Introduction

In the past two decades, the advancement of implanted electrode arrays has drastically shifted the field of neuroscience from theories based on individually recorded neurons to those based on the systems-level activity of neuron populations [1–3]. Specifically, the 96-channel Utah array (Blackrock Neurotech, Salt Lake City, UT) has been used to make pivotal discoveries in motor systems neuroscience and develop clinically impactful brain-computer interfaces [2, 4–8]. These electrode arrays are widely used in neuroprosthetics research for patients with motor disorders and have been used for a variety of purposes, such as movement and speech decoding [9, 10]. Much of this work in humans relies on the translation of results discovered in non-human primates (NHPs), a crucial animal model in neuroscience research [11, 12]. One useful application of these arrays in animal studies is microstimulation and electrolytic lesion experiments [13]. Electrolytic lesioning results in small, controllable, and precise amounts of neuron loss. This makes it a helpful technique to study damage to targeted areas of the brain, serving as a model for clinical injury-based neuron loss events at a small scale.

Electrolytic lesion has been experimented with since the 19th century [14], and methods have been researched in many species, including cats, dogs, monkeys, and humans [15, 16]. In electrolytic lesion experiments, researchers pass current through nearby electrodes on arrays in order to mark the position of electrodes or create lesions in brain tissue [17, 18]. In particular, recent work has established a novel lesioning technique to create consistent, controllable electrolytic lesions in NHPs using already-implanted Utah arrays, without damaging the ability to record from the array [19]. However, while recording capabilities seem to remain stable, the physical effect of electrolytic lesioning on the recording array needs further characterization.

Previous studies have shown temporally decreasing performance of collected signals from long-term implanted arrays, thought to be a consequence of the immune response, mechanical movement of the array relative to surrounding tissue, and physical degradation of the array [20–22]. This is also reflected in studies that specifically examine the damages that occur to electrode arrays while they are implanted in NHPs [23, 24]. In particular, the use of scanning electrode microscopy (SEM) has been used to visually examine and analyze the effect of neural implantation on arrays with extremely high definition [24–28]. These studies, which span both NHPs and humans, provide direct evidence that the observable physical damage incurred by implanted arrays has direct impact on the array’s performance.

The Utah array is a silicon-based array manufactured with several layers of different materials [29]. Silicon electrodes are first etched out of a glass and silicon base. Each individual electrode is then metallized at the tip. Finally, a layer of parylene C coats all but these metallized tips to help isolate and insulate each electrode [30, 31]. Both in vitro and in vivo testing of electrode arrays revealed environmental damage to these materials, such as cracking, textural defects, and degradation in response to the brain’s temperature and salinity [32]. The immune response of the brain also damages the electrodes due to effects like glial scarring (gliosis) and inflammation [33, 34]. This damage may be exacerbated by the surgical techniques used during implantation, which include pushing the electrode array into cortex and tethering the implant to the skull [33, 35, 36].

Additionally, the Utah array is available with both platinum and iridium oxide-based coatings. Aggressive electrical stimulation is known to dissolve platinum-based electrodes [37, 38]. Other studies have shown iridium oxide to be more resistant to stimulation-related damage, but not completely insusceptible [39, 40]. However, the physical electrode damage associated with electrolytic lesioning remains to be studied.

This work comprises images of eleven different multielectrode Utah arrays (ten 96-channel, one 64-channel). Of these imaged arrays, eight arrays chronically implanted in regions of the motor cortex for varying amounts of time were analyzed further (monkey H = 2242 days, F = 1875, U = 2680, C = 594). Four out of eight of these arrays were used to perform electrolytic lesioning experiments (n = 4 monkeys; H = 10 lesions, F = 1, U = 4, C = 1). Additionally, the image set includes images of a broken 96-channel Utah array, which was shattered during explant but remained intact while implanted. This set of images represents the largest publicly available collection of high-quality, single-electrode SEM images of explanted multielectrode Utah arrays. A total of 938 individual electrodes are available to view, along with 11 additional images from the broken array. Specific details for each imaged array are available in Table 1 (arrays.csv).

Methods

Array Preparation

Monkey H, F, and C were each implanted with two 96-channel Utah arrays in M1 and PMd. H’s M1 array was used for electrolytic lesioning experiments. F’s and C’s PMd arrays were used for electrolytic lesioning experiments Monkey U was implanted with three 96-channel arrays; two in M1 and one in PMd. U’s lateral M1 array was used for electrolytic lesioning experiments. U’s medial M1 array was not used for recording and was severely damaged during extraction. Monkey H’s arrays were implanted for 2242 days (electrolytic lesions on days 2088, 2129, 2136, 2164, 2172, 2180, 2187, 2192, 2221, 2228). Monkey F’s arrays were implanted for 1875 days (electrolytic lesion on day 1875). Monkey U’s arrays were implanted for 2680 days (electrolytic lesions on days 1215, 1250, 1264, 1286). Monkey C’s arrays were implanted for 594 days (electrolytic lesion on day 304). Further details on each monkey’s arrays, as well as any additional imaged arrays, are available in Supplemental Table 1.

All arrays used in this work were multielectrode Utah arrays (Blackrock Neurotech, Salt Lake City, UT). Monkey C was implanted with an array with electrode shank lengths of 1.5 mm, while Monkeys H, F, and U were implanted with arrays with electrode shank lengths of 1 mm. Monkey U and C’s arrays were manufactured with an iridium oxide metal coating, while H and F’s arrays were manufactured with platinum coatings.

Arrays were first soaked in either a 0.9% saline or 10% formalin solution immediately following explant surgery, then stored separately in either water, formalin, ethanol, or dry in a closed container. Prior to imaging, all arrays were allowed to dry without any further cleaning or preparation.

Scanning Electron Microscopy

Scanning Electron Microscopy (SEM) images were collected using a Hitachi TM4000+ Tabletop SEM (Hitachi Ltd., Tokyo, Japan). Arrays were mounted on PELCO SEMClip™ Pin Mounts (Ted Pella, Redding, CA) using conductive copper tape, carbon tape, and included clip mounts. Arrays were not sputter-coated for imaging. Due to the amount of organic material left on the arrays, images were collected in low-vacuum mode to avoid distortion from accumulated electrical charge of the non-conductive material. Full array images were collected with the arrays flat on the mount surface, while single-electrode images were collected with the arrays positioned at an approximate 45° angle. Most images were collected using a mixed backscatter/secondary electron (BSE/SE) mode with a low accelerating voltage of 5kV, although modes and voltage values were adjusted for certain electrodes to optimize image clarity. Specific settings for each collected image are included in Figure 1.

Figure 1:

Interactive Display: Interactive display of all captured SEM array images. Once an array has been selected, a diagram of each array’s anatomical position (derived from surgical drawings and notes) appears, along with an SEM image of the array. All array images are displayed with the wire bundle to the right side. Specific electrodes may be selected from the array image; SEM images of each electrode and their scores in each of the five damage categories will appear along with any additional notes. Furthermore, examples of specific scores for each damage type are selectable through an additional drop-down menu for each array. Similarly, electrodes used for electrolytic lesioning are also selectable through an additional drop-down menu for each array.

Damage Scoring

A prior study categorized damage observable with SEM into six distinct types: abnormal debris, tip breakage, coating cracks (cracking of the metal coating around the silicon core of an electrode), parylene C cracks, parylene C delamination, and shank fracture [26]. Similarly, damage to the arrays in this study were scored in five categories: abnormal debris (AB), silicon tip breakage (TB), platinum coating cracks (CC), parylene C coating cracks (PC), and parylene C coating delamination (PD). Each recording electrode was scored from 0 to 3 in five different categories, as described above. Electrodes with shank fractures (SF) were not scored, as these fractures may have reasonably occurred during explantation and handling of the arrays separate from the damage incurred while in the brain. Each electrode was scored between 0-3, where 0 indicated no visible effects, 1 indicated visible effects but likely no impact on recording ability, 2 indicated possible impact on recording ability, and 3 indicated likely impact on recording ability. Scores were given without prior knowledge of which electrodes had been used for electrolytic lesioning. Although this scoring system is subjective and based on visual observation, all images used to compute these scores are publicly available in Figure 1 for independent evaluation.

Statistical Tests

First, the Pearson’s correlation coefficient was calculated between each subcategory of damage to determine whether any types of damage were related. To explore potential spatial relationships, this analysis was repeated for each radial ring of electrodes, as shown in Supplemental Figure 4.

Statistical significance was also assessed between electrodes used in the electrolytic lesion experiments versus those which had not. The Shapiro-Wilk test indicated that scores did not follow a normal distribution; therefore, nonparametric methods such as the Mann-Whitney U test and the Levene test were used to compare the underlying statistics of each array’s electrode population. The null hypothesis of the Mann-Whitney U test is that sample means of two sets of scores are the same. The null hypothesis of the Levene test is that the sample variance of two sets of scores are the same.

Similarly, statistical significance was assessed as described above between lesioning versus normal electrodes pooled across all four arrays used for electrolytic lesioning. Finally, this analysis was repeated between electrode scores across pairs of lesioning and non-lesioning arrays implanted in the same subject for the same length of time.

Results

All collected images, along with their associated scores, are available in interactive Figure 1. Each of the twelve arrays and individual electrodes may be selected and viewed. Representative electrodes for each category and damage score are also viewable for each array when available. Electrodes used for lesioning are also viewable when available. Rough anatomical layouts of the array’s implant location are also included. Electrode numbering maps are available in Supplemental Figure 1.

The first set of analyses specifically considers the four arrays used in electrolytic lesioning experiments. Scores for each damage category, along with total scores, are visualized as heatmaps in Figure 2. Median total scores for each radial section of the array show that the outermost electrodes tend to have the greatest total observed damage across all four arrays. Generally, electrode damage decreases when moving to the center of the array. Shank fractures tend to be spatially close on each array, suggesting that shared mechanical trauma caused these sections of fracture. As mentioned in the Methods section, these fractures may have reasonably occurred during explantation and handling, as no diminished recording performance of these fractured electrodes was observed while the array was actively implanted. Additional heatmaps are available for the four additional arrays implanted in NHP but not used for lesioning in Supplemental Figure 2.

Figure 2:

Histograms of damage scores in each category, along with average score distributions per array, are shown in Figure 3. For each category, histograms of electrodes used for lesioning (black) are stacked on top of histograms of normal electrodes (gray). Additionally, the average distribution of scores is shown for each array. These plots indicate that there is no observable difference in the distribution of scores given to electrodes used for lesioning versus those not. Additional histograms are available for the four additional arrays implanted in NHP but not used for lesioning in Supplemental Figure 3.

Figure 3:

The Shapiro-Wilk test for normality returned a significant p-value for almost all sets of scores for each array and type of damage. Similarly, the Shapiro-Wilk test for normality returned a significant p-value for almost all sets of scores for each array and type of damage when split into populations used and not used for lesioning, but was not performed where electrode count was too small to carry out the test (C and F; lesioning electrodes = 2). In the majority of cases that returned an insignificant p-value, certain types of damage were not present on the array and thus scores were uniformly zero (monkey U’s M1 tip breakage, parylene C delamination, and parylene C cracking scores; monkey C’s PMd parylene C delamination scores and M1 parylene C cracking scores). In all other cases where the Shapiro-Wilk test returned an insignificant p-value (monkey H’s total damage scores on lesioning electrodes, monkey U’s coating cracks scores), low electrode counts may reduce the statistical power of the test (H = 18 lesioning electrodes, U = 8). This indicates that scores are largely not normally distributed and that nonparametric statistical tests should be used to evaluate differences between populations. All Shapiro-Wilk test p-values are available in Supplemental Table 2.

When comparing lesioning and non-lesioning electrodes within the same array, each of the two nonparametric statistical tests (Mann-Whitney U-test, Levene Test) returned insignificant p-values for each category of damage as well as for total damage scores for all four arrays used in lesioning experiments. Similarly, when comparing pooled lesioning and non-lesioning electrodes across all four lesioning arrays, each of the two tests returned insignificant p-values for each category of damage and total damage scores. This indicates that the electrode populations used for lesioning and those not used for recording only do not have statistically different sample means or variances, suggesting they are samples of the same underlying population. These p-values are included in Supplemental Table 3. However, again, it is important to note that these tests done on individual arrays have limited power due to low electrode counts (H = 18 lesioning electrodes, F = 2, U = 8, C = 2).

This next set of analyses compares entire population statistics of all eight NHP arrays. Both tests resulted in inconsistently significant/non-significant p-values when comparing lesioning arrays against non-lesioning arrays within the same implanted animal. In monkey H, the Mann-Whitney U test resulted in an insignificant p-value for coating cracks and parylene C delamination scores, while the Levene test resulted in an insignificant p-value for abnormal debris, coating cracks, and parylene C cracking scores. In monkey F, the Mann-Whitney U test resulted in an insignificant p-value for parylene C delamination scores, while the Levene test resulted in an insignificant p-value for coating cracks, parylene C delamination, and parylene C cracking scores. In monkey U, the Mann-Whitney U test resulted in significant p-values for all scores, while the Levene test resulted in an insignificant p-value for abnormal debris, tip breakage, and coating cracks scores. Finally, in monkey C, the Mann-Whitney U test resulted in an insignificant p-value for parylene C delamination and parylene C cracking scores, while the Levene test resulted in an insignificant p-value for abnormal debris, parylene C delamination, and parylene C cracking scores. This indicates that there are likely large differences between the amount and intensity of material damage dealt to implanted arrays even when implanted in the same subject. These results do not determine whether or not electrolytic lesioning experiments contribute to the magnitude of these differences, but the inconsistency in the significance of the above results may indicate that other factors, such as the brain’s varied immune response, differences in mechanical forces, and disparities in external handling, may play a large role in damage to these arrays. These p-values are included in Supplemental Table 4.

The Pearson correlation matrix across all four arrays for the five different damage types is shown in Figure 4. This plot demonstrates that there is overall low pairwise correlation between each type of damage, with the exception of coating cracks and tip breakage, which display moderate positive correlation (r = 0.47, Bonferroni-corrected p < 0.05). These results align with with previous comparisons on the distribution of damage scores, which showed that damage score tended to vary across the five different types. Additional correlation plots for each of the five array layout rings are available in Supplemental Materials. Raw r and p-values for each test are also available in Supplemental Table 5.

Figure 4:

Discussion

Overall, these results collectively demonstrate that there is no obvious, statistically significant difference between observed damage to electrodes used versus not used for electrolytic lesioning. Because this method of lesioning involves passing limited direct current into the brain over relatively short time periods, its physical effects on the array electrodes are similar to alternating current stimulation. A recent study found that SEM-visible damage caused by stimulation on iridium oxide electrodes is highly variable between arrays, where one imaged array demonstrated noticeable stimulation-related damage and the other array had no such damage [28]. Furthermore, the study found that array recording quality was not compromised by stimulation. Similarly, other work did not show significant differences in SEM-visible degradation between both platinum and iridium oxide coated electrodes used for stimulation [24, 25]. All of these previous studies used different stimulation protocols: a pulse train delivered at 20-300 Hz at amplitudes 1-100 μA [28], a pulse train delivered at 300 Hz at amplitudes 1-210 μA [24], and an undisclosed low-current stimulation [25]. Additionally, dissolution of electrodes due to current is most salient at aggressive levels of stimulation; one study reviewing electrode dissolution delivered variable amplitudes of current at 50 Hz for seven hours[40]; another study utilized currents at least one order of magnitude greater than the above protocols [37]. The electrolytic lesioning in this work primarily used direct currents delivered at around 150 μA for 45 seconds, though parameters ranged between 50-450 μA and 12-600 seconds. Thus, despite the differences between specific lesioning and stimulation protocols, the findings presented here are consistent with previous studies that found no unusual damage to the electrodes used for stimulation. This aligns with previous findings that electrolytic lesioning does not significantly impact the recording ability of electrodes [19].

The patterns of damage found in this work also largely reflect previous SEM analyses of explanted arrays. A prior study also found that the electrodes closest to the perimeter of the array suffered the most damage [25]. This pattern is also visually observable in other published results [26, 28]. This is likely because the edges of the arrays serve as a physical shield for the innermost electrodes. Edge electrodes are subject to greater access to brain tissue, as well as to mechanical damage incurred during surgery and handling.

The frequency of damage types identified in this work differ from previous studies, but is not unusual. In this work, the most common damage type within arrays used for lesioning is abnormal debris (n=276, 71.9%), followed by metal coating cracks (n=214, 55.7%), tip breakage (n=102, 26.6%), parylene C cracks (n=95, 24.7%), and parylene C delamination (n=48, 12.5%). On these lesioning arrays, there were n=53, or 13.8%, shank fractures. Similarly, across all eight NHP arrays imaged, the most common damage type is abnormal debris (n=518, 67.4%), followed by metal coating cracks (n=480, 62.5%), tip breakage (n=233, 30.3%), parylene C cracks (n=224, 29.2%), and parylene C delamination (n=60, 7.8%). Across all eight NHP arrays, there were n=117, or 15.2%, shank fractures. However, a prior study found that coating cracks were the most common type of observed damage, followed closely by parylene C cracks, then tip breakage, abnormal debris, and parylene C delamination [26]. While the frequency of damage types are not exactly aligned, previous work included a cleaning step during array preparation for SEM imaging; this work does not explicitly attempt to remove excess brain tissue from imaged arrays, which likely led to the relative increase we observed in abnormal debris. Additionally, variation in individual researcher’s scoring standards may also contribute to differences. As mentioned above, the overall spatial pattern of damage between these two studies is consistent.

This study found a moderate positive correlation between coating cracks and tip breakage. This reflects the fact that damage to the silicon core of an electrode often occurs in tandem with cracking and flaking of the surrounding metal coating; the silicon cannot break without also breaking its coating. This correlation is visible in the collected images presented in Figure 1.

Conclusion

This work analyzed high-definition SEM images of previously implanted electrode arrays used for electrolytic lesioning. Across all eight arrays previously implanted in NHPs, damage disproportionately occurred to the outer edges of arrays. Additionally, no statistically significant difference was found between the damage experienced by normal and lesioning electrodes within the same array. Furthermore, statistical testing between arrays used and not used for lesioning experiments did not indicate consistent, significant differences. Finally, this work also presents the largest publicly available set of SEM images of explanted arrays, consisting of eleven different multielectrode Utah arrays (ten 96-channel, one 64-channel) and one broken 96-channel array. These results from this dataset align with previous SEM studies of chronically implanted arrays used for both recording and stimulation, while providing further evidence for electrolytic lesioning as a safe and useful technique for experimental neuroscience.

Supplemental Materials

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Supplemental Table 1:

Supplemental Table 2:

Supplemental Table 3:

Supplemental Table 4:

Supplemental Table 5:

Acknowledgements

We thank S Baker for veterinary support, and K Chin and M Truong for administrative support. The members of the Brain Interfacing Laboratory are Michelle S Wechsler, Mackenzie Risch, Alexandra Paraskevopoulou, Stephen I Ryu, Alissa S Ling, Elizabeth Jun, Michael P Silvernagel, Yuxin Wu, Kenji Y Marshall, Muhammad Abdulla, and Sydney Hunt. MS Wechsler, A Paraskevopoulou, K Lebedev, and MJ Risch were responsible for animal care and surgical support. SI Ryu was responsible for nonhuman primate array implantation. AS Ling, E Jun, MP Silvernagel, Y Wu, K Marshall, MU Abdulla, and S Hunt assisted in animal care. Part of this work was performed at the Stanford Nano Shared Facilities (SNSF), supported by the National Science Foundation under award ECCS-2026822. A Tor was supported by the Department of Defense (DoD) through the National Defense Science & Engineering Graduate (NDSEG) Fellowship Program and by a training grant from the National Science Foundation (1828993). SE Clarke was supported by a Stanford School of Medicine’s Dean’s Posdoctoral Fellowship. IE Bray was supported by an American Heart Association Predoctoral Fellowship (828653) and the National Science Foundation GRFP (1656518). This work was supported by a Stanford Human-Centered AI Seed Research Grant to SE Clarke and P Nuyujukian. This work was additionally supported by the following to P Nuyujukian: the National Institutes of Health (R01NS123517, R01NS130789, U19NS118284) and the Stanford Wu Tsai Neurosciences Institute.

Additional information

Author Contributions

A Tor was responsible for data curation, formal analysis, investigation, methodology, software, validation, visualization, writing, and review/editing. SE Clarke was responsible for conceptualization, data curation, supervision, validation, writing, and review/editing. IE Bray was responsible for conceptualization and review/editing. P Nuyujukian was responsible for conceptualization, funding acquisition, methodology, project administration, resources, supervision, writing, and review/editing. The members of the Brain Interfacing Laboratory provided additional support in review/editing.