Adaptive intrathecal nanorobotics in mice and nonhuman primates for navigated CNS therapy

Yumeng Lu; Xinjian Fan; Chengjuan Fan; Gongzi Zhang; Hui Cui; Yiran Jiang; Shuwei Zhang; Yuxin Jia; Renfeng Dong; He Tan; Lihai Zhang; Zhiguang Wu

doi:10.1126/sciadv.aec2918

. 2026 Jul 31;12(31):eaec2918. doi: 10.1126/sciadv.aec2918

Adaptive intrathecal nanorobotics in mice and nonhuman primates for navigated CNS therapy

Yumeng Lu ^1,^2,^3,^†, Xinjian Fan ^4,^†, Chengjuan Fan ^5,^†, Gongzi Zhang ⁶, Hui Cui ^1,^2,³, Yiran Jiang ^1,^2,³, Shuwei Zhang ⁶, Yuxin Jia ⁷, Renfeng Dong ⁸, He Tan ⁹, Lihai Zhang ^6,^*, Zhiguang Wu ^1,^2,^3,^*

^¹School of Medicine and Health, Harbin Institute of Technology, Harbin 150001, China.

^²State Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin 150001, China.

^³Chongqing Research Institute, Harbin Institute of Technology, Chongqing 401120, China.

^⁴School of Mechanical and Electrical Engineering, Soochow University, Suzhou 215131, China.

^⁵Department of Oncology, the Second Affiliated Hospital of Harbin Medical University, Harbin 150001, China.

^⁶China Department of Orthopedics, Chinese PLA General Hospital, Beijing 100141, China.

^⁷Department of Mathematics, Imperial College London, London SW7 2AZ, UK.

^⁸School of Chemistry and Chemical Engineering, Lingnan Normal University, Zhanjiang 524048, China.

^⁹Comprehensive Cancer Centre, Denmark Hill Campus, London SE5 9NU, UK.

Corresponding author. Email: zhanglihai74@qq.com (L.Z.);

Corresponding author. zhiguangwu@hit.edu.cn (Z.W.)

^†

These authors contributed equally to this work.

Roles

Yumeng Lu: Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing

Xinjian Fan: Data curation, Formal analysis, Funding acquisition, Methodology, Software, Validation, Visualization, Writing - original draft, Writing - review & editing

Chengjuan Fan: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing

Gongzi Zhang: Formal analysis, Investigation

Hui Cui: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing

Yiran Jiang: Formal analysis, Investigation, Validation, Visualization

Shuwei Zhang: Resources, Validation, Writing - review & editing

Yuxin Jia: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing - review & editing

Renfeng Dong: Investigation

He Tan: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - review & editing

Lihai Zhang: Conceptualization, Formal analysis, Methodology, Project administration, Resources, Validation, Visualization, Writing - review & editing

Zhiguang Wu: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing

PMCID: PMC13426435 PMID: 42536754

Abstract

Intrathecal delivery for central nervous system (CNS) therapy lacks spatiotemporal control due to spinal canal anatomical constraints, especially at human scale. We developed an adaptive magnetic nanorobotic platform enabling multimode swarm mobility for in vivo spinal canal navigation in mice and nonhuman primates. Programmable chain-like, vortex-like, and ribbon-like swarm were accomplished, and vortex swarm was indicated as optimal strategy for adaptive intrathecal navigation. Driven by a robotic arm–generated magnetic field, the vortex swarms indicate navigation within spinal canal over centimeter distances, which magnetically enhanced accumulation at CNS target sites and prolonged their in vivo retention, thereby substantially enhancing their therapeutic efficacy compared with conventional passive intravascular delivery, with minimal side effects. Tests in nonhuman primates indicate that the nanorobotic swarm migration is from the gap between the bundles of cauda equina and then the spin cord, revealing their morphological adaption at the cauda equina interface to circumvent anatomical constraints.

An adaptive nanorobotic platform with swarm navigation in nonhuman primates solves CNS delivery challenges.

INTRODUCTION

The spectrum of diseases in central nervous system (CNS) has been steadily expanding from classical autoimmune disorders such as CNS injuries to far more diverse diseases (1, 2). Evidence now suggests that syndromes such as CNS injuries involves the intricate interplay between acute neuroinflammation and chronic neural degeneration (3, 4). The uncontrolled immune responses lead to persistent neurological deficits and neuroinflammation with limited surgical and internal treatment options (5–10). Although various therapeutic agents and biomimetic nanocomplexes such as resolvin D2 (RvD2) neutrophil membrane–derived vesicles (NMVs), and stem cell–derived exosomes (EXO), involving the regulation of inflammatory cytokines and interactions between neutrophils and endothelial cells have been used through intravascular and intrathecal routes (11–19). Compared with the systemic delivery such as oral and intravascular administrations, which can be limited by various biological barriers (e.g., blood-brain barrier and blood-spinal cord barrier), intrathecal delivery is an emerging delivery strategy for the CNS therapy, which provides a reliable dosage of the infusate into the intrathecal space. While effective in alleviating inflammation and promoting nerve regeneration, the therapeutic performance of these therapeutic nanocomplexes suffer from long diffusion period and limited spatial targeting toward the targeted sites in CNS diseases (20–22).

By contrast, swimming nanorobots, with capability of navigation in nanoscale environments via diverse propulsion mechanisms, represent a remarkable convergence of robotics and medicine. They are engineered to traverse challenging anatomical landscapes, from the constricted channels of narrow vasculature to the tightly confined tissue microenvironments (23–38). Inspired by motile microorganisms in nature, various propulsion mechanisms for cutting-edge nanorobots have been developed for low–Reynolds number environments using acoustic and magnetic actuation to offer elevated delivery of various therapeutic agents to the hard-to-reach area (39–47). In particular, magnetically actuated nanorobots have evolved efficient and sophisticated propulsion in diverse complex biofluids including ocular vitreous medium, the gastrointestinal tract, and the extracellular matrix (48–56). In addition, owing to the limited capabilities of a single nanorobot, swarm navigation entitles the multimode control to accomplish high environmental adaptability and elevate their tasking capabilities (57–59). For example, emergence and control of magnetic micro/nanorobots and various swarm configurations have been accomplished through the manipulation of different magnetic inputs (60). Their therapeutic applications have covered various diseases ranging from glioblastoma to thrombolysis (61, 62). Combined with acoustomagnetic or gradient-assisted control also provides additional capability such as upstream locomotion for the realization of micro- and nanosystem magnetic navigation against various biological flow (63–65). Therefore, it is proposed that the approach to turn the therapeutic nanocomplexes into nanorobotic swarm may accomplish the active long-distance intrathecal delivery and controllable release toward targets in CNS, offering an innovative technology for adaptive nanorobotic therapy toward CNS diseases.

Nevertheless, translating biomimetic nanocomplexes into intrathecal nanorobots for navigated CNS therapy remains a formidable challenge. First is the substantial transportation capacity for effective therapy. The clinical administration dosage for intrathecal delivery is mainly in the range of 0.1 to 10 μg/kg tissue, implying the challenges in swarm navigation with therapeutic scale. Second are the anatomical barriers in human scale. The swarm movement in CNS, characterized by intricate anatomical barriers, particularly that anatomical scale disparities between mice and human pose a critical translational gap, remains underexplored. The major in vivo investigations were conducted in mice model, where the diameter spinal canals (diameter of ∼1 mm) were filled with limited bundles (∼5% of human) of cauda equina and laminar cerebrospinal fluid (CSF). Their dynamics have limit resemblance to the centimeter-scale, human spinal canal filled with cauda equina, where the gap is ∼0.05 to 0.4 mm (66). Third is the safety concern. High biocompatible characteristics eliminate the requirement for subsequent surgical removal. An ideal nanorobotic platform for navigated intrathecal CNS therapy is expected to accomplish the therapeutic scale adaptive swarm motion in CSF along the gap between the bundles of cauda equina and, subsequently, perform large amount release at the targeted sites in CNS with minimal toxicity.

Here, we developed a biomimetic reconfigurable nanorobotic complex (BRaNC), an adaptive nanorobotic system that accomplishes multimode swarm locomotion and self-organization in spinal canal for navigated intrathecal delivery in vivo in CNS therapy (Fig. 1). The biomimetic nanocomplex robots were developed by fusing magnetic nanoparticles (NPs), RvD2 as a model drug, and NMVs and stem cell–derived EXO as model biomimetic unities, respectively (Fig. 1A). This adaptive nanorobotic system, in which magnetic biomimetic nanocomplexes assemble into three distinct swarm modes including chain, vortex, and ribbon, is achieved by manipulating the frequency of the rotating magnetic field and its polarization in a three-dimensional space. Upon intrathecal administration, BRaNCs, which could be visualized with computed tomography imaging technique, enables controllable intrathecal locomotion via robotic arm–assisted magnetic fields (RARMFs) (Fig. 1B). Upon programming the magnetic field inputs, BRaNCs form adaptive swarm along the gaps between bundles of cauda equina (Fig. 1C) and then conduct adaptive swarm navigation with the migration of the robotic arm–mounted magnetic field system (Fig. 1D). Upon reaching the targeted sited in the CNS, BRaNCs with high frequencies of magnetic field inputs trigger accelerated release toward the CNS tissue (Fig. 1E). This integrated BRaNC system thus represents a substantial advancement in CNS therapies, bridging the gap between adaptive nanorobotics and targeted CNS interventions. By demonstrating unprecedented control over therapeutic scale swarm in mice and nonhuman primate models, we establish BRaNC as a transformative platform for CNS therapy, one that leverages reconfigurable robotics to overcome the fundamental limitations of fixed-form delivery systems. This work not only advances the frontiers of adaptive nanorobots in biomedicine but also opens previously unidentified avenues for precision interventions in complex physiological environments, where adaptability and multifunctionality are essential for therapeutic success.

Fig. 1. — (A) Schematic illustration of the composition of BRaNCs, including RvD₂, Fe₃O₄, NMVs, and stem cell–derived EXOs, and their assembly into three distinct swarm modes (vortex, chain, and ribbon) under different magnetic fields (H_v, H_c, and H_r). (B) Intrathecal injection of BRaNC, guided by an RARMF system toward the targeted site of CNS (highlighted in red) in vivo. (C) Vortex swarm formation and adaption after intrathecal injection upon magnetic field input of H_v. (D) Long distance swarm intrathecal migration of BRaNCs with the manipulation of robot arm–mounted magnetic field system. (E) Swarm-triggered release for targeted CNS therapy.

RESULTS

Development and magnetic swarm navigation of BRaNCs

Fabrication process of BRaNCs mainly involves three steps as follows: (i) synthesis of Fe₃O₄ and RvD2-loaded NPs (noted as NP cores) through a typical emulsion/solvent evaporation method (67, 68), (ii) obtaining vesicles from neutrophils membrane (noted NMVs) and stem cell EXOs, and (iii) fusion of these vesicles to produce the resulting BRaNCs (fig. S1A). The size of the NP cores could be manipulated from ∼160 nm (1000 W) to ∼250 nm (1800 W) through the operation of the power of the ultrasound field (fig. S1B). On the other hand, the NMV and EXO hybrid vehicles were fabricated via a sonication protocol (67, 69). The resulting hybrid vehicles were then sequentially extruded through 1000-, 800-, and 400-nm porous membranes to create the hybrid nanovesicles with NMVs and EXOs. The NP cores were subsequently fused with the hybrid nanovesicles using a coextrusion method through polycarbonate membranes with successive pores of 1, 800, 400, and 200 nm. The mechanical force during extrusion promoted the fusion of hybrid vesicles with NP cores, leading to the formation of BRaNCs. The geometry of the resulting BRaNC was visualized with transmission electron microscopy after uranyl acetate staining. Compared with the bare spherical geometry of NP cores (fig. S1C), the BRaNC exhibited a spherical core-shell structure, reflecting the successful enclosure of NP cores with hybrid nanovesicles (NMVs and EXOs) and showed a spherical core-shell structure (Fig. 2A). The NMVs and EXOs on NP cores were further monitored through the study of the changes in NP cores size and surface charge before and after the fusion process. The sizes of NP cores increased from 178.7 ± 4.3 nm to 199.0 ± 3.3 nm because of the wrapping of NMVs and EXOs after the fusion (Fig. 2B). The surface zeta potential also changed from −22.2 ± 1.4 mV in the NP cores to −11.6 ± 3.8 mV in BRaNCs. This shift could be attributed to the charge screening effect of NMVs and EXOs as neutrophil membranes and EXOs under different conditions have less negative surface charge than citrate-stabilized Fe₃O₄ NPs.

Fig. 2. — (A) Transmission electron microscopy image shows the morphology of BRaNC. Scale bar, 100 nm. (B) Size analysis and zeta potential of the NP core, NMV + EXO hybrid nanovesicles, BRaNCs (pH 7.2), and BRaNC (pH 6.5). (C) Drug release profiles of NP core and BRaNC in simulated CSF (pH 7.2) and simulated inflammatory CSF (pH 6.5) over 48 hours (h). (D) Magnetization loop from superconducting quantum interference device analysis of NP cores and BRaNCs. (E) Western blot analysis of CXCR4 and CD81 on NP core, NMV, EXO, and BRaNC, respectively. (F) Microscopic images of cellular uptake of BRaNCs and NP core after their cocultivation with macrophages for 1 hour. Scale bar, 20 μm. (G) Quantitative analysis of macrophage uptake of BRaNC and NP core measured using ICP-MS measurements. (H) Schematic illustration and time-lapse microscopic images showing the multimode swarm behaviors of BRaNCs under their corresponding magnetic field inputs, including vortex swarm formation under H_v, chain-like swarm under H_c, and ribbon-like swarm under H_r. Scale bar, 200 μm.

Moreover, the release kinetics of RvD2 from BRaNCs in simulated CSF (pH 7.2) and in simulated inflammatory CSF [simulated CSF with pH 6.5 according to previous reports (70)], were characterized by high-performance liquid chromatography (HPLC). In comparison of the release from NP cores (29.9% in 48 hours), BRaNCs in CSF showed the negligible release of RvD2 with a release rate of 6.4% in 48 hours, indicating its stability through the coating of NMVs and EXOs. In contrast, under the acidic condition in inflammatory CSF, RvD2 was rapidly released from BRaNCs (31.2% in 1 hour, 47.6% in 3 hours, and 73.8% in 48 hours) (Fig. 2C). Next, the magnetization of BRaNCs was measured using a superconducting quantum interference device. The magnetization curve in fig. S2 shows that both BRaNCs and NP cores exhibited typical superparamagnetic behavior, with saturation magnetization (M_s) values of 11.7 and 12.8 emu/g, respectively. The corresponding magnetic dipole forces were calculated as ∼0.17 pN using Derjaguin-Landau-Verwey-Overbeek theory. Noted that the M_s of BRaNCs and NP cores could be elevated to 39.4 and 43.6 emu/g (Fig. 2D) by replacing the magnetic NPs with M_s of ∼60 emu/g. The corresponding magnetic dipole forces were calculated as 2.33 pN, which is ∼13-fold of BRaNCs with M_s of 11.7 emu/g, offering the considerable performance for their actuation upon various external magnetic field.

Furthermore, note that CXCR4 is a representative adhesion protein of neutrophils (71), and CD81 is a tetraspanin protein of stem cell EXOs (72). The Western blot analysis of BRaNCs displayed the presence of CXCR4 and CD81, verifying the coating of NMVs and EXOs (Fig. 2E). To further investigate the biocompatibility of BRaNCs, an antiphagocytic study against macrophages was conducted. As shown in the microscopic images in Fig. 2F, the macrophages showed negligible uptake of BRaNCs upon the cocultivation after 1 hour. In contrast, coincubation of macrophages with NP cores led to a substantial number of dark spots in the intracellular and perinuclear areas of macrophages, indicating that the NP cores were phagocytosis taken up by the cells.

Inductively coupled plasma mass spectrometry (ICP-MS) analysis was performed to quantify iron uptake by macrophages. As shown in Fig. 2G, NP cores had an uptake of 17.8 ng of iron per 1000 macrophage cells, while BRaNCs had an uptake of 3.8 ng per 1000 macrophage cells. The more than fourfold reduction in iron uptake clearly demonstrates that the coating of NMVs and EXOs could effectively inhibit macrophage uptake and thus elevate the biocompatiblity of BRaNCs. This inhibition is largely due to the immunosuppressive antigens of the neutrophil membrane present on BRaNCs, also suggesting the full coverage and right-side-out orientation of NMVs and EXOs after their fusion onto the BRaNCs (73).

To fulfill the efficient navigation of BRaNCs toward the targeted CNS injuries, the movement behaviors of BRaNCs upon magnetic field were investigated. Conventional strategy to actuate the magnetic micro/nanoscale robots mainly uses the three orthogonal Helmholtz coil pairs that create a uniform rotating magnetic field. However, in case of magnetic navigation in spinal canal in human scale for the efficient delivery in CNS, it is required the swarm control of the therapeutic scale (∼10 ng RvD2) of nanorobots within its working area. Meanwhile, the colloidal nanoscale particles upon uniform rotating magnetic field are isotropic in fluid, leading to the toughness for active propulsion. To address these issues, a hybrid magnetic actuation, which integrates electromagnetic coils and permanent magnets, was developed to simultaneously generate a high-magnitude magnetic field (B) and magnetic field gradient (∇B) in a fully programmable manner.

As shown in Fig. 2H, taken from movies S1 to S3, the multimode controls of the BRaNCs, were accomplished through programmable manipulation of magnetic field inputs. As an instance, BRaNCs at 0 s without external magnetic actuation were well dispersed in fluid, and then they exhibited different behaviors including spinning, tumbling, and rolling in fluid upon different magnetic field inputs. Subsequently these BRaNCs formed corresponding swarms with different configurations including vortex-like swarm, chain-like swarm, and ribbon-like swarm upon their magnetic field inputs, noted as H_v, H_c, and H_r, respectively. One can observe that the vortex swarms bestow the aggregation at a high density for adaptive swarm manipulation, and ribbon and chain swarms are suitable for large area spread to accomplish release in short period. Together. BRaNCs, as an example of nanorobotic platform with on-demanded therapeutic agents and biomimetic untities, were successfully developed, and their multimode swarm were accomplished upon the remoted manipulation of external magnetic field inputs.

Multimode swarm transformation and adaptive intrathecal swarm behavior of BRaNCs

Further characterization of the different swarm behaviors of BRaNCs were carried out to compare their features for optimal navigation strategy within spinal canal. Primarily, we were concerned that the vortex swarm could result in aggregation with a high density, which is desirable to the manipulation of BRaNCs with a high dose. The emergence of the vortex upon different frequencies of H_v input was quantified using the normalized projected area S/S₀, where S₀ is the projected area occupied by the vertically rotating BRaNCs upon magnetic input of H_v. The normalized projected area initially decreased from 4.0 at 6 Hz to 1.0 at 24 Hz and then increased to 4.9 at 54 Hz (Fig. 3A), which is attributed to the angle between the long axis and the z axis of BRaNCs experienced the tendency of initial decreasing and subsequent increasing as the function of frequencies (74). Correspondingly, the dependance of frequencies of H_v input on the size of the vortex swarm of BRaNCs exhibited that the vortex swarm reached a size of 132.0 μm at 24 Hz and 163.6 μm at 18 Hz (Fig. 3B). The vortex swarm with size of ∼130 μm exhibited the optimal velocity of 44.7 μm/s, while the vortex swarm with size of 163.6 μm attained a velocity of 27.3 μm/s (Fig. 3B, inset), which may be attributed to the changed angle between the long axis and the z axis of BRaNCs. Moreover, the velocity of the vortex swarm (noted as V_v), chain swarm (noted as V_c), and ribbon swarm (noted as V_r) of BRaNCs upon different frequencies of H_v and H_c exhibited an increasing then decreasing trend, reaching a maximum velocity at 24 Hz of 44.7 μm/s (V_v), 37.2 μm/s (V_c), and 27.1 μm/s (V_r), respectively (Fig. 3C). Similarly, the formation time of these swarms exhibit the decreasing-increasing trend, reaching the shortest period at 24 Hz of 6.0, 5.9, and 6.9 s, respectively (Fig. 3D). The above data indicate the functional advantage and trade-offs of different swarm configurations (summarized as table S1), including high speed/stability but narrow coverage for vortex swarm, wide coverage but low speed for ribbon swarm, and balanced speed/coverage but low stability for chain swarm, respectively.

Fig. 3. — (A) The dependance of normalized projected area S/S₀ of vortex swarm on the frequencies of H_v input. (B) The relationship between the size of the vortex swarm and the frequencies of H_v input. Inset the size and corresponding velocity of vortex swarm (inset). (C and D) The velocity (C) and formation period (D) of vortex swarm, chain swarm, and ribbon swarm under different frequencies of their corresponding inputs. (E) Time-lapse microscopic images showing the reversible transformation of different swarms through the manipulation of magnetic field inputs. Scale bar, 200 μm. (F) Schematic and time-lapse microscopic images of the fusion of vortex swarm into a giant swarm of BRaNCs with therapeutic dose upon magnetic field input of H_v (frequency of 24 Hz). Scale bar, 200 μm. (G) Schematic diagram (top) and time-lapse images (bottom) exhibiting the dynamics of BRaNCs with therapeutic dose upon different magnetic inputs. Scale bar, 1 cm. (H) In vivo imaging system (IVIS) images illustrating the movement of BRaNCs upon different magnetic inputs in model spinal canal. Scale bar, 1 cm. (I) Corresponding normalized number of BRaNC counted from colorful spots in (H), displaying the distribution of BRaNCs in spinal canal upon various magnetic field input. All data are normalized to their maxima. (J) Drug release profiles of RvD2 from BRaNCs under different magnetic field inputs in simulated CSF (pH 7.2) and simulated inflammatory CSF (pH 6.5) over 3 hours.

Apart from the formation of various swarms, programmable manipulation of magnetic field inputs bestows reversible transformation between these swarms, which involve the break of the original swarm state and become a different dynamic steady state for the system. For example, BRaNCs upon the magnetic field input of H_c gradually formed chain swarms in 3 s, then transferred toward vortex swarm by switching to the H_v of magnetic input over 39 s, and subsequently returned to chain swarm through the operation of H_c input again (Fig. 3, E, a). Accordingly, the reversible swarm transformation between vortex and ribbon and ribbon and chain were also realized with the operation of magnetic field of H_v and H_r and H_r and H_c, respectively (Fig. 3, E, b and c). In addition, the swarms of BRaNCs returned to separation in fluid (noted as liquid state) once the magnetic fields were turned off. The fast and reversible transformations could be achieved between the above swarm patterns, including a total of six transformations (fig. S3 and movies S4 to S9). In addition, having considered the effect of CSF compositions, the swarm formation and transformation in model CSF were also investigated. As shown in figs. S4 and S5, BRaNCs in model CSF were mainly maintained the swarm behavior as that in phosphate-buffered saline (PBS). Furthermore, the model CSF with amplified concentration were used, and the red blood cell (RBC) membrane–coated magnetic particles and uncoated magnetic particles were served as additional controls. As shown in fig. S6, the formation period and size of the RBC membrane–coated magnetic particles swarm were affected upon the high concentrated ions in amplified CSF. By contrast, BRaNCs swarm behaviors were mainly maintained even upon the 3× CSF, and the obvious changes occurred upon 5× CSF, which is relative far beyond CSF with physiological conditions. The velocity of BRaNCs with different swarm configurations in these model CSF with amplified concentration maintained the similar performance as that in PBS, which is considerably higher than that of both RBC-coated magnetic particles and uncoated magnetic particles (fig. S7). The above data indicate that CSF composition could alter dipole coupling and hydrodynamic behavior, but that may be not played a dramatic impact on the BRaNCs, which may be attributed to the biosurface chemistry to the particles.

Among the three types of swarms, BRaNCs with a therapeutic scale concentration (1 mg/ml for BRaNCs and 1 μg/ml for encapsulated RvD2) upon magnetic field input of H_v generated a tiny vortex fluid field with itself as the vortex core, which cause hydrodynamic interaction and the time-averaged magnetic dipole forces to attract the neighboring BRaNCs, while repulsion prevent these BRaNCs from getting too close, resulting in the fusion of the dynamic aggregation with a large scale. For example, as shown in time-lapse images in Fig. 3F, taken from movie S10, the merging process of the vortex swarm typically consists of three phases: proximity, fusion, and stabilization. First, once the vortices getting close, the nearing parts of the two vortices start exchanging members. Second, the neighboring vortices enter the fusion zone, forming an irregularly shaped vortex. Eventually, the neighbor vortices fuse and thus create a large, stable, circular vortex. Under the above magnetic H_v input, BRaNCs with a therapeutic dose (10 μg of BRaNCs, carrying 10 ng of RvD2, in 10 μl) can be formed the dense giant vortex with size of ∼0.2 mm.

To investigate the movement behavior in spinal canal in human scale, where spinal cord with a bundle at its base (cauda equina) located inside and CSF fill in the space. Considered that the position for intrathecal injection is in the area of cauda equina, a model spinal canal structure that spinal cord with bundles of cauda equina inside was developed. To accomplish long distance movement of BRaNCs, a robotic arm with the velocity of migration of ∼10 cm/hour was integrated onto the magnetic field system. Time-lapse images in Fig. 3G, taken from movie S11, display the dynamics of BRaNCs (therapeutic dose) with different magnetic motions including nonactuation (NA), chain swarm (chain), ribbon swarm (ribbon), and vortex swarm (vortex) groups. Upon the intrathecal injection in the area of model cauda equina, minor change in the NA group occurred in 122 s. BRaNCs upon robotic arm–assisted magnetic actuation (chain, ribbon, and vortex groups) in model spinal canal exhibited obvious movement along the spinal cord through the cauda equina. Despite the clear directional movement of BRaNCs in chain and ribbon groups in 122 s, their distributions decay along the model spinal cord and the major BRaNCs in both chain and ribbon groups located with their travel distance of less than 4 cm. By contrast, the major BRaNCs in vortex group formed an aggregation with a high density in 23 s, and then the aggregation enabled directional swarm migration with the guidance of robotic arm over 100 s, reaching a velocity of 0.4 mm/s. Accounting that the CSF exhibits dynamic flow in vivo, a microfluidic chip was used to the model spinal canal to further mimic the dynamic flow of CSF. Time-lapse images in fig. S8 showed the similar swarm behavior of BRaNCs in model spinal canal with static flow, offering the promise for the adaption of BRaNCs in spinal canal with dynamic flow of CSF. The magnetic attraction from gradient magnetic field leading to the main portion of BRaNCs displayed negligible net displacement over time (fig. S9). Noted that the bundles representing cauda equina played a minor impact on the formation and migration of vortex swarm, visualizing the adaptive behavior of BRaNCs with vortex swarm. The uncoated magnetic NPs in model spinal canal were mainly adhered onto the cauda equina (fig. S10A). Further quantification in fig. S8B indicated that BRaNCs swarm enable adaptive mobility in model spinal canal against flow and CSF compositions, and the vortex swarm facilitate optimal performance compared with ribbon swarm and chain swarm. The above data indicate that the coating of NMVs and EXOs also plays an important role in adaptive navigation of BRaNCs. Furthermore, the dynamics of BRaNCs upon different magnetic field inputs in excised murine spiny canal were investigated ex vivo. To visualize their intrathecal distribution overtime, in vivo fluorescence imaging was used, and Cy7–N-hydroxysuccinimide (NHS), a fluorescence dye for in vivo fluorescence imaging, was functionalized on RvD2 of BRaNCs. The initial position of injection was set as 0 cm for comparison. The fluorescence signal representing BRaNCs exhibited that the major BRaNCs in NA group located in the position of 0 cm and BRaNCs upon magnetic navigation resulted in the directional motion along the spinal canal, particularly that the vortex group showed an ∼5-cm intrathecal trajectory in 40 min (Fig. 3H), which is similar as the results in Fig. 3G. Furthermore, the dynamic distribution of BRaNCs in the spinal cord was quantified by counting the fluorescence signals of BRaNCs. As shown in Fig. 3I, the majority (73.7%) of BRaNCs in NA groups was maintained at the 0-cm position, and only 3.6% arrived at the position of 4 cm over 40 min. By contrast, the chain and ribbon groups exhibited higher distributions at position of 2 cm (11.3% in the chain group and 12.8% in the ribbon group) and 4 cm (7.7% in the chain group and 13.1% in the ribbon group) over 40 min. The distribution in vortex group exhibited the highest level (19.1% at position of 2 cm and 39.3% at position of 4 cm) compared with other groups. The 10-fold elevation of periodic change in distribution compared with that in NA groups indicates that the magnetic actuation enhances the delivery in spinal canal, and vortex swarm offers an optional strategy for magnetic intrathecal navigation.

In addition, concerned that the mobility of BRaNCs upon magnetic field is mainly from the generation of local fluid field, it may also improve the drug release of BRaNCs. As shown in Fig. 3J, compared with the passive delivery in CSF (4.5% for 3 hours) and inflammatory CSF (47.6% for 3 hours), the drug release of BRaNCs upon vortex swarm, chain swarm, and ribbon swarm exhibited similar tendency overtime, indicating the minor impact of swarm types on the drug release while the frequencies played an important role in drug release. Consisting with frequency of magnetic field inputs with the mobility, the drug releases of BRaNCs first increases and then decreases with frequency ranging from 6 to 24 Hz. The drug release in CSF upon frequency of magnetic field input is relatively low within 1 hour, which may be attributed to the encapsulation of NMVs and EXOs, ensuring the leakage of drug before their arrival at the targeted sites in CNS. By comparison, the release upon magnetic actuation in inflammatory CSF was dramatically enhanced toward 92.7% upon magnetic field input of H_v for 3 hours, accomplishing magnetic actuation–triggered controllable release. The above results indicated that magnetic swarm navigation not only confirms the adaptive intrathecal mobility of the therapeutic-scale nanorobots addressing the anatomical challenges of centimeter-scale spinal canals with submillimeter gaps but also facilitates the controllable release through their generation of local fluid.

Dynamics of BRaNCs upon intrathecal navigation in vivo

After having investigated the swarm behaviors of BRaNCs in spinal canal ex vivo, the intrathecal navigation of the BRaNCs in vivo was further accessed using the mice model of spinal cord injury (SCI). The SCI mice model were established with the contusive SCI operation at thoracic vertebral (T) 10 level, and the intrathecal injections toward SCI mice were conducted between lumbar vertebra (L) 3 and L4 levels. The position of T10 was thus used as targeted site for magnetic intrathecal navigation of the BRaNCs and stained with fluorescence dye of Cy5 for the following fluorescence imaging in vivo. The SCI mice subsequently separated into different groups, including intravenous injection of BRaNCs (IV group), intrathecal injection of RvD2 (RvD2 group), intrathecal injection of BRaNCs with NA (NA group), and intrathecal injection of BRaNCs with magnetic control input of H_c (chain group), H_r (ribbon group), and H_v (vortex group) with robotic arm through intrathecal injection, respectively. The velocity of migration of robotic arm and the frequency of magnetic field were set as ∼10 cm/hour and 24 Hz, respectively. For better exhibition of the intrathecal distribution in vivo, the area in spinal canal from intrathecal injection to CNS injury were sequentially noted as L4-L6, L1-L3, T12-T13, and T10-T11, respectively. As shown in Fig. 4A, the fluorescence signal in the IV group exhibited wide distribution in the whole body, low distribution in the spinal canal, and minor location in CNS injury after intravenous injection over 30 min, indicating its low delivery efficiency toward CNS injury. By comparison, the fluorescence signals in both the RvD2 and NA groups were mainly located at the intrathecal injection site over 30 min, implying the slow diffusion within the CSF in spinal canal in vivo. Upon the magnetic navigation with different inputs, the distribution of BRaNCs firstly exhibited the decreased coverage area and increase intensity at 3 min (fig. S11), suggesting the dynamic process of population assembly in vivo.

Fig. 4. — (A) IVIS images illustrating the SCI mice in different groups over 30 min. The position in CNS injury is T10, and the position for intrathecal injection is L5-L6. T, thoracic vertebra; L, lumbar vertebra; T10, the position between T10 and T11; T12, the position between T12 and T13; L3, the position between L1 and L3; L6, the position between L4 and L6. (B) Distribution of fluorescence intensity in different spinal regions. (C) Quantification of overlayed fluorescence signal of RvD2 from BRaNCs and CNS injury to estimate the targeted capability toward CNS injury overtime. (D) IVIS images of excised spinal canal over 24 hours. (E) Distribution of RvD2 from BRaNCS in spinal canal over 24 hours. (F) Distribution in vortex group at different organs overtime. (G) IVIS images of different organs (heart, liver, spleen, lung, kidney, and brain) over 24 hours. (H) Quantitative analysis of fluorescence intensity in different organs overtime. (I) Concentrations of RvD2 in major organs at 6 hour (up) and 24 hour (down). Data are presented as means ± S.D. (n = 5).

In chain and ribbon group, the fluorescence signals displayed the relative long trajectories in spinal canal, suggesting the migration of BRaNCs upon magnetic field inputs of H_c and H_r, respectively, while the intensity of the fluorescence signal in the chain and ribbon groups decreased as the migration of BRaNCs and then nearly vanished before position of T10-T11. In contrast, the vortex group displayed fluorescence signal along the spinal canal with minor decrease along the migration over 30 min and finally reached the targeted site in the CNS injury (T10-T11). Note that the intrathecal behaviors of uncoated magnetic NPs and RBC membrane–coated magnetic NPs were also served as control for comparison of the BRaNC structure. As shown in fig. S12, the main uncoated magnetic NPs are still located in the area of intrathecal injection, resulting in the low targeted capability (∼1%) at the CNS injury site. Meanwhile the survival periods of the mice after intrathecal injection of uncoated magnetic NPs with the same dose as BRaNCs also suggest their high toxicity through intrathecal administration. The RBC-coated magnetic particles exhibited better performance compared with the uncoated magnetic NPs, with prolonged period in spinal canal, and the SCI mice did not exhibit additional symptom after their intrathecal injection (fig. S13), verifying the surface chemistry of membrane elevate the concentration of RVD2 in vivo. The uncoated Fe₃O₄ NPs and RBC membrane–coated Fe₃O₄ NPs in these control experiments were evaluated under nonactuated intrathecal conditions. These controls were therefore used to assess the effect of surface coating on intrathecal compatibility and retention, rather than to directly compare different membrane coatings under identical vortex magnetic navigation. Thus, the present data do not exclude the possibility that RBC membrane–coated Fe₃O₄ NPs may also be redistributed under vortex magnetic actuation.

The dynamic distribution in different groups were further quantified through the analysis of fluorescence from RvD2 in spinal canal. As shown for the portion of fluorescence signal in spinal canal in Fig. 4B, the main portion in the IV group was located in L4-L6 (68.4%) over 30 min while low accumulation in T10-T11 (2.0%), T12-T13 (2.4%), and L1-L3 (2.8%), indicating the low efficiency toward spinal canal using the intravascular injection. Although the distribution in L1-L3 substantially elevated in the RvD2 group (65.0%) and NA group (57.6%) over 30 min, they still showed low distribution in T10-T11 (15.0% in the RvD2 group and 16.0% in NA group) and T12-T13 (20.2% in RvD2 group and 23.4% in NA group), suggesting the low delivery toward targeted CNS injury through passive diffusion. The distribution in the chain group and ribbon group exhibited the elevated portion in T10-T11 (22.0% in the chain group and 26.0% in the ribbon group) and T12-T13 (36.8% in the chain group and 47.4% in the ribbon group) compared with that in NA group. By comparison, distribution of fluorescence signal in vortex group exhibited the relative higher portion of 58.1% in T10-T11 and 19.0% in T12-T13, respectively. To further quantify the targeting capability toward CNS injury in vivo, the fluorescence signal that overlapped with RvD2 (labeled with Cy7-NHS) and the targeted injury site (labeled with Cy5) was calculated. As shown in Fig. 4C, overlayed fluorescence signal in the IV group, RvD2 group, and NA group was mainly maintained with the minor change over 30 min (0.6% in the IV group, 1.0% in the RvD2 group, and 1.1% in the NA group). By comparison, the chain group and ribbon group exhibited 2.4 and 3.7% over 30 min. In particular, the vortex group reached 8.9% in 30 min, which is ∼eightfold increase in overlayed fluorescence signal compared with that in the NA group, demonstrating that vortex swarm can effectively enhance the targeted capability toward CNS injury. These in vivo distribution results indicate that the rapid lesion-directed accumulation of BRaNCs is primarily driven by magnetic navigation rather than by passive inflammatory tropism alone. In the absence of magnetic actuation, BRaNCs showed only limited enrichment at the injured site within the observation window, whereas vortex-swarm actuation markedly increased their redistribution toward the T10-T11 injury region. Therefore, the NMV/EXO membrane coating should not be interpreted as the sole or dominant source of short-term targeting in the spinal canal. Instead, the NMV/EXO coating mainly contributes to biointerface regulation, including reduced macrophage uptake, improved compatibility with CSF, reduced nonspecific adhesion, prolonged retention after magnetic delivery, and therapeutic functionality.

In case of safety concern, the distributions of BRaNCs in long period in vivo were further studied through the euthanasia of the SCI mice after the treatments and following harvest of vital tissues for analysis. The fluorescence signals of the spinal canals in all groups decreased after the treatment for 3 hours and then nearly disappeared over 24 hours except in the vortex group (Fig. 4D). Corresponding quantitative analysis indicate that the proportion in the spinal canal in the vortex group decrease from 9.0% at 3 hours to 3.1% at 24 hours, while that in the IV group was only 1.2% at 3 hours and 0.6% at 24 hours (Fig. 4E). In the following, fluorescence images of the excised vital tissues including the heart, liver, spleen, lung, kidney, and brain reveal that the IV group reached its maximum accumulation in the liver and kidneys at 3 and 6 hours, respectively (Fig. 4G). Fluorescence signal in RvD2 group reaches its maximum accumulation in the liver and kidney at 3 hours and then decayed over 24 hours, suggesting its shorter metabolic time compared with that in the IV group. By contrast, the fluorescence signals in BRaNCs in the chain, ribbon, and vortex groups exhibited prolonged period. Particularly, the vital organs in vortex group still displayed obvious fluorescence over 24 hours. The fluorescence biodistribution of BRaNCs in the vortex group over 24 hours indicated that the liver reached its highest portion of 48.4% over 6 hours, and the highest portion in kidney occurred at 12 hours (Fig. 4F). The liver in the IV group reached its higher portion of 47.5% at 3 hours, and the periodic highest portion of kidney is 34.2% at 12 hours (fig. S14). As shown in Fig. 4H, most of fluorescence from vital organs in the chain, ribbon, and vortex group are higher than that in the NA group, and the vortex group exhibited the maximum accumulation in all vital organs at 24 hours. It suggests that the magnetic navigation elevated their accumulation at the targeted injury area and prolonged the circulation time in vivo.

Last, the actual concentrations of RvD2 in the liver, kidney, and spinal canal were quantified using the triple quad mass spectrometer at 6 and 24 hours, respectively. Compared with the IV group (0.11 ng/g tissue), the RvD2 group and NA group was elevated to 0.21 ng/g tissue and 0.30 ng/g tissue at 6 hours (Fig. 4I). Magnetic intrathecal navigation further enhanced the accumulation in spinal canal (0.49 ng/g tissue in the chain group, 0.65 ng/g tissue in the ribbon group, and 1.25 ng/g tissue in the vortex group) at 6 hours, and RvD2 in these groups remained in the spinal canal over 24 hours. Collectively, the above data verified that the magnetic intrathecal navigation elevates the accumulation at targeted area in the CNS; notably, it also prolonged the circulation period for better biocompatibility in vivo. In comparison with different magnetic inputs, vortex swarm (input of H_v) is identified as optimal strategy for intrathecal navigation and used for the following in vivo studies.

Adaptive swarm behavior of BRaNCs within spinal canal in nonhuman primate in vivo

While effective in swarm navigation within spinal canal in mice in vivo, the intrathecal navigation using adaptive nanorobots in practical medicine remain unclear because of the different anatomical landscapes between mouse-scale and human-scale. Therefore, movement behavior of BraNCs in nonhuman primate was investigated. To validate the clinical scalability of BRaNCs, we engineered an intrathecal navigation platform integrating digital subtraction angiography (DSA) imaging, RARMF, and anesthesia apparatus (Fig. 5A). RARMF is a hybrid system consisting of a robotic arm, rotatable magnet, and Helmholtz multi-coil system, as exhibited in fig. S15. The Helmholtz multicoil is to rotate the magnet to create the rotating magnetic field, aiming to the swarm control of the BRaNCs, and the robotic arm was integrated with the system, and thus the swarm motion of the BRaNCs could accompany with the robotic arm with magnetic system. Experiment and simulation data indicate that the magnetic flus density in RARMF system is stable and estimated to be >5 mT in the region with diameter of ∼20 mm (fig. S16). In nonhuman primates, vortex swarm configuration, which is optimized for long-distance intrathecal navigation in mice models, was leveraged to address anatomical challenges of centimeter-scale spinal canals with submillimeter gaps (0.05 to 0.4 mm). Upon aseptic surgical procedures and lumbar puncture under anesthesia, BRaNCs were injected into the intrathecal space (the position between L6 and L7) for subsequent magnetic navigation using RARMF (fig. S17 and movie S12). As shown in time-lapse images in Fig. 5B, the signal from BRaNCs after intrathecally injection remain in position between L6 and L7, and the change of their distribution was exhibited in a relative symmetric manner and then nearly disappeared in 20 min. In contrast, the signal from BRaNCs upon RARMF (velocity of migration: ∼10 cm/hour) exhibited the change from the area of L6-L7 toward the area of L4-L3 in 20 min (Fig. 5C), verifying the migration of BRaNCs with travelling distance of ∼8 cm. Intrathecal navigation of BRaNCs in nonhuman primate displayed more anatomical details compared with that in mice. To this end, the images were transformed into colorful images as the signal intensities to further investigate the intrathecal behavior of BRaNCs. As shown in Fig. 5D, BRaNCs upon RARMF in 0 to 10 min exhibited the upward distribution along the right side of spinal canal between L6 and L5. Subsequently, the distribution was changed to both sides of spinal canal between L4 and L3 from 10 to 20 min. Note that the subsequent behavior of BRaNCs was tough for tracking and monitoring because of the limited area for observation.

Fig. 5. — (A) Schematic (left) and actual (right) setup for the application of BRaNCs strategy in nonhuman primates. (B) DSA images showing the diffusion of BRaNC with NA within the spinal canal over 20 min. (C) DSA imaging–guided swarm control of BRaNC within spinal canal in nonhuman primate. The swarm formation and migration were accomplished upon the magnetic field input of H_v. (D) Illustrations (top) and colorful DSA images (bottom) showing the dynamics of BRaNC in different positions over 20 min. (E) Corresponding normalized number of BRaNCs counted from the changed signals in (C), exhibiting the distribution of BRaNCs in spinal canal. (F and G) Illustration (left), enlarged colorful DSA images (middle), and quantitative hot spots (right) at L3-L4 level at 0 min [(F) yellow line] and 20 min [(G) green line], respectively, which were indicated as dashed-line rectangle. Scale bars, 1 cm.

Considered were the bundle nerves (e.g., sensory and motor) of the cauda equina, which are located at the end of the spinal cord; roots at each disc level within the thecal sac were filled with CSF in the subarachnoid space (75). The adaptive swarm motion within spinal canal are indicated as schematic illustration in Fig. 5D. BraNCs upon magnetic field input of (H_v) exhibit vortex swarm in CSF on the side of the spinal canal at L6-L5 level, and then the vortex swarm conducted the adaption motion toward both side of spinal canal at L4-L3 level because the anatomy of spinal cord become more crowded at the L3-L4 level. As shown in Fig. 5E, BRaNCS upon vortex swarm input exhibit migration from the L6 to the L3 of the spinal canal over 20 min. At 0 min, there were 78.3 and 5.9% BRaNCS distributions at the L6 and L3 levels, respectively. At 10 min, the distribution of BRaNCS in the L5 area was the highest, reaching 54.9%. At 20 min, the total distribution of BRaNCS at L4 and L3 were 18.2 and 58.0%, respectively. To further investigate the adaptive swarm of BRaNCs within spinal canal, the area between L4 and L3 at 0 and 20 min were enlarged. Noted that the dynamics of BRaNCs was also studied (fig. S18), showing the similar behavior as that in nonhuman primate in vivo. The profile of the spinal cord cannot be observed without the arrival of BRaNCs, and the quantitative analysis with hot spots displayed the relative uniform distribution within spinal canal, indicating that minor contrast from the spinal cord (Fig. 5F). By contrast, the profile of the spinal cord was clear after the arrival of BraNCs with vortex swarm at 20 min, and the corresponding quantitative statistical heatmap exhibited the contrast with nearly linear trajectory, suggesting that BRaNCs with vortex swarm carried out the adaption behavior because of the limited space between spinal cord and spinal canal (Fig. 5G). Such adaptive swarm navigation in nonhuman primate spinal canal demonstrated substantial promises in addressing anatomical constraints for navigated intrathecal delivery in human scale.

Therapeutic evaluation using adaptive nanorobotic delivery with BRaNCs in vivo

Followed by the verification of the adaptive swarm motion within spinal canal in both mice and nonhuman primate, the therapeutic effects using adaptive nanorobotic delivery with BRaNCs upon vortex swarm navigation in vivo were thus evaluated with SCI mice. The SCI model in vivo was established with surgical operation toward CNS, and the therapy protocol for thoracic spine injury as one example of CNS injury and the test of relevant indicators were conducted in the following days for therapy (fig. S19). The mice were randomly divided into seven groups with CNS injury treatment, namely, healthy mice as background control (sham group), SCI mice with equivalent amounts of saline (SCI group), BRaNCs through intravascular administration (IV group), RvD2 through intrathecal administration (RvD2 group), NMVs through intrathecal administration (NMV group), intrathecal BRaNCs with NA (BRaNC-NA group), and intrathecal BRaNCs with magnetic adaptive vortex swarm motion (BRaNC-vortex group), respectively. Compared with the sham group, hematoxylin and eosin (H&E) staining for pathological examination showed obvious inflammatory cell infiltration and vacuolization of the mice in the SCI group, which indicate the generation of inflammatory and nerve injury. The IV, RvD2, NMV, and BRaNC-NA groups showed attenuated vacuolization and necrosis of the spinal cord. By comparison, the optimal nerve recovery and negligible necrosis in the mice occurred in BRaNC-vortex group (Fig. 6A), suggesting that enhanced delivery of BRaNCs with vortex swarm navigation has improved the pathological condition of nerve inflammation and attenuated spinal cord tissue damage (76). The injured area of the spinal cord formed cavities, which was used to evaluate the lesion area. Nearly the same appearance with severe cavitation were observed in IV group, suggesting the low therapeutic effect through the intravascular administration (fig. S20). In contrast, the groups using passive intrathecal administration including RvD2, NMV, and BRaNC-NA groups showed the smaller size of lesion area. In particular, the lesion area in BRaNC-vortex group showed the smallest area, which is comparable to that in the sham group.

Fig. 6. — (A) H&E staining images of the spinal tissues after the treatment of different groups. Scale bars, 200 μm (top) and 20 μm (bottom), respectively. (B) Immunofluorescence staining images of IBA-1, TUJ1, and 4′,6-diamidino-2-phenylindole (DAPI) in the spinal cords after various treatments. Scale bar, 100 μm. (C) Quantification of neuronal cells (TuJ1 + cell counts, left) and microglia cells (IBA-1 + cell counts, right) through the analysis from (B) using ImageJ, showing inflammation suppression and neurological functional recovery. (D) Number of neutrophils in CNS injury of SCI mice after various treatments. (E) Relative expression of proinflammatory TNF-α, IL-1β, IL-6, and IL-10 in Sci mice after different treatments. (F to H) BMS scores (F), body weight (G), the survival rate of SCI mice after various treatments. Data are means ± SEM; (n = 5); (*****P <* 0.0001, ***P < 0.001, **P < 0.01).

To further evaluate the neuroinflammation of CNS injury mice upon different treatments, the spinal cords were taken for immunofluorescent assay and analysis after the treatment for 15 days. The green fluorescence from IBA-1 staining channel and red fluorescence from TUJ1 staining, representing the microglia (IBA-1–positive cells) and neurons (TUJ1-positive cells) in the spinal cord tissue of SCI mice, occurred (Fig. 6B). The obvious fluorescence in IBA-1 channel in SCI group indicate that the neuroinflammation and the low fluorescence in TUJ1 imply the injury of nerve. Notably, the SCI mice in the BRaNC-vortex group exhibited minor fluorescence in IBA-1, meanwhile obvious signal from TUJ1 channel, implying its alleviation of neuroinflammation and neurons recovery of CNS injury through the adaptive nanorobotic intrathecal delivery. The number of TUJ1-positive cells and IBA-1–positive cells in BRaNC-vortex group is comparable to that in the sham group (Fig. 6C), suggesting their suppression of neuroinflammation and neurons recovery. Among various roles in neuroinflammation, neutrophils are the first-line cells of the immune response, and their elevation indicate the inflammation or infection in the body (77, 78). To further investigate the mechanisms of adaptive nanorobotic therapy in neuroprotective and anti-inflammatory, the blood from different therapy groups for hemocytometry examination. As shown in Fig. 6D, the number of neutrophils in SCI group is ∼threefold than that of healthy mice, which is attributed to the massive recruitment of neutrophils by the immune response. In contrast, BRaNC-vortex treatment brought the counts of neutrophil close to that in the Sham group, and it may be attributed to the vortex swarm–triggered release of the compositions including RvD2, neutrophil membranes, and EXOs in large quantities at the injury site. In addition, macrophages are heterogeneous cells with extensive functional plasticity that have been divided into M1 and M2 types (79). Most macrophages in the injured spinal cord are M1 macrophages, and only a few transient M2 macrophages exist (80). The dominance of M1 macrophages and the decreased number of M2 macrophages after SCI aggravates the injury (81). M1 macrophages produce proinflammatory cytokines that promote tissue inflammation and injury. In contrast, M2-type macrophages typically produce anti-inflammatory factors that reduce the ability to produce proinflammatory molecules at the site of injury, leading to tissue remodeling. Macrophages can switch from one phenotype to another, which is induced by inflammatory factors following injury or infection (82, 83). Such alleviated inflammation upon the treatment in the BRaNC-vortex group also resulted in decreased secretion of proinflammatory cytokines including tumor necrosis factor–α (TNF-α), interleukin-1β (IL-1β), IL-6, as well as increased expression of anti-inflammatory cytokine IL-10 (Fig. 6E). This suggests that the protective functions of BRaNCS and the expression of inflammatory factors are related to the regulation of the microglia/macrophage-activated phenotype (76). The above results suggest that the treatment can effectively inhibit the neuroinflammation of spinal cord injured tissues, promote nerve repair, and improve the motor function of mice in CNS injury.

Next, the recovery of neurologic functions, including sensory-motor and cognitive function, are essential for CNS injury treatment. Basso mouse scale (BMS) scores to assess the recovery of motor function in SCI mice were performed at 1, 3, 5, 7, 9, 11, 13, and 15 days after CNS injury (Fig. 6F). Compared with 2.3 point in the SCI group, the BRaNC-vortex group had the substantially highest score (9.5) than the other groups, indicating BRaNCs upon vortex swarm navigation could substantially improve behavioral function and promote the recovery of neurological function in mice with CNS injury. The body weights of mice in each group were closely monitored daily to reveal the protective effect of the CNS after SCI. All groups showed substantial weight loss in the first 3 days after SCI. This may be due to the decline of motor function caused by CNS injury. The average weight in SCI group showed a steady decline, which decreased to 12.6 g, while other groups showed moderate weight recovery (Fig. 6G). In contrast, BRaNC-vortex–treated SCI mice showed the considerable weight recovery among all groups, with weight increasing to 22.1 g. The survival rates showed all SCI mice in BRaNC-vortex group alive in 15 days, compared with that in other groups (37.5% in SCI group, 50.0% in IV group, 57.1% in RvD2 group, 64.3% in NMV group, and 83.3% in BRaNC-NA group, respectively), further verifying the protective effects of adaptive nanorobots using vortex swarm of BRaNCs against CNS injury (Fig. 6H).

Therapy and safety comparison using adaptive nanorobotics with BRaNCs

Considering the elevated delivery efficiency with adaptive nanorobotic delivery strategy using BRaNCs, the therapeutic efficacy toward CNS injury using BRaNCs upon vortex swarm were compared with different doses through conventional intravascular administration. To this end, we divided the SCI mice into six groups including intrathecal injection of BRaNCs upon vortex swarm (vortex), intravascular administration of BRaNCs with equal dose (1× IV), 3-fold doses (3× IV), 6-fold doses (6× IV), 9-fold doses (9× IV), and 12-fold doses (12× IV), respectively. Evaluation of the effect of the different treatments using H&E staining, which displayed the spinal cord vacuolation and inflammatory infiltration, revealed that the lesion size in CNS injury gradually recovered with the elevated doses from 1× IV to 9× IV (Fig. 7A and fig. S21). Negligible difference between 9× IV, 12× IV, and vortex group suggest their similar therapeutic efficacy.

The spinal cords in different groups were further conducted immunofluorescence detection. As shown in Fig. 7B, the distribution of microglia and neuronal cells in the spinal cord tissues in vortex group is similar as that in 9× IV group and 12× IV group. Moreover, the level of TUJ1, IBA-1, and Neu in vortex group is better than that in 1× IV group, 3× IV group, and 6× IV group, and was close to healthy mice (Fig. 7, C to E). TNF-α, IL-1β, IL-6, and IL-10 in vortex group is obvious lower than that in 1× IV group, 3× IV group, and 6× IV group, and close to that in 9× IV group and 12× IV group (Fig. 7, F to I). Moreover, as shown in Fig. 7J, BMS scores in vortex group reached 9.5 after the treatment for 15 days, which is higher than that other groups (3.2 in 1× IV group, 4.0 in 3× IV group, 6.0 in 6× IV group, 8.5 in 9× IV group, and 5.7 in 12× IV group, respectively). Noted that the therapeutic effect in 12× IV group is lower than that in 9× IV group, suggesting their side effects with high dose. In addition, the SCI mice in vortex group exhibited the highest stability in survival period and fastest growth in body weight compared with other groups (Fig. 7, K and L).

Furthermore, tissue and blood after the different treatments were collected, and the biosafety was evaluated using serum and major organs. H&E staining analysis for the major organs showed minor pathological abnormalities or inflamed cells in the brain, heart, liver, spleen, lung, or kidney in the treated BRaNC-vortex groups (Fig. 7M and fig. S22). In contrast, the elevated activities of aspartate transferase, alanine transaminase (ALT), and alkaline phosphatase in RvD2 group and 9× IV group indicate their hepatotoxicity from intravascular injection with high dose (Fig. 7N and fig. S23). The higher level of urea and urine acid in 9× IV group compared with that in other groups reply their renal toxicity (Fig. 7O and fig. S24). Also, the highest level of the creatine kinase (CK), white blood cell, and lymphocyte in 9× IV group among all groups suggest the immune response caused by their high dose through intravascular administration (Fig. 7, P to R). The above data indicate that adaptive nanorobotic navigation induces a substantial therapeutic efficacy to CNS injury, which is comparable to 9-fold to 12-fold dose of BRaNCs without magnetic actuation through intravascular administration. However, the administration of BRaNCs with more than ninefold dose lead to dramatic hepatotoxicity and renal toxicity. Conversely, accomplishing substantial therapeutic effects, the intrathecal delivery of BRaNCs with vortex swarm shows negligible toxicity. Combined, the adaptive nanorobotic intrathecal delivery can not only facilitate therapeutic effect compared with passive BRaNCs with ∼10-fold dose but also cause minimal side effects to the body.

DISCUSSION

CNS diseases remain a major therapeutic challenge largely due to the impermeability of the blood-brain barrier and the complex anatomical constraints of the spinal canal. Both barriers have long hindered the development of targeted intrathecal delivery systems. While nanomedicine has shown promise in preclinical models for enhancing drug bioavailability, its translation to clinical practice has been hindered by two critical gaps: the inability to replicate human-scale spinal anatomy in rodent models, and the lack of programmable, adaptive platforms that can navigate the dynamic CSF environment while maintaining therapeutic efficacy. This study addresses these unmet needs by developing an adaptive magnetic nanorobotic platform, and its findings offer transformative insights into the future of CNS therapeutic delivery.

The core strength of this work lies in its ability to bridge preclinical validation and clinical relevance, supported by three interconnected discoveries. First, the validation in nonhuman primate, coupled with DSA imaging tracking, resolves the longstanding “mouse-to-human” scale gap in nanorobotic research. The nanorobotic swarms in nonhuman primate migrate along the gaps between cauda equina bundles before reaching the spinal cord, which directly mimics the anatomical constraints of the human spinal canal, proving that the platform can adapt to the complex, nonuniform spaces that would be encountered in clinical settings. The DSA imaging tracking further adds value by enabling noninvasive monitoring of swarm navigation, a critical feature for future clinical translation as it allows for dose adjustment and safety oversight. Although the feedback control is limited in living imaging, the current magnetic swarm control can still offer enough precision for intrathecal navigation considered the one-way direction in spinal canal. Second, the identification of vortex swarms as the optimal configuration for intrathecal navigation addresses a key limitation of earlier nanorobotic systems: the lack of context-dependent mobility. While chain-like and ribbon-like swarms were achievable, only vortex swarms balanced high-density aggregation (to avoid dilution in CSF) with flexible navigation (to bypass anatomical barriers like nerve bundles), which is crucial for intrathecal navigation of nanorobots. The space in the spinal canal is limited, and tissue damage must be minimized; a one-size-fits-all swarm mode would fail to adapt to varying anatomical features (e.g., narrower regions near the cervical spine versus wider lumbar regions). The capability of vortex swarm for self-organization in response to both magnetic inputs and local anatomy ensures that therapeutic payloads reach target sites rather than being trapped in nontargeted tissues. Third, the substantially elevated therapeutic efficacy compared to conventional passive delivery, which is coupled with prolonged in vivo retention and minimal side effects, directly balance efficacy with safety. Passive intrathecal delivery often requires high doses to achieve therapeutic levels, increasing the risk of neurotoxicity or systemic side effects.

In addition, the biomimetic design also offer structural advantages in enhanced targeting efficacy and neuroprotection compared with simple magnetic particles and RBC membrane–coated magnetic NPs controls. The NMVs and EXOs as typical biomimetic components exert various structural advantages including reduction of immunogenicity by inhibiting phagocytosis, enhances targeting, and prolongs payload retention in CNS. They are attributed to the neutrophil adhesion protein CXCR4 mediates binding to inflamed endothelial cells in CNS injury and exosome tetraspanin CD81 facilitate the penetration of neural tissue, respectively. In addition, the biomimetic design offers biointerfacial advantages compared with simple magnetic particles and RBC membrane–coated magnetic NPs controls under nonactuated conditions. The therapeutic advantage of BRaNCs arises from the coupling of magnetic navigation with biomimetic surface engineering. Magnetic actuation provides the main directional driving force for rapid long-distance intrathecal transport, whereas the NMV/EXO coating improves biological compatibility, reduces macrophage uptake and nonspecific adhesion, supports CSF-compatible swarm behavior, and contributes to payload retention and therapeutic function after magnetic delivery, rather than acting as the sole driver of short-term lesion targeting. These functions may be associated with neutrophil membrane proteins such as CXCR4 and exosomal tetraspanins such as CD81, but the rapid lesion-directed accumulation observed here is mainly enabled by magnetic navigation. Combined with biomimetic swarm with NMV/EXO coating for CNS targeting, multimode topology switching for adaptive navigation, validation in nonhuman primates for intrathecal delivery, which was summarized as table S2 (58, 74, 84), our platform enhance accumulation at CNS targets and extending retention time and reduces the required dose while maintaining or improving efficacy.

Despite the progress, this study has limitations that warrant further investigation. First, the number of nonhuman primate for investigation was limited, and, thus, the long-term safety data (e.g., nanorobot retention over months) remain to be collected. Future work will expand nonhuman primate studies to include larger cohorts and long-term follow-up, with a focus on nanorobot clearance and tissue compatibility. Second, the magnetic control system, including the imaging system, while effective in preclinical settings, is still required the miniaturized design. Third, future studies should test the platform in nonhuman primate models of specific CNS diseases (e.g., spinal cord contusion and glioma) to evaluate performance in pathological environments, where tissue edema or tumor mass may alter spinal canal anatomy. Another limitation is that RBC membrane–coated Fe₃O₄ NPs were not evaluated under the same vortex magnetic navigation condition in vivo. Therefore, the present study cannot determine whether RBC-coated magnetic NPs would also exhibit lesion-directed accumulation when exposed to identical magnetic actuation. Future studies should include magnetically actuated RBC-coated controls to further distinguish the contributions of magnetic force, membrane composition, circulation or retention behavior, and therapeutic payload.

This study represents a critical step toward clinical translation of nanorobotics for CNS therapy. By resolving the mouse-to-human scale gap, identifying an optimal swarm mode, and demonstrating enhanced therapeutic efficacy, the platform addresses key barriers that have slowed nanomedicine translation. Thoracic spine injury in the study is served as an example for therapy of various disease toward CNS using active nanorobotics. The units consisting of BRaNCs includes the neutrophil membrane vesicles, stem cell EXOs, and RvD2, which are widely acknowledged as therapeutic agents for CNS injury. In this study, they were used as example of biomimetic nanocomplex, and these units could be substituted as specific requirements. Their applications could be expanded to a wide range of CNS diseases such as Alzheimer’s disease, meningitis, cerebral infarction, and virus infection. Together, merging adaptive nanorobotics with biomimetic nanotechnology, BRaNCs bestow swarm intelligence for navigated intrathecal delivery, offering hope for improved outcomes in CNS diseases and potentially other complex medical conditions.

MATERIALS AND METHODS

Synthesis of Fe₃O₄ particles

Initially, 250 ml of deoxidized water was combined with nitrogen. Subsequently, 0.9 g of FeCl₂ and 1.4 g of FeCl₃ were dissolved in 40 ml of deoxidized water and heated to 80°C with stirring in a nitrogen pan. Subsequently, 5 ml of ammonia was added, and the mixture was stirred for 30 min. The sample was rinsed with a magnet thrice, and 2 ml of a sodium citrate solution (0.5 g/ml) was added. The sample was heated at 95°C for 90 min and cooled to room temperature under a nitrogen atmosphere. The sample should be rinsed with a magnet, and the nitrogen should be replenished and then transferred to a small tube for storage.

Synthesis of NP core

The NP cores were synthesized by a typical emulsion/solvent evaporation method (67, 68). A solution of 50 mg of lactic-co-glycolic acid in 2 ml of ethyl acetate was prepared, and 1 μg of Cy7-NHS fluorescently stained RvD2 was added. Once complete dissolution has been achieved, 1 ml of iron tetroxide NPs should be added, and the solution should be shaken at 2000 rpm for 1 min. The solution should then be sonicated in an ice bath with an ultrasonic cell crusher (XM-650 T, Xiaomei Ultrasound Instrument, China) for 3 min at 1000 W output. The solution is then added to a 3% poly (vinyl alcohol) solution, fixed at 15 ml, and sonicated in an ice bath at 1000 W for 15 min. Subsequently, 1 ml of isopropanol is added to remove all air bubbles. The solution was mechanically stirred at 600 rpm for 48 hours. Then, NP cores were washed by centrifugation at 10,000g for 10 min, sealed, and stored at 4°C.

Isolation of neutrophils

Following the induction of anesthesia and euthanasia of mice, the femur and tibia were separated, muscles were dissected, ends of bones were trimmed, and marrow was dislodged from the bones using a needle. The cell suspension was filtered into a 50-ml centrifuge tube and centrifuged at 500g for 5 min. The supernatant was carefully removed, and 1.5 ml of PBS was added to resuspend the cell pellet. Subsequently, 2 ml of 80% Percoll (Biosharp, BS909, China), 1.5 ml of 65% Percoll, and 1.5 ml of 55% Percoll were added to a 15-ml centrifuge tube from the bottom to the top, and the tube was left for 5 min. The bone marrow cell suspension was then added to the top layer and centrifuged at 1200g for 30 min. The 55% Percoll layer was removed, and the cells between the 80 and 65% layers were collected. A total of 14 ml of PBS was added, and the mixture was thoroughly mixed. Centrifugation at 500g for 5 min was then performed, after which the supernatant was discarded. Three milliliters of prediluted erythrocyte lysate was added, and after 5 min, 14 ml of PBS was added. The mixture was then centrifuged at 500g for 5 min, after which the supernatant was discarded. The cells were then suspended in the neutrophil medium for cell counting.

Isolation of NMV

To obtain activated neutrophils, lipopolysaccharide (100 ng/ml) was added to neutrophil cultures, incubated for 4 hours, and washed with PBS. The cells were then suspended in ice-cold extraction buffer-1 (IB-1), which contained 30 mM tris-HCl (pH 7.5), 225 mM d-mannitol, 75 mM sucrose, 0.2 mM EDTA, and a mixture of protease and phosphatase inhibitors (Beyotime, P1045, China). The mixture was homogenized 50 to 100 times in a double homogenizer. Subsequently, it was centrifuged at 800g for 10 min at 4°C to remove nondestructed cells and nuclei. Last, the mixture was centrifuged at 10,000g for 20 min at 4°C, with the sediment discarded and mitochondria removed. The supernatant should be centrifuged once more at 100,000g for 1 hour at 4°C. The precipitate comprising cytoplasmic membranes is then washed with 10 mM tris-HCl and 0.5 mM EDTA, containing protease inhibitors, and subsequently frozen, weighed, and stored at −80°C. The protein content of the membranes was quantified using the BCA protein assay kit (Beyotime, P0010, China).

Isolation and culture of MSCs

First, mice aged between 3 and 4 weeks were euthanized. Subsequently, the bilateral femurs and tibias were excised and immersed in Hank’s solution. The muscle and connective tissue surrounding the femur and tibia were meticulously excised and rinsed in Hank’s solution. Before placement in the cell culture chamber, the bones were sterilized in 75% ethanol. The bones were then placed in a 10-cm-diameter petri dish and washed twice with Hank’s solution. Once more, the muscle and connective tissue were removed to ensure cleanliness. The ends of the femur and tibia were excised with scissors to expose the marrow cavity. The bone marrow was repeatedly washed with a 1-ml syringe until the bone was white. Bone marrow fluid was collected and centrifuged at 300g for 5 min. The supernatant was discarded, and if the precipitate was present in large amounts, it was first dispersed by tapping lightly. The precipitate was suspended in 2 ml of PBS and mixed thoroughly to ensure uniform distribution of the cells. A prechilled erythrocyte lysate (PBS to erythrocyte lysate ratio 4:1) was immediately added, resuspended, and shaken vigorously three to five times over 2 min. After 2 hours, 5 ml of PBS was added to the lysate to achieve a dilution. The solution was inverted, resuspended, and then centrifuged at 300g for 5. A volume of 5 ml of prepared medium, containing 15% serum and 5% double antibody (100-fold concentration of double antibody), was added, and the contents were vortexed gently to resuspend the cells. The suspension was then subjected to centrifugation at 300g for 3 min. The washing process was repeated twice. Last, the cells were transferred to a fresh medium and incubated. After 24 to 48 hours, the suspended cells were removed, and the medium was replaced every 3 days until the cells had undergone several passages.

Isolation of EXO

Once the adipose stem cells reached 80% confluency, the serum was replaced with exosome-free serum. After 48 hours of incubation, the cell culture supernatant was collected. To remove live and dead cells, the cells were first centrifuged at 3000g for 15 min at 4°C. The supernatant was removed, and 1.25 ml of EXO extraction kit (Rengen Biosciences, EXORG24B-1, China) was added, mixed thoroughly, and allowed to stand at 4°C for at least 30 min. Subsequently, the samples underwent centrifugation at 12,000g for 30 min at 4°C. The supernatant was subsequently removed. A white precipitate was observed at the bottom of the test tube after discarding the supernatant. This precipitate was then resuspended in an appropriate amount of PBS. The resuspended EXOs were subsequently transferred to a purification column and placed in a 1.5-ml recovery column. The purification column was then discarded, and the recovery column, containing the isolated EXOs, was stored at −80°C as a reserve.

Synthesis of hybrid vesicles (NMV + EXO)

For the synthesis of hybrid vesicles (NMV + EXO), the mixture of NMV (1 mg/ml) and EXO (1 mg/ml) was sonicated for 30 s at 100-W output on ice. The mixture was sonicated for 30 s (sonication for 1 s at 2-s intervals) with an ultrasonic cell crusher (XM-650T, Xiaomei Ultrasound Instrument, China). The hybrid vehicles were then sequentially extruded through 1000-, 800-, and 400-nm porous membranes by liposome extruder (Mini-extruder, Avanti Polar Lipids, USA) to create neutrophil membrane and stem cell exosome hybrid nanovesicles.

Synthesis of BRaNCs

The NP cores were mixed with hybrid vesicles at a 1:1 polymer-to-protein weight ratio. The NP cores were fused with the hybrid nanovesicles using a coextrusion method by liposome extruder (Mini-extruder, Avanti Polar Lipids, USA) through polycarbonate membranes with successive pores of 1 μm, 800 nm, 400 nm, and 200 nm. The mechanical force during extrusion promoted the fusion of hybrid vesicles with NP cores, leading to the formation of BRaNCs.

Characterization of BRaNCs

BRaNCs were isolated by centrifugation at 10,000g for 10 min and washed three times to fully remove free fluorescence dye and membrane vesicles. The hydrodynamic size and zeta potential of the NP core and BRaNCs were measured both before and after hybrid vesicle coating by dynamic light scattering using nanometrics (Zetasizer Nano ZS, Malvern Panalytica, England). The morphology of NP core and BRaNCs was acquired and visualized by transmission electron microscope (TEM-H7650, Hitachi, Japan), stained with 0.2 wt % uranyl acetate dihydrate (Acmec, U25690, China). The magnetic hysteresis curve of three samples (NP core and BRaNCs) was obtained using the vibrating sample magnetometer (VSM 7404, Lake Shore, USA).

Drug release study

The drug release profiles from BRaNCs were assessed in PBS at pH 6.5 and pH 7.2 and incubated for 3, 6, 12, 24, and 48 hours, respectively. Subsequently, the supernatant was collected and centrifuged at 3000 rpm for 5 min to remove any unrelated substances. Last, RvD2 concentrations in the supernatant were quantified by HPLC (E3000, Thermo Fisher Scientific, USA). Furthermore, the drug release of BRaNCs swarms of different magnetic modalities was also conducted.

Identification of proteins of NMV and EXO

BRaNCs were purified from unbound proteins or membrane fragments by centrifugation at 16,000g. For protein analysis, NP core, NMV, EXO, and BRaNCs were centrifuged at 13,000g for 5 min at 4°C, and an extended BCA protein assay was performed to quantify the total protein content. The protein supernatants were then mixed with SDS buffer and heated at 100°C for 5 min. The electrophoresis was conducted using a gel electrophoresis system on a 6% tris/glycine-SDS–polyacrylamide gel. For blotting analysis, the proteins were transferred to cellulose acetate membranes and sealed in Tris-buffered saline with Tween-20 (containing 5% skim milk) for 1 hour at room temperature. The membranes were incubated at 4°C overnight with antibodies against CXCR4 (Bioswamp, PAB31318, China) and CD81 (Baijia, IPB0164, China), followed by incubation with the corresponding horseradish peroxidase–conjugated rabbit antibody (Bioss, bs-0295G-HRP, China). Subsequently, images were obtained by gel imaging system (Amersham Imager 600, GE, USA).

Macrophage uptake study

Mouse RAW 264.7 cells in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum and penicillin-streptomycin. The NP cores and BRaNCs were incubated with the cells at a final concentration of 0.5 mg ml⁻¹ for 30 min at 37°C. The media were then replaced with fresh media, and the cells were incubated for another 30 min. Afterward, the cells were washed three times with PBS and detached by scraping. The figures were observed by using optical microscopy (Axio Vert A1, Zeiss, Germany). After the observation, macrophages were collected and lysed, and the content of engulfing iron was determined by plasma-coupled mass spectrometry (ICP-MS 7700×, Agilent, Australia).

Multimodal emergence and transformations of the BRaNCs swarm

Model artificial CSF (aCSF) was prepared following the previously established experimental protocol (85). By increasing the concentration of ions, threefold ionic strength aCSF (3× CSF) and fivefold ionic strength aCSF (5× CSF) were generated. Conventional Fe₃O₄ NPs were used as control, and the motion of BRaNCs was evaluated in aCSF and PBS in the microporous structures. For motility characterization, the BRaNC motion can be generated by using the three-axis Helmholtz coil and be recorded with an inverted microscope (IX73P1F, Olympus, Japan). The BRaNC was placed in the center of the Helmholtz coil, and the emergence and dynamic conversion of different swarms were realized by adjusting the magnetic mode. The transition of different swarm configuration was conducted through the manipulation of magnetic field inputs. Briefly, the transition from vortex swarm to chain swarm was accomplished by changing the magnetic inputs from H_v to H_c and the frequency of 24 Hz. In addition, the transition from chain swarm to ribbon swarm is achieve through the manipulation of magnetic field input from H_c to H_v. All transitions of different swarm configurations were completed within 5 s, and no hysteresis was observed during the adjustment of magnetic field inputs. The velocity and formation time of swarms were changed by adjusting the frequency of the magnetic field. ImageJ was used to accurately calculate the characteristic parameters of the robot, such as movement velocity and swarm size.

Controllable intrathecal behavior of BRaNCs

To display the intrathecal behavior of BRaNCs swarm, an ex vivo spinal cord model was constructed using microchannels and isolated spinal cords, respectively. The microchannel is composed of transparent plastic pipes. The outer ∼4.5-cm pipe simulates the spinal canal, and the inner ∼2-cm pipe simulates the spinal cord. About 6 cm of wire is placed in it to simulate the spinal cauda equina. The microchannel was prefilled with aCSF or PBS to create an aqueous environment. Both traditional Fe₃O₄ NPs and Fe₃O₄ NPs coated with RBC membranes were used as controls. A 100 μl solution of BRaNC was added to one side of the microchannel. The added BRaNC was a robotic-arm integrated onto the magnetic field system to accomplish long-distance movement of BRaNCs. The motion display was divided into the intrathecal injection of the robot, swarm formation, and adaptive swarm migration. Furthermore, the isolated spinal cords were observed in mice, and controllable intrathecal behavior of BRaNCS was recorded at different time points using an in vivo imaging system (IVIS Spectrum, PerkinElmer, USA).

Sources of experimental animals

The experimental animals are Balb/c female mice and nonhuman primates. All the animals were conducted in accordance with the Guide for Care and Use of Laboratory Animals. Mice and nonhuman primates experiments were permitted by the Experimental Animal Welfare Ethics Committee of Harbin Institute of Technology (no.: IACUC-2024114) and Beijing Life Biosciences Co., LTD (no.: LFSW-DWSY-2025-03-21-01), respectively.

CNS injury model establishment

The thoracic spine injury model as one example of CNS injury was established according to our previous report (86). Animals were anesthetized using 2.5% isoflurane gas with oxygen. A laminectomy was performed to expose the spinal cord at the T10-T11 spinal level. A severe contusive injury was performed using the infinite horizon spinal cord impactor system and a dwell time of 60 s. After the lesion, the skin was sutured using 9-mm wound clips, and the animals were recovered on a heating pad to maintain body temperature. Bladders were manually expressed daily.

Biodistribution and spinal cord–targeting studies of BRaNCs

For the biodistribution and spinal cord–targeting studies, the Cy5 fluorescence was used to mark the injury site. After the CNS injury, the mice were injected intravenously with 10 μl of CY7-NHS–loaded BRaNCs (1 mg/ml). The six independent in vivo experiment groups: intravenous injection of BRaNCs (IV group); intrathecal injection of RvD2 (RvD2 group); intrathecal injection of BRaNCs with NA (NA group); and intrathecal injection of BRaNCs with magnetic control input of H_c (chain group), H_r (ribbon group), and H_v (vortex group) with robotic-arm through intrathecal injection. Each of the experiments had five animals of the exact same age. The dynamic changes of BRaNCs in mice at different times were studied through the in vivo imaging system (IVIS Spectrum, PerkinElmer, USA). At specified time points after the injection, the animals were euthanized. After perfusion with PBS, organs including heart, lung, liver, spleen, kidney, and spinal cord were photographed photos by in vivo imaging system (IVIS Spectrum, Perkin Elmer, USA). Afterward, these organs were carefully collected and homogenized for biodistribution analysis of RvD2 with an HPLC (E3000, Thermo Fisher Scientific, USA). All replicates represent different rats subjected to the same treatment (n = 5).

Controlled behavior of BRaNC-vortex in nonhuman primates

To address anatomical challenges of centimeter-scale spinal canals with submillimeter gaps (0.05 to 0.4mm), we engineered a closed-loop intrathecal navigation platform integrating DSA imaging, RARMF, and anesthesia apparatus. First, the monkey was anesthetized, and treatment began after the monkey’s breathing was stable. All treatments were performed in a sterile environment. After SCI was established, a magnetic field was turned on, and treatment was performed. Sodium pantothenate was incorporated into BraNCs, attempting to elevate the performance of DSA imaging. In nonhuman primates, vortex swarm configuration is optimized for long-distance intrathecal navigation in murine models.

CNS therapy of adaptive nanorobotic delivery with BRaNCs

After the SCI model was established, the mice were randomly assigned to seven groups (n = 5 per group): Sham, SCI, IV, RvD2, NMV, BRaNC-NA, and BRaNC-vortex. On the 3rd, 5th, 7th, 9th, 11th, 13th, and 15th days, the drug was repeatedly injected. The dose of 1 mg/ml BRaNC per mouse was 10 μl. After undergoing various treatments, the mice were euthanized and dissected to assess disease progression. Each mouse’s whole blood was collected. Plasma samples were used for neutrophil detection, while serum samples were used to test the levels of inflammatory cytokines, such as IL-1β, IL-6, IL-10, and TNF-α. Spinal cords were collected and preserved in 4% paraformaldehyde for histological analysis. Furthermore, the body weights and survival of the mice were monitored every 2 days. Furthermore, the therapeutic efficacy toward CNS injury using BRaNCs upon vortex swarm was compared with different doses through intravascular administration. The mice with CNS injury into six groups, including intrathecal injection of BRaNCs upon vortex swarm (vortex), intravascular administration of BRaNCs with 3-fold doses (3× IV), 6-fold doses (6× IV), 9-fold doses (9× IV), and 12-fold doses (12× IV), respectively. Treatment protocols were consistent with those previously described.

In vivo safety studies of BRaNCs

For in vivo biosafety studies, 8-week-old female Balb/c mice were divided into groups: intrathecal injection of PBS (sham group), intrathecal injection of RvD2 (RvD2 group), intrathecal injection of BRaNCs with NA (BRaNC-NA group), and intrathecal injection of BRaNCs with magnetic control input of H_v (vortex group) with robotic-arm through intrathecal injection, intravenous injection of 1× BRaNCs (1× IV group), and intravenous injection of 9× BRaNCs (9× IV group). Each of the experiments had five animals of the exact same age. From days 0 to 7, mice in different groups received daily administration per day. At the end of the study (day 8), the mice were euthanized for sample collection. For the histological analysis, the collected heart, liver, spleen, lungs, kidneys, and brain samples were carefully sectioned and stained with H&E. After staining, the samples were imaged using an inverted microscope (Axio Vert A1, Zeiss, Germany). Each mouse’s whole blood was collected. Plasma samples were used for immune cell detection, while serum samples underwent detailed analysis by a biochemical analyzer (BS-240 Vet, Mindray, China). The concentration of alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, uric acid, urea nitrogen, and CK in the serum samples was quantified.

Statistical analysis

Quantitative data were illustrated as means ± SD. Comparison between multiple groups was performed using one-way analysis of variance (ANOVA) with a Tukey post hoc test. Statistical significance of the results was represented as *P < 0.05, **P < 0.01, and ***P < 0.001

Acknowledgments

We thank X. Chen from Harbin Institute of Technology for laboratory assistance.

Funding:

This work was financially supported by National Key Research and Development Program of China, grant no. 2024YFB4708500 (Z.W.); National Natural Science Foundation of China, grants 52375565 (Z.W.) and 62422313 (X.F.); State Key Laboratory of Robotics and Systems (HIT), grant SKLRS202502B (Z.W.); State Key Laboratory of Systems Medicine for Cancer, grant KF2420-93 (Z.W.); and Fundamental Research Funds for Central Universities, grant HIT.OCEF.2022034 (Z.W.).

Author contributions:

Conceptualization: Z.W., H.T., Y. Jia, H.C., C.F., and L.Z. Writing—review and editing: Z.W., X.F., S.Z., H.T., C.F., Y. Jia, H.C., Y.L., and L.Z. Writing—original draft: Z.W., Y.L., C.F., H.C., and X.F. Investigation: Y.L., G.Z., H.T., Y. Jiang, C.F., Y. Jia, H.C., R.D., and Z.W. Methodology: Y.L., H.T., C.F., Y. Jia, H.C., Z.W., L.Z., and X.F. Resources: Z.W., S.Z., H.T., C.F., H.C., Y.L., and L.Z. Funding acquisition: Z.W., H.T., H.C., and X.F. Data curation: H.T., C.F., Y. Jia, H.C., Y.L., Z.W., and X.F. Validation: Z.W., Y.L., S.Z., H.T., Y. Jiang, C.F., Y. Jia, H.C., L.Z., and X.F. Supervision: Z.W., H.T., H.C., Y.L., and C.F. Formal analysis: Y.L., X.F., G.Z., H.T., Y. Jiang, C.F., Y. Jia, H.C., Z.W., and L.Z. Software: X.F., H.T., C.F., Y. Jia, and H.C. Project administration: Z.W., H.T., C.F., H.C., Y.L., and L.Z. Visualization: H.T., Y. Jiang, C.F., H.C., Y.L., Z.W., L.Z., and X.F.

Competing interests:

The authors declare that they have no competing interests.

Data, code, and materials availability:

All data and code needed to evaluate and reproduce the results in the paper are present in the paper and/or the Supplementary Materials. This study did not generate new materials.

Supplementary Materials

The PDF file includes:

Figs. S1 to S24

Tables S1 to S3

Legends for movies S1 to S12

References

sciadv.aec2918_sm.pdf^{(2.4MB, pdf)}

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S12

sciadv.aec2918_movies_s1_to_s12.zip^{(49.9MB, zip)}

REFERENCES

1.Hu W., Lucchinetti C. F., The pathological spectrum of CNS inflammatory demyelinating diseases. Semin. Immunopathol. 31, 439–453 (2009). [DOI] [PubMed] [Google Scholar]
2.Wraith D. C., Nicholson L. B., The adaptive immune system in diseases of the central nervous system. J. Clin. Invest. 122, 1172–1179 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Lucas S.-M., Rothwell N. J., Gibson R. M., The role of inflammation in CNS injury and disease. Br. J. Pharmacol. 147, S232–S240 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Allan S. M., Rothwell N. J., Inflammation in central nervous system injury. Philos. Trans. R. Soc. B 358, 1669–1677 (2003). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.McDonald J. W., Sadowsky C., Spinal-cord injury. Lancet 359, 417–425 (2002). [DOI] [PubMed] [Google Scholar]
6.Ahuja C. S., Wilson J. R., Nori S., Kotter M. R. N., Druschel C., Curt A., Fehlings M. G., Traumatic spinal cord injury. Nat. Rev. Dis. Primers. 3, 17018 (2017). [DOI] [PubMed] [Google Scholar]
7.Anjum A., Yazid M. D., Fauzi Daud M., Idris J., Ng A. M. H., Selvi Naicker A., Ismail O. H. R., Athi Kumar R. K., Lokanathan Y., Spinal cord injury: Pathophysiology, multimolecular Interactions, and underlying recovery mechanisms. Int. J. Mol. Sci. 21, 7533 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Zuidema J. M., Gilbert R. J., Gottipati M. K., Biomaterial approaches to modulate reactive astroglial response. Cells Tissues Organs 205, 372–395 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Corrigan F., Mander K. A., Leonard A. V., Vink R., Neurogenic inflammation after traumatic brain injury and its potentiation of classical inflammation. J. Neuroinflammation 13, 264 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Zhao Y., Li Q., Niu J., Guo E., Zhao C., Zhang J., Liu X., Wang L., Rao L., Chen X., Yang K., Neutrophil membrane-camouflaged polyprodrug nanomedicine for inflammation suppression in ischemic stroke therapy. Adv. Mater. 36, e2311803 (2024). [DOI] [PubMed] [Google Scholar]
11.Nong J., Glassman P. M., Myerson J. W., Zuluaga-Ramirez V., Rodriguez-Garcia A., Mukalel A., Omo-Lamai S., Walsh L. R., Zamora M. E., Gong X., Wang Z., Bhamidipati K., Kiseleva R. Y., Villa C. H., Greineder C. F., Kasner S. E., Weissman D., Mitchell M. J., Muro S., Persidsky Y., Brenner J. S., Muzykantov V. R., Marcos-Contreras O. A., Targeted nanocarriers co-opting pulmonary intravascular leukocytes for drug delivery to the injured brain. ACS Nano 17, 13121–13136 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Spang M. T., Middleton R., Diaz M., Hunter J., Mesfin J., Banka A., Sullivan H., Wang R., Lazerson T. S., Bhatia S., Corbitt J., D’Elia G., Sandoval-Gomez G., Kandell R., Vratsanos M. A., Gnanasekaran K., Kato T., Igata S., Luo C., Osborn K. G., Gianneschi N. C., Eniola-Adefeso O., Cabrales P., Kwon E. J., Contijoch F., Reeves R. R., DeMaria A. N., Christman K. L., Intravascularly infused extracellular matrix as a biomaterial for targeting and treating inflamed tissues. Nat. Biomed. Eng. 7, 94–109 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Yu H., Luo H., Chang L., Wang S., Geng X., Kang L., Zhong Y., Cao Y., Wang R., Yang X., Zhu Y., Shi M.-J., Hu Y., Liu Z., Yin X., Ran Y., Yang H., Fan W., Zhao B.-Q., The NEDD8-activating enzyme inhibitor MLN4924 reduces ischemic brain injury in mice. Proc. Natl. Acad. Sci. U.S.A. 119, e2111896119 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Chen W.-F., Chen C.-H., Chen N.-F., Sung C.-S., Wen Z.-H., Neuroprotective effects of direct intrathecal administration of granulocyte colony-stimulating factor in rats with spinal cord injury. CNS Neurosci. Ther. 21, 698–707 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Ajeeb R., Clegg J. R., Intrathecal delivery of macromolecules: Clinical status and emerging technologies. Adv. Drug Deliv. Rev. 199, 114949 (2023). [DOI] [PubMed] [Google Scholar]
16.Taylor M. J., Thompson A. M., Alhajlah S., Tuxworth R. I., Ahmed Z., Inhibition of Chk2 promotes neuroprotection, axon regeneration, and functional recovery after CNS injury. Sci. Adv. 8, eabq2611 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Takata F., Nakagawa S., Matsumoto J., Dohgu S., Blood-brain barrier dysfunction amplifies the development of neuroinflammation: Understanding of cellular events in brain microvascular endothelial cells for prevention and treatment of BBB dysfunction. Front. Cell. Neurosci. 15, 661838 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Alam A., Thelin E. P., Tajsic T., Khan D. Z., Khellaf A., Patani R., Helmy A., Cellular infiltration in traumatic brain injury. J. Neuroinflammation 17, 328 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Brett B. L., Gardner R. C., Godbout J., Dams-O’Connor K., Keene C. D., Traumatic brain injury and risk of neurodegenerative disorder. Biol. Psychiatry 91, 498–507 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Jimenez Hamann M. C., Tsai E. C., Tator C. H., Shoichet M. S., Novel intrathecal delivery system for treatment of spinal cord injury. Exp. Neurol. 182, 300–309 (2003). [DOI] [PubMed] [Google Scholar]
21.Danhier F., Feron O., Préat V., To exploit the tumor microenvironment: Passive and active tumor targeting of nanocarriers for anti-cancer drug delivery. J. Control. Release 148, 135–146 (2010). [DOI] [PubMed] [Google Scholar]
22.Nance E., Pun S. H., Saigal R., Sellers D. L., Drug delivery to the central nervous system. Nat. Rev. Mater. 7, 314–331 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Rubenstein M., Cornejo A., Nagpal R., Programmable self-assembly in a thousand-robot swarm. Science 345, 795–799 (2014). [DOI] [PubMed] [Google Scholar]
24.Gross R., Bonani M., Mondada F., Dorigo M., Autonomous self-assembly in swarm-bots. IEEE Trans. Robot. 22, 1115–1130 (2006). [Google Scholar]
25.Garattoni L., Birattari M., Autonomous task sequencing in a robot swarm. Sci. Robot. 3, eaat0430 (2018). [DOI] [PubMed] [Google Scholar]
26.Liu L., Li X., Chen Y., Gao J., Jiang Y., Ye Y., Wang P., Peng F., Tu Y., Gout management using uricase and sodium citrate hollow mesoporous nanomotors. Nat. Commun. 16, 2339 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Peter F., Kadiri V. M., Goyal R., Hurst J., Schnichels S., Avital A., Sela M., Mora-Raimundo P., Schroeder A., Alarcón-Correa M., Fischer P., Degradable and biocompatible magnesium zinc structures for nanomedicine: Magnetically actuated liposome microcarriers with tunable release. Adv. Funct. Mater. 34, 2314265 (2024). [Google Scholar]
28.Zhao H., Zheng Y., Cai Y., Xu T., Dong R., Zhang X., Intelligent metallic micro/nanomotors: From propulsion to application. Nano Today 52, 101939 (2023). [Google Scholar]
29.Dong R., Cai Y., Yang Y., Gao W., Ren B., Photocatalytic micro/nanomotors: From construction to applications. Acc. Chem. Res. 51, 1940–1947 (2018). [DOI] [PubMed] [Google Scholar]
30.Huang Y., Wu C., Dai J., Liu B., Cheng X., Li X., Cao Y., Chen J., Li Z., Tang J., Tunable self-thermophoretic nanomotors with polymeric coating. J. Am. Chem. Soc. 145, 19945–19952 (2023). [DOI] [PubMed] [Google Scholar]
31.Deng Y., Paskert A., Zhang Z., Wittkowski R., Ahmed D., An acoustically controlled helical microrobot. Sci. Adv. 9, eadh5260 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Aghakhani A., Yasa O., Wrede P., Sitti M., Acoustically powered surface-slipping mobile microrobots. Proc. Natl. Acad. Sci. U.S.A. 117, 3469–3477 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Schmidt C. K., Medina-Sánchez M., Edmondson R. J., Schmidt O. G., Engineering microrobots for targeted cancer therapies from a medical perspective. Nat. Commun. 11, 5618 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Wu Z., Li L., Yang Y., Hu P., Li Y., Yang S.-Y., Wang L. V., Gao W., A microrobotic system guided by photoacoustic computed tomography for targeted navigation in intestines in vivo. Sci. Robot. 4, eaax0613 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Jeon S., Kim S., Ha S., Lee S., Kim E., Kim S. Y., Park S. H., Jeon J. H., Kim S. W., Moon C., Nelson B. J., J.-y. Kim, Yu S.-W., Choi H., Magnetically actuated microrobots as a platform for stem cell transplantation. Sci. Robot. 4, eaav4317 (2019). [DOI] [PubMed] [Google Scholar]
36.Yan X., Zhou Q., Vincent M., Deng Y., Yu J., Xu J., Xu T., Tang T., Bian L., Wang Y.-X. J., Kostarelos K., Zhang L., Multifunctional biohybrid magnetite microrobots for imaging-guided therapy. Sci. Robot. 2, eaaq1155 (2017). [DOI] [PubMed] [Google Scholar]
37.Wang X., Gong Z., Wang T., Law J., Chen X., Wanggou S., Wang J., Ying B., Francisco M., Dong W., Xiong Y., Fan J. J., MacLeod G., Angers S., Li X., Dirks P. B., Liu X., Huang X., Sun Y., Mechanical nanosurgery of chemoresistant glioblastoma using magnetically controlled carbon nanotubes. Sci. Adv. 9, eade5321 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Zhang F., Li Z., Duan Y., Abbas A., Mundaca-Uribe R., Yin L., Luan H., Gao W., Fang R. H., Zhang L., Wang J., Gastrointestinal tract drug delivery using algae motors embedded in a degradable capsule. Sci. Robot. 7, eabo4160 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Yu J., Jin D., Chan K.-F., Wang Q., Yuan K., Zhang L., Active generation and magnetic actuation of microrobotic swarms in bio-fluids. Nat. Commun. 10, 5631 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Law J., Wang X., Luo M., Xin L., Du X., Dou W., Wang T., Shan G., Wang Y., Song P., Huang X., Yu J., Sun Y., Microrobotic swarms for selective embolization. Sci. Adv. 8, eabm5752 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Zhang Z., Sukhov A., Harting J., Malgaretti P., Ahmed D., Rolling microswarms along acoustic virtual walls. Nat. Commun. 13, 7347 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Li W., Wu C., Xiong Z., Liang C., Li Z., Liu B., Cao Q., Wang J., Tang J., Li D., Self-driven magnetorobots for recyclable and scalable micro/nanoplastic removal from nonmarine waters. Sci. Adv. 8, eade1731 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Hu Z., Tan H., Ye Y., Xu W., Gao J., Liu L., Zhang L., Jiang J., Tian H., Peng F., Tu Y., NIR-actuated ferroptosis nanomotor for enhanced tumor penetration and therapy. Adv. Mater. 36, 2412227 (2024). [DOI] [PubMed] [Google Scholar]
44.Gwisai T., Mirkhani N., Christiansen M. G., Nguyen T. T., Ling V., Schuerle S., Magnetic torque–driven living microrobots for increased tumor infiltration. Sci. Robot. 7, eabo0665 (2022). [DOI] [PubMed] [Google Scholar]
45.de Ávila B. E.-F., Angsantikul P., Li J., Angel Lopez-Ramirez M., Ramírez-Herrera D. E., Thamphiwatana S., Chen C., Delezuk J., Samakapiruk R., Ramez V., Obonyo M., Zhang L., Wang J., Micromotor-enabled active drug delivery for in vivo treatment of stomach infection. Nat. Commun. 8, 272 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Chen C., Ding S., Wang J., Materials consideration for the design, fabrication and operation of microscale robots. Nat. Rev. Mater. 9, 159–172 (2024). [Google Scholar]
47.Cuntín-Abal C., Bujalance-Fernández J., Yuan K., Arribi A., Jurado-Sánchez B., Escarpa A., Magnetic bacteriophage-engineered Janus micromotors for selective bacteria capture and detection. Adv. Funct. Mater. 34, 2312257 (2024). [Google Scholar]
48.Fan X., Chen Q., Li M., Wu Z., Tong D., Xie H., Yang Z., Sun L., Sitti M., Machining swarf formation–inspired fabrication of ferrofluidic helical miniature robots with multimodal locomotion capability. Sci. Adv. 11, eads4411 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Shen J., Wang Y., Yao M., Liu S., Guo Z., Zhang L., Wang B., Long-span delivery of differentiable hybrid robots for restoration of neural connections. Matter 8, 101942 (2025). [Google Scholar]
50.Yoo J., Tang S., Gao W., Micro- and nanorobots for biomedical applications in the brain. Nat. Rev. Bioeng. 1, 308–310 (2023). [Google Scholar]
51.Wu Z., Chen Y., Mukasa D., Pak O. S., Gao W., Medical micro/nanorobots in complex media. Chem. Soc. Rev. 49, 8088–8112 (2020). [DOI] [PubMed] [Google Scholar]
52.Tu J., Gao W., Spray-on magnetic skin for robotic actuation. Sci. Robot. 5, eabf1390 (2020). [DOI] [PubMed] [Google Scholar]
53.Li J., Esteban-Fernández de Ávila B., Gao W., Zhang L., Wang J., Micro/nanorobots for biomedicine: Delivery, surgery, sensing, and detoxification. Sci. Robot. 2, eaam6431 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Peng X., Tang S., Tang D., Zhou D., Li Y., Chen Q., Wan F., Lukas H., Han H., Zhang X., Gao W., Wu S., Autonomous metal-organic framework nanorobots for active mitochondria-targeted cancer therapy. Sci. Adv. 9, eadh1736 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Wang J., Xing Y., Ngatio M., Bies P., Xu L. L., Xing L., Zarea A., Makela A. V., Contag C. H., Li J., Engineering magnetotactic bacteria as medical microrobots. Adv. Mater. 37, e2416966 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Yu H., Liu X., Zhang Y., Shen J., Liu X., Liu S., Wang X., Sun B., Du H., Xu L., Zou B., Ding J., Xu Q., Zhang L., Wang B., Photocatalytic microrobots for treating bacterial infections deep within sinuses. Sci. Robot. 10, eadt0720 (2025). [DOI] [PubMed] [Google Scholar]
57.Wang C., Wang T., Li M., Zhang R., Ugurlu H., Sitti M., Heterogeneous multiple soft millirobots in three-dimensional lumens. Sci. Adv. 10, eadq1951 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Fan X., Jiang Y., Li M., Zhang Y., Tian C., Mao L., Xie H., Sun L., Yang Z., Sitti M., Scale-reconfigurable miniature ferrofluidic robots for negotiating sharply variable spaces. Sci. Adv. 8, eabq1677 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Han H., Ma X., Deng W., Zhang J., Tang S., Pak O. S., Zhu L., Criado-Hidalgo E., Gong C., Karshalev E., Yoo J., You M., Liu A., Wang C., Shen H. K., Patel P. N., Hays C. L., Gunnarson P. J., Li L., Zhang Y., Dabiri J. O., Wang L. V., Shapiro M. G., Wu D., Zhou Q., Greer J. R., Gao W., Imaging-guided bioresorbable acoustic hydrogel microrobots. Sci. Robot. 9, eadp3593 (2024). [DOI] [PubMed] [Google Scholar]
60.Xia X., Ni M., Wang M., Wang B., Liu D., Lu Y., Artificial intelligence-assisted multimode microrobot swarm behaviors. ACS Nano 19, 12883–12894 (2025). [DOI] [PubMed] [Google Scholar]
61.Zhang L., Chen Z., Ran H., Azechi M., Yang X., Huang W., Tian H., Shen L., Peng F., Tu Y., Biomimetic magnetobacterial microrobots for active pneumonia therapy. Nat. Commun. 16, 7856 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Jiang J., Liang H., Ye Y., Huang W., Miao J., Tan H., Hu Z., Tian H., Qin H., Zhang X., Zhang L., Gao J., Shen X., Wang S., Peng F., Tu Y., Stem cell secretome armed magneto-actuated micromotors as spatio-temporal manipulators for wound healing acceleration. Nat. Commun. 16, 6754 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Ahmed D., Sukhov A., Hauri D., Rodrigue D., Maranta G., Harting J., Nelson B. J., Bioinspired acousto-magnetic microswarm robots with upstream motility. Nat. Mach. Intell. 3, 116–124 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Wang T., Chen Z., Huang Q., Arai T., Liu X., Advanced microrobots driven by acoustic and magnetic fields for biomedical applications. Cyborg Bionic Syst. 6, 0386 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Feng X., Xue Y., Gonca S., Ji K., Zhang M., García-García F. R., Li Q., Huang Y., Kamenev K. V., Chen X., Ultrasmall superparamagnetic iron oxide nanoparticles for enhanced tumor penetration. J. Mater. Chem. B 11, 3422–3433 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Saifuddin A., Burnett S. J. D., White J., The variation of position of the conus medullaris in an adult population: A magnetic resonance imaging study. Spine 23, 1452–1456 (1998). [DOI] [PubMed] [Google Scholar]
67.Dehaini D., Wei X., Fang R. H., Masson S., Angsantikul P., Luk B. T., Zhang Y., Ying M., Jiang Y., Kroll A. V., Gao W., Zhang L., Erythrocyte–platelet hybrid membrane coating for enhanced nanoparticle functionalization. Adv. Mater. 29, 1606209 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Zhang H., Li Z., Gao C., Fan X., Pang Y., Li T., Wu Z., Xie H., He Q., Dual-responsive biohybrid neutrobots for active target delivery. Sci. Robot. 6, eaaz9519 (2021). [DOI] [PubMed] [Google Scholar]
69.Zhao Y., Li A., Jiang L., Gu Y., Liu J., Hybrid membrane-coated biomimetic nanoparticles (HM@BNPs): A multifunctional nanomaterial for biomedical applications. Biomacromolecules 22, 3149–3167 (2021). [DOI] [PubMed] [Google Scholar]
70.Xi K., Gu Y., Tang J., Chen H., Xu Y., Wu L., Cai F., Deng L., Yang H., Shi Q., Cui W., Chen L., Microenvironment-responsive immunoregulatory electrospun fibers for promoting nerve function recovery. Nat. Commun. 11, 4504 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Eash K. J., Greenbaum A. M., Gopalan P. K., Link D. C., CXCR2 and CXCR4 antagonistically regulate neutrophil trafficking from murine bone marrow. J. Clin. Invest. 120, 2423–2431 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Yaghoubi Y., Movassaghpour A., Zamani M., Talebi M., Mehdizadeh A., Yousefi M., Human umbilical cord mesenchymal stem cells derived-exosomes in diseases treatment. Life Sci. 233, 116733 (2019). [DOI] [PubMed] [Google Scholar]
73.Pillay J., Tak T., Kamp V. M., Koenderman L., Immune suppression by neutrophils and granulocytic myeloid-derived suppressor cells: Similarities and differences. Cell. Mol. Life Sci. 70, 3813–3827 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Xie H., Sun M., Fan X., Lin Z., Chen W., Wang L., Dong L., He Q., Reconfigurable magnetic microrobot swarm: Multimode transformation, locomotion, and manipulation. Sci. Robot. 4, eaav8006 (2019). [DOI] [PubMed] [Google Scholar]
75.Orendácová J., Cízková D., Kafka J., Lukácová N., Marsala M., Sulla I., Marsala J., Katsube N., Cauda equina syndrome. Prog. Neurobiol. 64, 613–637 (2001). [DOI] [PubMed] [Google Scholar]
76.Xue M. T., Sheng W. J., Song X., Shi Y. J., Geng Z. J., Shen L., Wang R., Lü H. Z., Hu J. G., Atractylenolide III ameliorates spinal cord injury in rats by modulating microglial/macrophage polarization. CNS Neurosci. Ther. 28, 1059–1071 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Liew P. X., Kubes P., The neutrophil’s role during health and disease. Physiol. Rev. 99, 1223–1248 (2019). [DOI] [PubMed] [Google Scholar]
78.Mayadas T. N., Cullere X., Lowell C. A., The multifaceted functions of neutrophils. Annu. Rev. Pathol. 9, 181–218 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Gordon S., Alternative activation of macrophages. Nat. Rev. Immunol. 3, 23–35 (2003). [DOI] [PubMed] [Google Scholar]
80.Kigerl K. A., Gensel J. C., Ankeny D. P., Alexander J. K., Donnelly D. J., Popovich P. G., Identification of two distinct macrophage subsets with divergent effects causing either neurotoxicity or regeneration in the injured mouse spinal cord. J. Neurosci. 29, 13435–13444 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
81.David S., Kroner A., Repertoire of microglial and macrophage responses after spinal cord injury. Nat. Rev. Neurosci. 12, 388–399 (2011). [DOI] [PubMed] [Google Scholar]
82.Mosser D. M., Edwards J. P., Exploring the full spectrum of macrophage activation. Nat. Rev. Immunol. 8, 958–969 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
83.Wolfs I. M., Donners M. M., de Winther M. P., Differentiation factors and cytokines in the atherosclerotic plaque micro-environment as a trigger for macrophage polarisation. Thromb. Haemost. 106, 763–771 (2011). [DOI] [PubMed] [Google Scholar]
84.Chen Z., Wang Y., Chen H., Law J., Pu H., Xie S., Duan F., Sun Y., Liu N., Yu J., A magnetic multi-layer soft robot for on-demand targeted adhesion. Nat. Commun. 15, 644 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
85.McNay E. C., Sherwin R. S., From artificial cerebro-spinal fluid (aCSF) to artificial extracellular fluid (aECF): Microdialysis perfusate composition effects on in vivo brain ECF glucose measurements. J. Neurosci. Methods 132, 35–43 (2004). [DOI] [PubMed] [Google Scholar]
86.Álvarez Z., Kolberg-Edelbrock A. N., Sasselli I. R., Ortega J. A., Qiu R., Syrgiannis Z., Mirau P. A., Chen F., Chin S. M., Weigand S., Kiskinis E., Stupp S. I., Bioactive scaffolds with enhanced supramolecular motion promote recovery from spinal cord injury. Science 374, 848–856 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figs. S1 to S24

Tables S1 to S3

Legends for movies S1 to S12

References

sciadv.aec2918_sm.pdf^{(2.4MB, pdf)}

Movies S1 to S12

sciadv.aec2918_movies_s1_to_s12.zip^{(49.9MB, zip)}

Data Availability Statement

All data and code needed to evaluate and reproduce the results in the paper are present in the paper and/or the Supplementary Materials. This study did not generate new materials.

[R1] 1.Hu W., Lucchinetti C. F., The pathological spectrum of CNS inflammatory demyelinating diseases. Semin. Immunopathol. 31, 439–453 (2009). [DOI] [PubMed] [Google Scholar]

[R2] 2.Wraith D. C., Nicholson L. B., The adaptive immune system in diseases of the central nervous system. J. Clin. Invest. 122, 1172–1179 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R3] 3.Lucas S.-M., Rothwell N. J., Gibson R. M., The role of inflammation in CNS injury and disease. Br. J. Pharmacol. 147, S232–S240 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R4] 4.Allan S. M., Rothwell N. J., Inflammation in central nervous system injury. Philos. Trans. R. Soc. B 358, 1669–1677 (2003). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R5] 5.McDonald J. W., Sadowsky C., Spinal-cord injury. Lancet 359, 417–425 (2002). [DOI] [PubMed] [Google Scholar]

[R6] 6.Ahuja C. S., Wilson J. R., Nori S., Kotter M. R. N., Druschel C., Curt A., Fehlings M. G., Traumatic spinal cord injury. Nat. Rev. Dis. Primers. 3, 17018 (2017). [DOI] [PubMed] [Google Scholar]

[R7] 7.Anjum A., Yazid M. D., Fauzi Daud M., Idris J., Ng A. M. H., Selvi Naicker A., Ismail O. H. R., Athi Kumar R. K., Lokanathan Y., Spinal cord injury: Pathophysiology, multimolecular Interactions, and underlying recovery mechanisms. Int. J. Mol. Sci. 21, 7533 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R8] 8.Zuidema J. M., Gilbert R. J., Gottipati M. K., Biomaterial approaches to modulate reactive astroglial response. Cells Tissues Organs 205, 372–395 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R9] 9.Corrigan F., Mander K. A., Leonard A. V., Vink R., Neurogenic inflammation after traumatic brain injury and its potentiation of classical inflammation. J. Neuroinflammation 13, 264 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R10] 10.Zhao Y., Li Q., Niu J., Guo E., Zhao C., Zhang J., Liu X., Wang L., Rao L., Chen X., Yang K., Neutrophil membrane-camouflaged polyprodrug nanomedicine for inflammation suppression in ischemic stroke therapy. Adv. Mater. 36, e2311803 (2024). [DOI] [PubMed] [Google Scholar]

[R11] 11.Nong J., Glassman P. M., Myerson J. W., Zuluaga-Ramirez V., Rodriguez-Garcia A., Mukalel A., Omo-Lamai S., Walsh L. R., Zamora M. E., Gong X., Wang Z., Bhamidipati K., Kiseleva R. Y., Villa C. H., Greineder C. F., Kasner S. E., Weissman D., Mitchell M. J., Muro S., Persidsky Y., Brenner J. S., Muzykantov V. R., Marcos-Contreras O. A., Targeted nanocarriers co-opting pulmonary intravascular leukocytes for drug delivery to the injured brain. ACS Nano 17, 13121–13136 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R12] 12.Spang M. T., Middleton R., Diaz M., Hunter J., Mesfin J., Banka A., Sullivan H., Wang R., Lazerson T. S., Bhatia S., Corbitt J., D’Elia G., Sandoval-Gomez G., Kandell R., Vratsanos M. A., Gnanasekaran K., Kato T., Igata S., Luo C., Osborn K. G., Gianneschi N. C., Eniola-Adefeso O., Cabrales P., Kwon E. J., Contijoch F., Reeves R. R., DeMaria A. N., Christman K. L., Intravascularly infused extracellular matrix as a biomaterial for targeting and treating inflamed tissues. Nat. Biomed. Eng. 7, 94–109 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] 13.Yu H., Luo H., Chang L., Wang S., Geng X., Kang L., Zhong Y., Cao Y., Wang R., Yang X., Zhu Y., Shi M.-J., Hu Y., Liu Z., Yin X., Ran Y., Yang H., Fan W., Zhao B.-Q., The NEDD8-activating enzyme inhibitor MLN4924 reduces ischemic brain injury in mice. Proc. Natl. Acad. Sci. U.S.A. 119, e2111896119 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] 14.Chen W.-F., Chen C.-H., Chen N.-F., Sung C.-S., Wen Z.-H., Neuroprotective effects of direct intrathecal administration of granulocyte colony-stimulating factor in rats with spinal cord injury. CNS Neurosci. Ther. 21, 698–707 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R15] 15.Ajeeb R., Clegg J. R., Intrathecal delivery of macromolecules: Clinical status and emerging technologies. Adv. Drug Deliv. Rev. 199, 114949 (2023). [DOI] [PubMed] [Google Scholar]

[R16] 16.Taylor M. J., Thompson A. M., Alhajlah S., Tuxworth R. I., Ahmed Z., Inhibition of Chk2 promotes neuroprotection, axon regeneration, and functional recovery after CNS injury. Sci. Adv. 8, eabq2611 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R17] 17.Takata F., Nakagawa S., Matsumoto J., Dohgu S., Blood-brain barrier dysfunction amplifies the development of neuroinflammation: Understanding of cellular events in brain microvascular endothelial cells for prevention and treatment of BBB dysfunction. Front. Cell. Neurosci. 15, 661838 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R18] 18.Alam A., Thelin E. P., Tajsic T., Khan D. Z., Khellaf A., Patani R., Helmy A., Cellular infiltration in traumatic brain injury. J. Neuroinflammation 17, 328 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] 19.Brett B. L., Gardner R. C., Godbout J., Dams-O’Connor K., Keene C. D., Traumatic brain injury and risk of neurodegenerative disorder. Biol. Psychiatry 91, 498–507 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] 20.Jimenez Hamann M. C., Tsai E. C., Tator C. H., Shoichet M. S., Novel intrathecal delivery system for treatment of spinal cord injury. Exp. Neurol. 182, 300–309 (2003). [DOI] [PubMed] [Google Scholar]

[R21] 21.Danhier F., Feron O., Préat V., To exploit the tumor microenvironment: Passive and active tumor targeting of nanocarriers for anti-cancer drug delivery. J. Control. Release 148, 135–146 (2010). [DOI] [PubMed] [Google Scholar]

[R22] 22.Nance E., Pun S. H., Saigal R., Sellers D. L., Drug delivery to the central nervous system. Nat. Rev. Mater. 7, 314–331 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] 23.Rubenstein M., Cornejo A., Nagpal R., Programmable self-assembly in a thousand-robot swarm. Science 345, 795–799 (2014). [DOI] [PubMed] [Google Scholar]

[R24] 24.Gross R., Bonani M., Mondada F., Dorigo M., Autonomous self-assembly in swarm-bots. IEEE Trans. Robot. 22, 1115–1130 (2006). [Google Scholar]

[R25] 25.Garattoni L., Birattari M., Autonomous task sequencing in a robot swarm. Sci. Robot. 3, eaat0430 (2018). [DOI] [PubMed] [Google Scholar]

[R26] 26.Liu L., Li X., Chen Y., Gao J., Jiang Y., Ye Y., Wang P., Peng F., Tu Y., Gout management using uricase and sodium citrate hollow mesoporous nanomotors. Nat. Commun. 16, 2339 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R27] 27.Peter F., Kadiri V. M., Goyal R., Hurst J., Schnichels S., Avital A., Sela M., Mora-Raimundo P., Schroeder A., Alarcón-Correa M., Fischer P., Degradable and biocompatible magnesium zinc structures for nanomedicine: Magnetically actuated liposome microcarriers with tunable release. Adv. Funct. Mater. 34, 2314265 (2024). [Google Scholar]

[R28] 28.Zhao H., Zheng Y., Cai Y., Xu T., Dong R., Zhang X., Intelligent metallic micro/nanomotors: From propulsion to application. Nano Today 52, 101939 (2023). [Google Scholar]

[R29] 29.Dong R., Cai Y., Yang Y., Gao W., Ren B., Photocatalytic micro/nanomotors: From construction to applications. Acc. Chem. Res. 51, 1940–1947 (2018). [DOI] [PubMed] [Google Scholar]

[R30] 30.Huang Y., Wu C., Dai J., Liu B., Cheng X., Li X., Cao Y., Chen J., Li Z., Tang J., Tunable self-thermophoretic nanomotors with polymeric coating. J. Am. Chem. Soc. 145, 19945–19952 (2023). [DOI] [PubMed] [Google Scholar]

[R31] 31.Deng Y., Paskert A., Zhang Z., Wittkowski R., Ahmed D., An acoustically controlled helical microrobot. Sci. Adv. 9, eadh5260 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] 32.Aghakhani A., Yasa O., Wrede P., Sitti M., Acoustically powered surface-slipping mobile microrobots. Proc. Natl. Acad. Sci. U.S.A. 117, 3469–3477 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R33] 33.Schmidt C. K., Medina-Sánchez M., Edmondson R. J., Schmidt O. G., Engineering microrobots for targeted cancer therapies from a medical perspective. Nat. Commun. 11, 5618 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R34] 34.Wu Z., Li L., Yang Y., Hu P., Li Y., Yang S.-Y., Wang L. V., Gao W., A microrobotic system guided by photoacoustic computed tomography for targeted navigation in intestines in vivo. Sci. Robot. 4, eaax0613 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] 35.Jeon S., Kim S., Ha S., Lee S., Kim E., Kim S. Y., Park S. H., Jeon J. H., Kim S. W., Moon C., Nelson B. J., J.-y. Kim, Yu S.-W., Choi H., Magnetically actuated microrobots as a platform for stem cell transplantation. Sci. Robot. 4, eaav4317 (2019). [DOI] [PubMed] [Google Scholar]

[R36] 36.Yan X., Zhou Q., Vincent M., Deng Y., Yu J., Xu J., Xu T., Tang T., Bian L., Wang Y.-X. J., Kostarelos K., Zhang L., Multifunctional biohybrid magnetite microrobots for imaging-guided therapy. Sci. Robot. 2, eaaq1155 (2017). [DOI] [PubMed] [Google Scholar]

[R37] 37.Wang X., Gong Z., Wang T., Law J., Chen X., Wanggou S., Wang J., Ying B., Francisco M., Dong W., Xiong Y., Fan J. J., MacLeod G., Angers S., Li X., Dirks P. B., Liu X., Huang X., Sun Y., Mechanical nanosurgery of chemoresistant glioblastoma using magnetically controlled carbon nanotubes. Sci. Adv. 9, eade5321 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R38] 38.Zhang F., Li Z., Duan Y., Abbas A., Mundaca-Uribe R., Yin L., Luan H., Gao W., Fang R. H., Zhang L., Wang J., Gastrointestinal tract drug delivery using algae motors embedded in a degradable capsule. Sci. Robot. 7, eabo4160 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R39] 39.Yu J., Jin D., Chan K.-F., Wang Q., Yuan K., Zhang L., Active generation and magnetic actuation of microrobotic swarms in bio-fluids. Nat. Commun. 10, 5631 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R40] 40.Law J., Wang X., Luo M., Xin L., Du X., Dou W., Wang T., Shan G., Wang Y., Song P., Huang X., Yu J., Sun Y., Microrobotic swarms for selective embolization. Sci. Adv. 8, eabm5752 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R41] 41.Zhang Z., Sukhov A., Harting J., Malgaretti P., Ahmed D., Rolling microswarms along acoustic virtual walls. Nat. Commun. 13, 7347 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R42] 42.Li W., Wu C., Xiong Z., Liang C., Li Z., Liu B., Cao Q., Wang J., Tang J., Li D., Self-driven magnetorobots for recyclable and scalable micro/nanoplastic removal from nonmarine waters. Sci. Adv. 8, eade1731 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] 43.Hu Z., Tan H., Ye Y., Xu W., Gao J., Liu L., Zhang L., Jiang J., Tian H., Peng F., Tu Y., NIR-actuated ferroptosis nanomotor for enhanced tumor penetration and therapy. Adv. Mater. 36, 2412227 (2024). [DOI] [PubMed] [Google Scholar]

[R44] 44.Gwisai T., Mirkhani N., Christiansen M. G., Nguyen T. T., Ling V., Schuerle S., Magnetic torque–driven living microrobots for increased tumor infiltration. Sci. Robot. 7, eabo0665 (2022). [DOI] [PubMed] [Google Scholar]

[R45] 45.de Ávila B. E.-F., Angsantikul P., Li J., Angel Lopez-Ramirez M., Ramírez-Herrera D. E., Thamphiwatana S., Chen C., Delezuk J., Samakapiruk R., Ramez V., Obonyo M., Zhang L., Wang J., Micromotor-enabled active drug delivery for in vivo treatment of stomach infection. Nat. Commun. 8, 272 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R46] 46.Chen C., Ding S., Wang J., Materials consideration for the design, fabrication and operation of microscale robots. Nat. Rev. Mater. 9, 159–172 (2024). [Google Scholar]

[R47] 47.Cuntín-Abal C., Bujalance-Fernández J., Yuan K., Arribi A., Jurado-Sánchez B., Escarpa A., Magnetic bacteriophage-engineered Janus micromotors for selective bacteria capture and detection. Adv. Funct. Mater. 34, 2312257 (2024). [Google Scholar]

[R48] 48.Fan X., Chen Q., Li M., Wu Z., Tong D., Xie H., Yang Z., Sun L., Sitti M., Machining swarf formation–inspired fabrication of ferrofluidic helical miniature robots with multimodal locomotion capability. Sci. Adv. 11, eads4411 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] 49.Shen J., Wang Y., Yao M., Liu S., Guo Z., Zhang L., Wang B., Long-span delivery of differentiable hybrid robots for restoration of neural connections. Matter 8, 101942 (2025). [Google Scholar]

[R50] 50.Yoo J., Tang S., Gao W., Micro- and nanorobots for biomedical applications in the brain. Nat. Rev. Bioeng. 1, 308–310 (2023). [Google Scholar]

[R51] 51.Wu Z., Chen Y., Mukasa D., Pak O. S., Gao W., Medical micro/nanorobots in complex media. Chem. Soc. Rev. 49, 8088–8112 (2020). [DOI] [PubMed] [Google Scholar]

[R52] 52.Tu J., Gao W., Spray-on magnetic skin for robotic actuation. Sci. Robot. 5, eabf1390 (2020). [DOI] [PubMed] [Google Scholar]

[R53] 53.Li J., Esteban-Fernández de Ávila B., Gao W., Zhang L., Wang J., Micro/nanorobots for biomedicine: Delivery, surgery, sensing, and detoxification. Sci. Robot. 2, eaam6431 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R54] 54.Peng X., Tang S., Tang D., Zhou D., Li Y., Chen Q., Wan F., Lukas H., Han H., Zhang X., Gao W., Wu S., Autonomous metal-organic framework nanorobots for active mitochondria-targeted cancer therapy. Sci. Adv. 9, eadh1736 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R55] 55.Wang J., Xing Y., Ngatio M., Bies P., Xu L. L., Xing L., Zarea A., Makela A. V., Contag C. H., Li J., Engineering magnetotactic bacteria as medical microrobots. Adv. Mater. 37, e2416966 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] 56.Yu H., Liu X., Zhang Y., Shen J., Liu X., Liu S., Wang X., Sun B., Du H., Xu L., Zou B., Ding J., Xu Q., Zhang L., Wang B., Photocatalytic microrobots for treating bacterial infections deep within sinuses. Sci. Robot. 10, eadt0720 (2025). [DOI] [PubMed] [Google Scholar]

[R57] 57.Wang C., Wang T., Li M., Zhang R., Ugurlu H., Sitti M., Heterogeneous multiple soft millirobots in three-dimensional lumens. Sci. Adv. 10, eadq1951 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] 58.Fan X., Jiang Y., Li M., Zhang Y., Tian C., Mao L., Xie H., Sun L., Yang Z., Sitti M., Scale-reconfigurable miniature ferrofluidic robots for negotiating sharply variable spaces. Sci. Adv. 8, eabq1677 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R59] 59.Han H., Ma X., Deng W., Zhang J., Tang S., Pak O. S., Zhu L., Criado-Hidalgo E., Gong C., Karshalev E., Yoo J., You M., Liu A., Wang C., Shen H. K., Patel P. N., Hays C. L., Gunnarson P. J., Li L., Zhang Y., Dabiri J. O., Wang L. V., Shapiro M. G., Wu D., Zhou Q., Greer J. R., Gao W., Imaging-guided bioresorbable acoustic hydrogel microrobots. Sci. Robot. 9, eadp3593 (2024). [DOI] [PubMed] [Google Scholar]

[R60] 60.Xia X., Ni M., Wang M., Wang B., Liu D., Lu Y., Artificial intelligence-assisted multimode microrobot swarm behaviors. ACS Nano 19, 12883–12894 (2025). [DOI] [PubMed] [Google Scholar]

[R61] 61.Zhang L., Chen Z., Ran H., Azechi M., Yang X., Huang W., Tian H., Shen L., Peng F., Tu Y., Biomimetic magnetobacterial microrobots for active pneumonia therapy. Nat. Commun. 16, 7856 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] 62.Jiang J., Liang H., Ye Y., Huang W., Miao J., Tan H., Hu Z., Tian H., Qin H., Zhang X., Zhang L., Gao J., Shen X., Wang S., Peng F., Tu Y., Stem cell secretome armed magneto-actuated micromotors as spatio-temporal manipulators for wound healing acceleration. Nat. Commun. 16, 6754 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R63] 63.Ahmed D., Sukhov A., Hauri D., Rodrigue D., Maranta G., Harting J., Nelson B. J., Bioinspired acousto-magnetic microswarm robots with upstream motility. Nat. Mach. Intell. 3, 116–124 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R64] 64.Wang T., Chen Z., Huang Q., Arai T., Liu X., Advanced microrobots driven by acoustic and magnetic fields for biomedical applications. Cyborg Bionic Syst. 6, 0386 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R65] 65.Feng X., Xue Y., Gonca S., Ji K., Zhang M., García-García F. R., Li Q., Huang Y., Kamenev K. V., Chen X., Ultrasmall superparamagnetic iron oxide nanoparticles for enhanced tumor penetration. J. Mater. Chem. B 11, 3422–3433 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R66] 66.Saifuddin A., Burnett S. J. D., White J., The variation of position of the conus medullaris in an adult population: A magnetic resonance imaging study. Spine 23, 1452–1456 (1998). [DOI] [PubMed] [Google Scholar]

[R67] 67.Dehaini D., Wei X., Fang R. H., Masson S., Angsantikul P., Luk B. T., Zhang Y., Ying M., Jiang Y., Kroll A. V., Gao W., Zhang L., Erythrocyte–platelet hybrid membrane coating for enhanced nanoparticle functionalization. Adv. Mater. 29, 1606209 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] 68.Zhang H., Li Z., Gao C., Fan X., Pang Y., Li T., Wu Z., Xie H., He Q., Dual-responsive biohybrid neutrobots for active target delivery. Sci. Robot. 6, eaaz9519 (2021). [DOI] [PubMed] [Google Scholar]

[R69] 69.Zhao Y., Li A., Jiang L., Gu Y., Liu J., Hybrid membrane-coated biomimetic nanoparticles (HM@BNPs): A multifunctional nanomaterial for biomedical applications. Biomacromolecules 22, 3149–3167 (2021). [DOI] [PubMed] [Google Scholar]

[R70] 70.Xi K., Gu Y., Tang J., Chen H., Xu Y., Wu L., Cai F., Deng L., Yang H., Shi Q., Cui W., Chen L., Microenvironment-responsive immunoregulatory electrospun fibers for promoting nerve function recovery. Nat. Commun. 11, 4504 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R71] 71.Eash K. J., Greenbaum A. M., Gopalan P. K., Link D. C., CXCR2 and CXCR4 antagonistically regulate neutrophil trafficking from murine bone marrow. J. Clin. Invest. 120, 2423–2431 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R72] 72.Yaghoubi Y., Movassaghpour A., Zamani M., Talebi M., Mehdizadeh A., Yousefi M., Human umbilical cord mesenchymal stem cells derived-exosomes in diseases treatment. Life Sci. 233, 116733 (2019). [DOI] [PubMed] [Google Scholar]

[R73] 73.Pillay J., Tak T., Kamp V. M., Koenderman L., Immune suppression by neutrophils and granulocytic myeloid-derived suppressor cells: Similarities and differences. Cell. Mol. Life Sci. 70, 3813–3827 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R74] 74.Xie H., Sun M., Fan X., Lin Z., Chen W., Wang L., Dong L., He Q., Reconfigurable magnetic microrobot swarm: Multimode transformation, locomotion, and manipulation. Sci. Robot. 4, eaav8006 (2019). [DOI] [PubMed] [Google Scholar]

[R75] 75.Orendácová J., Cízková D., Kafka J., Lukácová N., Marsala M., Sulla I., Marsala J., Katsube N., Cauda equina syndrome. Prog. Neurobiol. 64, 613–637 (2001). [DOI] [PubMed] [Google Scholar]

[R76] 76.Xue M. T., Sheng W. J., Song X., Shi Y. J., Geng Z. J., Shen L., Wang R., Lü H. Z., Hu J. G., Atractylenolide III ameliorates spinal cord injury in rats by modulating microglial/macrophage polarization. CNS Neurosci. Ther. 28, 1059–1071 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R77] 77.Liew P. X., Kubes P., The neutrophil’s role during health and disease. Physiol. Rev. 99, 1223–1248 (2019). [DOI] [PubMed] [Google Scholar]

[R78] 78.Mayadas T. N., Cullere X., Lowell C. A., The multifaceted functions of neutrophils. Annu. Rev. Pathol. 9, 181–218 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R79] 79.Gordon S., Alternative activation of macrophages. Nat. Rev. Immunol. 3, 23–35 (2003). [DOI] [PubMed] [Google Scholar]

[R80] 80.Kigerl K. A., Gensel J. C., Ankeny D. P., Alexander J. K., Donnelly D. J., Popovich P. G., Identification of two distinct macrophage subsets with divergent effects causing either neurotoxicity or regeneration in the injured mouse spinal cord. J. Neurosci. 29, 13435–13444 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R81] 81.David S., Kroner A., Repertoire of microglial and macrophage responses after spinal cord injury. Nat. Rev. Neurosci. 12, 388–399 (2011). [DOI] [PubMed] [Google Scholar]

[R82] 82.Mosser D. M., Edwards J. P., Exploring the full spectrum of macrophage activation. Nat. Rev. Immunol. 8, 958–969 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R83] 83.Wolfs I. M., Donners M. M., de Winther M. P., Differentiation factors and cytokines in the atherosclerotic plaque micro-environment as a trigger for macrophage polarisation. Thromb. Haemost. 106, 763–771 (2011). [DOI] [PubMed] [Google Scholar]

[R84] 84.Chen Z., Wang Y., Chen H., Law J., Pu H., Xie S., Duan F., Sun Y., Liu N., Yu J., A magnetic multi-layer soft robot for on-demand targeted adhesion. Nat. Commun. 15, 644 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R85] 85.McNay E. C., Sherwin R. S., From artificial cerebro-spinal fluid (aCSF) to artificial extracellular fluid (aECF): Microdialysis perfusate composition effects on in vivo brain ECF glucose measurements. J. Neurosci. Methods 132, 35–43 (2004). [DOI] [PubMed] [Google Scholar]

[R86] 86.Álvarez Z., Kolberg-Edelbrock A. N., Sasselli I. R., Ortega J. A., Qiu R., Syrgiannis Z., Mirau P. A., Chen F., Chin S. M., Weigand S., Kiskinis E., Stupp S. I., Bioactive scaffolds with enhanced supramolecular motion promote recovery from spinal cord injury. Science 374, 848–856 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Adaptive intrathecal nanorobotics in mice and nonhuman primates for navigated CNS therapy

Yumeng Lu

Xinjian Fan

Chengjuan Fan

Gongzi Zhang

Hui Cui

Yiran Jiang

Shuwei Zhang

Yuxin Jia

Renfeng Dong

He Tan

Lihai Zhang

Zhiguang Wu

Roles

Abstract

INTRODUCTION

Fig. 1. Schematic of BRaNCs in CNS injury in vivo.

RESULTS

Development and magnetic swarm navigation of BRaNCs

Fig. 2. Development and magnetic actuation of BRaNCs.

Multimode swarm transformation and adaptive intrathecal swarm behavior of BRaNCs

Fig. 3. Multimode swarm transformation and adaptive swarm locomotion of BRaNCs within spinal canal ex vivo.

Dynamics of BRaNCs upon intrathecal navigation in vivo

Fig. 4. Distribution of BRaNCs upon intrathecal navigation with prolonged period in mice in vivo.

Adaptive swarm behavior of BRaNCs within spinal canal in nonhuman primate in vivo

Fig. 5. Visualization and validation of adaptive swarm of BRaNCs within spinal canal in nonhuman primate.

Therapeutic evaluation using adaptive nanorobotic delivery with BRaNCs in vivo

Fig. 6. Therapeutic evaluation in CNS injury using adaptive nanorobotic delivery with BRaNCs in vivo.

Therapy and safety comparison using adaptive nanorobotics with BRaNCs

Fig. 7. Evaluation of therapeutic efficacy and biosafety of adaptive nanorobotics using BRaNCs toward CNS injury in vivo.

DISCUSSION

MATERIALS AND METHODS

Synthesis of Fe3O4 particles

Synthesis of NP core

Isolation of neutrophils

Isolation of NMV

Isolation and culture of MSCs

Isolation of EXO

Synthesis of hybrid vesicles (NMV + EXO)

Synthesis of BRaNCs

Characterization of BRaNCs

Drug release study

Identification of proteins of NMV and EXO

Macrophage uptake study

Multimodal emergence and transformations of the BRaNCs swarm

Controllable intrathecal behavior of BRaNCs

Sources of experimental animals

CNS injury model establishment

Biodistribution and spinal cord–targeting studies of BRaNCs

Controlled behavior of BRaNC-vortex in nonhuman primates

CNS therapy of adaptive nanorobotic delivery with BRaNCs

In vivo safety studies of BRaNCs

Statistical analysis

Acknowledgments

Funding:

Author contributions:

Competing interests:

Data, code, and materials availability:

Supplementary Materials

The PDF file includes:

Other Supplementary Material for this manuscript includes the following:

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases

Synthesis of Fe₃O₄ particles