Post-explant profiling of subcellular-scale carbon fiber intracortical electrodes and surrounding neurons enables modeling of recorded electrophysiology

Joseph G. Letner, Paras R. Patel, Jung‐Chien Hsieh, Israel M. Smith Flores, Elena della Valle, Logan A. Walker, James D. Weiland, Cynthia A. Chestek, Dawen CaiJournal of Neural Engineering 2023 · DOI

TL;DR

This paper presents a novel method for precisely localizing the recording sites of tiny carbon fiber electrodes within brain tissue after implantation. This allows researchers to accurately map nearby neurons and their health, revealing that these electrodes maintain a naturalistic neuron distribution. This detailed mapping enabled a computational model that suggests the proximity of neurons, not just electrode properties, significantly impacts how many individual neuron signals can be reliably extracted from recordings.

What Problem Does This Paper Address?

Understanding how neural signals are recorded by implanted electrodes is crucial for both deciphering brain function and developing effective brain-machine interfaces (BMIs). A key challenge is precisely knowing where the electrode is recording from and which neurons are nearby. Traditional silicon electrodes can cause significant tissue damage (foreign body response, FBR), leading to neuron loss and altered neural activity, making it hard to interpret recordings. While smaller, more biocompatible electrodes like carbon fibers show promise, accurately correlating their recordings with specific neurons, especially after chronic implantation, has been difficult. This paper tackles this by developing a post-explant histological technique to precisely map electrode locations and surrounding neurons, enabling a more accurate understanding of the electrode-neuron relationship and its impact on signal interpretation.

Methods

The core innovation lies in the post-explant histological localization of carbon fiber electrode tips. After implanting rats with subcellular-scale (6.8 µm diameter) carbon fiber electrode arrays for extended periods (6-12+ weeks), the researchers explanted the arrays. They then used a combination of immunohistochemistry (staining for neurons, astrocytes, and microglia) and confocal microscopy to create high-resolution 3D reconstructions of the implant site.

Key steps included:

  • Tip Localization: Electrode tracts were identified as dark holes in transmitted light images, often surrounded by glial scarring (GFAP). By following these tracts in 3D and observing where they filled with surrounding tissue, the precise tip location was estimated. This was independently verified by three reviewers to ensure reproducibility.
  • Neuron Segmentation: Neuron somata (cell bodies) within a 50 µm radius of the localized electrode tips were manually segmented in 3D using NeuN staining. This allowed for the measurement of neuron positions, health, and density.
  • Electrophysiology Modeling: A simple point source model was used to predict extracellular spike waveforms. This model was fit using recorded electrophysiology and the measured positions of nearby neurons. The model aimed to understand how neuron placement influences the amplitude of recorded spikes and, consequently, the ability to distinguish individual neuron signals.

Datasets

The study utilized data from five adult male Long-Evans rats implanted with carbon fiber electrode arrays. Recordings were collected weekly for 10-minute sessions. Histological data was generated from these implanted brains. For control comparisons, data from contralateral hemispheres (the opposite side of the brain from the implant) were used, where hypothetical fiber locations were simulated. No new benchmark datasets were introduced; rather, the study generated a unique dataset of precisely localized electrode tips and surrounding neuron populations.

Key Insights and Ideas

  • Subcellular Precision in Histology: The ability to localize electrode tips with subcellular precision after implantation is a significant methodological advance. This overcomes the limitations of in-vivo localization and provides a crucial "ground truth" for correlating recordings with neural anatomy.
  • High Biocompatibility of Carbon Fibers: Despite chronic implantation, the carbon fiber electrodes showed minimal signs of severe FBR. Neuron density near the tips was remarkably similar to that of healthy, contralateral brain tissue, suggesting these electrodes are highly biocompatible.
  • Neuron Placement Matters for Spike Sorting: The study's modeling revealed that the spatial distribution of neurons around an electrode is a critical factor in determining how many individual neuron spike clusters can be reliably sorted. This challenges the assumption that only electrode properties or spike sorting algorithms limit cluster counts.
  • Redefining the "Single Unit" Radius: The modeling suggests that the radius within which single units can be reliably distinguished from others is closer to the fourth closest neuron (around 30.7 µm), rather than the commonly assumed 50 µm. This implies that neuron density and arrangement play a more significant role than previously appreciated.

Key Results

  • Electrode Tip Localization: The researchers achieved a median absolute difference of 9.6 µm in tip localization across three reviewers, demonstrating high precision.
  • Neuron Density Preservation: Neuron density near implanted carbon fibers was 82% of that measured in contralateral control sites, significantly higher than reported for traditional silicon electrodes (which can lose up to 60% of neurons within 50 µm).
  • Neuron Morphology Alterations: Neurons near implanted fibers showed increased soma distortion (CSSI) (111% larger than controls) and were 23% smaller in volume, indicating mechanical stress. However, this distortion was predicted to normalize at distances beyond ~74 µm.
  • Neuron Distribution Similarity: The distances to the nearest six neurons around implanted fibers were statistically similar to those around hypothetical fibers in the contralateral hemisphere. The first closest neuron was only 1.0 µm farther away on average.
  • Spike Amplitude Modeling: The model predicted that spike amplitudes become indistinguishable for sorting beyond the fourth closest neuron (30.7 ± 4.6 µm), suggesting this is a critical boundary for single-unit isolation.

Limitations and Future Work

  • Neuron Elongation: While neuron density was preserved, neurons near the electrodes were stretched. The study acknowledges this morphological change and suggests it's likely due to chronic micromotion or insertion trauma.
  • FBR Components: While overall FBR was low, some components like astrocyte enrichment were observed along electrode tracts, suggesting potential for further optimization of array architecture.
  • Model Simplification: The point source model, while effective, is a simplification. The authors note that incorporating dendritic morphology and axon initial segment contributions could improve model accuracy.
  • Brain Region Dependence: The identified "single unit" radius is specific to layer V motor cortex. Future work could investigate how this radius varies in brain regions with different neuron densities and cell types.
  • Further Validation: While the model's predictions align with observed spike sorting limitations, further experimental validation with more complex models and diverse recording sites would be beneficial.

Summary updated

Abstract

<i>Objective.</i>Characterizing the relationship between neuron spiking and the signals that electrodes record is vital to defining the neural circuits driving brain function and informing clinical brain-machine interface design. However, high electrode biocompatibility and precisely localizing neurons around the electrodes are critical to defining this relationship.<i>Approach.</i>Here, we demonstrate consistent localization of the recording site tips of subcellular-scale (6.8<i>µ</i>m diameter) carbon fiber electrodes and the positions of surrounding neurons. We implanted male rats with carbon fiber electrode arrays for 6 or 12+ weeks targeting layer V motor cortex. After explanting the arrays, we immunostained the implant site and localized putative recording site tips with subcellular-cellular resolution. We then 3D segmented neuron somata within a 50<i>µ</i>m radius from implanted tips to measure neuron positions and health and compare to healthy cortex with symmetric stereotaxic coordinates.<i>Main results.</i>Immunostaining of astrocyte, microglia, and neuron markers confirmed that overall tissue health was indicative of high biocompatibility near the tips. While neurons near implanted carbon fibers were stretched, their number and distribution were similar to hypothetical fibers placed in healthy contralateral brain. Such similar neuron distributions suggest that these minimally invasive electrodes demonstrate the potential to sample naturalistic neural populations. This motivated the prediction of spikes produced by nearby neurons using a simple point source model fit using recorded electrophysiology and the mean positions of the nearest neurons observed in histology. Comparing spike amplitudes suggests that the radius at which single units can be distinguished from others is near the fourth closest neuron (30.7 ± 4.6<i>µ</i>m,X-± S) in layer V motor cortex.<i>Significance.</i>Collectively, these data and simulations provide the first direct evidence that neuron placement in the immediate vicinity of the recording site influences how many spike clusters can be reliably identified by spike sorting.
BibTeX
@article{letner2023post,
  author = {Joseph G. Letner and Paras R. Patel and Jung‐Chien Hsieh and Israel M. Smith Flores and Elena della Valle and Logan A. Walker and James D. Weiland and Cynthia A. Chestek and Dawen Cai},
  title = {{Post-explant profiling of subcellular-scale carbon fiber intracortical electrodes and surrounding neurons enables modeling of recorded electrophysiology}},
  journal = {Journal of Neural Engineering},
  year = {2023},
  doi = {10.1088/1741-2552/acbf78},
}
Figure