Spike Code Flow in Cultured Neuronal Networks

Shinichi Tamura, Yoshi Nishitani, Chie Hosokawa, Tomomitsu Miyoshi, Hajime Sawai, Takuya Kamimura, Yasushi Yagi, Yuko Mizuno-Matsumoto, Yen-Wei ChenComput. Intell. Neurosci 2016 · DOI

TL;DR

This paper introduces a novel method to visualize and quantify the "flow" of information within cultured neuronal networks by analyzing spike trains. They identified specific "codes" (short spike sequences) that appear to propagate across electrodes, suggesting a mechanism for information transfer that is robust to noise and timing variations.

What Problem Does This Paper Address?

Understanding how neuronal networks process and transmit information is a fundamental challenge in neuroscience. While we can record spike trains, deciphering the "code" – how information is encoded in these electrical signals – remains elusive. Traditional methods like rate coding or analyzing spatiotemporal patterns have limitations. This paper tackles the problem of how information might be conveyed across a network, proposing that specific, recognizable patterns of spikes (codes) could act as dynamic markers of information flow, similar to how signals propagate in artificial systems.

Methods

The researchers stimulated cultured hippocampal neuronal networks with an 8x8 microelectrode array and recorded the resulting spike trains. They developed a method to extract short, binary "codes" (sequences of "1" for a spike and "0" for no spike within a defined time bin) from these spike trains. These codes were identified based on their occurrence more frequently than expected by chance, particularly M-sequence-related patterns.

To visualize the propagation of these codes, they constructed "code flow maps," essentially time-series movies of the electrode array showing where specific codes appeared over time. To quantify this flow, they calculated the "maximum cross-correlation" between the presence of a code on one electrode and its potential appearance on neighboring electrodes in the subsequent time frame. This metric, denoted as $\Phi_N(C, F)$, was further normalized to account for code duration and compared across original, shuffled, and randomly generated spike trains to assess the significance of the observed flow.

Datasets

The study used cultured hippocampal neuronal networks derived from Wistar rat embryos, grown on microelectrode array dishes (MED-P515A). The data collected from these cultures served as the primary dataset. The paper also mentions using "shuffled spike train interval" and "electrode shuffled" data, as well as "randomly generated trains" as control datasets for comparison.

Key Insights and Ideas

  • Spike Trains as "Codes": The paper proposes that specific, short sequences of spikes ("codes") can represent information units within a neuronal network.
  • Code Flow as Information Transfer: The visualization of these codes propagating across the electrode array suggests a mechanism for information transfer that is not solely dependent on individual neuron firing rates but on the movement of these patterned sequences.
  • Robustness to Timing: The identified codes appear to maintain their shape to some extent as they propagate, indicating a degree of robustness to precise timing variations, which is characteristic of biological neural networks.
  • Distinguishing Real Flow from Noise: By comparing the code flow in real neuronal data with shuffled and random data, the researchers could statistically demonstrate that the observed flow was a genuine network phenomenon and not a result of random chance or simple timing artifacts.
  • Local Circuitry Clues: The analysis of code propagation and cross-correlations could potentially offer insights into the structure and connectivity of local neuronal circuits.

Key Results

  • Specific binary codes, such as "1101" and "1011," were extracted from spike trains with a nominal time accuracy of 1%.
  • Code flow maps visualized these codes propagating from one electrode to neighboring ones, often maintaining their shape.
  • Quantitative analysis using maximum cross-correlations showed that the normalized cross-correlations were relatively constant, irrespective of the specific code.
  • Shuffling spike train intervals or electrodes significantly reduced the observed code flow, indicating that the structured propagation was a property of the intact network.
  • The flow of codes in the original, un-shuffled spike trains was significantly higher than in shuffled or random trains, supporting the idea of directed information transfer.

Limitations and Future Work

  • Spike Sorting: The study acknowledges that spikes were not sorted, meaning each electrode's signal could originate from multiple neurons. This limits the ability to definitively attribute codes to single neurons.
  • Short-Term Coherency: The observed phenomena were primarily analyzed within a short time frame (e.g., 20 ms after stimulation). The paper notes that longer time epochs might not exhibit such synchronous or coherent behavior, and trials may run asynchronously.
  • Statistical Nature: The code flow is described as a "stochastic phenomenon," requiring statistical measurement.
  • Future Work: The authors suggest that this approach can help estimate local circuit shapes and that the analysis can be extended to investigate communication between remote brain positions. They also plan to discuss the simulation of code spectrum and flow in future publications. The paper also hints at future work on using these findings for pattern classification and brain intelligence research.

Summary updated

BibTeX
@article{tamura2016spike,
  author = {Shinichi Tamura and Yoshi Nishitani and Chie Hosokawa and Tomomitsu Miyoshi and Hajime Sawai and Takuya Kamimura and Yasushi Yagi and Yuko Mizuno-Matsumoto and Yen-Wei Chen},
  title = {{Spike Code Flow in Cultured Neuronal Networks}},
  journal = {Comput. Intell. Neurosci.},
  year = {2016},
  doi = {10.1155/2016/7267691},
}
Figure