Use of a Neural Circuit Probe to Validate in silico Predictions of Inhibitory Connections

Honglei Liu, Daniel Bridges, Connor Randall, Sara A. Solla, Bian Wu, Paul K. Hansma, Xifeng Yan, Kenneth S. Kosik, Kristofer E. BouchardbioRxiv 2017 · DOI

TL;DR

This paper introduces a novel computational and experimental pipeline to accurately identify inhibitory connections within neural circuits. By combining in silico simulations with a custom-built Neural Circuit Probe (NCP) for targeted drug delivery, the researchers can validate predicted inhibitory interactions, paving the way for more complete mapping of neural circuits.

What Problem Does This Paper Address?

Understanding how information flows through neural networks is fundamental to neuroscience. While identifying excitatory connections between neurons is relatively straightforward by analyzing spike train correlations, pinpointing inhibitory connections is significantly more challenging. This difficulty arises because low firing rates can be due to natural biological processes or active inhibition, making it hard to distinguish between them. Existing methods struggle to reliably identify these crucial inhibitory links, which are essential for maintaining network stability and enabling complex computations.

Methods

The core of this work lies in a two-pronged approach:

  • In Silico Interventional Experiments: The researchers first model the spiking activity of neurons using a Generalized Linear Model (GLM). This model is trained on spontaneous firing data. Then, they simulate "interventions" within this model, effectively silencing or activating individual neurons. By observing the predicted downstream effects on other neurons in the simulation, they can identify potential inhibitory connections. This is done by looking for neurons that show a negative correlation in their activity when a specific neuron is "intervened" upon.
  • Neural Circuit Probe (NCP): To experimentally validate these in silico predictions, the team developed a custom Neural Circuit Probe (NCP). This device allows for highly localized and transient delivery of drugs (like Tetrodotoxin, TTX, to silence neurons) to individual, identified neurons on a multi-electrode array (MEA). This precise control enables them to mimic the in silico interventions and observe the real-world impact on neuronal activity.

The GLM itself uses a history window of spikes and is enhanced with cosine basis functions to smooth the inferred interaction filters. An L1 regularization term is added to prevent overfitting.

Datasets

The study utilized recordings of spontaneous neuronal activity from cultured mouse hippocampal neurons. These recordings were obtained using Multi-electrode Arrays (MEAs). Specific details on the number of units and duration of recordings are provided for different experiments:

  • A 20-second recording from 17 units was used for the first GLM fitting and in silico experiments.
  • A 60-second recording from 120 electrodes was used for subsequent GLM fitting and in silico experiments.

The paper does not introduce a new benchmark dataset but rather uses these recordings for both model training and validation.

Key Insights and Ideas

  • Bridging the In Silico/In Vivo Gap: The most significant insight is the successful integration of computational modeling with precise experimental intervention. The in silico approach provides testable hypotheses, and the NCP provides the means to rigorously test them in a biologically relevant context.
  • The Power of Simulated Interventions: The paper demonstrates that simulating the silencing or activation of neurons can reveal inhibitory connections that might not be obvious from analyzing spontaneous firing patterns alone. This is particularly true for indirect inhibitory effects.
  • Targeted Drug Delivery is Key: The NCP's ability to deliver drugs to single neurons with high spatial and temporal resolution is crucial. This precision allows for direct validation of predicted interactions without widespread, confounding effects.
  • Inhibitory Connections are Complex: The study highlights that inhibitory effects can be indirect, mediated through other neurons, making their identification even more challenging. The in silico approach is shown to be effective in uncovering these indirect pathways.

Key Results

  • The researchers successfully identified three putative inhibitory connections using their in silico method.
  • These three predicted inhibitory connections were experimentally validated using the Neural Circuit Probe (NCP) and TTX delivery. For example, silencing neuron 10 led to a rebound in the activity of neuron 8, confirming an inhibitory link.
  • The study showed that the in silico experiments could accurately predict inhibitory connections, even those that were not readily apparent from the GLM filters alone.
  • The NCP's ability to deliver drugs locally and reversibly was demonstrated, allowing for transient silencing of neurons and observation of subsequent network responses.

Limitations and Future Work

  • GLM Assumptions: The GLM assumes a Poisson process for spike generation, which is a simplification of biological reality. The paper acknowledges that GLM-based analysis is not as accurate as intracellular recordings but offers significant advantages in scalability and ease of use.
  • Indirect Connections: While the paper explores indirect inhibitory connections, fully mapping all such pathways in a complex network remains a significant challenge. The current work focuses on validating direct and some indirect predictions.
  • Future Work: The authors suggest that their approach provides a foundation for mapping complete neural circuits. Future work could involve applying this pipeline to larger and more complex neural systems, exploring different types of interventions, and further refining the computational models to capture more nuanced neuronal dynamics. The development of the NCP also opens doors for other applications requiring precise neuromodulation.

Summary updated

Abstract

Abstract Understanding how neuronal signals propagate in local network is an important step in understanding information processing. As a result, spike trains recorded with Multi-electrode Arrays (MEAs) have been widely used to study behaviors of neural connections. Studying the dynamics of neuronal networks requires the identification of both excitatory and inhibitory connections. The detection of excitatory relationships can robustly be inferred by characterizing the statistical relationships of neural spike trains. However, the identification of inhibitory relationships is more difficult: distinguishing endogenous low firing rates from active inhibition is not obvious. In this paper, we propose an in silico interventional procedure that makes predictions about the effect of stimulating or inhibiting single neurons on other neurons, and thereby gives the ability to accurately identify inhibitory causal relationships. To experimentally test these predictions, we have developed a Neural Circuit Probe (NCP) that delivers drugs transiently and reversibly on individually identified neurons to assess their contributions to the neural circuit behavior. With the help of NCP, three inhibitory connections identified by our in silico modeling were validated through real interventional experiments. Together, these methods provide a basis for mapping complete neural circuits.
BibTeX
@misc{liu2017use,
  author = {Honglei Liu and Daniel Bridges and Connor Randall and Sara A. Solla and Bian Wu and Paul K. Hansma and Xifeng Yan and Kenneth S. Kosik and Kristofer E. Bouchard},
  title = {{Use of a Neural Circuit Probe to Validate in silico Predictions of Inhibitory Connections}},
  year = {2017},
  doi = {10.1101/204594},
  howpublished = {bioRxiv},
}
Figure