Magnetoelectric nanodiscs enable wireless transgene-free neuromodulation

Materials and methods

Synthesis of MENDs

Synthetic procedures for MENDs were reproduced across two institutions (Massachusetts Institute of Technology and Friedrich-Alexander University of Erlangen–Nuremberg). The Fe₃O₄ MNDs were synthesized by reducing haematite nanodiscs synthesized via previously established protocols^11,53,54. Haematite nanodiscs were first produced by heating a uniform mixture of 0.273 g of FeCl₃·6H₂O (Fluka), 10 ml ethanol and 600 μl of deionized (DI) water in a sealed Teflon-lined steel vessel at 180 °C for 18 h. After washing the red haematite nanodiscs with DI water and ethanol three to five times, the dried haematite was dispersed in 20 ml of trioctyl-amine (Sigma-Aldrich) and 1 g of oleic acid (Alfa Aesar/Thermo Fisher Scientific). For the reduction of haematite to magnetite, the mixture was transferred into a three-neck flask connected to a Schlenk line, evacuated for 20 min at room temperature and then heated to 370 °C (20 °C min⁻¹) in H₂ (5%) and N₂ (95%) atmosphere for 30 min.

The CFONDs were formed by nucleation and growth of a CoFe₂O₄ layer on the surface of MNDs. For this procedure, 120 mg of MNDs (cores) were dispersed uniformly in a precursor solution of 20 ml diphenyl ether (Aldrich), 1.90 ml oleic acid (Sigma-Aldrich), 1.97 ml oleylamine (Aldrich), 257 mg cobalt acetylacetonate (Co(acac)₂, Aldrich) and 706 mg iron acetylacetonate (Fe(acac)₃, Aldrich). A three-neck flask including the solution of MND cores and the shell precursors was connected to a Schlenk line. The solution was evacuated and then heated to 100 °C (7 °C min⁻¹) for 30 min in an N₂ atmosphere while magnetically stirring at 400 rpm. After closing the N₂ line, the temperature was increased to 200 °C (7 °C min⁻¹), maintained for 30 min and then increased to 230 °C (7 °C min⁻¹) and maintained for 30 min. The solution was cooled to room temperature (~30 min), and the resulting CFONDs were washed with ethanol and n-hexane and subjected to centrifugation at 6,869 rcf for 8 min; the washing process was repeated two to three times. The thickness of the CoFe₂O₄ layer is controlled by repeating the organometallic synthesis and washing steps described above. To obtain a 5 nm CoFe₂O₄ layer, the synthesis was repeated three times.

The Fe₃O₄–CoFe₂O₄–BaTiO₃ MENDs were made by formation of BaTiO₃ shell on the surface of CFOND via the sol–gel method. A mixture comprising 16 mg of CFONDs dispersed in n-hexane, 30 ml of DI water, 6 ml of ethanol and 2 g of poly(vinylpyrrolidone) (Sigma-Aldrich) was sonicated for 20 min, which led to segregation of the oil phase. The oil phase and other insoluble solids were removed with a spatula. The hydrophilic CFOND dispersions were then transferred to a three-neck flask connected to a Schlenk line, and then dried in vacuum at 80 °C until amber-coloured gel was formed on the bottom of the flask. The gel was redispersed in the BaTiO₃ shell precursor solution that was prepared by mixing 0.5 g citric acid (Sigma-Aldrich) and 24 µl titanium isopropoxide (Aldrich) dissolved in 15 ml of ethanol and 0.1 g citric acid and 0.0158 g barium carbonate (Aldrich) dissolved in DI water. The solution of CFONDs and BaTiO₃ precursors were moved to the three-neck flask connected to the vacuum line and kept at 80 °C for 12–14 h. The powders were then moved to a clean ceramic container and heated at 600 °C for 2 h, 700 °C for 2 h, then 800 °C for 1 h, sequentially. To prevent breaking the BaTiO₃ shell, the furnace door was kept closed until the temperature slowly cooled down to room temperature. The MENDs were dispersed in Tyrode and PBS before being used for in vitro and in vivo experiments.

Structural and magnetic characterization of magnetic nanomaterials

Structural imaging of MNDs, CFONDs and MENDs and energy-dispersive X-ray spectroscopy mapping on MEND-including neurons was performed via SEM (Zeiss Merlin). TEM imaging and single-particle electron diffraction analysis was performed using a FEI Tecnai G2 Spirit TWIN TEM. The diameter and thickness of MNDs, CFONDs and MENDs were estimated from the ensemble averages of particles in TEM and SEM images. Powder X-ray diffraction patterns of as-synthesized MNDs, CFONDs and MENDs were collected by a three-circle diffractometer coupled to a Bruker-AXS Smart Apex charge-coupled device detector with graphite-monochromated Mo K_α radiation (\(\lambda =0.71073\,{\text{\AA }}\)), and the data were processed with PANalytical HighScore Plus software. Room-temperature hysteresis curves were generated using the combined superconducting quantum interference device and vibrating sample magnetometer mode of a Quantum Design MPMS-3 at 300 K. An Agilent 5100 inductively coupled plasma-optical emission spectrometer was used to quantify the elemental concentration for the calculation of saturation magnetization. For inductively coupled plasma-optical emission spectrometry analysis, nanoparticles were digested in 37% v/v HCL (Sigma-Aldrich) overnight and diluted in 2 wt% HNO₃ (Sigma-Aldrich).

Micromagnetic simulations

Full magnetoelastic simulations were performed with MuMax3 magnetoelastic extension using literature values of magnetic and elastic constants for the materials (Supplementary Table 1) and a cell size of 2 × 2 × 2 nm³. The simulation was performed at discrete MFs corresponding to the maxima of an oscillating field with an amplitude of 10 mT around an offset at 220 mT. Increasing the strength of the elastic damping constant, a steady-state elastic displacement could be achieved within a simulation time of 72–168 h running on a GeForce RTX 3060 GPU. The steady state emerged after ~35 ns of simulation time following the initial field application, and 10 ns was required for subsequent field steps. Hence, all simulations were performed with the initial field value for 40 ns, and each subsequent field value for 15 ns. To quantify the normalized displacement of each particle at the maxima of an oscillating field with an amplitude of 10 mT around an offset at 220 mT, we averaged root-mean-square displacement normalized to the solid diagonal of each simulation cell. To calculate the deformation of each particle under the oscillating field with an amplitude of 10 mT around an offset at 220 mT, we averaged the difference in normalized displacements at two sequent maxima. Note that, due to limits of MuMax3 Magnetoelastic extension, we had to use the same sign in the anisotropy of the Fe₃O₄ and CoFe₂O₄ for the core–shell particles. This is an acceptable approximation as a lower bound of the strain in the composite particle. With a difference in anisotropy sign, the very large magnetostriction could not be simulated with our existing computational infrastructure as the resultant pressure waves took microseconds to dissipate.

Design and fabrication of electromagnets

For ME coupling coefficient measurements and Ca imaging in vitro, TEMCo 14 AWG copper magnet wire was wound around a C-shaped magnetic core with an 8 mm gap. The coil was connected to a power supply (Crown DC-300A Series II) and signal generator (Picoscope 2204A) to generate a MF combining a static OMF (magnitude 0–320 mT) and an AMF (frequency 0–1,000 Hz, amplitude 0–14 mT). For in vivo experiments, we separated the apparatuses that generated the OMF and AMF to generate a uniform, time-varying, MF over a larger volume. To generate the AMF (10 mT, 150 Hz) for c-Fos expression and fibre photometry experiments, we used a TEMCo 14 AWG copper magnet wire solenoid with a 10 cm inner diameter connected to the power supply and the signal generator. To generate the d.c. MF (220 mT), we used a ring-shaped permanent magnet (3.81 cm outer diameter × 1.905 cm inner diameter × 0.3175 cm thick, K&J Magnetics, RX8C2) that was positioned in the centre of the AMF coil. To apply MF during behaviour experiments, we designed a custom arena composed of two-connected cylindrical chambers (FixtureDisplays clear acrylic tube, 75 mm diameter, 2 mm wall thickness and 205 mm length) that was divided into two stimulation chambers (11 cm) separated by a neutral area (19 cm) that contained the animal entry port. The entry port was covered with 3D-printed acrylic curved plate after the entrance of mouse. To generate OMF during behavioural experiments, two rectangular permanent magnets (7.62 cm × 7.62 cm × 1.27 cm thick, K&J Magnetics, BZ0Z08-N52) were placed at the top and bottom of each stimulation chamber. To generate AMFs during behavioural experiments, solenoids (TEMCo 14 AWG copper magnet wire) were wound around each stimulation chamber and connected to the power supply and the signal generator.

ME coupling coefficient measurements

To experimentally evaluate \({\alpha }_{{{\mathrm{ME}}}}\) of the MENDs, we expanded on the prior work that leveraged electrochemical means to determine ME coupling from nanoparticles dispersed in an electrolyte⁵⁵ and applied this approach to MEND films^56,57,58,59. A three-electrode electrochemical cell was used to measure the potential required to maintain the surface charge, which fluctuated according to the electric polarization of the MENDs in the presence of a MF. To fit within the 8 mm gap of the C-shaped electromagnet described above, a nuclear magnetic resonance tube (8 mm outer diameter, 0.5 mm thickness) was used as the electrochemical cell. Within the cell, the Ag/AgCl reference electrode, Pt counter electrode and working electrode were immersed in Tyrode’s solution, PBS 1× or PBS× 10. The working electrode was prepared by drop-casting MEND solution onto a 0.3 × 1.2 cm² indium tin oxide glass substrate and connecting it to a copper wire using conductive silver epoxy (MG Chemicals, 842AR-15ML). The electrical connection was then sealed with epoxy (H.B. Fuller, 10010217) for insulation. The thickness of the MEND film on the indium tin oxide substrate was measured via a profilometer (Bruker Dektak DXT-A Stylus). All electrochemical measurements were performed using a potentiostat (Gamry Interface 1010E).

Fluorescent imaging in cultured hippocampal neurons

All animal procedures were approved by the Massachusetts Institute of Technology Committee on Animal Care (protocol #2305000529). Hippocampal neurons were extracted from neonatal rat (Sprague-Dawley, 001) pups (P1) and dissociated with Papain (Worthington Biochemical). The cells were then seeded on glass slides (5 mm diameter, Bellco Glass 1943-00005) in 24-well plates at a density of 112,500 cells ml⁻¹. Before seeding, the glass slides were cleaned by evaporating ethanol with an alcohol lamp and then coated with Matrigel (Corning). The cells were maintained in 1 ml Neurobasal medium (Invitrogen). Glial inhibition was conducted with 5-fluoro-2′-deoxyuridine (F0503 Sigma) 3 days after seeding; this step was omitted for experiments evaluating the effects of glia on neuronal responses to MEND-mediated modulation. Four days following seeding, the neurons were transduced with 1 µl of an adeno-associated virus serotype 9 (AAV9) carrying a fluorescent calcium ion indicator GCaMP6s under a pan-neuronal human synapsin (hSyn) promoter (AAV9-hSyn::GCaMP6s, Addgene viral prep #100843-AAV9, >1 × 10¹³ IU ml⁻¹). For simultaneous voltage and calcium imaging, AAV9-CAG::Voltron2, home-packaged, >5 × 10¹¹ IU ml⁻¹) and AAV9-CAG::GCaMP6s, (Addgene viral prep #100844-AAV9, >1 × 10¹³ IU ml⁻¹). After 5 days of incubation, calcium (Ca²⁺) imaging with GCaMP6s was performed at 1 fps imaging speed. Voltage imaging with Voltron 2.0^60,61 combined with Janelia Fluor 585 (Promega HT1040) was performed at 10 fps; these experiments were followed by GCaMP6s imaging of the same region and with the same frame speed. To confirm the latency using Ca²⁺ indicator with faster kinetics, calcium imaging at 10 fps was performed again with pAAV-Syn::GCaMP6f (Addgene viral prep #100837-AAV9, >1 × 10¹³ IU ml⁻¹)⁶²

ME neuromodulation with MENDs was expected to be most effective when the particles were in direct contact with neuronal membranes. Consequently, all experiments in vitro were conducted following a 1 h incubation period to allow the particles to precipitate onto the cells. The MEND density on the cell membranes was measured by quantifying their mass on each sample and calculating the area occupied by the neuronal network. For Ca²⁺ imaging, the hippocampal neurons were washed once with Tyrode’s solution and then immersed into 0.1 mg ml⁻¹ MEND solution within a well of a 24-well plate for 1 h to obtain ~0.75 g_MEND mm⁻² density of the particles on the neuronal surfaces. To change the MEND density on neuronal surfaces, we varied the concentration and volume of the MEND incubation solution. The hippocampal neurons on the glass coverslips were then transferred into a custom sample holder containing 200 µl Tyrode’s solution, which was introduced into an electromagnet as described above. The fluorescence changes in GCaMP6s were recorded at a rate of 1 fps. The fluorescence intensity of each cell was analysed with ImageJ software, and F₀ values were determined as the average fluorescence intensity during 30 s before the initial application of the AMF. Averaged ΔF/F₀ was estimated from 300 randomly selected neurons from 3 plates. The number of the peaks in \(\Delta\)F/F₀ was counted when the peak height is larger than the half of mean standard deviation of ΔF/F₀ for the overall imaging time of each video. The cells were defined as responsive if the ΔF/F₀ value exceeded three times the standard deviation (3σ) of the baseline (collected over 30 s before the initial application of the AMF) within 15 s from MF onset; hence, the responsiveness was defined as the fraction of responsive cells per MF epoch. The ratio of MF epoch number inducing average ΔF/F₀ ≥ 3σ of average ΔF/F₀ baseline to the overall number of MF epoch application in an individual video was defined to be the spiking probability. The latency was defined to be the time taken from the MF onset to the highest intensity point of the spikes.

To examine the cell viability in the presence of MENDs following MF stimuli, we performed analysis with a Live/Dead Viability/Cytotoxicity Kit (Invitrogen, L3224). The live and dead cells were indicated with Calcein AM (green) and ethidium homodimer-1 (red). The proportion of viable cells was calculated by normalizing the viable cell number to the total number of cell nuclei based on Hoechst staining (Thermo Scientific, Hoechst 33342 Solution (20 mM)). The neurons were incubated with the assay reagents for 20 min in Tyrode’s solution at 37 °C and then imaged before and after three cycles of MF field via a fluorescent microscope (Olympus IX73, 20× objective lens).

Stereotactic surgeries

All animal procedures were approved by the Massachusetts Institute of Technology Committee on Animal Care (protocol #2208000413). Surgeries were conducted under aseptic conditions within a stereotaxic frame (David Kopf Instruments) with 6–8-week-old WT mice. Approximately equal numbers of male and female mice were used (c-Fos expression analysis: three females and three males in each group; place preference assays: five females and six males for MENDs, four females and three males for MNDs, three females and four males for PBS; cylinder test: five females and four males for the MEND and MND groups, three females and three males for the PBS group; for photometry, Fig. 5a, seven females and seven males, Fig. 5b, two females and two males, Fig. 5d, three females and three males, Fig. 5k, four females and four males, Fig. 5l, four females and three males, Fig. 5m, four females and one male). Mice were anaesthetized under isoflurane (0.5–2.5% in O₂) using an anaesthesia machine (VET EQUIP). During the surgery, the eyes were covered with ophthalmic ointment, and a heat pad was used to maintain the animals’ core temperature. Mice were provided with subcutaneous injections of 0.6 ml of sterile Ringer’s solution for hydration and extended-release buprenorphine (1 mg kg⁻¹) for analgesia at the start of the procedure. The head was fixed into position using ear bars, then the fur on the top of the head was removed using depilatory cream. Following sterilization of the skin with betadine and ethanol, a midline incision was made along the scalp. Coordinates for the injection/implantation site were established on the basis of the Mouse Brain Atlas⁶³. A small craniotomy was drilled through the skull using a rotary tool (Dremel Micro 8050) and a carbon steel burr (Heisinger, 19007-05), and the dura was gently removed. Particles were injected into the VTA (anterior–posterior (AP) −2.9 mm, medial–lateral (ML) −0.5 mm, dorsal–ventral (DV) −4.5 mm) or STN (AP −2.06 mm, ML −1.50 mm, DV −4.50) using a microinjection apparatus (10 µl Hamilton syringe #80308, UMP-3 syringe pump and Micro4 controller; all from World Precision Instruments).

For c-Fos expression analysis, behavioural assays, MRI imaging and TEM analysis, 1.5 µl of MEND particles, MND control particles or PBS were injected at a rate of 500 nl min⁻¹. For MENDs and MND nanoparticle solutions, we used 1.5 mg ml⁻¹ concentration for all experiments, except for an additional 0.5 mg ml⁻¹ low concentration of MENDs tested in the c-Fos expression experiments. For PBS controls, the same volume of sterile PBS was injected instead of particles. After injection, the syringe was lifted up by 0.1 mm from the initial DV coordinate and left in place for 10 min before slowly withdrawing (0.5 mm min⁻¹), and then the skin was sutured (Ethicon, Ethilon 661H, polyamide 6, 19 mm needle length).

For the mice used for fibre photometry experiments, 300 nl of AAV9 hSyn::GCaMP6s, Addgene viral prep #100843-AAV9, >1 × 10¹³ IU µl⁻¹) was injected along with the 1.3 µl particles or PBS. The AAV solution was loaded into the syringe tip after the particles or control PBS, such that it was injected first and immediately followed by the particle or PBS injection into the brain. Following the injection, the syringes were lifted as described above. At 0.1 mm above the injection coordinates, a 200-µm-diameter silica fibre (Thorlabs FT200EMT) with a 2.5-mm-diameter stainless-steel ferrule (Thorlabs SF230-10) or the new fibre described in the following section was implanted and cemented in place with adhesive acrylic (C&B-Metabond, Parkell) followed with dental cement (Jet Set-4).

Following surgery, a subcutaneous injection of carprofen (5 mg kg⁻¹) was provided as an anti-inflammatory and analgesia agent, and the animals were placed in a clean recovery cage, part of which was positioned on a heating pad with ad libitum access to water and diet gel wet food.

Fibre photometry

Fibre photometry recordings were performed using a Neurophotometrics fibre photometry system (FP3002, Neurophotometrics). This system utilized a blue (peak wavelength λ = 470 nm) light-emitting device (LED) to excite GCaMP6s and a violet (peak λ = 415 nm) LED as an isosbestic wavelength control, with a fluorescence light path that includes a dichroic mirror to pass emitted green fluorescence (passband 495–530 nm) onto a complementary metal-oxide semiconductor camera (FLIR BlackFly). The 470 nm and 415 nm LEDs were each calibrated to provide 75 µW of optical power out of the tip of a 200 µm silica fibre matching the type implanted into the animals (Thorlabs FT200EMT). The system was coupled to a low-autofluorescence branching bundle patch cord (400 µm core, 0.57 numerical aperture, Doric), which was connected to the animal’s implant using a ceramic split mating sleeve (Thorlabs ADAF1). All photometry recordings were performed under isoflurane anaesthesia (0.5–2.5% in O₂, VET EQUIP) using a nose cone. After induction, the patch cord was connected to the animal’s fibre implant, and the animal’s head was placed in the centre of the custom magnetic apparatus described above: 10-cm-diameter solenoid AMF coil with a ring-shaped static magnet inside to provide OMF, whose centres are aligned. Fibre photometry recordings were performed at 130 Hz sampling rate.

After recording a 5 min baseline to account for rapid photobleaching at the start of the experiment, each animal received 10–20 total pulses of magnetic stimulation. Data were collected in Bonsai software and exported to MATLAB (MathworksR2022a) for analysis. Photometry data were analysed as follows: data were low-pass filtered below 25 Hz (second-order Butterworth filter), both isosbestic (405 nm excitation wavelength) and Ca²⁺-sensitive (470 nm excitation wavelength) signals were fitted with MATLAB exp2 fitting function, and then the fitting functions were subtracted from the original signals to correct for photobleaching. To remove the motion artifacts, the baseline-corrected isosbestic signal was subtracted from the baseline-corrected 470 nm signal. For the experiments using 2 s pulses of MF (OMF 220 mT, AMF 150 Hz, 10 mT), a 200 µm silica fibre was implanted in the mice and MF epochs were applied every 180s. The fluorescence average of 30 s before every stimulation epoch was taken as F₀, and ΔF/F₀ was segmented corresponding to each stimulation epoch (30 s pre-stimulation and 150 s post-stimulation).

Experiments comparing MEND-mediated stimulation and electrode DBS as well as repeated recordings 2 weeks, 4 weeks, 2 months and 3 months after the surgery were performed with the fibre described in Supplementary Note 3 and Supplementary Fig. 38. The F₀ was calculated from 15 s before stimulation, and ΔF/F₀ was segmented corresponding to each stimulation epoch (15 s pre-stimulation and 85 s post-stimulation). In these experiments, the DBS current was between 2 and 10 μA with a frequency of 100 Hz. MF stimuli consisted of 5 s epochs of OMF 220 mT combined with AMF at 100 Hz, 10 mT. For all GCaMP6s traces, we quantified spiking probability, peak intensity, peak width and peak position. When a MF or DBS triggered ∆F/F₀ ≥ 2σ (where σ is a standard deviation of baseline) within 20 s from the onset of the stimulation, the trial was classified as responsive. The fraction of responsive trials in each animal was defined as spiking probability. The peak intensity and position refer the maximum value of ∆F/F₀ transient and its position with respect to stimulation onset, respectively. The peak width is the interval between the times when the ∆F/F₀ = σ/2 before and after the maximum.

Immunohistochemical quantification of biomarkers

For the c-Fos expression experiments and biocompatibility assessment, mice were anaesthetized via an intraperitoneal injection of ketamine (100 mg kg⁻¹) and xylazine (10 mg kg⁻¹) mixture. The heads of the anaesthetized mice were placed into the centre of a custom-built apparatus with a 10-cm-diameter solenoid a.c. coil outside and ring-shaped static magnet inside, whose centres are aligned. Each mouse received three 2 s pulses of magnetic stimulation (220 mT OMF; 150 Hz, 10 mT AMF) separated by 90 min rest epochs before killing. Following the 90 min c-Fos induction period, the anaesthetized mice were killed by a lethal intraperitoneal injection of a sodium pentobarbital (Fatal-plus, 50 mg ml⁻¹, dose 100 mg kg⁻¹). The mice were then transcardially perfused with PBS and 4% paraformaldehyde, and their brains were extracted and kept in 4% paraformaldehyde overnight at 4 °C. After moving the fixed tissue to PBS and storing at 4 °C for 24 h, the brains were sectioned into 50 µm coronal slices with vibrating blade microtome (Leica VT1000S). The permeabilization was done on the slices for 30 min in the dark at room temperature in a 0.3% v/v Triton X-100 solution in PBS, and then the blocking was done for 1 h with 0.3% v/v Triton X-100 and normal donkey serum 5% v/v blocking serum solution in PBS with an orbital shaker. Following PBS washing three times, the brain slices were incubated in the first antibody solution overnight at 4 °C. After three washes with PBS, the brain slices were immersed in a secondary antibody solution for 2 h at room temperature on the orbital shaker in a dark room. After another three washes with PBS, the brain slices were stained with 4′,6-diamidino-2-phenylindole (DAPI), washed and then transferred onto glass slides with mounting medium (Fluoromount G, Southern Biotech). Rabbit anti-c-Fos (1:500, Cell Signaling Technology, 2250s) primary antibodies and donkey anti-rabbit Alexa Fluor 488 (1:1,000, Invitrogen, A-21206) secondary antibodies were used for the c-Fos expression analysis. For tyrosine hydroxylase (TH) staining, sheep-anti-tyrosine hydroxylase antibody (1:500, Novus Biologicals NB300-110SS) and donkey anti-sheep IgG (H + L), Alexa Fluor 568 (1:1,000, Thermo Scientific A-21099) were used as primary and secondary antibodies. Three pairs of primary (goat anti-Iba1 antibody (1:500, Abcam ab107159), goat anti-GFAP antibody (1:1,000, Abcam ab53554), rabbit anti-CD68 antibody (1:250, Abcam ab125212)) and secondary (donkey anti-goat IgG (H + L), Alexa Fluor 633 (1:1,000, Fisher Scientific A-21082) and donkey anti-rabbit Alexa Fluor 488 (1:1,000, Invitrogen, A-21206)) antibodies were used for the toxicity assessment.

Behavioural assays

For behavioural experiments, the mice were tested during the light phase of the 12 h light/dark cycle.

For the place preference assay, a 71-mm-inner-diameter acrylic cylinder was divided into three chambers: two 11-cm-long stimulation chambers (solid and stripe-patterned bottom) separated by a 19-cm-long neutral chamber (purple-coloured bottom). The stimulation chambers were equipped with custom-made electromagnets and static magnets outside the stimulation chambers, as shown in Fig. 4h and Supplementary Fig. 30. For 3 days (day −2 to day 0; Fig. 4h and Supplementary Fig. 30) before the baseline place preference test, the mice were habituated daily for 15 min in the setup and to the researchers. On day 1, the mice were placed into the chamber for 5 min habituation during which they could explore the chambers freely without any stimulation, and their place preference was tested for the following 10 min without any magnetic stimulation. From day 2 to day 4, following a 5 min habituation in the setup, the mice were stimulated with the MF (220 mT OMF, 10 mT, 150 Hz AMF, 2 s epochs separated by 90 s rest periods) when they entered the less preferred area for 10 min daily. On test day 5, the mice explored the entire arena in the absence of MF stimuli in either chamber. The mouse location in the arena was recorded with two cameras (Logitech HD Pro Webcam C920) positioned at the ends of each stimulation chamber and Logitech Capture Bio Recording and Streaming Software (2.08.11) was used to record the videos. Analysis of the place preference assay was done manually, where the observer was blinded to the subject type and marked the time spent in each chamber. Mice that showed more than 500 s (total test time is 600 s) preference to a chamber during pre-test or stayed 0 s at the stimulated chamber on day 2 were eliminated from the subsequent analyses.

For the rotational behaviour assay, an acrylic cylinder with an inner diameter of 71 mm and a height of 8 cm height was used as an arena. The AMF was applied via by the custom-made electromagnet surrounding the chamber, and the OMF was provided by two O-shaped permanent magnets (7.5 mm outer diameter, 4 mm inner diameter) attached to the top and bottom of the cylinder (Supplementary Fig. 33). A 3 min baseline video was first acquired. Then, another video was recorded over a 3 min period, during which AMF was applied in 5 s epochs separated by 25 s intervals. The videos were scored manually by a researcher blinded to subject group, and the number of rotations was compared with the baseline.

MRI

MRI was performed on a 7 T magnetic resonance imager operated by Bruker AV4 NeoBioSpec70-20USR console, equipped with a 114 mm 660 mT m⁻¹ actively shielded gradient and a QSN075/040 RF coil (Bruker BioSpin). T2-weighted images were obtained using the TurboRARE protocol with (Repetition Timer)/(Time to Echo) = 3,400/35 ms, echo spacing 11.667 ms, number of averages 4 and RARE factor 8. Both axial and coronal datasets were obtained with the geometric parameters of 196 × 196 matrix, field of view (FOV) of 18 mm × 18 mm, and interleaved slice thickness of 0.3 mm with no gap. The number of slices was adjusted to ensure the entire sample was covered.

Statistical analyses

OriginPro 2019 was used for assessing the statistical significance of all comparisons in this study except the post-hoc analysis for non-parametric data, which was performed with Matlab2023b. Although sample sizes were not determined with the power analysis, the group sizes for immunohistochemistry and behaviour tests were decided to be similar with previous research performed in the same brain circuit. Shapiro–Wilk test was performed to test normality of data distribution. For c-Fos expression analysis, the data distribution was tested for normality, and then analysed with analysis of variance (ANOVA) followed by Tukey’s post-hoc comparison test (*P < 0.05, **P < 0.01, ***P < 0.001). For the immunohistochemistry analyses relevant to the toxicity assessment, unpaired t-test was used to assess the differences between two groups, where significance threshold was indicated with non-significant (n.s.) P > 0.05, *P < 0.05, **P < 0.01, ***P < 0.001. In comparisons between c-Fos, Iba1, CD68 and TH expression in right and left hemispheres, paired t-test and Wilcoxon signed-rank test were used for parametric and non-parametric datasets, respectively. For place preference behavioural experiments, a Kruskal–Wallis test followed by a Tukey’s post-hoc comparison test was applied to compare three groups simultaneously with thresholds of n.s. P > 0.05, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. For the comparison of pre- and post-learning days within each group, a paired t-test was performed when the data distribution was found normal. For non-normal distribution, Wilcoxon signed-rank test was performed. For the preference change data comparing all three groups, the data followed normality and, thus, one-way ANOVA was used followed by Tukey’s post-hoc comparison test. The same method was applied to compare the number of the ipsilateral and contralateral rotations in 3 min assays with and without MF.

For comparison of spiking probability in the photometry across mice injected with MENDs, MNDs, PBS or implanted with electrodes, one-way ANOVA followed by Tukey’s post-hoc comparison test employed applied to compare five groups simultaneously (***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05, P > 0.05 is not indicated). For the peak intensity comparison between the MEND and electrode stimulations, a two-sample t-test for MEND data was performed. The peak width and position in the MEND group did not follow normal distribution, and a Mann–Whitney test was performed. The comparison of biomarkers in the brains of mice subjected to MF stimulation 2 weeks and 2 months after the MEND injections was performed via paired t-test as all data followed the normal distribution. ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05, n.s. P > 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.