Umbrella-type stretchable electrode arrays for long-term neural recording

Abstract

Flexible implantable neural electrodes have been studied extensively due to their chronic single-unit recording capability over weeks or months. However, high-density stable recording of the same neuronal ensembles over years-long periods remains challenging. Here we address this challenge by developing umbrella-type stretchable electrode arrays, which are anchored to surrounding brain tissue, forming a seamless neuroelectronic interface. We show that the umbrella-type stretchable electrode arrays create a neuroelectronic interface with reduced probe drift and gliosis that can stably track the same neuronal ensembles in head-fixed mice over 80 weeks (approximately 1.5 years). We also demonstrate that orientation-selective neuronal populations in the mouse primary visual cortex can be consistently recorded over 20 weeks. The umbrella-type stretchable electrode arrays with long-term recording capabilities hold significant promise for both fundamental neuroscience research and clinical brain-machine interface systems.

Stable recording of high-density neuronal ensembles at single-unit resolution over years-long timescales is challenging. Here, the authors develop umbrella-type stretchable electrode arrays anchored to surrounding brain tissue, forming a neuroelectronic interface that tracks the same neuronal ensembles in head-fixed mice over 1.5 years.

Introduction

Implantable neural probes play an important role in both basic neuroscience research^1–3 and clinical applications^4,5. Flexible implantable neural electrodes have been widely studied for chronic recordings because of their tissue-compatible mechanical properties^6–14. Using flexible neural probes with ultralow bending stiffness, researchers could stably record the neuronal action potentials over a months-long period. However, high-density stable recording of the same neuronal ensembles at single-unit resolution over longer timescales (e.g., years) remains challenging. Key factors leading to signal degradation during long-term recording include the micromotion-related probe drift and excessive gliosis^15,16, which might hinder the years-long stable recording with flexible neural probes. The open mesh electronics could form an interwoven structure with surrounding neural tissue, enabling stable recording of the same neurons from mice over ~ 1 year¹¹. Nonetheless, their limited channel count (16 or 32) could not satisfy the need of high-density recording of neuronal activity for precise decoding^10,17,18. Increasing electrode density usually comes with increased bending stiffness and larger surgical footprints of flexible probes, exacerbating the probe drift and gliosis at electrode-tissue interfaces. Therefore, how to avoid the probe drift and excessive gliosis of high-density electrodes during long-term implantation is critical for stable tracking of the same neuronal ensembles over a years-long period.

When an umbrella is closed, all the ribs fold in the same direction. Inspired by this phenomenon, here we develop the high-density umbrella-type stretchable electrode arrays (USEs), which consist of 8 ultraflexible electrode filaments (mimicking the umbrella ribs). All the electrode filaments are connected to the same central reference electrode, allowing for simultaneous implantation of all the recording sites in vivo by only one-time insertion and retraction of the rigid shuttle. After implantation, 8 electrode filaments form the open structure with a cylinder-like geometry in the brain. After recovery of brain tissues around the USEs probe, a seamless neuroelectronic interface without excessive gliosis or glial scar can be formed. Furthermore, chronically implanted USEs probes are anchored to surrounding brain tissue due to the central reference electrode at the bottom, preventing probe drift in the brain. Using 120-channel USEs probes, spiking activities from the same neuronal ensembles in the mouse were stably tracked over 80 weeks (~ 1.5 years), which spanned from young adulthood to old age. Moreover, the orientation-selective neuronal populations in mouse primary visual cortex could be consistently recorded over 20 weeks, demonstrating the potential of USEs probes for fundamental neuroscience research.

Results

Design and fabrication of the USEs probe

The USEs probes were fabricated with a standard micro-fabrication process (Supplementary Fig. 1). For both 120-channel and 240-channel devices, they consisted of three polyimide insulating layers and two gold conducting layers (Fig. 1a and Supplementary Fig. 2). With a total thickness of ~ 3 μm, the USEs probes exhibit ultraflexible characteristics. Ten 120-channel devices and six 240-channel devices could be fabricated on a single 4-inch wafer (Supplementary Fig. 3a). The as-fabricated USEs devices are composed of a bonding zone, a connecting zone, and a recording zone (Fig. 1b and Supplementary Fig. 3b). For the rodent recording, the length of the connecting zone is designed to be ~ 11 mm. With a diameter of ~ 5 mm, the recording zone could be entirely positioned on the mouse skull. As shown in Fig. 1c and Supplementary Fig. 3c, there are 8 electrode filaments in the recording zone. Each filament includes the stretchable serpentine part for structural stretching and a linear part with recording sites (Supplementary Fig. 4). Center-to-center spacing between recording sites is 50 μm for 120-channel probes and 30 μm for 240-channel probes, respectively. For each filament, all the recording sites are distributed with a total length of 600 μm (120-channel) and 750 μm (240-channel), respectively. The width of electrode filaments is 38 μm (120-channel) and 70 μm (240-channel), respectively. At the end of the linear part, all the 8 electrode filaments are connected to the central reference electrode (diameter 210 μm), which could be used to assist the implantation process besides serving as the reference channel during recording. Using the focused ion beam (FIB) and scanning electron microscopy (SEM), we could clearly observe the internal structure of 120-channel (Fig. 1d, e and Supplementary Fig. 5) and 240-channel (Supplementary Fig. 6) USEs probes. Two gold conducting layers were completely encapsulated by the polyimide insulating layers without any leakage. To minimize the total width of each electrode filament, the width of the thinnest conducting wire was designed to be ~ 2 μm. 120-channel flexible printed circuits (FPCs) were electrically connected to the bonding zone of the USEs probes through anisotropic conductive film (ACF) using flip-chip bonding⁹. For 240-channel USEs probes, two 120-channel FPCs were used to perform the electrical bonding. After being released from the silicon substrate, the USEs probes exhibited high flexibility in the air and water (Fig. 1f and Supplementary Fig. 7). To reduce the impedance and thermal noise^19,20, platinum (Pt) or poly(3,4-ethylenedioxythiophene):poly(styrene sulfonate) (PEDOT:PSS) was deposited on the surface of recording sites (Supplementary Fig. 8). The average impedance of effective channels was decreased to 50 ~ 60 kiloohms (at 1 kHz) after electrodeposition. To avoid being contaminated or broken before the implantation surgery, the flexible USEs probes were coated with polyvinyl alcohol (PVA) as a protective layer. After packaging with a 3D-printed shell, the total weight of USEs probes is ~ 1.7 g (Supplementary Fig. 9).

Fig. 1. Structural characterization and implantation method of the USEs probe.

a Schematics showing the overall structural design of the 240-channel USEs probe. b Photograph of an as-fabricated 240-channel USEs probe on the substrate. Scale bar, 2 mm. c Zoom-in view of the recording zone of the USEs probe as marked by the red dashed box in (b). Scale bar, 1 mm. d SEM image showing the structural design of the USEs probe. Scale bar, 500 μm. Inset, cross-sectional SEM image of an electrode filament. Inset scale bar, 10 μm. e SEM image showing the recording sites and central reference electrode. Scale bar, 200 μm. The SEM characterization experiments were conducted in two probes. f Photograph showing the flexibility of USEs probe after critical point drying. Scale bar, 2 mm. g Mechanical simulation of the USEs probe after stretching for a displacement of ~ 2500 μm. h Optical images showing the structure of the USEs probe after implantation into agarose gel. Scale bar, 1 mm (left) and 200 μm (right). The implantation experiment in agarose was repeated three times with similar results. i Schematics showing the implantation procedure in the mouse brain. j Photographs of the USEs probes before (left) and after (right) implantation into the mouse brain with the assist of an optical fiber. Scale bar, 1 mm. k Mouse with an implanted 240-channel USEs probe. Scale bar, 2 cm. l Impedance of the recording sites at 1 kHz before and after implantation into the mouse brain. Blue dots and red dots represent 120-channel probes (n = 367 electrodes) and 240-channel probes (n = 390 electrodes), respectively. Box plot in (l): Lower and upper box boundaries indicate the 25th and 75th percentiles, center lines indicate medians, and whiskers extend to the outermost data points within 1.5 × the interquartile range. Source data are provided as a Source Data file.

Ultralow bending stiffness of flexible neural probes precluded their direct implantation into brain tissue^21,22. Shuttle-assisted implantation of flexible electrode filaments has been regarded as an effective method for clinical brain-machine interface (BMI) studies^7,23. However, the distributed implantation of multiple electrode filaments will cause successive insertion injuries and increase the surgical risk. By contrast, all electrode filaments of the USEs probe can be simultaneously implanted into brain tissue with only one-time insertion process. This electrode design holds potential to reduce the implantation time and surgical risk. To evaluate the maximum insertion depth of the electrode filament, we performed finite element analysis (FEA) to calculate the stress distribution on the filament while being stretched and moving down by a specific depth (δz). As shown in Fig. 1g and Supplementary Fig. 10, the outermost circle of the recording zone was fixed in the mechanical simulation. To simulate the insertion process (Supplementary Video 1), the top of the rigid shuttle was subjected to a longitudinal force, leading to a downward displacement δz of the central reference electrode. Stress concentrations were mainly located at the bent locations of electrode filaments. The maximum von Mises stress had a positive correlation with displacement δz. When δz was ~ 2500 μm, the maximum stress was ~ 68 MPa, which was smaller than the yield strength of polyimide²⁴. The mechanical simulation results indicated that the USEs probes would not be broken within the safe insertion depth.

We first tested the feasibility of electrode implantation using 0.6% (wt/vol) agarose brain phantom²⁵. Before implantation, the USEs probe was uniformly placed on the moist agarose surface. A flat-tipped optical fiber (~ 190 μm in diameter) was first aligned with the USEs probe’s central reference electrode, then vertically lowered to fully implant all the recording sites into agarose gel (Fig. 1h, Supplementary Fig. 11 and Video 2). The insertion speed could be adjusted manually in the range of 15 ~ 20 μm/s. Serpentine parts of the electrode filaments underwent gradual stretching on the agarose surface during implantation. After implantation, the linear parts of the 8 electrode filaments assembled into a cylinder-like configuration. Based on the implantation experiments in agarose, we developed in vivo insertion protocols for rodent models. As shown in Fig. 1i and Supplementary Fig. 12, the in vivo implantation procedure comprised three sequential steps. Firstly, a rigid shuttle (e.g., flat-tipped optical fiber) was aligned with the central reference electrode of the USEs probe, which had been positioned on the mouse brain and aligned with the cranial window. To minimize friction and cutting forces exerted by the electrode filaments during the stretching process, an alginate hydrogel film was pre-applied to the brain surface²⁶. Subsequently, the rigid shuttle was moved downward, causing the serpentine parts to stretch and enabling the implantation of recording sites into the brain tissue. During the implantation process, we didn’t observe significant lateral cutting of electrode filaments through the alginate hydrogel layer and their subsequent contact with the cranial window border, demonstrating the mechanical flexibility of the USEs probe. Given the mechanical similarity between hydrogel and brain tissue, it can be inferred that a minimal planar lesion in brain tissue was induced during implantation. Finally, the rigid shuttle was retracted, leaving all the recording sites in situ within the brain tissue. The effectiveness of aforementioned implantation methods has been demonstrated in mice (Fig. 1j, k, Supplementary Fig. 13 and Video 3, 4). The average impedance of the USEs probes increased from ~ 50 kiloohms to several hundred kiloohms after implantation into brain tissue (Fig. 1l).

Interfaces between the USEs probe and brain tissues

The biocompatibility of neuroelectronic interfaces directly influences the chronic recording properties of implanted electrodes²⁷. To characterize the interfaces between the USEs probe and brain tissues, we implanted the USEs probes into the hippocampal region of Thy1-YFP transgenic mice. A 1 mm-thick brain slice with an intact USEs probe was sectioned and rendered transparent via the PEGASOS tissue clearing method²⁸. Consistent with agarose implantation, the electrode filaments of implanted USEs probes assembled into open cylinder-like structures within brain tissue (Fig. 2a, b and Supplementary Fig. 14). 3D reconstructed images showed the interfaces between the chronically implanted USEs probe and surrounding neurons (Fig. 2c, d). For both 120-channel (at 16 wpi) and 240-channel USEs probes (at 21 wpi), minimal neuronal loss was observed around the electrode filaments. Moreover, the interpenetration of USEs probes with surrounding neurons confirmed the formation of seamless neuroelectronic interfaces. The bending behavior of electrode filaments demonstrated their flexibility within brain tissue. The recording sites of USEs probes were comparable to neuronal somas in size, permitting the neural signal detection with high spatial resolution.

Fig. 2. Interfaces between the USEs probe and brain tissues.

a A 240-channel USEs probe in a 1 mm-thick brain slice after tissue clearing. Scale bar, 200 μm. b Micro-CT image of the 240-channel USEs probe in brain tissue. Scale bar, 200 μm. c 3D reconstructed image of a 1 mm-thick brain slice with 120-channel USEs probe at 16 weeks post-implantation. Green: YFP-expressing neurons; Red: polyimide insulating layers. Scale bar, 100 μm. d 3D reconstructed image of a 1 mm-thick brain slice with 240-channel USEs probe at 21 weeks post-implantation. Scale bar, 100 μm. e Confocal fluorescence and bright-field images of a 50 μm-thick brain slice with 120-channel USEs probe at 16 weeks post-implantation (zoom-in view of the electrode-tissue interface in Supplementary Fig. 15), which was labeled for nuclei (DAPI, blue), microglia (Iba1, purple), reactive astrocytes (GFAP, red), and mature neurons (NeuN, green), respectively. Scale bar, 50 μm. The immunohistochemical experiment was repeated four times with similar results.

Besides neurons, glia cells also play a pivotal role in neuroelectronic interfaces²⁹. We evaluated the immune responses of chronically implanted USE probes by performing the immunohistochemical staining. The USEs probes were implanted into the prefrontal cortex (PFC) of C57BL/6 J mice. After 8 ~ 16 weeks post-implantation, the mouse brains with implanted USEs probes were removed and then sectioned into horizontal slices with a thickness of 30 or 50 μm. Bright-field imaging revealed that cross-sections of 8 electrode filaments in the brain slices exhibited circular or oval profiles (Fig. 2e and Supplementary Figs. 15–18), consistent with the open cylinder-like configuration. DAPI, Iba1, GFAP and NeuN were used to label the nuclei, microglia, reactive astrocytes, and mature neurons, respectively. The immunohistochemical results showed a significant increase in the density of glia cells (including microglia and reactive astrocytes) inside the open cylinder compared to outside. A growing body of literature indicates that both microglia^30,31 and reactive astrocytes^32–35 promote neurological recovery after injury. During chronic implantation of 120-channel USEs probes, neurons inside the open cylinder revealed a good recovery of neuronal tissue. For 240-channel USEs probes at 16 weeks post-implantation, a neuron ‘kill zone’ inside the open cylinder persisted, although partial recovery of the neuronal tissue surrounding the electrode filaments was observed (Supplementary Fig. 19). Moreover, a significant reduction of neuronal density outside the open cylinder was detected over a range of 0-100 μm away from the 240-channel USEs probes. In contrast, 120-channel USEs probes showed minimal neuronal loss outside the open cylinder at 16 weeks post-implantation (Supplementary Fig. 20). This difference in tissue recovery rates between 120-channel and 240-channel probes might be attributed to their mechanical structures and resulting interfacial strains (stresses) at the electrode-tissue interfaces³⁶. For both 120-channel and 240-channel USEs probes, we observed significant activation of astrocytes near the electrode filaments, suggesting the relevance of reactive astrocytes to tissue recovery. Notably, the recovered neuronal tissue inside the open cylinder effectively prevented electrode drift, facilitated by the central reference electrode at the bottom. This anchor effect facilitated the long-term mechanical stability of electrode filaments in the brain parenchyma. A slight decrease in neuronal density was observed within ~ 50 μm of some electrode filaments, though they remained close enough to detect neuronal spike firing³⁷. Moreover, the chronically implanted electrode filaments showed no glial encapsulation, indicating the absence of excessive gliosis at electrode-tissue interfaces.

Chronic neural recordings in both head-fixed and free-moving mice

To test the chronic in vivo recording capabilities, the USEs probes were implanted into the PFC or motor cortex of C57BL/6 J mice (Fig. 3a and Supplementary Fig. 21). Given that 120-channel probes exhibited faster tissue recovery rates, we primarily used them for subsequent recordings in this study. After a 2-week recovery from the implantation surgery, neural recordings were performed once a week. Before recording, the head-fixed mice with implanted USEs probes were connected to the electrophysiological recording system via customized 120-channel printed circuit board (PCB) (Supplementary Fig. 22). For a representative mouse (Mouse-2) at 2 weeks post-implantation of the 120-channel USEs probe, ~ 60 channels could detect the neuronal spike firing (Fig. 3b). The action potential (AP) signals after band-pass (250–5000 Hz) filter indicated that the USEs probes could record neuronal activities with millisecond-scale temporal resolution (Fig. 3c). Figure 3d showed the average spike waveforms from ~ 91 neuronal spike units with a recorded unit yield of ~ 0.78. Despite primarily using 120-channel USEs probes for recordings, our preliminary data indicated that 240-channel USEs probes (Mouse-4) could also detect single-neuron spike signals from large neuronal ensembles (Supplementary Figs. 23, 24). Statistical analysis of signal data from three mice with 120-channel probes (Mouse-1, Mouse-2, Mouse-3) showed that the spike amplitude, firing rate, and signal-to-noise ratio (SNR) of recorded units remained relatively stable from 2 to 10 weeks post-implantation (Fig. 3e–g and Supplementary Fig. 25). The in vivo impedance further indicated the chronic recording stability of USEs probes during the tissue recovery period.

Fig. 3. Chronic neural recordings using 120-channel USEs probes in both head-fixed and free-moving mice.

a Schematics showing the mouse brain with an implanted USEs probe. b Spike rasters recorded by a 120-channel USEs probe at 2 weeks post-implantation. Scale bar, 1 s. c AP traces recorded by the USEs probe in (b). Scale bars, 100 ms (horizontal) and 100 μV (vertical). d Average spike waveforms detected from the AP traces in (c). Scale bars, 1 ms (horizontal) and 100 μV (vertical). e–g Statistical analysis of spike amplitude (e), firing rate (f), and SNR (g) for all the recorded units as a function of time post-implantation. N = 3 mice (represented with 3 colors). The sample size at each timepoint is available in Supplementary Table 1. Box plot in (e–g): Lower and upper box boundaries indicate the 25th and 75th percentiles, center lines indicate medians, center squares indicate mean values, and whiskers extend to the outermost data points within 1.5 × the interquartile range. h Schematics showing the experimental setup for neural recording in a freely moving mouse in a circular arena. i Trajectory of the free-moving mouse in the circular arena. j Average spike waveforms recorded by 7 example channels from 3 to 8 weeks post-implantation. Scale bars, 1 ms (horizontal) and 100 μV (vertical). k Spike firing of a representative neuron is phase-locked to the theta oscillation (4–10 Hz) of LFP. Scale bars, 100 ms (horizontal) and 200 μV (vertical). l ISI histograms and spike waveforms of the representative neuron in (k) from 3 to 8 weeks post-implantation. Colored lines indicate the mean spike waveforms, with raw spike waveforms overlaid in gray. Scale bars, 500 μs (horizontal) and 50 μV (vertical). m Polar plots showing the phase locking of neuronal firing to theta oscillation from 3 to 8 weeks post-implantation. n Phase value distribution of neuronal spikes in theta oscillation from 3 to 8 weeks post-implantation. Source data are provided as a Source Data file.

Single-neuron recording in freely moving animals is critical for neuroscience research^38–40. However, long-term stable recording of neuronal action potentials in behaving animals remains challenging owing to progressive signal degradation^41,42. Here, we demonstrated the chronic recording capability of USEs probes in freely moving mice. 120-channel USEs probes were implanted into the hippocampus of mouse, which was put into a circular arena during neural recording (Fig. 3h). Customized PCB was used to electrically connect the USEs probes in mouse brain and signal acquisition system (Supplementary Fig. 26a, b). Simultaneous trajectory tracking and neural recording were performed once a week in freely moving mice (Supplementary Fig. 26c and Fig. 3i). Action potentials from a neuronal ensemble (Mouse-5) were recorded from 3 to 8 weeks post-implantation (Fig. 3j). Spike timing of individual neurons relative to LFP oscillations is critical for stimuli representation, memory formation, and social perception^43,44. Nevertheless, the temporal evolution of single-neuron phase locking to LFP oscillations over an extended timescale (e.g., months) is not well studied. We observed significant phase locking of single-unit spikes relative to theta (4–10 Hz) oscillation in hippocampus (Fig. 3k). A representative neuron was stably tracked from Channel-46. Correlation coefficient (r) was calculated to quantify the mean waveform similarity of single units recorded at different timepoints⁴⁵. High waveform similarity (r > 0.91) and similar ISI distributions across time indicated that the spiking activities at different timepoints were from the same neuron (Fig. 3l and Supplementary Table 2). The polar plots of this neuron exhibited consistent phase value distributions during the entire recording period (Fig. 3m, n). These results confirmed that the USEs probes could stably record single-neuron spiking activities in freely moving mice over months, holding potential for the study of neural mechanisms underlying animal behaviors.

Ultra-long-term stable recording of the same neuronal ensembles

Ultra-long-term stable mapping of the same neuronal populations over months to years could advance our understanding of learning and aging^6,9. To further evaluate the recording stability of the USEs probes in brain tissue over extended periods (beyond 3 months), we chronically recorded the neural activities in two representative mice (Mouse-1 and Mouse-3). At 12 weeks post-implantation, high-SNR action potentials were recorded from Mouse-1 with an implanted 120-channel USEs probe (Fig. 4a–c), yielding 73 isolated neuronal spiking units. Spiking activities of neuronal ensembles were continuously recorded from 12 to 80 weeks post-implantation with a total of ~60 channels (Fig. 4d). As shown in Supplementary Table 3, 19.2% (14/73) neuronal units in 23.3% (14/60) channels were stably recorded from 12 to 80 weeks post-implantation (wpi). Although the ratio of stably recorded units gradually decreased as the timespan length increased (Supplementary Fig. 27), it remained relatively consistent or gradually increased over time post-implantation (Fig. 4e, f). 39.7% (29/73), 44.0% (37/84), 50.0% (32/64), 68.3% (43/63), 67.1% (47/70), and 77.8% (49/63) neuronal units were stably recorded for 12-24 wpi, 24-36 wpi, 36-48 wpi, 48-60 wpi, 60-72 wpi, and 72-80 wpi, respectively (Supplementary Fig. 28). The specific quantity of channels detecting spike firing was 47 ± 4 with a recorded unit yield of 0.61 ± 0.08 over time (Fig. 4g). Despite variations in recording results across time, around 30 channels could consistently detect neuronal spike firing throughout the entire period (Supplementary Fig. 29). To further assess the ultra-long-term recording stability of USEs probes, we statistically analyzed multiple recording parameters, including peak-to-peak spike amplitude, noise level, SNR, and firing rate. The average spike amplitude (> 100 μV) and SNR (> 10) of recorded neuronal units remained stable from 12 to 80 weeks post-implantation (Fig. 4h–j), suggesting that the electrodes formed intimate interfaces with surrounding neurons. The average firing rate of neuronal populations exhibited slight fluctuations but maintained relatively stable overall (Fig. 4k), potentially due to a global homeostatic mechanism⁴⁶. Since electrode impedance is a key factor in recording quality⁴⁷, we monitored the in vivo impedance of electrode channels throughout the long-term recording period. An average impedance of several hundred kiloohms indicated no degradation in electrochemical properties at electrode-tissue interfaces after ~ 1.5 years post-implantation (Fig. 4l). In addition to Mouse-1, a neuronal ensemble in Mouse-3 was also stably recorded from 14 to 28 weeks post-implantation (Supplementary Fig. 30). These results demonstrated the capability of the USEs probe for ultra-long-term recording of neuronal populations.

Fig. 4. Ultra-long-term stable recording of the same neuronal ensembles in Mouse-1 with 120-channel USEs probe from 12 to 80 wpi.

a Photograph of a representative mouse (Mouse-1) head-fixed on the stereotaxic apparatus for neural recording. Scale bar, 10 mm. b Representative AP traces recorded at 12 weeks post-implantation. Scale bars, 100 ms (horizontal) and 200 μV (vertical). c Zoom-in view of the AP traces as marked by the black dashed box in (b). Scale bars, 10 ms (horizontal) and 200 μV (vertical). d Average spike waveforms of the same neuronal ensembles recorded from 12 to 80 wpi. Scale bars, 1 ms (horizontal) and 200 μV (vertical). e, f Ratio of stably recorded units over implantation time. g Number of channels with spike (red) and recorded unit yield (blue) as a function of time. h Average spike amplitude of the recorded units as a function of time. i Noise level of the AP traces after bandpass (250–5000 Hz) filtering. j, k SNR (j) and firing rate (k) of the recorded units as a function of time. l Impedance of channels (with impedance below 1.5 MΩ at 1 kHz) as a function of time. The sample size in (h–l) at each timepoint is available in Supplementary Table 4. Box plot in (h–l): Lower and upper box boundaries indicate the 25th and 75th percentiles, center orange lines indicate medians, center red squares indicate mean values, and whiskers extend to the outermost data points within 1.5 × the interquartile range. Source data are provided as a Source Data file.

Ultra-long-term stable recording of individual neurons

The balance between stability and plasticity in neuronal processing is important for brain function. To understand this kind of balance, longitudinal tracking of neural activities from individual neurons is necessary⁴⁸. However, consistent tracking of the same individual neurons over ultra-long-term timescales is challenging due to probe drift and glial scarring. Considering that chronically implanted USEs probes exhibited reduced electrode drift and gliosis, we next demonstrated the capabilities of USEs probes for ultra-long-term tracking of individual neurons. For Channel-80 in Mouse-1, principal component analysis (PCA) consistently showed two well-isolated spike clusters from 12 to 32 weeks post-implantation, indicating that spike firing activities from two distinct neurons (Neuron 1 and Neuron 2) were stably recorded (Fig. 5a–c). For both Neuron-1 and Neuron-2, the mean waveform similarity of recorded neuronal spikes was high (r > 0.99) across time (Supplementary Table 5). The average spike amplitude of Neuron-1 and Neuron-2 at 12 weeks post-implantation was 177 μV and 94 μV, respectively. Compared to Neuron-2, Neuron-1, with a larger spike amplitude exhibited greater amplitude fluctuations over time (Fig. 5d), which might be due to its closer electrode-neuron distance and higher micromotion sensitivity⁴⁹. With intrinsic changes over time, the firing rates of Neuron-1 and Neuron-2 gradually exhibited increased coherence (Fig. 5e), reflecting potential plasticity of individual neurons. Inter-spike interval (ISI) distributions of Neuron-1 and Neuron-2 exhibited distinct patterns (Fig. 5f), revealing their disparity in firing feature. However, the ISI distributions of each neuron were relatively stable across time (Fig. 5g). Similar analysis indicated that two neurons were consistently recorded by Channel-116 in Mouse-6 from 20 to 26 weeks post-implantation (Supplementary Fig. 31 and Table 6). In addition, two representative single units were stably detected by Channel-74 in Mouse-1 from 32 to 80 weeks post-implantation (Fig. 5h). The largely overlapping spike clusters and high waveform similarity (r > 0.93) of each single unit across time (Fig. 5i, j and Supplementary Table 7) suggested that they were recorded from the same individual neurons (Neuron-3 and Neuron-4). The ultra-long-term recording capability of USEs probes enabled us to study the firing properties of individual neurons for almost one year. During the entire recording period, the single-neuron signal quality (including spike amplitude and SNR) exhibited high fidelity (Fig. 5k). At 80 weeks post-implantation, the average spike amplitude of Neuron-3 and Neuron-4 was 306 μV (SNR ~ 26.6) and 139 μV (SNR ~ 11.9), respectively. Intrinsic evolution of firing rate also occurred in these two neurons over time (Fig. 5l). The shape of ISI histograms for both Neuron-3 and Neuron-4 remained relatively stable from 32 to 80 weeks post-implantation (Fig. 5m, n).

Fig. 5. Ultra-long-term stable recording of individual neurons using 120-channel USEs probe.

a Time evolution of PC clusters for two representative neurons (Neuron 1 and Neuron 2) from 12 to 32 wpi. b Centers (dots) and 2σ (ovals) of the PC clusters for Neuron 1 and Neuron 2. c Spike waveforms of Neuron 1 and Neuron 2. Scale bars, 500 μs (horizontal) and 50 μV (vertical). d Spike amplitude of Neuron 1 and Neuron 2 as a function of time. n = 1000 spike waveforms at each timepoint for each neuron. e Firing rate of Neuron 1 and Neuron 2 as a function of time. n = 12 epochs. Data in (d, e) are presented as mean ± SD. f ISI histograms of Neuron 1 and Neuron 2 recorded at 32 wpi. g Time evolution of ISI histograms for Neuron 1 and Neuron 2 from 12 to 32 wpi. h Time evolution of PC clusters for two representative neurons (Neuron 3 and Neuron 4) from 32 to 80 wpi. i Centers (dots) and 2σ (ovals) of the PC clusters for Neuron 3 and Neuron 4. j Spike waveforms of Neuron 3 and Neuron 4. For (c, j), colored lines indicate the mean spike waveforms, with raw spike waveforms overlaid in gray. Scale bars, 500 μs (horizontal) and 50 μV (vertical). k Spike amplitude of Neuron 3 and Neuron 4 as a function of time. n = 1000 spike waveforms at each timepoint for each neuron. l Firing rate of Neuron 3 and Neuron 4 as a function of time. n = 12 epochs. Data in (k, l) are presented as mean ± SD. m ISI histograms of Neuron 3 and Neuron 4 recorded at 80 wpi. n Time evolution of ISI histograms for Neuron 3 and Neuron 4 from 32 to 80 wpi. o Polar plots showing the phase locking of neuronal firing to gamma oscillation (30–90 Hz) from 32 to 80 wpi. p Length of resultant vector for Neuron 3 and Neuron 4 as a function of time. Source data are provided as a Source Data file.

Based on the ultra-long-term recording data from Neuron-3 and Neuron-4 in PFC, we studied the phase locking properties of these two neurons relative to gamma oscillation (30–90 Hz) from 32 to 80 weeks post-implantation. The polar plots of both neurons showed similar phase value distributions over the entire period (Fig. 5o and Supplementary Fig. 32). The mean preferred phases of these two neurons were close to the trough (~ 180°) of gamma oscillation. The lengths of mean resultant vectors remained relatively stable during the one-year recording period (Fig. 5p), demonstrating the ultra-long-term stability of neuronal phase-locking properties to gamma oscillation.

Ultra-long-term neural recording in the mouse visual cortex

Stable recording of the neuronal activities in the visual cortex is important for the mechanism study of the balance between plasticity and stability in visual system^50,51. Here, we used the USEs probes to detect the spiking activities from neuronal ensembles in mouse primary visual cortex (V1) and study their dynamic responses to grating stimuli over an extended period of several months. Simultaneous spike recording and visual stimulation were performed in head-fixed mice with chronically implanted USEs probes (Fig. 6a). Orientation selectivity was clearly observed in partial V1 neurons upon drifting grating stimulation (Fig. 6b, c). For example mouse (Mouse-7), action potentials from the V1 neuronal populations were consistently detected from 16 to 35 weeks post-implantation (Fig. 6d and Supplementary Fig. 33). The detected spiking neuronal populations exhibited some dynamic evolution over time. While there were some fluctuations in the recorded unit number, 30 ~ 60 spike units were detected during this recording period (Fig. 6e). We analyzed the signal metrics of these recorded units in V1. Average spike amplitude (~ 80 μV), SNR (~ 10), and trough-to-peak duration (~ 0.4 ms) maintained stable from 16 to 35 weeks post-implantation (Fig. 6f–h), demonstrating the ultra-long-term signal stability of USEs probes in mouse V1. To evaluate the orientation selectivity (OS) of recorded neurons, we calculated their orientation selectivity index (OSI)^52,53. As shown in Fig. 6i, ~ 30% spike units in V1 exhibited significant orientation selectivity (OSI ≥ 0.3) at week 16 post-implantation. The proportion of neuronal units exhibiting OS fluctuated, ranging from 22% to 42% across the entire recording period. The average OSI of orientation-selective neurons ranged from 0.37 to 0.56 (Fig. 6j), which was basically consistent with previous reports⁵⁴. Temporal fluctuations in the distribution of orientation preferences (Fig. 6k) may be attributed to two factors: (1) dynamic evolution of the recorded V1 neuronal populations, and (2) absence of orientation maps in rodents, where neighboring neurons often exhibit different orientation preferences^55,56. Throughout the recording period with visual stimuli, average firing rates of OS neuronal populations were robustly modulated by preferred orientations relative to orthogonal ones (Fig. 6l). Among the V1 neuronal populations, some orientation-selective neurons were stably tracked over months, and their orientation selectivity profiles remained highly consistent (Fig. 6m–o and Supplementary Fig. 34 and Table 9). These results demonstrated the ultra-long-term recording capability of USEs probes in the rodent visual cortex, highlighting their potential for fundamental neuroscience research.

Fig. 6. Ultra-long-term neural recording in the mouse visual cortex from 16 to 35 wpi.

a Experimental setup for simultaneous neural recording and visual stimulation in head-fixed mice. Drifting grating stimuli (12 orientations) were presented in random order. b, c Voltage traces (b) and spike rasters (c) of an example neuron responding to 12 grating orientations (40 trials each). Red and blue arrows indicate stimulus onset and offset, respectively. Scale bars in (b), 200 ms (horizontal) and 50 μV (vertical). d Average spike waveforms of the V1 neuronal populations recorded from 17 to 35 wpi. Scale bars, 1 ms (horizontal) and 100 μV (vertical). e–h Recorded unit number (e), average spike amplitude (f), SNR (g), and trough-to-peak duration (h) as a function of time from 16 to 35 wpi. Data in (f–h) are presented as mean ± SEM. i Proportion of orientation-selective neurons as a function of time. j OSI of orientation-selective neurons as a function of time. Sample size at each timepoint is available in Supplementary Table 8. Box plot in (j): Lower and upper box boundaries indicate the 25th and 75th percentiles, center orange lines indicate medians, center red squares indicate mean values, and whiskers extend to the outermost data points within 1.5 × the interquartile range. k Distribution of orientation preferences for the OS neuronal populations. Different colors indicate different orientations. l Firing rates of the OS neuronal populations upon preferred (red) and orthogonal (blue) visual stimuli. The baseline (200 ms before stimulus onset) FR was subtracted. Data are presented as mean ± SEM. m An example V1 neuron was detected from 29 to 35 wpi by Channel-93 in Mouse-7. Blue lines indicate the mean spike waveforms, with raw spike waveforms overlaid in gray. Scale bars, 500 μs (horizontal) and 100 μV (vertical). n Firing rates of the example neuron in (m) upon visual stimuli with varied orientation angles. n = 40 trials per orientation angle. Data are presented as mean ± SEM. o Polar plots showing the mean values of firing rates in (n) from 29 to 35 wpi. Source data are provided as a Source Data file.

Discussion

Shuttle-assisted implantation of flexible electrode threads holds great potential for both neuroscience research and clinical translation^10,23. The USEs probes developed in this study enabled simultaneous implantations of multiple electrode filaments into brain tissue with a one-time insertion process, providing an efficient implantation method for high-density electrode arrays. The open cylinder-like structures of USEs probes within brain tissue elicited no excessive gliosis or glial scar. Furthermore, the anchor effect between chronically implanted USEs probes and recovered neuronal tissue facilitated the formation of a robust neuroelectronic interface with reduced probe drift. As a result, the USEs probes could stably track the same neuronal ensembles in head-fixed mice over a years-long timescale, paving the way for ultra-long-term neural recordings during brain development and the aging process. Besides rodent models, the USEs probes also hold great potential for the long-term chronic recording in large animal brains or spinal cords, a task that is technically challenging due to dynamic pulsations and deformations. To expand the spatial coverage across distributed brain regions, multiple hexagonal USEs probes can be integrated into a single hierarchical device using a honeycomb-like configuration. Considering that the USEs probes could be implanted into brain tissue with the assist of optical fiber, integrating the optogenetics manipulation with electrophysiological recording is totally doable in the future. Bi-directional neural interfaces based on USEs probes offer promising tools for dissecting neural circuits.

Methods

Animals and ethics

All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the Chinese Institute for Brain Research, Beijing (CIBR) under protocol CIBR-IACUC-057 and CIBR-IACUC-187. Male C57BL/6 J wild-type mice (Jackson Laboratory stock No. 000664) and Thy1-YFP transgenic mice were acquired from the Laboratory Animal Resource Center (LARC) at CIBR. Sex was not considered in this study, because the main aim here is to develop robust neuroelectronic interfaces and validate their ultra-long-term recording stability, rather than to investigate sex-dependent issues. C57BL/6 J mice were used to evaluate the chronic recording capability of the USEs probes and perform the immunohistochemical experiments. Thy1-YFP transgenic mice were used to characterize the interfaces between the USEs probe and brain tissues. Mice were 8 ~ 10 weeks old at surgery. Before implantation surgery, mice were housed in groups (≤ 5 mice per cage). After implantation surgery, mice were housed individually. Housing conditions are as follows: 12 h light/dark cycle; 21.5 ~ 23.5 °C temperature; 30 ~ 70% humidity; food and water ad libitum.

Fabrication of the USEs probes

All devices used in this work were fabricated on the 4-inch silicon wafers. The micro-fabrication processes of the USEs probes were shown in Supplementary Fig. 1. Main steps are as follows: (1) To form the aluminum (Al) sacrificial layer under the flexible devices, S1813 photoresist (Microposit, Dow Chemical Co.) was spin-coated (2000 r.p.m.) and then baked (115 °C, 3 min). The MA6 mask aligner (SUSS MicroTec) was used to perform the photolithography and pattern the photoresist layer. Al sacrificial layer with a thickness of 100 nm was deposited with magnetron sputtering (Lab 18, Lesker), followed by lift-off process. (2) The bottom insulating layer with a thickness of ~ 1.6 μm was formed by spin-coating of polyimide (U-Varnish S, UBE Industry, Ltd.) and baking in vacuum (200 °C, 2 h). (3) LOR 3 A/S1813 double-layer resist was used to form the gold conducting layer. LOR 3 A (MicroChem. Corp., now Kayaku Advanced Materials, Inc.) was spin-coated (2000 r.p.m.) and then baked (170 °C, 10 min). S1813 was spin-coated (5000 r.p.m.) and then baked (115 °C, 3 min). Photolithography was performed to pattern the LOR 3 A/S1813 double-layer resist, followed by deposition of Cr (10 nm)/Au (100 nm) using e-beam evaporation. After the lift-off process, the first metal conducting layer was formed. (4) The middle-layer polyimide insulating layer with a thickness of ~0.8 μm was formed using the same method in step 2. (5) The second metal conducting layer was formed by repeating step 3. (6) The top-layer polyimide insulating layer with a thickness of ~ 0.8 μm was formed using the same method in step 2. (7) AZ4620 photoresist (Merck Electronic Materials Ltd.) was spin-coated (2000 r.p.m.) and baked (120 °C, 12 min). Photolithography was performed to pattern the AZ4620 photoresist, followed by oxygen (O₂) reactive ion etching (RIE) (Etchlab 200, SENTECH Instruments GmbH) to form the umbrella-type structure of the USEs probe. (8) To connect the USEs probes with customized flexible printed circuits (FPCs), Cr (10 nm)/Au (200 nm) was deposited as the metal input/output (I/O) pad layer using the similar method with step 3. (9) A SU-8 (MicroChem. Corp., now Kayaku Advanced Materials, Inc.) layer with a thickness of ~ 2 μm was patterned by photolithography to structurally reinforce the junction regions between central reference electrode and electrode filaments. (10) The wafer was cut into individual USEs devices, followed by flip-chip bonding between the I/O pad layer on the USEs probe and customized FPCs using the anisotropic conductive film (ACF, DP3342MS, Sony Chemical & Device Corporation). (11) To release the USEs probe from the Si substrate, the Al sacrificial layer was etched by immersing in the FeCl₃ (012357.22, Alfa Aesar, now Thermo Scientific) solution (1 mol/L). (12) Platinum or PEDOT was electrodeposited on the recording sites of USEs probes by using the three-electrode system with an electrochemical workstation (Gamry Interface 1010E). For the electrodeposition of platinum, the plating solution contained 12 mmol/L chloroplatinic acid (C805194, Shanghai Macklin Biochemical Co., Ltd.) and 1.05 mmol/L lead acetate (L196609, Aladdin). For the electrodeposition of PEDOT:PSS, the plating solution contained 10 mmol/L 3,4-ethylenedioxythiophene (EDOT, 483028, Sigma-Aldrich) and 2 wt.% polystyrene sulfonate (PSS, 527483, average Mw ~70000, Sigma-Aldrich). A constant voltage of − 0.1 V and 1 V was applied for the electrodeposition of platinum and PEDOT, respectively. (13) PMMA (950 C9, Kayaku Advanced Materials, Inc.) was spin-coated (4000 r.p.m.) on the Si wafer and then baked (180 °C, 10 min) to form the PMMA/Si substrate. Si substrate under the released USEs probe was trimmed and removed, after which the USEs probe was transferred onto the PMMA/Si substrate. (14) 5% (wt/vol) PVA solution (363170, Mw 13,000 ~ 23,000, 87–89% hydrolyzed, Sigma-Aldrich) was used to coat the connecting zone and recording zone of the USEs probe and then baked (45 °C, 60–90 min). Subsequently, another layer of 5% (wt/vol) PVA (363081, Mw 85,000 ~ 124,000, 87–89% hydrolyzed, Sigma-Aldrich) was used to coat the connecting zone of the USEs probe for structural support and then baked (45 °C, 60–90 min). (15) After removing the PMMA layer with acetone, the PVA-supported USEs probe was released from the substrate. (16) The bonding zone of the device was packaged using copper foil to reduce the electromagnetic interference.

Structural characterization of the USEs probes

The optical images of the devices were acquired by using a light microscope (BX53, Olympus) and an upright materials microscope (DM4 M, Leica). A SEM/FIB dual beam system (Nova 200 Nanolab, FEI) was used to observe the internal structure of the USEs probes with ultra-high resolution. Morphology of the USEs probes after implantation into brain tissue was investigated with the 3D X-ray microscope (Xradia 520 Versa, Carl Zeiss) and the stereo microscope (SMZ745T, Nikon).

Finite element analysis

The three-dimensional finite element analysis (FEA) was performed by using Abaqus software (2022). The recording zone of the USEs probe (thickness ~ 3 μm) was modeled with a scale of 1:1. To simplify the finite element model, 100 nm-thick gold conducting layers were not included. The Young’s modulus and Poisson ratio of the polyimide were set as 3.15 GPa and 0.34, respectively. To simulate the probe insertion process, an alginate hydrogel layer and brain tissue were also modeled. 400-μm-diameter cylindrical holes were made in the models of alginate hydrogel and brain tissue, providing an implantation path for the rigid shuttle (190 μm diameter) and USEs probe. The outermost circle of the recording zone was fixed in the mechanical simulation. The top of the rigid shuttle was subjected to a longitudinal force, leading to a downward displacement δz of the central reference electrode.

Implantation of the USEs probe into agarose brain phantom

0.6% (wt/vol) agarose (A8190, Solarbio) gel-based brain phantom was used to investigate the structure of the USEs probe after stretching and implantation in vitro. A PVA-supported USEs probe was fixed on the stereotaxic apparatus (RWD Life Science Co. Ltd.). The recording zone was placed on the surface of agarose gel, after which the PVA protective layer was dissolved. An optical fiber (diameter: ~190 μm) with a smooth bottom surface was aligned with the central reference electrode of the USEs probe, and then slowly moved down, implanting the electrodes into the agarose gel. After implantation, the optical fiber was retracted, while the flexible electrode filaments were left in the agarose gel.

Animal surgeries

For the implantation surgery, mice were anesthetized by intraperitoneal injection of tribromoethanol (1.25%, 0.02 mL/g weight, Nanjing Aibei Biotechnology Co., Ltd.) and then head-fixed on the stereotaxic apparatus (RWD Life Science Co., Ltd.). A metal head plate was horizontally attached to the skull surface with tissue adhesive (1469SB, 3 M Vetbond). Two stainless-steel screws (diameter: 1 mm) were implanted into the cerebellum as grounding channels. A cranial window (diameter: 0.8 ~ 1 mm) was drilled above the target brain area, and the dura was carefully removed. Alginate hydrogel film (thickness: 0.3 ~ 0.5 mm) was prepared by crosslinking 5% (wt/vol) sodium alginate (S817373, Macklin) with 0.1 mol/L calcium chloride (C805228, Macklin) solution, and then placed on the brain surface. The recording zone of the PVA-supported USEs probe was slowly moved down using the micromanipulator and then placed on the surface of the alginate hydrogel film. After dissolving the PVA protective layer with deionized water, the recording zone of the ultraflexible USEs probe was completely exposed. The location of the USEs probe was adjusted to ensure that the central reference electrode was aligned with the cranial window. An optical fiber (diameter: ~ 190 μm) with a smooth bottom surface was aligned with the central reference electrode of the USEs probe, and then implanted the electrodes into the target brain region. To reduce the friction force during the stretching process of the flexible filaments, the surface of the alginate hydrogel film was kept moist. After implantation, the optical fiber was carefully retracted, while the flexible electrode filaments were left in the brain tissue. Redundant alginate hydrogel film outside the USEs probe was removed. Dental cement was used to fix the probe and FPC onto the mouse skull.

Tissue clearing and imaging

Thy1-YFP transgenic mice with implanted USEs probes were anesthetized by intraperitoneal injection of tribromoethanol (1.25%, 0.02 mL/g weight), after which ~ 30 mL normal saline (0.9%, SJZ No.4 Pharmaceutical) and ~ 30 mL paraformaldehyde (PFA, 4%, BL539A, Biosharp) was transcardially perfused. After decapitation, the brains with USEs probes were carefully extracted and incubated in 4% PFA (4 °C, ~ 24 h) for postfixation. 1 mm-thick brain slices with embedded USEs probes were acquired by vibratome (Leica VT1200S, Leica Biosystems) sectioning along the sagittal direction. The brain slices were washed with PBS three times (20 min every time), after which tissue clearing was performed with the previously reported PEGASOS method²⁸. Multiple steps, including decolorization, delipidation, dehydration, and clearing, were completed successively within 1 week. After turning transparent, the brain slices were imaged using the upright confocal microscopes (Leica Stellaris 8 DIVE with 20 × /0.95 oil-immersion objective for the local imaging, and Leica TCS SP8 with 10 × /0.40 objective for the global imaging). 561 nm and 488 nm lasers were used to image the polyimide insulating layers and YFP-expressing neurons, respectively. Imaging analysis, as well as 3D reconstruction and visualization, were performed with the Imaris (Bitplane, Oxford Instruments) and LAS X (Leica) softwares.

Immunohistochemical staining

C57BL/6 J mice with implanted USEs probes were anesthetized, and then transcardially perfused with normal saline (0.9%, SJZ No.4 Pharmaceutical) and PFA (4%, BL539A, Biosharp). After decapitation, the brains with USEs probes were extracted and immersed in 4% PFA (4 °C, ~ 24 hours) to complete the postfixation. The brains were immersed in 30% (g/mL) sucrose solution in PBS (4 °C) for dehydration until they sank to the bottom. Dehydrated brains were embedded in the OCT compound (Tissue-Tek, 4583, Sakura Finetek Japan Co., Ltd.), and then frozen at − 20 °C for 1 ~ 2 h. 30 μm-thick or 50 μm-thick brain slices were sectioned perpendicular to the implanted USEs probe using the freezing microtome (Leica CM3050S, Leica Biosystems). 5% (g/mL) bovine serum albumin (BSA, Fraction V) solution (in PBS) was prepared for the follow-up experiments. The brain slices were washed with PBS two times (10 min every time) to remove the OCT compound, and then incubated with 0.3% Triton X-100 (T8200, Solarbio) solution (in BSA solution) at room temperature (RT) for 15 min to increase the cell-membrane permeability. After being washed with PBS three times (10 min every time), the brain slices were blocked with the BSA solution at RT for 1 h. After being washed with PBS three times (10 min every time), the brain slices were incubated with primary antibodies (in BSA solution) at 4 °C for ~ 12 h . Primary antibodies used here included: neurons, Rabbit polyclonal to NeuN (1:1000, ABN78, Sigma-Aldrich); astrocytes, Chicken polyclonal to GFAP (1:1000, PA1-10004, Thermo Fisher); microglia, Mouse monoclonal to Iba1 (1:1000, ab283319, Abcam). After being washed with PBS three times (10 min every time), the brain slices were incubated with secondary antibodies (in BSA solution) at RT in the dark for 2 h. Secondary antibodies used here included: neurons, Goat anti-Rabbit IgG (H + L) Alexa Fluor 488 (1:1000, A-11008, Thermo Fisher); astrocytes, Goat anti-Chicken IgY (H + L) Alexa Fluor 633 (1:1000, A-21103, Thermo Fisher); microglia, Goat anti-Mouse IgG (H + L) Alexa Fluor 555 (1:1000, ab150114, Abcam). After being washed with PBS three times (10 min every time), the brain slices were incubated with 1 μg/mL DAPI (C10002, Beyotime) solution (in PBS) at RT in the dark for 10 min. After being washed with PBS five times (10 min every time), the brain slices were covered with Fluoromount-G compound (0100-01, SouthernBiotech) to avoid fluorescence quenching. The brain slices were imaged using the inverted confocal microscopes (Zeiss LSM 880). 405 nm, 488 nm, 555 nm, and 633 nm lasers were used to image the nuclei, mature neurons, microglia, and reactive astrocytes, respectively. Imaging analysis was performed with the ZEN microscopy software (Zeiss). The fluorescence intensity of NeuN, GFAP and Iba1 was analyzed using QuPath software.

Long-term neural recording and signal analysis

Electrophysiological signals were recorded by using a 256-channel signal acquisition system (CerePlex Direct, Blackrock Microsystems) starting from 1 week after implantation. The implanted USEs probe was connected to the preamplifier headstages (Blackrock Microsystems) through a customized 120-channel PCB with Omnetics connectors. The central reference electrode of the USEs probe was used as the reference channel. Local field potentials (LFP) were analyzed after filtering the raw data (1 kHz sampling rate) with a low-pass filter (250 Hz). Action potential (AP) traces were acquired after filtering the raw data (30 kHz sampling rate) with a bandpass filter (250–5000 or 300–5000 Hz). Spike sorting was performed using Offline Sorter software (Plexon Inc.) or the Kilosort 2.5 algorithm⁵⁷ with postprocessing. During spike detection with Offline Sorter software, the amplitude threshold was set as -4σ or -5σ, where σ was the standard deviation. The K-Means algorithm was used to perform the PCA clustering. After spike sorting, the data of timestamps and waveforms for all the recorded units (including single-unit and multi-unit) were exported (Matlab file), and then analyzed using custom-written Matlab scripts. Spike rasters of the recorded units were plotted using the timestamp data. ISI histograms were acquired using the NeuroExplorer software. Spike amplitude of a unit was the peak-to-peak amplitude of the mean spike waveform. SNR is defined by the formula: SNR = A/N, where A is the spike amplitude of recorded units, and N is the noise level of the corresponding AP traces. Noise level N is calculated by N = MAD/0.6745, where MAD is the median absolute deviation of the AP traces.

Recording of neuronal activity in freely moving mice

C57BL/6 J mice with implanted 120-channel USEs probes were put into the circular arena with a diameter of 35 cm. The USEs probes in the mouse brain were electrically connected to the signal acquisition system via miniaturized 120-channel PCB. To balance the weight of the apparatus (e.g., PCB and headstages) on mouse head, a floating helium balloon was tied to the recording cables⁵⁸. Recording and analysis of electrophysiological signals were performed with the methods mentioned above. Trajectories of freely moving mice in the circular arena were tracked using a camera and custom-written algorithms.

Phase locking analysis

To analyze phase locking to theta oscillation and gamma oscillation, the LFP raw traces were band-pass filtered in the theta band (4–10 Hz) and gamma band (30–90 Hz), respectively⁵⁹. The Hilbert transform was applied to extract the instantaneous phase component of filtered LFP signals. For a given neuronal unit, each spike was then assigned to a specific phase value according to the firing timepoint. The polar histograms (bin size 10°) were plotted to intuitively show the distribution of phase values for all spikes. The mean resultant vector was obtained from the polar histograms. The statistical significance of spike-LFP phase locking could be assessed using the Rayleigh test (p < 0.05). The Rayleigh statistic was calculated as Z = nr², where n is the number of spikes, and r is the length of the mean resultant vector. Exact n, r, Z and p-values are provided in the Source Data file. All the phase locking analyses were conducted using custom-written Matlab scripts.

Visual stimulation

The distance between the mouse eye and stimulus presentation monitor (Samsung S27AG524PC, 60 × 34 cm, 2560 × 1440 pixels, 165 Hz refresh rate) was ~ 16 cm. Visual stimuli were generated using Matlab and Psychtoolbox-3 toolbox⁶⁰. In each trial, a circular patch (radius: 58 degree) of drifting square-wave gratings (spatial frequency: 0.1 cycle/degree; temporal frequency: 1.6 Hz; contrast: 100%) was presented for 600 ms, followed by 200 ms blank (mean luminance) period. The orientation angles of drifting gratings ranged from 0° to 330° in 30° increments (total 12 orientations at random). Each orientation angle of the drifting grating stimuli included 40 trials. The orientation of the drifting gratings is orthogonal to their drift direction. The orientation selectivity index (OSI) was calculated as follows⁶¹: OSI = (R_pref - R_orth)/(R_pref + R_orth), where R_pref is the mean firing rate of neuronal response to the preferred orientation θ_pref, and R_orth is the mean firing rate of neuronal response to the orthogonal orientation θ_orth (the minimum between θ_pref + 90° and θ_pref - 90°).

Statistics and reproducibility

Chronic electrophysiological recordings were performed on seven independent mice. Immunohistochemical staining experiments were performed on six independent C57BL/6 J mice. 3D characterization of interfaces between USEs probes and brain tissues were performed on two independent Thy1-YFP mice. No statistical method was used to predetermine sample sizes. Sample sizes were chosen based on previous works employing flexible neural electrodes for chronic recording in mice. The sample sizes used to derive statistics are provided in the corresponding figure legends and supplementary tables. For the spike sorting with Kilosort, spike units with firing rate < 0.2 Hz or amplitude < 40 μV were excluded. Exact p-values for Rayleigh statistic are provided in the Source Data file. For the visual stimulation experiment, drifting grating stimuli were presented in random order. The investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

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

Author contributions

L.G. conceived the idea and supervised the project. L.G. designed the experiments. W.L., J.B., J.L., and L.G. fabricated and characterized the devices. J.B., J.L., C.Y., X.Z., and L.G. performed the animal surgeries. J.L., X.Z., J.B., and L.G. performed tissue imaging, immunohistochemistry experiment, and corresponding analysis. J.B., C.Y., Y.J., J.L., and L.G. performed the electrophysiological recordings. L.G., C.Y., X.Z., and J.B. conducted the neural electrophysiological signal analysis. C.Y. conducted the simultaneous visual stimulation and neural recording experiments, and the corresponding data analysis. L.G. wrote the manuscript with input from all authors.

Peer review

Peer review information

Nature Communications thanks Dingchang Lin and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China (32500998, to L.G.), CIBR start-up fund (to L.G.), fund from NeuCyber (NC-2023-HE-05-a, to L.G.), and Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM504).

Data availability

Raw data for electrophysiological recordings from Mouse-1 are available via Zenodo at https://doi.org/10.5281/zenodo.21833298 (ref. ⁶²), https://doi.org/10.5281/zenodo.21833361 (ref. ⁶³), https://doi.org/10.5281/zenodo.21833371 (ref. ⁶⁴), and https://doi.org/10.5281/zenodo.21866081 (ref. ⁶⁵). Raw data for electrophysiological recordings from Mouse-2 are available via Zenodo at https://doi.org/10.5281/zenodo.21833296 (ref. ⁶⁶). Raw data for electrophysiological recordings from Mouse-3 are available via Zenodo at https://doi.org/10.5281/zenodo.21833339 (ref. ⁶⁷). Raw data for electrophysiological recordings from Mouse-4 are available via Zenodo at https://doi.org/10.5281/zenodo.21833383 (ref. ⁶⁸). Raw data for electrophysiological recordings from Mouse-7 are available via Zenodo at https://doi.org/10.5281/zenodo.21821147 (ref. ⁶⁹). Any additional requests for information can be directed to, and will be fulfilled by, the corresponding author. Source data are provided with this paper.

Code availability

The custom MATLAB code used for phase-locking analysis is available via GitHub at https://github.com/Gaolab-cibr/Spike-LFP-Phase-Locking, and via Zenodo at https://doi.org/10.5281/zenodo.21428945 (ref. ⁷⁰).

Competing interests

Chinese Institute for Brain Research (CIBR) and NeuCyber NeuroTech (Beijing) Co., Ltd. have filed a patent application regarding the design and method of USEs probes (application no. 202411437968.0), in which L.G., W.L., J.B., and J.L. are inventors. All the other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77145-4.