Structural and functional changes of deep layer pyramidal neurons surrounding microelectrode arrays implanted in rat motor cortex

Abstract

Devices capable of recording or stimulating neuronal signals have created new opportunities to understand normal physiology and treat sources of pathology in the brain. However, it is possible that the tissue response to implanted electrodes may influence the nature of the signals detected or stimulated. In this study, we characterized structural and functional changes in deep layer pyramidal neurons surrounding silicon or polyimide-based electrodes implanted in the motor cortex of rats. Devices were captured in 300 μm-thick tissue slices collected at the 1 or 6 week time point post-implantation, and individual neurons were assessed using a combination of whole-cell electrophysiology and 2-photon imaging. We observed disrupted dendritic arbors and a significant reduction in spine densities in neurons surrounding devices. These effects were accompanied by a decrease in the frequency of spontaneous excitatory post-synaptic currents, a reduction in sag amplitude, an increase in spike frequency adaptation, and an increase in filopodia density. We hypothesize that the effects observed in this study may contribute to the signal loss and instability that often accompany chronically implanted electrodes.

Graphical abstract

Statement of Significance

Implanted electrodes in the brain can be used to treat sources of pathology and understand normal physiology by recording or stimulating electrical signals generated by local neurons. However, a foreign body response following implantation undermines the performance of these devices. While several studies have investigated the biological mechanisms of device-tissue interactions through histology, transcriptomics, and imaging, our study is the first to directly interrogate effects on the function of neurons surrounding electrodes using single-cell electrophysiology. Additionally, we provide new, detailed assessments of the impacts of electrodes on the dendritic structure and spine morphology of neurons, and we assess effects for both traditional (silicon) and newer polymer electrode materials. These results reveal new potential mechanisms of electrode-tissue interactions.

INTRODUCTION

Electrodes implanted in the brain enable the detection and stimulation of electrical and chemical signals from local neurons. Recent years have seen rapid growth in the design and application of neurotechnology in research and clinical settings[1–6]. New advances in the field have been driven by the movement to design softer, smaller, higher-density arrays with minimized insertional trauma[1,7]. Spikes from thousands of individual neurons can be simultaneously recorded from implants in rodents using state-of-the-art technology[8], and hundreds of units have been isolated from arrays implanted in human subjects[9]. Ideally, implanted electrodes would serve as a “silent observer” of localized activity, allowing stable signal detection over many years without initiating a loss or functional impairment of signal-generating neurons. However, reductions in signal quality over time are often observed, and an expanding set of reported observations indicates substantial changes in the density, morphology, and gene expression of cells surrounding electrodes following implantation[10].

The degree and time course of signal quality loss vary widely across literature, with some reports indicating relative stability, and others reporting pronounced and rapid losses. Chestek and colleagues[11] reported an average 2.4% per month decline in spike amplitude recorded in non-human primates over variable observation periods (9–31.7 months), consistent with the observation that devices often lose activity over 6–12 months in non-human primates[12]. Comparatively more rapid decay in signal quality has been reported in the days and weeks following implantation in rodents[13]. In addition to chronic effects, signal instability has been reported within the span of a single day[14],[15]. Non-stationarity introduces added complexity for systems that use decoding algorithms which rely on accurate spike detection, studies which interpret changes in firing rate to draw conclusions, and closed-loop strategies which deliver stimulation conditioned on spike detection. Various underlying causes of these effects have been proposed, including natural physiological variability, mechanical/electrical sources of electrode failure, and the tissue response to electrodes post-insertion[10,16–18].

Although the tissue response to implanted electrodes has long been hypothesized to be a key contributor to signal loss over long periods of time, defining the underlying mechanisms driving these responses, the link to recording quality, and the relationship to electrode design features all remain active areas of inquiry[19]. Electrode implantation results in an immediate and traumatic disruption of the neuropil and vasculature, followed by fluid extravasation and cellular infiltration into the implant site. The subsequent encapsulation of the device by responding reactive glia, and a loss of neuronal density, are commonly observed and have been used as metrics to assess the tissue response to implanted electrodes[10,20]. Changes in the structure and function of neurons are known effects of other forms of traumatic brain injury, yet these potential impacts have been comparatively less well-characterized in studies of implanted electrodes. Focal injuries of neocortex are known to result in a reorganization of network connectivity, local upregulation of glutamatergic transporters, changes in the expression of repolarizing potassium currents, an altered balance of neurotransmitter receptors, and changes in excitability[21],[22,23]. Likewise, pronounced loss of dendritic spines is a known consequence of cortical impact[24]. Disruption of dendrites surrounding electrodes has been observed previously[25], but the impact of devices on dendritic spine densities and morphologies is unknown. Recent reports have indicated changes in the expression of ion channels and synaptic markers at the gene and protein level surrounding implanted electrodes[26,27], but effects on the intrinsic excitability of surrounding neurons are unclear.

In this study, we explored the hypothesis that electrode implantation alters the structure and function of local neurons. To do this, we captured single-shank, Michigan-style microelectrode arrays (MEAs) within thick sections of live rat motor cortex tissue collected at 1- and 6-weeks post-implantation. Whole-cell electrophysiology was used to assess changes in passive properties, spiking characteristics, and afferent synaptic activity in neurons “near” (<100 μm) or “distant” (~500 μm) from devices within individual tissue slices. During recording sessions, neurons were filled with a fluorescent dye to assess changes in dendritic architecture and spine densities via two-photon imaging and morphometric analysis. We tested both polyimide and silicon-based devices to assess material-based effects on responses. We observed several effects, which were pronounced in the neurons immediately surrounding the implant (<100 μm): (1) fluorescence microscopy showed that neurons displayed reduced dendritic length and branching at 1- and 6-weeks post-implant, (2) whole-cell intracellular recordings revealed that neurons surrounding devices showed reduced sag amplitude, increased spike frequency adaptation, and reduced frequency of spontaneous excitatory post-synaptic currents (sEPSCs) at 6 weeks, and (3) assessments of spine morphology indicated a loss of larger, more mature spine types accompanied by an increase in filopodia. The results suggest an emergence of a hypoexcitable network surrounding implanted electrodes which could contribute to observations of signal loss, while reductions in sag amplitude and increases in adaptation could contribute to variability in firing rates.

MATERIALS AND METHODS

Electrodes and Surgical Implantation

Two styles of single shank, 16-channel microelectrode arrays (MEAs) were chosen for comparison, one silicon-based and one polyimide-based. Non-functional, A1×16-style silicon devices were purchased from a commercial vendor (3 mm shank length, 15 μm thickness, tapered width measuring 123 μm at maximum; Neuronexus Technologies, Ann Arbor, MI). Polyimide devices were custom-fabricated and supplied courtesy of Dr. John Seymour (University of Texas Health Science Center) based on methods previously described[28]. The dimensions of the polyimide devices were size-matched to those of the silicon devices, with the exception of a reduced thickness (4.4 μm). Devices were gas-sterilized prior to use. Either silicon (21 rats, 42 devices) or polyimide (30 rats, 60 devices) devices were bilaterally implanted into primary motor cortex (M1) of ~12-week old male Sprague-Dawley rats based on methods and stereotaxic coordinates previously described[29–31]. Prior to implantation, the silicon devices were attached to a dummy connector and polyimide devices were temporarily fixed to a size-matched silicon shuttle via sterile 20% sucrose solution in PBS applied to the dorsal edge of the interface between the shuttle and device. Following insertion, the silicon devices were released from the dummy connector by severing the connection point with surgical scissors. The polyimide devices were released from the silicon shuttle following insertion by liberal application of sterile saline and withdrawal of the shuttle. The silicon shuttles were identical devices to those used for indwelling implants (i.e., Neuronexus devices as described above). Each device was inserted perpendicular to the cortical layers to a depth of 1.8–2.0 mm from the cortical surface of M1. The dorsal end of the device was flush or slightly protruding from the brain surface, and surgical closure was achieved with a combination of packed gel-foam and sterile suture of the incision site. Sham insertion/stab injury (28 rats) was achieved by brief insertion and immediate withdrawal of a silicon-based device (same device type, dimensions, and depth of penetration as for indwelling implants) following the craniotomy. Methods for sham insertions otherwise followed the same surgical procedures as for chronically implanted devices. In addition, naïve (non-implanted) animals (4 rats) were included in this study. All experimental procedures were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Michigan State University Institutional Animal Care and Use Committee.

Brain Slice Preparation

Animals were sacrificed at 1 week (10 rats silicon, 15 rats polyimide, 15 rats sham) or 6 weeks (11 rats silicon, 15 rats polyimide, 13 rats sham) post-surgery to prepare brain slices targeting the implanted or sham-inserted region of M1. Naïve animals (4 rats) were sacrificed at ~12 weeks of age and non-implanted M1 was targeted. Briefly, rats were deeply anesthetized with 3% isoflurane and pentobarbital (100–200 mg/kg), transcardially perfused with slicing solution containing (in mM): 2.5 KCl, 1.25 NaH₂PO₄, 10.0 MgSO₄, 0.5 CaCl₂, 26.0 NaHCO₃, 11.0 glucose, and 234.0 sucrose, and decapitated. Brains were quickly removed and placed into chilled (<4 °C), oxygenated (95% O₂/5% CO₂) slicing solution. Coronal slices (300 μm thickness) surrounding the implanted device or sham were obtained using a vibrating slicer (Leica Biosystems). The slices were then hemi-sectioned, and transferred to a chamber for 30 min of incubation in heated (36 °C) and oxygenated (95% O₂/5% CO₂) physiological solution containing (in mM): 126.0 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 2.0 MgCl₂, 2.0 CaCl₂, 26.0 NaHCO₃, and 10.0 glucose.

Electrophysiology

Slices of the implanted region were transferred to a submersion-type recording chamber and super-fused (2.5 ml/min) with oxygenated physiological solution maintained at 32 °C. Recording pipettes were pulled from 1.5 mm outer diameter capillary tubing and had tip resistances of 3–6 MΩ when filled with solution containing (in mM): 117.0 K-gluconate, 13.0 KCl, 1.0 MgCl₂, 0.07 CaCl₂, 0.1 EGTA, 10.0 HEPES, 2.0 Na₂–ATP, 0.4 Na-GTP, and 50 μM Alexa Fluor 594. The pH of this solution was adjusted to 7.3 and osmolarity was adjusted to 290 mOsm. The use of this intracellular solution resulted in an 8 mV junction potential that has been corrected for in all voltage measurements.

Whole-cell recordings were obtained from deep layer pyramidal neurons in M1 with the visual aid of a BX51WI fixed-stage microscope (Olympus) equipped with Dodt contrast optics. A low-power objective was used to identify specific cortical layers and a high-power water immersion objective was used to visualize individual neurons. The device was visualized under the microscope and the distance between the MEA/tract and recorded neuron was determined under high-power (near device) or low-power objectives (distant device). Electrophysiological data were acquired using a MultiClamp 700B amplifier filtered at 4 kHz and digitized at 10 kHz using a Digidata 1440A digitizer in combination with pCLAMP 10 software (Molecular Devices, San Jose, CA). During the recordings, pipette capacitance was neutralized, and access resistance was continuously monitored. Voltage clamp recordings were limited to neurons that had a stable access resistance of <20 MΩ. Only neurons with stable membrane potentials, input resistances, and action potential amplitudes were included in the analyses. Current and voltage protocols were generated using pCLAMP software, and data were digitized and stored on a computer for off-line analyses.

Intrinsic properties and action potential (AP) output were collected in current-clamp configuration from neurons at resting membrane potential. Synaptic activity was recorded in voltage-clamp configuration while holding at −58 mV to isolate excitatory postsynaptic currents. Input resistance was calculated from the linear slope of the voltage-current relationship obtained from applying a series of depolarizing and hyperpolarizing constant current pulses from rest (−80 to +80 pA, 40 pA increments, 1s duration). Sag amplitude was calculated from a current step protocol that produced an initial hyperpolarization to ~−100 mV (1 s duration), and calculating the difference in voltage between the initial peak hyperpolarization and steady state membrane potential at the end of the current step. Rheobase is measured as current (5 pA increments, 1 s duration) necessary to reliably evoke an action potential from resting membrane potential. This current is then used to repeatedly evoke AP discharge (20 iterations). Action potential characteristics (threshold, half-width, and max amplitude) are measured for each iteration and averaged for each neuron. Firing properties of neurons were analyzed from a current step protocol that depolarized neurons from rest (0 to 2400 pA, 80 pA increment, 1s duration). Firing rate is reported as average over 1s depolarizing step. Spike frequency adaptation was calculated as the ratio of the mean of last 2 interspike intervals to the mean of the 3^rd and 4^th interspike interval. This calculation avoided the initial doublet AP discharge characteristic of neocortical pyramidal neurons. The slopes of the frequency-intensity relationship were calculated from the linear portion of the graph featuring at least 4 consecutive intensities greater than 0 pA. All population data are expressed as mean ± standard deviation (SD). Statistical analyses were performed using GraphPad Prism (DotMatics, San Diego, CA) and Mini Analysis (Synaptosoft, Fort Lee NJ).

Imaging

Deep layer pyramidal neurons filled via recording pipette with Alexa Fluor 594 (50 μM; Molecular Probes, Eugene, OR) were imaged by laser excitation (820 nm) using a two-photon laser-scanning microscopy system (Ultima, Bruker, Madison, WI) coupled with a Ti:Sapphire laser (Mai Tai HP, MKS-Spectra-Physics, Milpitas, CA). After at least 15 min of filling, each neuron was measured for consistent fluorescent levels, then basal dendrites were identified, and the entire length was recorded in 50 μm sections as a Z-stack of images for analysis of dendritic spines. Neurons were then imaged as Z-stacks of 200 μm × 200 μm sections overlapping by ~20% to reconstruct the neuron’s full morphology by stitching the stacks together using Image J (NIH).

Dendritic arbors were reconstructed using stitched Z-stacks. Sholl analysis is a well-established technique used to assess the complexity of dendritic arbors[32,33]. Using the Simple Neurite Tracer tool (Image J, NIH), the traced dendrite arbors are skeletonized, and a Sholl analysis was used to quantify dendritic length and branching patterns. Basal dendritic spines were reconstructed using the 50 μm sections taken as a Z-stack. Using NeuronStudio[34], first the dendrite sections were reconstructed without dendritic spines using 3D voxels fit to the fluorescence region, then spines were identified by human observation and double-checked by a second observer. Spine morphology subtypes were determined (i.e., thin, mushroom, stubby and filopodia) based on the spine’s head to neck diameter ratio and spine head diameter to spine neck length ratio[34]. Analysis for spine density and subtypes was carried out using exported data from NeuronStudio and GraphPad Prism (DotMatics, San Diego, CA).

Statistics

Unless otherwise indicated, results are reported as mean ± SD, and significance was tested using a non-parametric, one-way ANOVA Kruskal-Wallis test, followed by Dunn’s test for multiple comparisons. Sholl intersections were compared using a Kolmogorov-Smirnov test. Significance was defined as p<0.05.

RESULTS

Dendritic Arborization is Altered in Near-Device Neurons

Visual inspection revealed an asymmetric distribution of dendritic arborization in Near-device neurons (defined as neurons within 100 μm of the device, Fig. 1A–B). Dendrites were preferentially lost on the implant-facing side of the neuron, likely due to some combination of initial insertional injury and the subsequent chronic tissue response (Supp. Figs. 1–2). Quantification of these effects at the 1-week time point using Sholl analysis revealed reduced arborization (Fig. 1C). Near-device neurons displayed a significant reduction in intersections in comparison to unimplanted tissue (p<0.01), irrespective of material or time point. Near-insertion neurons showed a significant reduction in Sholl intersections at the 1 week time point (p<0.01), but not at the 6 week time point (p>0.05). Distant-device and distant-insertion neurons did not display a statistically significant difference from unimplanted tissue (p>0.05 for these conditions at each time point). This was accompanied by a statistically significant loss in basal dendrite length in both Near-Silicon and Near-Polyimide neurons in comparison to distant neurons for both materials, as well as neurons in naïve (non-implanted) tissue (Fig. 1D). At 6 weeks, Near-Silicon and Near-Polyimide neurons maintained the loss of basal dendrite length in comparison to naïve tissue (Fig. 1D). However, differences between Near-device and Distant-device neurons (defined as > ~500 μm from the device) no longer reached significance, which may be attributable to a slight decrease in the basal dendrite length of Distant-device neurons at the 6-week time point. Neurons near and distant from the insertional stab injury site in sham treatments displayed no significant difference from any other condition at either time point, although the data indicated a trend toward reduced dendritic length. Likewise, both silicon- and polyimide-based electrodes produced similar effects. Near-Silicon and Near-Polyimide neurons also exhibited a reduction in dendritic branching, as indicated by decreased intersections with rings of increasing radii in the Sholl analysis (Fig. 1B–C). Likewise, an asymmetric effect on branching was observed, where loss of branching was exacerbated on the implant-facing side of the neuron (Supp. Fig. 3).

Fig. 1. Electrode implantation is associated with disruption of the dendritic arbor in nearby neurons.

(A) Schematic of implantation scheme and representative images of layer 5 regular-spiking pyramidal neurons in naïve tissue, as well as soma located near (<100 μm) or distant (~500 μm) from implanted electrodes or sham insertion sites (electrode can seen in top left corner of Near-MEA image). Scale bar, 50 μm. (B) Qualitative observation revealed an asymmetric loss of dendrites surrounding implants (top panel), which was quantified using Sholl analysis (skeletonized image of arbor, bottom panel). Scale bar, 50 μm. (C) Near-device neurons show a clear reduction in dendritic branching, as indicated by a reduction of intersecting points detected in Sholl analysis. (D) Overall dendritic length is significantly reduced in near-silicon and near-polyimide devices, which is not evident in sham insertion controls.

Spine Loss and Immaturity in Near-Device Neurons

In addition to the loss of dendritic length and branching, we also noted a decrease in spine density (normalized to dendritic length) in Near-device neurons. At the 1-week time point, spine density decreased by ~50% for both Near-Silicon and Near-Polyimide electrodes (Fig. 2AC). Distant-device neurons also displayed a significant, albeit somewhat lesser, reduction in spine densities for both electrode materials. While Near-Insertion neurons showed a reduction in spine density, no significant difference was observed between Distant-stab insertion neurons and naïve controls. A modest improvement in spine density loss surrounding devices was observed by the 6-week time point, but reductions for Near-device, Distant-device, and Near-stab insertion neurons remained significant in comparison to naïve controls (Fig. 2D). Spine losses for Near-stab insertion neurons were relatively consistent between the 1- and 6-week time points in comparison to the slight recovery exhibited by neurons surrounding devices.

Fig. 2. Dendritic spines are less dense and more morphologically immature surrounding devices.

(A) Representative images of basal dendrites from Near-device, Distant-device, and Near-insertion pyramidal neurons after 1 week (scale bar = 10 μm). (B) Images illustrating the differences between each of the dendritic spine categories. Scale bar, 1 μm. Near-device pyramidal neurons exhibited a lower total density of spines at both 1 Week (C) and 6 Weeks (D). Losses included reductions in spines with mushroom, stubby, and thin morphologies as well as filopodia. (E) Alternatively, the density of filopodia was increased surrounding devices at the 6-week time point. Filopodia density was not significantly affected in insertion injury controls.

We analyzed the morphology of individual spines and classified them according to common metrics reported in literature[35]. Mushroom spines, defined by a large head and a small neck, were significantly decreased surrounding devices[36]. Stubby spines have similar characteristics but lack a well-defined neck. Stubby spines were significantly decreased at both time points in comparison to naïve neurons. These losses were evident at both time points, irrespective of distance or electrode material. Thin spines are implicated in plasticity and are similarly shaped to mushroom spines, but with a smaller head. Spine losses were also observed in thin spines surrounding both devices and stab injury sites.

In contrast to the other spine types, filopodia were uniquely increased in density surrounding devices at the 6-week time point (Fig. 2E). Filopodia are long and thin, without a well-defined head. These structures are often observed in developing neurons, and glutamate uncaging studies indicate that they lack the AMPA receptors required for functional glutamatergic neurotransmission[37]. As such, they may mature into spines and are a putative site of “silent” synapses: synapses that are ultrastructurally normal, yet non-functional. Filopodia density was initially decreased surrounding devices in comparison to naive neurons at the 1-week time point. However, this effect was reversed at the 6-week time point, when significant increases in filopodia density were observed. These effects were not recreated by the insertion stab injury. The emergence of filopodia surrounding devices could result from the loss in overall spine density at the 6-week time point.

Reduced sEPSC Frequency in Near-Device Neurons

Spontaneous excitatory post-synaptic currents (sEPSCs) are caused by neurotransmitter release from a presynaptic neuron in the absence of an applied stimulus. Initially, no detectable changes in the frequency or amplitude of sEPSCs were detected between any conditions at the 1-week time point (Fig. 3). At the 6-week time point, neurons near either silicon or polyimide-based electrodes displayed similar, significant reductions in sEPSC frequency compared to naïve neurons. These effects required the presence of the device and were not recreated by the insertional injury. Similarly to the 1-week time point, this observation was decoupled from an effect on sEPSC amplitude at the 6-week time point, which could otherwise implicate changes in the activation of postsynaptic receptors.

Fig. 3. The frequency of spontaneous excitatory post-synaptic currents (sEPSCs) is reduced surrounding devices at 6 weeks.

(A) Representative current traces from 6-week conditions. Spontaneous excitatory postsynaptic currents (sEPSCs) were measured with a V_hold of −58 mV. (B) The sEPSC frequencies were unaffected for Near-Device neurons at 1 week post-surgery, but at 6 weeks a significant decrease was observed only in neurons near devices (both polyimide and silicon). (C) sEPSC amplitudes appear unaltered in all conditions.

While structural changes in the dendritic arbor (Figs. 1–2) were accompanied by functional changes in sEPSC frequency in Near-device neurons (Fig. 3), the timing of these effects was somewhat discordant. Significant losses in dendritic length, branching, and spine densities were observed at both the 1- and 6-week time points, but the reduction in sEPSC frequency was solely observed at the 6-week time point. The overall loss of spine density was the most pronounced at the 1-week time point, when no effects on sEPSC frequency were detected. Thus, although both structural and functional data indicate a generalized reduction in the underpinnings of synaptic transmission in Near-device neurons, the observations implicate a contribution of alternative mechanisms in addition to losses in postsynaptic contact sites. Salatino et al. (2017) observed an initial increase in the expression of vesicular glutmate transporter 1 (VGLUT1) surrounding electrodes in rat primary motor cortex at three days post-implantation.[26] It is possible that increased VGLUT1 expression contributed to preserved excitatory tone near the device at the 1-week time point.

Increased Adaptation and Reduced Sag Amplitude

Whole-cell electrophysiology was used to investigate the impacts of the implanted electrode on the intrinsic excitability of surrounding neurons. Measured passive and active properties included resting potential, input resistance, rheobase, sag amplitude, spike frequency adaptation, firing frequency, the slope of the frequency-current relationship (FI Slope) and the characteristics of AP shape (Fig. 4). Complete results are summarized in Table 1. These measurements largely did not indicate any impacts of device implantation on neuronal intrinsic excitability, with two notable exceptions: the sag amplitude and spike frequency adaptation were both significantly affected at the 6- week time point in Near-device neurons. Sag amplitude, which is a depolarization in response to a hyperpolarizing stimulus, was significantly decreased in neurons near both silicon and polyimide-based electrodes at the 6-week time point. These effects were not observed at the 1-week time point or in stab injury neurons. Increased spike frequency adaptation, which indicates broadening of the inter-spike interval during a sustained depolarizing stimulus, was observed in neurons near silicon and polyimide electrodes at the 6-week time point in comparison to naïve controls. Neurons near stab-wound insertion sites also displayed increased spike frequency adaptation at the 6-week time point, highlighting the importance of initial insertional damage in this effect. The reduction in sag amplitude, which controls the regularity of firing, as well as increased spike frequency adaptation, indicate that there are changes in the rhythmicity of neuronal firing immediately surrounding chronically implanted electrodes.

Fig. 4. Assessing intrinsic excitability: an overview of electrophysiological characteristics analyzed.

The features measured to test device-based impacts on active and passive electrophysiological properties are shown schematically (top panels), and representative images of firing responses are shown for neurons surrounding devices and insertion sites from 6-week conditions (bottom panels). Traces revealed qualitative effects of devices on spike frequency adaptation and sag amplitude. Quantitative data are summarized in Table 1.

Table 1.

Summary of device-based impacts on electrophysiological properties of surrounding neurons.

		N	RMP (mV)	Rin (MΩ)	Sag Amplitude (mV)	Rheobase (pA)	AP Threshold (mV)	AP Max Amplitude (mV)	AP Half-Width (ms)	Max Firing Frequency (AP/s)	FI Slope (AP/pA)	FI Adaptation (Last/First IEI)
Naive	Mean	44	78.0	99	5.9	165	−45.0	84.5	0.72	49.6	0.056	1.49
	SD		±5.7	±48	±3.7	±92	±5.4	±11.4	±0.16	±17.7	±0.022	±0.48
1W Near-Silicon	Mean	15	77.5	103	3.5	164	−42.0	79.1	0.91	47.7	0.050	2.07
	SD		±12.1	±50	±2.6	±113	±11.2	±17.2	±0.62	±21.6	±0.020	±0.95
1W Distant-Silicon	Mean	15	78.2	116	4.8	163	−44.4	87.5	0.73	50.6	0.049	1.99
	SD		±4.4	±41	±2.2	±107	±4.1	±5.8	±0.09	±15.9	±0.020	±0.74
1W Near-Polyimide	Mean	23	74.5	109	5.4	164	−40.9	79.4	0.78	32.7	0.052	1.61
	SD		±15.3	±95	±4.1	±101	±8.3	±19.8	±0.24	±20.6	±0.021	±0.58
1W Distant-Polyimide	Mean	18	80.3	115	5.1	165	−42.9	81.7	0.75	48.6	0.048	1.67
	SD		±5.7	±52	±1.7	±100	±3.1	±10.1	±0.12	±17.8	±0.015	±0.39
1W Near-Insertion	Mean	15	72.9	112	6.2	161	−42.6	81.8	0.81	42.6	0.054	1.71
	SD		±10.3	±64	±4.7	±204	±3.4	±12.1	±0.15	±35.9	±0.042	±0.97
1W Distant-Insertion	Mean	13	78.5	104	5.5	176	−42.0	88.4	0.67	59.6	0.070	1.43
	SD		±4.2	±52	±3.2	±101	±10.6	±5.4	±0.09	±17.8	±0.036	±0.32
6W Near-Silicon	Mean	18	74.0	119	* 2.9	150	−45.8	91.6	0.75	60.0	0.058	* 2.34
	SD		±11.7	±63	±2.3	±103	±4.9	±6.5	±0.09	±8.3	±0.025	±1.00
6W Distant-Silicon	Mean	11	79.6	93	3.4	173	−45.3	91.4	0.80	52.7	0.049	2.03
	SD		±3.4	±33	±1.7	±59	±4.8	±8.2	±0.10	±13.2	±0.013	±0.50
6W Near-Polyimide	Mean	17	81.5	147	* 2.7	143	−44.8	87.9	0.80	47.2	0.053	* 2.11
	SD		±8.7	±97	±1.9	±90	±3.4	±7.3	±0.12	±17.6	±0.030	±0.59
6W Distant-Polyimide	Mean	12	80.3	93	5.5	208	−46.0	82.5	0.75	44.3	0.040	1.89
	SD		±6.5	±46	±3.3	±134	±4.9	±6.9	±0.10	±23.2	±0.018	±0.94
6W Near-Insertion	Mean	11	75.6	125	5.2	152	−44.1	86.8	0.79	46.8	0.047	* 3.24
	SD		±13.7	±62	±3.7	±105	±5.0	±7.9	±0.10	±21.1	±0.030	±2.40
6W Distant-Insertion	Mean	16	79.9	131	5.2	149	−44.6	85.9	0.79	50.3	0.053	2.11
	SD		±4.6	±82	±2.3	±90	±3.5	±6.8	±0.11	±20.9	±0.016	±0.74

The majority of measurements showed no significant difference with neurons from naïve tissue. However, significant impacts of the device presence were observed at 6 weeks for the sag amplitude and the spike frequency adaptation: sag amplitude was significantly decreased, and adaptation was significantly increased versus controls (highlighted in gray, * P < 0.05).

DISCUSSION

While the biological response to electrodes implanted in the brain has long been viewed as a key contributor to signal loss and instability, a causal relationship between these phenomena has not been definitively established. Localized neuronal density loss and increased expression of glial markers are typically quantified to assess the degree of the tissue response to the electrode. However, these measurements are often misaligned in time and severity in comparison to the degree of signal loss, indicating that other factors are likely at play. While these effects may include electrical and mechanical failure points on the electrode, there is also a growing body of evidence that the biological response to electrodes is more complex than what can be captured by a few pre-selected markers[38–41]. The goal of our study was to add to current understanding of the impacts of electrodes on surrounding tissue, and our data illustrate nuanced structural and functional changes in the individual neurons within the recordable radius of the device interface[17].

The seminal paper by Biran et al. (2005) reported a ~40% loss of neuronal density within the first 100 μm of a “Michigan”-style, single shank electrode implanted for a period of one month in the motor cortex of rats[42]. McConnell et al. (2009) attributed recording failure to an observed local “neurodegenerative state” marked by dendritic loss, neuronal death, and tau pathology within 100 μm of Michigan electrodes implanted in rat cortices[43]. Loss in neuronal density has since become a commonly used metric to assess the integration between implants and surrounding tissue. Less focus has been placed on the state of the remaining neurons. Using in vivo calcium imaging, Kozai and colleagues (2018) observed an initial reduction in neuronal activity surrounding implanted electrodes followed by functional recovery during the first month post-implantation[44]. Welle et al. (2020) showed progressive atrophy of the dendritic arbor surrounding electrodes during a three month observation period, likewise using in vivo 2-photon imaging[45]. Similarly to previous reports, our data showed a generalized reduction in dendritic length surrounding implants. Our data additionally indicated that the loss of length, as well as reductions in branching, were asymmetrically driven by effects on the device-facing side.

We observed new evidence of a reduction in spine densities surrounding electrodes implanted in the rat motor cortex (normalized to dendritic length). Since spines are typically the site of excitatory synapses, a loss in density implies a reduction in the network input to individual neurons surrounding electrodes. Loss in spine density was not restricted to the first 100 μm surrounding the device: it was also observed, albeit to a lesser extent, at the 500 μm distance. Reduced excitatory input to neurons within 100 μm of the device could affect the spike output of individual neurons, posing a challenge to single unit detection. Loss of excitatory network input to neurons at the 500 μm distance could influence the characteristics of the local field potential (LFP) in multiple ways. Perhaps the simplest interpretation would be that a generalized reduction in input could dampen the amplitude of the LFP due to reduced activity. However, the LFP is heavily influenced by the degree of neuronal synchrony and the geometric arrangement of the sources and sinks of ionic currents relative to the location of the electrode[46]. Generalized loss of excitatory input distally on dendrites could amplify the influence of inhibitory, somatic inputs on the frequency content of the LFP signal, potentially increasing relative power in higher frequency bands.

Asymmetry in the loss of the dendritic arbor, coupled with spine density loss, could influence the amplitude and frequency content of the LFP in unexpected ways[47–49]. Linden et al. (2011) reported a biophysical modeling approach to simulate LFP signals from activated, morphologically distinct neuronal populations, and found that the LFP amplitude for pyramidal neurons depends on correlated synaptic input, synapse location and neuron morphology[48]. Hence, the loss of dendritic arbor could possibly interfere with the conduction of synaptic input and diminish the spatial reach and amplitude of LFP. Another modeling study by Gold and colleagues (2006) observed that the extracellular action potential (EAP) waveform (defined by a wideband signal encompassing both the LFP and spike band) in CA1 pyramidal neurons was influenced by ion channel distributions more heavily than dendritic morphology[50]. Therefore, the impact of asymmetric dendritic loss on the LFP could result from the associated change in the spatial distribution of ion channels on individual neurons. More recently, Ness et al. (2016) used biophysical modeling to confirm that the LFP power generated from pyramidal neurons is heavily impacted by active subthreshold dendritic currents, and more importantly, by hyperpolarization activated mixed cation currents (namely, the h-current)[51]. Assuming the loss of sag amplitude is reflective of a reduction in h-current, this study predicts a strong effect of electrode insertion on LFP power and frequency content. Predicting the impacts of our observations of neuronal restructuring on the LFP would benefit from dedicated modeling studies incorporating asymmetry and spine loss.

Morphological analysis suggested that remaining spines shifted toward a more immature, “silent” phenotype, as evidenced by an increase in filopodia surrounding electrodes. The increase in the density of filopodia can account for much of the recovery of spine densities between the 1- and 6-week time point. In glutamate uncaging studies, filopodia were associated with a lack of functional AMPA receptors, which were present on larger, stubby-type spines³³. Absence of AMPA receptors would render these structures functionally silent, which would reinforce the generalized loss of excitatory input expected due to overall spine loss. In work published by Barres and colleagues (2005), ultrastructurally normal, but functionally silent, synapses have been associated with thrombospondin production by reactive astrocytes[52]. More recently, Liddelow et al. (2017) reported a reduction in excitatory synapses upon exposure to soluble cues from reactive astrocytes induced by microglial-derived inflammatory cytokines[53]. While the exact mechanisms are yet to be determined, we hypothesize that reactive glia contribute to the observed losses in neuronal spine densities, as well as the enrichment of filopodia. Using in vivo multiphoton imaging of mouse pyramidal neurons, Miyamoto et al. (2016) observed the formation of filopodia upon microglial contact with dendrites[54]. Additionally, Weinhard and colleagues (2018) found that microglia facilitate synaptic remodeling in mouse hippocampal neurons by trogocytosis (i.e. selective, nonapoptotic partial phagocytosis[55]). Our recently reported gene expression data collected surrounding devices support the view that reactive glia are present at the device interface which are consistent with an inhospitable environment for synaptic transmission^35,36.

We did not observe obvious effects of the device on spike shape, passive properties, or standard metrics of the responsiveness to stimulation. However, we did observe three significant effects of the device on the electrophysiological properties of local neurons. First, we observed reduced sag amplitude at the 6-week time point in neurons near polyimide and silicon devices. These effects were not present for Distant-device neurons or stab insertion controls. To some extent, the reduction of sag amplitude may be related to generalized loss of the dendritic arbor, since the ion channels responsible for hyperpolarization-activated currents can be localized to the dendritic arbor. However, inspection of the morphological data indicates that there may be more to the story: while Near-device neurons at the 1-week time point displayed marked dendritic loss, no significant change in sag amplitude was detected. This could suggest that a per-cell reduction in ion channel expression occurred in concert with dendritic loss; our previous data reported changes in ion channel expression surrounding devices²⁸. Secondly, we observed increased adaptation in spike trains recorded in Near-device neurons, as indicated by increasing interspike intervals in response to a constant stimulus. This effect may be related to the reduction in sag amplitude since hyperpolarization-activated currents can contribute to the rhythmicity of firing[56]. Finally, we observed a reduction in the frequency of spontaneous EPSCs in Near-device neurons at the 6-week time point. While this may be partially attributable to spine loss, it is notable that sEPSC frequency is unaffected at the 1-week time point, when a more pronounced ~50% loss in spine density was observed. Again, a possible explanation involves a role for reactive glia: previous findings support that reactive microglia and astrocytes direct synaptic remodeling at device injury sites by releasing cytokines, glutamate, and adenosine triphosphate (ATP)[10,57,58]. Glial-derived adenosine produced by rapid hydrolysis of ATP can promote hyperpolarized post-synaptic contacts by downstream opening of K⁺ and Cl⁻ channels[10,59]. It is also possible that pre-synaptic mechanisms are also at play[60].

We chose to assess changes in the structure and function of neurons surrounding devices using a unique device-in-slice, whole-cell electrophysiology technique. There are several advantages to the approach, including the ability to assess both the intrinsic excitability and dendritic arbor of morphologically identified pyramidal neurons at defined distances from the implanted electrode interface. However, it is important to acknowledge that there are limitations that are broadly associated with brain slice electrophysiology. For example, alterations in the accumulation or clearance rates of local ionic concentrations, as well as a loss of long-range synaptic connections, are factors that could influence results obtained in the ex vivo preparation in comparison to the native in vivo environment[61]. An alternative approach would be to do in vivo whole-cell recordings surrounding implanted electrodes, but these are notoriously difficult and low-throughput, and most amenable to assessments in an acute implantation paradigm[62]. Finally, while standard quality criteria are employed to avoid contaminating experimental effects with artifacts of poor whole-cell recordings, it is also likely that some highly damaged neurons were excluded from the study due to these quality control procedures. Despite these limitations, our approach presents key advantages in allowing the interrogation of high-resolution structural and functional changes at chronic time points, which adds new information to the understanding of device-tissue interactions.

We did not observe any impacts of electrode material on our measured effects, despite a reduction in the Young’s modulus of polyimide in comparison to silicon. Several groups have pursued electrode designs which minimize the mismatch between the mechanical properties of devices and brain tissue; this mismatch has been hypothesized to contribute to poor integration [63,64]. We investigated the potential impact of material flexibility on our results by studying devices made of either silicon or polyimide: these materials are particularly relevant as two leading substrates used in neural recording research. There are possible reasons for the lack of effect. First, recent reports implicate bending stiffness, which is a function of both Young’s modulus and device dimensions, as a stronger determinant of tissue response than Young’s modulus alone[65]. As such, the reduced Young’s modulus may have been insufficient to produce an impact on the tissue response without a coordinated reduction in dimensions. Secondly, the devices were pseudo-tethered, that is, stabilized only by surrounding gel foam and connective tissue infiltrating the craniotomy. We chose not to connectorize the implants in order to facilitate easy retrieval of devices within brain slices in this study, but it is possible that a more rigid fixation point would draw out the benefits of the polyimide material[66]. Several reports indicate that the method of device fixation impacts the tissue response to neural implants: rigid fixation to a skull-mounted connector exacerbates gliosis in comparison to free-floating implants, presumably due to micromotion [67–69]. It is also possible that more rigid fixation would exacerbate the effects of device implantation on neuronal structure and function observed here. Investigating the impact of tethering forces will be a useful point of inquiry in future work.

In summary, we have reported a new characterization of structural and functional changes in neurons surrounding implanted electrodes in the brain. While neurons present at the device interface retain the ability to produce signals, alterations in the regularity of spiking, the restructuring of dendritic arbors, and disengagement from the surrounding network could contribute to common observations of variability, signal loss, and shifting stimulation thresholds. The many opportunities for further investigation include extension of additional time points, materials, surface coatings, insertion strategies, and tethering schemes, as well as connecting functional read-outs to neuronal transcriptional profiles and underlying glial mechanisms. Likewise, confirming morphological effects using alternative methods, and benchmarking results relative to histological outcomes, would be valuable follow-on studies. Additionally, further study of the potential effects of insertional techniques and damage would be a useful area of future exploration; some effects noted in this study were at least partially explainable by the stab wound injury. Future work is needed in this area to more completely characterize the effects of implanted electrodes on surrounding cells and predict effects of neuronal structural rearrangement on recorded signals via computational modeling.

DATA AVAILABILITY

Data will be made available on request.