Neuropathological effects of chronically implanted, intracortical microelectrodes in a tetraplegic patient

The participant was a 64 years-old, right-handed, tetraplegic male who was implanted with three MEAs 4 years after he sustained an incomplete, traumatic injury of the spinal cord at level C4 (ASIA Impairment Scale B) during a snowboarding accident.

The study was approved by the Institutional Review Boards at the University of Southern California (USC) (HS-13-00492) and Rancho Los Amigos National Rehabilitation Center. It was approved by California Institute of Technology under a reliance agreement with USC. The study received FDA clearance under investigative device exemption (IDE #G130100) and was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the participant after the details of the study and potential risks were fully explained.

Three 96-channel NeuroPort MEAs from Blackrock Microsystems were implanted into the cerebral cortex of the left hemisphere (figure 1) [5]. Two arrays had platinum-tipped microelectrodes for recording neuronal activity. In the 3rd array, the electrode tips were coated with SIROF for neural recording and microstimulation. One array with 1.0 mm long platinum-tipped microelectrodes was implanted into the putative human AIP, a grasp area at the junction of the left intraparietal and postcentral sulci; a 2nd MEA with platinum-tipped microelectrodes was implanted into BA5; and the 3rd array with 1.5 mm long SIROF-tipped microelectrodes was implanted into BA1 of the S1 (figures 1(C)–(E)). The electrode pitch of the MEAs is 400 µM. Each MEA had a gold wire bundle (arrowheads, figure 1(C)) connected to a titanium percutaneous pedestal affixed to the skull.

During surgery, the pneumatically actuated impulse inserter (used to insert the MEAs into the cortex) was applied twice on each of the recording arrays, and three times on the S1 array without removing MEAs between insertions, because the arrays failed to insert completely in the 1st one (or two) impulses. There appeared to be a thickened area of leptomeninges overlying the site of the S1 array, but it was inserted through the intact leptomeninges. The implantation surgery proceeded without clinical complication and the participant was monitored in the intensive care unit for 2 d before being discharged home to continue rehabilitation.

Microelectrodes in the corner positions of each MEA were electrically unconnected by design, leaving 96 microelectrodes active for recording and/or stimulation (figure 1(A)). Neural activity was recorded with the NSP system (Blackrock Microsystems). Neural signals were amplified, bandpass filtered to 0.3 (1st order Butterworth highpass)—7500 Hz (3rd order Butterworth lowpass) and sampled at 30 000 samples s⁻¹ with 16-bit resolution. Impedance measurements were conducted using hardware built into the Blackrock recording system (supporting measurement only at 1 kHz). Mean microelectrode impedances were calculated as the average of these measurements from all 96 microelectrodes in the MEA. Microelectrode impedance measurements were made at the time of manufacture, and throughout the study, are reported in supplementary material (available online at stacks.iop.org/JNE/18/0460b9/mmedia).

Recording of neuronal activity began 20 d after array implantation, with a total of 71 individual recording sessions over 205 d until the patient expired. During each session, neural recordings were made while the patient performed various behavioral tasks (for example, neural control of a computer cursor, neural representations of grasp of an object). Some tasks included the delivery of pulse trains through the SIROF-tipped MEA in S1 for the purpose of testing whether stimulation of S1 could elicit somatosensory percepts (CereStim stimulator, Blackrock Microsystems, Salt Lake City, UT, USA). These experiments did not produce the desired outcome, although similar experiments by our group and others have demonstrated this capability of intracortical microstimulation [5, 6].

Due to the prioritization of recording from the AIP and BA5 MEAs versus stimulating through the S1 MEA, fewer recordings were made through the S1 MEA-67 and 65 sessions from the AIP and BA5 MEAs, respectively, and 43 sessions for the S1 MEA.

Sessions with behavioral tasks on average included 7.8 ± 3.2 (mean ± SD) individual tasks that were each 5.3 ± 3.9 min in duration. Putative neuronal action potential waveforms were extracted from these data by bandpass filtering between 250 and 5000 Hz (4th order Butterworth bandpass) and thresholding the data at 3.5 times the root-mean-square estimate of the noise. Extracted waveforms were 48 samples long with the threshold crossing placed at the 11th sample (by convention with Blackrock Microsystems' recording systems). The threshold value was selected and verified manually, by visual inspection, to detect action potentials greater than the magnitude of background noise, thus minimizing the likelihood of spurious assignment of noise as neuronal action potentials. Spike sorting was used only to show sample putative single units on sample neural recordings from each array (i.e. figure 4); other analyses used unsorted threshold crossings. Spike sorting was performed independently for each data file by k-means clustering of the waveform data projected onto its 1st three principal components, with visual inspection to remove any likely noise units. A proxy for signal-to-noise ratio (SNR) of the action potential waveforms was calculated as the inverse of the coefficient of variation, i.e. the ratio between the mean and standard deviation of the hyperpolarization phase of the action potential waveform, with arbitrary units (au). Inter-spike interval (ISI) histograms were constructed as normalized histograms of the intervals between spike events. To highlight trends in impedance, neuronal unit counts, and SNR over time, a 3rd order polynomial was robustly fit to these data by linear regression (fitlm method, MATLAB, The Mathworks, Inc., Natick, MA, USA) and shown overlaid on the time series measurements.

A complete autopsy was performed to determine the cause of death. Acute and chronic bronchopneumonia, secondary to aspiration, with tetraplegia and a history of gastroesophageal reflux disease were contributing factors. Other pathological findings are described in the supplementary material.

The entire brain, with attached dura and electrodes, was removed and placed in 10% phosphate-buffered formalin for 1 month. After fixation of the brain, there was diffuse thickening of the leptomeninges covering each MEA site (figures 2(A)–(C)). Removal of the MEAs revealed the extent of meningeal encapsulation of the arrays (figures 2(D)–(F)) with the S1 site showing the greatest encapsulation (figure 2(E)). The sites of penetration of the individual microelectrodes into the underlying tissue were visible (figures 2(D)–(F)). The depth of penetration of the microelectrodes into the underlying tissues was determined for all three array sites. The tissue blocks measured approximately 1.5 × 1.5 × 1.0 cm (D × W × L) for the S1 array, 1.9 × 1.0 × 1.0 cm for the BA5 array, and 0.8 × 0.8 × 0.8 cm for the AIP array. Tissue from the AIP site was utilized for a novel histological study using a CLARITY-based method [33] and not examined further for this study. Normal (control) tissues from the contralateral anterior frontal cortex were also sampled. Those tissue blocks measured approximately 2.0 × 2.0 × 2.0 cm (data not shown). During tissue processing, the BA5 block became separated into two samples, a superficial portion (BA5-2A) and a deep portion containing white matter (WM) (BA5-2B).

Zoom In Zoom Out Reset image size

The BA5 and S1 implant sites were cut parallel to the pial surface to show the cross-section of the microelectrode tracks, while the anterior frontal cortex, without implant, was cut perpendicular to the pial surface to observe the normal layering of the cortex.

The sampled tissue areas were paraffin-embedded and processed for interval sectioning. The BA5 (BA5-2A and BA5-2B) and S1 tissue blocks were serially cut into five 5 µm thick sections at 100 µm or 300 µm intervals, respectively, with every level stained with H&E and remaining slides saved as unstained. H&E stained sections were analyzed at the BA5 (15 slides for each BA5 block) and S1 sites (24 slides) to assess tissue changes and penetration depth of microelectrodes. Once H&E sections established electrode penetration depth, the unstained slides for that approximate level were then processed for Masson's trichrome to assess fibrosis, cresyl violet to evaluate neuronal status, or immunostained.

Tissue sections stained with 0.75% cresyl violet were dehydrated through graded alcohols and cleared in xylene. Trichrome staining was performed using the Masson trichrome staining kit (Sigma–Aldrich, St Louis, MO, USA). Remaining sections were immunostained using the bond polymer refine DAB detection system (Leica Biosystems, Buffalo Grove, IL, USA). Antigen retrieval was performed with epitope retrieval solution 1 (ER1, pH 6) for 10 min. Sections were then incubated for 15 min with glial fibrillary acidic protein antibody (GFAP; AB5804, Chemicon, Temecula, CA, USA; 1:2500) for detection of gliosis, or ionized calcium-binding adaptor molecule (IBA-1; 019-19741, Wako, Osaka, Japan; 1:500) to define microglial response and foreign body giant cells (FBGCs). Images were obtained with a Nikon Eclipse E400 brightfield microscope (Nikon Instruments Inc., Melville, NY, USA). All stains were performed at Children's Hospital of Los Angeles.

Over 71 total recording sessions, neuronal action potential waveforms were recorded from the BA5 and S1 MEAs in 65 and 43 sessions, respectively, with BA5 recordings present in all sessions with S1 recordings. Figures 3(A) and (B) show waveforms of unsorted action potentials with estimates of SNR ⩾ 0.2. The S1 MEA recorded fewer channels with action potentials with SNR ⩾ 0.2 than the BA5 MEA (figures 3(C) and (D)). In figure 3(C), the 3rd order polynomial fits to the action potential counts from the channels with detectable action potentials and were significant for the BA5 MEA with respect to a constant model (F-statistic 19.1, p < 1 × 10⁻⁷) but were not significant for S1 MEA (F-statistic 0.807, p = 0.502). Of 96 electrically functioning channels in the BA5 MEA, 77 (80.2%) showed probable action potentials (unsorted waveforms with SNR ⩾ 0.2) in at least one recording session. Only two channels (2.1%) of the S1 MEA recording action potentials with SNR ⩾ 0.2 in at least 20% of recording sessions, versus 32 channels (33.3%) for the BA5 MEA.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

On the BA5 MEA, the electrodes of the lower-right (LR) quadrant of the array recorded the fewest action potential waveforms. On the S1 MEA, the best recordings of neuronal activity were from the electrodes near the upper and right edges of the array. Estimates of SNR were computed for unsorted action potential waveforms over the course of the study. The number of channels with detectable neural units increased over time for the BA5 MEA but not for the S1 MEA (figure 3(C)). SNR for the BA5 array increased over time (m = 0.0016 units d⁻¹, p < 0.001), but did not increase for the S1 array (m = 3.5 × 10⁻⁵ units d⁻¹, p = 0.54). Figure 3(D) shows mean SNR of unsorted action potential waveforms for each session with ten or more channels having SNR ⩾ 0.0, with higher values for BA5 MEA (Mann–Whitney, p < 1 × 10⁻⁹). Third-order polynomial curves shown in figure 3(D) were significant with respect to a constant model for the BA5 data (F-statistic 22.4, p < 1 × 10⁻⁸) but not S1 data (F-statistic 0.862, p = 0.482).

During each recording session, electrode impedance was measured at 1 kHz using the NeuroPort system. As expected, due to differences in the material properties of platinum and SIROF, the platinum-tipped microelectrodes exhibited higher impedances than the SIROF-coated microelectrodes, and microelectrode impedance decreased for both arrays over the duration of the implant (see supplementary material). Although most channels from the S1 MEA yielded poor-quality neural recordings, both the S1 and BA5 MEAs provided at least some examples of higher quality neuronal activity (figure 4). The recordings were largely free of electrical noise, apart from a small amount of line noise at 60 Hz and harmonics at 180 Hz. Bandpass-filtered data (250–5000 Hz) showed neuronal action potentials with amplitudes exceeding a threshold of 3.5 times the estimated root-mean-square of the noise (figures 4(B) and (F)). After spike sorting, the aggregate action potential waveforms for the displayed putative neural units and the ISIs showed features expected for intracortical, extracellular neuronal recordings, including the distribution shapes visible in the ISI histograms (figures 4(D) and (H)) and waveform shapes reflective of the action potential depolarization-repolarization cycle.

Visual inspection of the MEAs after removal from tissue did not reveal gross physical damage such as broken shafts (e.g. figure 2). High-resolution optical or SEM imaging was not performed.

We examined H&E-stained sections of the S1 and BA5 tissue blocks (cut parallel to the pial surface) to determine the depth of penetration of the microelectrodes and histological changes surrounding the electrodes. Twenty-four and 15 H&E-stained sections were analyzed for the S1 and BA5 array, respectively. The depth of penetration of the electrodes was approximately 0.5 mm for the BA5 array and up to 1.5 mm for the S1 array. The superficial aspect of the implant site consisted of neocortex and thickened leptomeninges (figures 5(A) and (B)) and leptomeningeal fibrosis as confirmed by Masson-trichrome staining (figures 5(C) and (D)). The BA5 array revealed proliferation of thin-walled vessels with luminal expansion and tissue dilation in the leptomeninges (figures 5(A) and (C)) that are absent in S1 sections.

Zoom In Zoom Out Reset image size

The ectatic blood vessels in the leptomeninges appear to be associated with a foreign body response [11, 24]. Depth of penetration of the S1 and BA5 electrodes varied due to the curvature of the cortex and/or angle of the histology sections relative to the orientation of the electrode shanks, as indicated by the variations in the size of the electrode tracks (figures 5(A)–(D)).

The tissue from the UL region nearest to both array's wire bundle includes a collagen-containing (fibrotic) scar suggesting a tissue reaction higher up along the electrode tracks and closer to the brain surface.

The tissue surrounding the electrode imprints show macrophage influx into the molecular layer (layer I). The BA5 sections at a depth of 0.25 mm reveal fibrous and gemistocytic astrocytes and activated microglia, as indicated by GFAP and IBA-1 immunohistochemistry (figures 6(A) and (D)). Gemistocytic astrocytes are characterized by hypertrophic cell bodies and short, thick processes (figures 6(A) and (B), inset) as opposed to non-hypertrophic cell bodies in the contralateral cortex (figure 6(C), inset). The cortex beneath the S1 array was replaced with extensive gliosis and necrosis, and the remaining cortex contained a diffuse distribution of reactive astrocytes and numerous microglia (figures 6(B) and (E)). Areas of necrosis and neuronal loss (acellular regions) also were observed beneath the BA5 (figure 6(G)) and S1 (figure 6(H)) arrays. Notably, areas of cell loss include a focus of hemorrhages with granulation tissue in the LR region of the BA5 array (figure 6(G)) which correlates with the low recording quality in the corresponding region of the BA5 array (figure 3(A)). Similar correlations between tissue damage and recording quality occurred at the S1 array site, where areas of decreased recordings of neuronal activity (figure 3(B)) in the LR and lower left (LL) correlate with areas of necrosis near electrode tips (figure 5(D)). Cresyl violet stain reveals cells with neuronal morphology (distinct pyramidal cell bodies as seen in control tissue) near the electrode tips (figure 6(J), inset) suggesting a possible correlation with the neuronal action potentials. In some areas there were numerous, small, hyalinized arterioles (figure 7(D)) beneath the S1 array and a few residual neurons near an electrode tip (figure 6(K), inset). However, the infarcted areas were mostly devoid of neurons.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In figure 7, the S1 MEA site showed FBGCs surrounding most of the electrode tracts, and infiltrates of lymphocytes and macrophages (figures 7(A) and (B)). The FBGC reaction consisted of syncytium-forming macrophages, as shown by positive IBA-1 staining (figure 6(E)). Deeper tissue sections in the LL region of the S1 array showed regions of cortical and WM necrosis (figure 7(C)) with numerous hyalinized, thickened arterioles (figure 7(D)) indicating vascular injury which likely contributed to persistent cortical ischemia. These changes also were seen at the BA5 site. H&E-stained sections of S1 MEA site revealed subacute inflammation with gemistocytic astrocytes (astrocytosis) in the cortex adjacent to the S1 site and destruction of the underlying WM that was engulfed by foamy microphages (figure 7(E)). The gliosis and subcortical necrosis are indicative of the intermediate duration of damage (weeks to months) due to the long-term presence of the arrays. In contrast, the BA5 MEA site (block BA5-2B) showed no significant histopathologic changes in the subcortical WM (data not shown). Near a superficial S1 electrode tract, diffusely thickened leptomeninges was observed, and around the subcortical WM was a rim of gemistocytic astrocytes and sparse, chronic inflammatory cells (figure 7(F)). One leptomeningeal small artery revealed recanalization suggestive of failed or successful focal restoration of blood flow and hemorrhage (figure 7(F)), the latter a possible cause of the surrounding infarction. Together, possible hemorrhagic- or ischemia-induced infarction, neuronal loss and subcortical WM necrosis likely explain the poor overall recording of neuronal activity (figure 3(B)) and absence of effective stimulation by the S1 array.