ex vivo test is conducted in the bovine eyeball to demonstrate the effectiveness of the rotating gradient magnetic field in driving microagents to the target site. No motion planning is needed for each microagent, and no guidance is required for real-time imaging. The experimental results show that the proposed method can enrich microagents in complex environments with good performance. Figure 1 shows the diagram of the proposed rotating gradient magnetic field actuation. For simplicity, a custom-designed gradient-based magnetic system with four orthogonal electromagnetic coils was considered in this research. As presented in Figure 1A, when the current passed through the magnetic coil, a static gradient magnetic field with a donut-shaped distribution was generated. By sequentially inputting DC to each coil to generate a rotating magnetic field, a swarm of magnetic microagents can be excited to accumulate at the center of the resulting concentration, which is called the aggregation center in this study. The position of the aggregation center can be adjusted by changing the input current of the coil. Figure 1B,C illustrates two scenarios of driving microagents in the blood vessels and the eyeball, respectively. The magnetic drive system mainly consists of four magnetic coils to generate the rotating gradient magnetic field and a platform containing microagents in a bio-environment located in the workspace. The height (h) of the platform with respect to magnetic coils can be adjusted. In this research, an open custommade chamber, a microfluidic chip, and a bovine eyeball were used to provide a liquid bio-environment for microagents. The microagents used included polyethylene glycol-coated (PEGylated) monodisperse magnetic microspheres (diameter: 10 nm), spherical-shaped hematite microparticles (diameter: 1 μm), fluorescein superparamagnetic iron oxide microparticles (diameter: 0.5 μm), burr-like porous spherical microrobots (diameter: 80 μm), and spherical-shaped microrobots (diameter: 300 μm). All these microagents are nonmagnetic in the absence Figure 1. Schematic of the gradient-based rotating magnetic field actuation. A) Microagents rotated by sequentially energizing each gradient-based electromagnetic coil. B) Microagents in blood vessels. C) Microagents on the surface of the retina of the eyeball. D) Experimental setup, including i) the magnetic coils and lifting platform, ii) the microfluidic chip, iii) the eye model used in the experiment, iv) microparticles captured in the microfluidic chip, and v) microparticles captured gathered in the retina of the eyeball.

Model of Rotating Gradient Magnetic Field Actuation
www.advancedsciencenews.com www.advintellsyst.com of external excitation. The motion of microagents in the liquid environment can be observed by the optical microscope with a charge-coupled device (CCD) camera or optical coherence tomography (OCT). The Reynolds (Re) number of the microagent was first determined to estimate the interrelationship between inertial and viscosity excitation effects. The Re number can be calculated as Re ¼ ρVL/μ, where ρ and μ are the density and viscosity of the fluid, respectively, and V and L are the speed and characteristic length of the microagent, respectively. For all microagents used in this study, their Re numbers were calculated to be around 5 Â 10 À[4] to 8 Â 10 À , [2] which is much less than 1. [7] The model of an individual microagent driven by the rotating gradient magnetic field in low-Reynolds number regimes (Re << 1) was established as follows: let μ 0 ¼ 4π Â 10 À7 Tm=A denote the permeability of a free space, m 0 denote the point dipole moment, E denote the identity matrix, r denote the position of the microagent in the field, and p denote the magnetized vector of point dipole starting from the center of workspace to the point dipole of coil. Under unit electric current, the magnetic field unit flux density B unit ðr 0 Þ of a single magnetic coil can be described as [36] B unit ðr 0 Þ ¼ μ 0 4π 3ðm 0 · r 0 Þr 0 kr 0 k 5 À m 0 kr 0 k 3 (1) where r 0 ¼ Àr þ p, representing the vector connecting the position of the microagent and the point dipole of the coil. Define a horizontal plane that includes the point dipole of coil as the reference plane (or zero-plane) and a plane higher than the reference plane by a distance of h as the h-plane. For a single magnetic coil, the flux density B unit ðr, 0Þ in the h-plane (where h ¼ 0) can be simplified as B unit ðr, 0Þ ¼ μ 0 m 0 =ð2πkÀ r þ pk 3 Þ. [36] Then, the unit flux density B unit ðr, hÞ in the h-plane can be derived by extending as follows The simulation results of B unit ðr, hÞ generated by a single magnetic coil are shown in Figure S1 Supporting Information. On the basis of Equation (2), the flux density Bðr, hÞ of the rotating gradient magnetic field is expressed by where B i ðr, hÞ is the flux density generated by the ith magnetic coil, I i is the current input of the ith coil, n is the number of coils, and R i is an orientation matrix of the ith coil expressed as Using Equation (3), the magnetic driving force F mag can be estimated as where m ¼ V c χ=μ 0 ð1 þ χÞ, representing the magnetic moment of the microagent, χ is the magnetic susceptibility of the material, and V c is the volume of the magnetic material contained in the microagent. The Nabla operator ∇ denotes the gradient of field. Given that the microagent always moves in a quasi-equilibrium state in a low Re number regime, the resistance force of the liquid, denoted by drag force F drag , is equal to F mag , namely, where R is the equivalent radius of the microagent, η is the dynamic viscosity of the fluid, and v is the velocity of the microagent. Then, the position of the microagent at time t, denoted by rðtÞ, can be derived as As time t ! ∞, rð∞Þ represents an equilibrium position to which the microagent converges, which is defined as the aggregation center, denoted by g(x, y), where x and y are coordinates of the working plane. The point dipole moment m 0 and the magnetic moment m can be solved by Genetic Algorithm (Python, Scikit-Opt solver). As a function of rðtÞ, p can be solved on the basis of a previous study. [36] Using rðtÞ in Equation (5), the velocity v of the microagent can be estimated; therefore, the drag force F drag can be determined. Consequently, magnetic force F mag , which is equal to F drag , can be estimated.
Denoting I as the current from power source, the input current of the ith coil is expressed as where f denotes the rotating frequency of the magnetic field. In real-time applications, parameter c i can be introduced to I i to adjust the current of the ith coil. In this way, the position of the aggregation center can be changed. The characterization of the relationship between the c i value and the location of the aggregation center is the key to ensure that the microagent is controllable in the entire workspace. The following rules apply when adjusting c i to determine the position of the aggregation center. First, c i for the magnetic coils on the same axis cannot be adjusted simultaneously. That is, when adjusting the c 1 of the first coil, the c 3 of the third coil in the opposite direction should keep the base value of 1, and c 2 and c 4 should be the same. Second, the current change is always conducted incrementally, which means that c i is not less than 1 when c i starts to change from a base value of 1. Coil pairs that change the coil current have four types, which are coils 1-2, 2-3, 3-4, and 4-1. c ¼ (c 1 , c 2 , c 3 , c 4 ) is considered as the parameter group for four coils. When c ¼ (1, 1, 1, 1), the aggregation center is located at the physical center of the workspace, namely, (x, y) ¼ (0, 0). By changing c i , the position of the new aggregation center can be calculated on the basis of Equation (5). Then, the reverse mapping relationship from g(x, y) to c can be established using the back propagation neural network (BPNN) model. Here, the BPNN model contains three layers, namely, the input layer g(x,y), the output layer c, and the hidden layer. The number of hidden layer units is 10. The rectified linear unit and mean squared error functions are used as the activation function and loss function, respectively. When c i is increased from 1 to 2 at a step of 0.01, 100 Â 100 data will be used to train the BPNN model with Python, and the reduced-state sequence estimation (RSSE) method will be used as the evaluation function. When the RSSE is less than 0.001 after 10 000 iterations, convergence is reached, and then the training model is established. In this way, the position of any desired aggregation center can be determined by setting the corresponding c. As an example, Table S1, Supporting Information shows the mapping relationship between the aggregation center position g(x,y) and c when the workspace of 15 Â 15 mm is divided into 15 Â 15 unit square areas. The unit square area with a length of 1 mm is accurate enough for most clinical treatments (e.g., tumor therapy). Further increasing the number of unit square areas can improve the resolution according to actual needs. Figure 2 illustrates the convergence control of a single microagent in the rotating magnetic field. Figure 2A shows the diagram of the magnetic system with a reference plane and a working plane. The reference plane is a horizontal plane where the point dipoles of four magnetic coils are located. The distance between the two planes is denoted by h. The initial parameters of the magnetic system are set to h ¼ 20 mm,

Convergence Control of a Single Microagent
, and f ¼ 20 Hz. The parameters of the microagent and its surrounding liquid environment are based on our previous studies. [7,28] The entire workspace can be divided into subregions, and the physical center of each subregion is treated as a possible aggregation center. Figure 2B presents a square workspace of 15 Â 15 mm in the working plane, which is divided into 15 Â 15 unit square areas. Figure 2C shows the simulation results of the converging trajectory of a single microagent from four directions. The microagent is initially located at (0, 3), (3, 0), (0, À3), and (À3, 0), and then converges to the center of the workspace, where The trajectory of the microagent is expressed with a red dotted line, and the central line of each trajectory is expressed with a black solid line. Given that the actual motion of the microagent is an orbital revolution around the centerline, it can be treated as a movement along the centerline. It is seen that the microagent converges to the center from four different positions successfully, showing that the rotating magnetic field can produce an equivalent centripetal force to attract the microagent to converge to the target position. Figure 2D displays the field distributions of the estimated magnetic force F mag with two different aggregation centers, which are calculated on the basis of Equation (5). The black arrow represents the direction of the magnetic force, and the contour maps and color distribution represent the value of the estimated magnetic force. According to the Poincaré-Brouwer theorem, there is no nonvanishing continuous tangent vector field on an even-dimensional n-sphere. Since the estimated force in Figure 2D is in an even-dimensional 2D space, zero-velocity points appear in the image. [40] Figure 2E shows the motion trajectory of the microagent moving from four directions to different aggregation centers under different parameter c ¼ (c 1 , c 2 , c 3 , c 4 ). The experimental results of the single microagent convergence are also shown in Figure S2, Supporting Information.
Equation (3) shows that the flux density Bðr, hÞ of the rotating gradient magnetic field is closely related to the h of the working plane. Figure 2F illustrates the motion trajectory of the microagent in the planes with different h ¼ 0, 10, 20, and 30 mm, calculated on the basis of Equation (5). In the planes of h ¼ 0 mm and h ¼ 10 mm, the microagent moves near the boundary of the working space, and the aggregation effect is not so obvious. In the planes of h ¼ 20 mm and h ¼ 30 mm, the microagent moves to the central position, indicating that the microagent can be attracted by the rotating magnetic field in these two planes. It is worth noting that when the distance of h is greater than a threshold, the microagent can aggregate to the desired aggregation center within the work area. The effect of the initial position p will be eliminated. Aggregation conditions (h > threshold) can be easily achieved by designing a lifting platform to adjust the height of the experimental setup. In the rest of this paper, unless otherwise noted, h of the working plane is set to 20 mm. Figure 3 illustrates the experimental results of microagent swarm aggregation (Movie S1, Supporting Information). The magnetic microagents used in these experiments included microparticles (diameter: 1 μm) and spherical microrobots (diameter: 80 μm). [7] The frequency of the rotating magnetic field was 10 Hz, and the current I from the power source was 2.5 A. Figure 3A shows the aggregation process of the microparticle swarm. At the beginning, the microparticles were distributed in a custom-made chamber with a density as low as 1μg mL À1 . When a magnetic field was applied, the microparticles converged to the resulting aggregation center. As time increased, more and more microparticles gathered and were densely distributed around the aggregation center. Figure 3B displays that by changing the parameter c, the microparticles gathered at different aggregation areas. The red dotted circle represents the boundary of the circle that covers 95% of the microparticles. Figure 3C shows the locomotion result of the microparticle swarm by adjusting c. At 1 s, the microparticle swarm was located at its original position by setting c ¼ (1.11, 1, 1, 1.12). After changing c i , the microparticle swarm moved to a new aggregation center at 160 s. By adjusting c i to set the new aggregation centers, the microparticle swarm continued to move to different target positions at 320 and 480 s. Figure 3D presents the locomotion result of the microrobot swarm with a low distribution density. The results in Figure 3 verified that the rotating gradient magnetic field can accumulate different types of magnetic microagents, despite their size, shape, and material characteristics, and the driving mechanism does not depend on the agent-agent interaction. The simulation results of agent-agent interaction in the magnetic field and fluid are shown in Figure S3 and S4, Supporting Information.

Motion Performance Analysis
To characterize the influence of the rotating frequency on the aggregation performance, experiments were performed to control a swarm of microparticles (diameter: 1 μm) under different rotating frequencies (10 and 50 Hz), as shown in Figure 4A-D.
Microparticles tend to move close to each other due to equivalent centripetal forces and particle-to-particle interactions, while centrifugal forces tend to separate microparticles as they rotate. Maximum centrifugal force occurs at the boundaries of particle swarm patterns. As the rotational frequency increases, the microparticles located in the surrounding layers become difficult to be captured by the dominant pattern and then escape, resulting in a decrease in the size of the dominant pattern. The separated microparticles will form multiple of vortices of smaller size. In contrast, the hydrodynamic interactions among microparticles www.advancedsciencenews.com www.advintellsyst.com also vary with frequency. It was reported in the literature that the repulsive force between rotating microrobots is proportional to ρω 2 r 7 /d 3 , where ω is the rotational speed. [41] It can be seen that the repulsive force increases with frequency, thereby dispersing the microparticle swarm. At 10 Hz, the distribution area of the microparticles was reduced from a larger area at 1 s to a smaller area at 120 s ( Figure 4A), indicating that the microparticles successfully aggregated. Then, changing the frequency from 10 to 50 Hz, the microparticles expanded to a larger area from 1 to 60 s ( Figure 4B), suggesting that the microparticles scattered when the frequency was increased to a certain high value. [41] Figure 4C,D shows the same aggregation and scattering phenomenon of microrobots (diameter: 80 μm) under the rotating frequencies of 10 and 50 Hz, respectively. Our experimental study also showed that when the rotating frequency ranged from 6 to 34 Hz, the microparticles aggregated, which is called "aggregation state." When the rotating frequency was increased above 48 Hz, the microparticles scattered, which is called "dispersion state." When the frequencies ranged from 34 to 48 Hz, the microparticles appeared unstable, which is called "unstable state." In addition to frequency, the working plane height also affects the microagent motion significantly. Figure 4E shows the aggregation of the microparticle swarm on working planes with different h values of 20, 25, 30, and 35 mm at a frequency of 10 Hz. When h increased, the size of the formed microparticle swarm increased, indicating that the aggregation ability decreased. Figure 4F displays the microparticle swarm on the working plane with h ¼ 20 mm, at different frequencies ranging from 6.66 to 50 Hz. The size of the formed microparticle swarm increased as the rotating frequency increased. When the frequency reached 50 Hz, the microparticles stayed together and scattered. Figure 4G presents the microparticle swarm tested on different working planes, with h values ranging from 0 to 60 mm at different rotation frequencies. Three different states of "aggregation state," "dispersion state," and "unstable state" are clearly displayed.

Microparticle Aggregation in the Constrained Area
Experiments of aggregating microparticles in a microfluidic chip were further conducted. The microfluidic chip was designed to simulate a blood vessel network for targeted delivery. Under the www.advancedsciencenews.com www.advintellsyst.com action of the rotating magnetic field, a specific agglomeration area is formed, attracting microparticles from different channels. When the location of the aggregation area was changed by adjusting parameter c to respond to other disease sites for treatment, the microparticles located in the existing aggregation area were released and accumulated in the new aggregation area. Two microfluidic chips with different internal structures were designed to mimic two vascular environments of blood vessels and capillary network. The blood vessels with large diameters are usually used for injection because they can be easily detected and pierced. In our research, a microfluidic chip with a diameter of 300 μm was used as a main channel, which was separated into two subchannels. Each subchannel was further separated into two branch channels (diameter: 50 μm). Figure 5A shows the simulation result of the flow that can be uniformly distributed (less than 5% dispersion) in the entire microfluidic channel. Here, the flow rate represents the injection velocity of the microinjection system, which is different from the flow rate in bio-tissues or organs. [42] When the flow rate of the injection system is 3 μL min À1 , the flow rate in the nested channels can reach 1.9 to 6.4 mm s À1 , which reflects the actual flow in capillaries of the brain and muscles. [43] Figure 5B presents the experimental results. The microparticles were evenly distributed at the beginning ( Figure 5B(i)), and then attracted to the upper and lower diamond-shaped microfluidic channels ( Figure 5B(ii) and (iii)). These results suggested that under a rotating magnetic field, magnetic microparticles can move to different aggregation areas by entering different channels. When the frequency was increased to 50 Hz, the microparticles first scattered and then were washed away by the fluid flow ( Figure 5B(iv)). A similar phenomenon can also be found in the nested channels of the simulated capillary network. By changing the aggregation center of the rotating gradient magnetic field, the magnetic particles were enriched in different areas of the channels, as shown in Figure 5C. The aggregation process in microfluidic channels is also shown in Movie S2, Supporting Information. Figure 5D displays the concentration degrees of microparticles in Figure 5B. The concentration degree represents the density of microparticles in aggregation, and its value can be calculated by image processing in Python OpenCV ( Figure S5, Supporting Information). Phase 0 represents the initial state where no microparticles pass through the fluid channel and can be used as a reference in image processing. When microparticles flowed into the binary channel, the concentration degree was displayed in Phase 1, which is higher than in Phase 0. Note that the concentration degrees of aggregation areas 1 and 2 were close to each other, indicating that the flow was evenly distributed. Phase 2 represents the process of Figure 5B(ii) where the microparticles were attracted by the rotating magnetic field in aggregation area 1. The concentration degrees of microparticles in aggregation area 1 increased sharply to a high level, whereas that in aggregation area 2 decreased. Phase 3 represents the www.advancedsciencenews.com www.advintellsyst.com process of Figure 5B(iii) in which the concentration degree in aggregation area 2 increased significantly, whereas that in aggregation area 1 decreased. Phase 4 represents the dispersion process where the concentration degrees of the two areas decreased at the initial level. Figure 5E shows the concentration degree of microparticles in the nested channel, where Phases 1-4 represent the processes (i)-(iv) in Figure 5C, respectively.

Ex Vivo Experiments of Microparticle Aggregation
Finally, the ex vivo experiments of controlling microparticles in the bovine eyeball were performed. After being injected into the eyeball, the drug-loaded microparticles begin to slowly degrade, and the drugs are released through passive diffusion and stain the retina. [44][45][46] However, this passive diffusion to the retina can cause side effects and reduce the efficiency of passive diffusion. This problem can be overcome by using actively propelled microparticles. [47] Previous research revealed that the physical properties of vitreous, such as density or viscoelasticity, are unaffected by the small-dose injection of particle suspension; the locomotion of microparticles is also not dependent on the partial or overall dilution. [48] In this research, a swarm of microparticles were injected into the bovine eyeball to demonstrate the feasibility of using rotating gradient magnetic field for www.advancedsciencenews.com www.advintellsyst.com microagent aggregation. The fluorescein bonded to the microparticles was used to demonstrate the drug release process. OCT and fluorescent microscopy were used to capture the images. [47] Microparticle suspension (0.1 mL, 1 mg mL À1 ) was injected into the posterior part of the eyeball ( Figure 6A(i)), and then subjected to 6 min of 2D mechanical vibration and 0.5 min of ultrasonic vibration dispersion ( Figure 6A(ii)) to simulate the rehabilitation operation between the two consecutive treatment courses. Subsequently, a rotating gradient magnetic field was applied to trigger the aggregation process of the microparticles. The current of each coil was 2.5 A, and the rotation frequency was 15 Hz. Figure 6A(iii)-(iv) shows the aggregation process of the microparticle swarm. The reconstructed OCT image with slice-array images of the gathered microparticles is illustrated in www.advancedsciencenews.com www.advintellsyst.com Figure S6 and Movie S3, Supporting Information. In Figure 6A(iv), when the microparticles were degraded after 30 min, the eyeball was stained by the released fluorescein, which can be used to simulate the drug release process for targeted therapy. Figure 6B shows the process of aggregation and dispersion, where the length of the OCT scan line is 5.8 mm. Under a rotating magnetic field, the uniformly distributed particles gathered into many clusters, as represented by the white pattern marked by the yellow dashed circle. At the same time, the microparticles began to move toward the center of the OCT view, which was the desired aggregation center. When the distribution area of the microparticles was reduced, the image contrast of OCT was improved. Finally, the microparticles were concentrated around the aggregation center at 400 s. After increasing the frequency of the rotating magnetic field to 66 Hz within a period from 400 to 750 s, the microparticles scattered. Figure 6C shows the comparison result of the stained area with and without the rotating magnetic field. When the rotating magnetic field was applied, the fluorescence area shrank from a large ellipse area at 5 min to a small ellipse area at 30 min. During this period, the brightness of the fluorescence stain increased, indicating that the microparticle swarm moved to the aggregation center and released fluorescence. Note that due to the difference in the aggregation characteristics of the microparticles in the tangential and radial cross-sectional directions of the eyeball, the microparticles formed an elliptical pattern instead of a circle. The results of the control group show that when no magnetic actuation control was applied, the passive diffusion of fluorescence was weaker than that of actively propelled microparticles. Moreover, the distribution area was large and irregular. Figure 6D shows the quantitative data of intensity and area of fluorescein used to simulate the drug release process. [5]

Discussions
In this study, we developed a new actuation mechanism that uses a rotating gradient-based magnetic field to drive magnetic microagents to the target site. By programming the microcontroller unit (MCU) of the custom-designed actuator to sequentially energize the electromagnetic coils, the gradient magnetic field can be rotated, and an equivalent centripetal force is generated on the microagents, converging the microagents to a common target position. By modifying the input currents of coils, the position of aggregation center can be adjusted. Experiments were performed to verify the feasibility of this new magnetic drive mechanism. The influence of rotating frequency and working plane height were also characterized. The effectiveness of the rotating gradient magnetic field was verified in the microfluidic chip network that simulated the vascular environment. The ex vivo experiment was also conducted successfully on the bovine eyeball model. All the previous results confirmed the feasibility of the proposed rotating gradient magnetic field in driving different kinds of microagents to a common site for targeted delivery.
Different from the other gradient field-based magnetic actuation methods, [49][50][51][52][53] the actuation mechanism proposed in this paper is a position-based swarm control, and the resulting target position can be simply adjusted by changing the coil current inputs. More importantly, this method does not rely on imaging feedback, thus greatly simplifying practical applications. With the method, [51] four external electromagnets generate magnetic force to directly drag ferrofluid droplets and direct the magnetic forces to the activated electromagnets according to the vector synthesis rule. Using the imaging system, the position of the droplet can be obtained and the electromagnetic actuation strategy can be adjusted to move the droplet. Several other works using varied magnetic fields have also been reported in the literature. [19,41,52,53] In these works, when the rotating magnetic field axis creates a pitch angle between the substrate and the rotating magnetic field plane, the unbalanced friction and fluid resistance will push the swarm forward. [54] Note that all these methods are velocity-based controls, which requires imaging feedback to adjust the control strategy by analyzing the deviation of the current position from the target position.
Compared with traditional gradient field-based actuation without rotation, the proposed rotational gradient field-based method can reduce viscous resistance and friction and make microagent actuation more efficient. The static viscous resistance and friction strongly resist the motion of the microagent when the gradient field is used directly, especially at the beginning of the motion. When the magnetic force reaches the static friction threshold, the microagent is dragged suddenly and accelerated to a high speed in a short time, which brings difficulty to the motion control. When the microagent is rotated by the magnetic field, the static friction force is eliminated and replaced by the kinetic friction force, which is much smaller than the static friction force. According to the research, [39] the energy consumption of rotating and dragging the microagent is much smaller than that of directly dragging the microagent.
In the current study, due to the hysteresis effect of the coil core, the energizing current of the electromagnetic coil is limited to a range of several amperes, which affects the driving ability of the rotating gradient magnetic field. To overcome this problem, the shape of the magnetic core can be optimized and materials with low hysteresis can be selected. Another challenge encountered is the realization of 3D aggregation of the microagent swarm. The current study used four horizontally distributed coils, which restricted the microagent motion to only on the 2D plane. In the future, we will improve the design of the system to realize a programmable 3D rotating gradient magnetic field, and test new type of microagents such as diamagnetic microagents.

Experimental Section
Experimental System Setup: The magnetic drive system ( Figure S7A, Supporting Information) consisted of four custom-designed magnetic coils to generate magnetic field in two orthogonal directions, four DC power supplies (AMETEK SGX300X17D-0ASAR) to generate high-power steady electrical voltage, one microcontroller unit (ATMEL MEGA32U4), and four voltage amplifiers controlled by MCU to transform DC current to the modified sinusoidal wave signals ( Figure S7B, Supporting Information). The other components include a custom-designed chamber, a microfluidic chip, an inverted optical CCD camera (THORLABS 1500M-GE with objective lens from 2Â to 10Â), and OCT (THORLABS, Telesto Series). The workspace of the magnetic field was located in the central area of the coil system, which was a square area with side length of 15 mm.
www.advancedsciencenews.com www.advintellsyst.com To initialize the system, the image coordinates of the OCT must be calibrated. Here, the standard grid calibration target could be used to mesh the image captured by the microscopy/OCT. With the grid calibration target fixed at the physical center of the magnetic coil, the position of the microscopy/OCT could be adjusted to ensure that the grid calibration target was in the center of the image. Then, the image center of the OCT coincided with the physical center of the magnetic coil, and the swarm would gather in the center of the image when c ¼ (1,1,1,1). Image grids could be created. When classifying, interpreting and diagnosing desired targets such as ophthalmic diseases by the OCT, the mapping table of c could be used to identify the location of these targets.
DC Current Input of the ith Coil: A typical 2D Helmholtz coil system with four coils was considered where the first and third coils were in the positive and negative directions of the x-axis, and the second and fourth coils were in the positive and negative directions of the y-axis, respectively. The flux density of the rotating magnetic field was described as BðtÞ ¼ B x cosð2πf tÞe x þ B y sinð2πf tÞe y , where e x and e y denoted the unit vectors in x and y directions, respectively. [53] For a Helmholtz coil system, B x was determined by the first and third coils. When currents I 1 and I 3 of the first and third coils were reciprocal functions, expressed as I 1 ¼ ÀI 3 , the superposition magnetic field B x was a unique vector field. Similarly, I 2 and I 4 were also reciprocal functions.
To generate the rotating gradient magnetic field, the two coils in the same direction should not be activated simultaneously. When I 1 > 0 and I 3 ¼ 0, the direction of B x would point to the positive direction of the x-axis; when I 1 ¼ 0 and I 3 > 0, the direction of B x would point to the negative direction of the x-axis. For B y on the y-axis, I 2 and I 4 were the same. This sequential activation process could be presented by Heaviside step function ( Figure S7B, Supporting Information). Then, the current I i of the ith coil is presented in Equation (6).
Microparticles and Microrobots: Commercial spherical-shaped hematite microparticles (MagbeStar MP150FG-Plain, BEAVER Co., Ltd., China), with superparamagnetic core and polymer outsourcing layer with a diameter of 1 μm, were used for the in vitro experiments. The microparticles could be loaded with various drugs via surface bonding. Commercial superparamagnetic iron oxide microparticles (SPIOm, Product 103 FG, NanoMicro Technology Co., Ltd, China) were used for the ex vivo experiments. The core of SPIOm was Fe 3 O 4 , the out layer of SPIOm was biocompatible polymers, the fluorescein λ max absorption and λ max emission were 488 and 515 nm, respectively. The particle size distribution (D50) was 0.5 μm. To adapt to the high-viscosity environment, the microparticles were processed with Pluronic F-127 (Sigma-Aldrich) in deionized water solution as hydrophobic surface functionalization. The microrobots had a burr-like porous spherical structure and a diameter of 80 μm. [7] They could be manufactured by a two-photon lithography system (Nanoscribe GmbH) using the degradable materials doped with superparamagnetic materials (74 vol% polyethylene glycol diacrylate, 24 vol% pentaerythritol triacrylate, and 2 vol% Fe 3 O 4 nanoparticles). [5] Functional cells could be carried and delivered by such microrobots. PEGylated monodisperse magnetic microspheres (Mag3100, Nanjing nanoeast Biological Technology Co., Ltd, China) were used in open area and eyeball aggregation experiments with a particle size distribution (D50) of 10 μm. The experimental results demonstrated that the rotating gradient magnetic field could also successfully manipulate and precisely control the nanoparticle swarm, as shown in Movie S1 and S3, Supporting Information. The images of the microparticles and microrobots are shown in Figure S8, Supporting Information.
Discussion on Biocompatibility Issue: The biocompatibility analysis of gradient field-based magnetic objects were well reported in the literature. Previous studies showed cytotoxicity results of pure Fe 3 O 4 on normal cell line (3T3 fibroblast cells) and cancer cell lines (SiHa cervical cancer cells). [55][56][57] 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT assay) showed 80% cell viability within 48 h even at Fe 3 O 4 concentration of 800μg mL À1 . [55] Microparticles could be noncytotoxic or low-cytotoxic by using biocompatible materials for the outer coating. [5,22,[55][56][57] Our previous studies also demonstrated the biocompatibility findings of burr-like porous spherical microrobots. [5,58] MTT assays showed that at ultrahigh concentrations of degradation products (equivalent to degrading 10 k microrobots in 1 μL of solution), cell viability was unaffected, with viability of all cell types above 80%, and the released cells could expand into a colony, while the microrobots degenerated. In addition, the growth and invasiveness of implanted orthotopic liver tumors were inhibited after intra-inferior vena cava injection in nude mice, and no rejection was observed.
Microfluidic Chip Design and Fabrication: The microfluidic chip was fabricated using soft-lithography technology. A 4 in. diameter silicon wafer was used as the substrate and spin-coated with a 100 μm thick layer of negative photoresist SU-8 2050 (Microchem Corp.). After several processes, such as prebake, exposure, post-bake, and development, an SU-8 mold with a pattern of molded vascular microchannels was obtained. Appropriate amounts of polydimethylsiloxane (PDMS) (Sylgard 184, Dow Corning) and curing agent were mixed at a ratio of 10:1 by weight and poured onto the SU-8 mold. The mold with PDMS mixture was placed in a vacuum oven and baked at 70 C for 2 h after removing air bubbles. Finally, the cured PDMS microchannel was peeled off from the mold, punched at the inlet and outlets, and bonded with a clean glass substrate to form the final chip.
Ex Vivo Experiment on the Bovine Eyeball: The bovine eyeball was bought from local market and kept at 0-4 C environment. The fluorescein microparticle suspension (1000 μL) was placed on a cell dish, mixed with phosphate-buffered saline (PBS) buffer solution (9 mL, pH ¼ 7.6). After injecting 100 μL PBS buffer solution containing fluorescein microparticles into the eyeball using a micropipette, the eyeball incision was sealed with glue and then fixed on a custom-designed vibration chamber. Then, the eyeball with uniformly distributed fluorescein microparticles was placed in the operating area of the magnetic coil system. Before observation with OCT, the upper part of the eyeball should be removed. The eyeball was cut at its equator, excluding all other tissues (lens, cornea, pupil, iris, etc.) except the vitreous body. These images were taken with a complementary metal oxide semiconductor camera at a rate of 9-12 frames s À1 . After magnetic propulsion, the eyeball was cut with a scalpel, then the vitreous body was removed mechanically, and the retina was cut into a square (2 Â 2 cm). The fluorescence of the microparticles was excited by a light-emitting diode (Ts2R, Nikon eclipse) with a center wavelength of 488 nm. The experiment was repeated three times.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.
H.C assisted the imaging and control experiments. L.F. assisted the microfluidic chip design and tests. L. Zheng assisted the rotating electromagnetic field design. L. Zhang provided guidance in design and experimental study. All authors reviewed and approved the manuscript.

Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
Keywords microrobots, rotating gradient magnetic field, swarm control, targeted therapy