Linking Brain and Behavior States in Zebrafish Larvae Locomotion using Hidden Markov Models

Mattéo Dommanget-Kott, Jorge Fernández-de-Cossío, Monica Coraggioso, Volker Bormuth, Rémi Monasson, Georges Debrégeas, Simona CoccobioRxiv 2024 · DOI

TL;DR

This paper introduces a unified Hidden Markov Model (HMM) framework to link zebrafish larval swimming behavior with neural activity in the ARTR brain circuit. By demonstrating that a 3-state HMM can accurately model both datasets and generate realistic synthetic trajectories, the study provides a powerful tool for understanding how neural states translate into self-generated actions.

What Problem Does This Paper Address?

Understanding the intricate relationship between brain activity and behavior is a fundamental challenge in neuroscience. Specifically, how collective neuronal activity orchestrates complex behaviors like locomotion remains an open question. Previous approaches often analyzed behavioral and neural data separately, hindering a unified interpretation. This paper aims to bridge this gap by developing a single modeling framework that can simultaneously capture the dynamics of both zebrafish larval swimming behavior and the neural activity in the Anterior Rhombencephalic Turning Region (ARTR), a key circuit for controlling swimming orientation. The challenge lies in accurately segmenting and modeling these complex, multimodal time-series data, especially given factors like temperature-dependent behavior and individual variability.

Methods

The core of this work is the application of Hidden Markov Models (HMMs). Unlike simpler Markov Chains (MCs) that rely on pre-defined thresholds for classifying behaviors, HMMs infer hidden states directly from the observed data. This allows for a more robust and less biased segmentation of behavioral and neural activity.

Key methodological aspects include:

  • 3-State HMM: Both behavioral (swimming bouts) and neural (ARTR activity) data are modeled using a 3-state HMM. For behavior, these states correspond to forward, left, and right turns. For neural data, they represent distinct patterns of ARTR activity.
  • Non-handedness Constraint: The behavioral HMM incorporates a non-handedness constraint, ensuring symmetry in left and right turning probabilities, reflecting the animal's general lack of a preferred turning direction.
  • Generative Modeling: The HMM's generative nature is exploited to create synthetic behavioral trajectories from the neural HMM. This allows for direct comparison of the statistical properties of generated and real data.
  • Temporal Re-scaling: A crucial step involves finding a temporal re-scaling factor to align the sojourn times of behavioral and neural states, accounting for differences in sampling rates and the inherent dynamics of each system.
  • Stubbornness Factor: A novel metric, the "stubbornness factor," is introduced to quantify behavioral persistence and detect deviations from Markovianity, particularly in distinguishing between correctly and incorrectly labeled bouts.

Datasets

The study utilizes two primary datasets:

  1. Behavioral Data: Video recordings of freely swimming zebrafish larvae at various temperatures (18°C, 22°C, 26°C, 30°C, 33°C). This dataset includes trajectories from multiple fish and a separate dataset of long-duration recordings from individual fish (N=18) at 26°C to assess inter-individual variability.
  2. Neuronal Data: Calcium imaging data of the ARTR circuit in immobilized zebrafish larvae, recorded at different temperatures. This data was previously published and made publicly available.

Key Insights and Ideas

  • Unified Neural Control: The study strongly suggests that the ARTR circuit, previously thought to primarily control left/right turns, also plays a role in selecting between forward and turning bouts. This implies a more integrated neural control of locomotion than previously understood.
  • HMMs Outperform Thresholding: HMMs provide a more accurate and less biased way to segment behavioral bouts compared to traditional threshold-based methods, especially in challenging conditions like temperature variations. This is evidenced by the HMM's ability to capture behavioral persistence more effectively.
  • Behavioral Memory and Temperature: Bout-type persistence decreases with increasing temperature, but this effect is more nuanced than previously thought. HMMs reveal that all bout types exhibit comparable persistence, and temperature primarily affects the amplitude of reorientation, not necessarily the underlying memory dynamics.
  • Individual Fingerprints: The HMM framework can capture individual differences in swimming patterns, allowing for the identification of specific fish based solely on their bout sequences. This opens doors for studying the neural basis of individuality.
  • Bridging Neural and Behavioral Time Scales: The temporal re-scaling factor is a critical insight, demonstrating that neural activity operates on a faster timescale than observable behavioral bouts, and quantifying this relationship is key to linking the two.

Key Results

  • A 3-state HMM accurately models both zebrafish larval locomotion and ARTR neural activity.
  • The HMM-based behavioral labeling shows significantly less temperature dependence than threshold-based methods, suggesting that previous observations of strong temperature effects on turning bout fractions were partly artifacts of the labeling method.
  • The stubbornness factor analysis reveals that HMMs largely resolve the non-Markovianity issues observed with threshold-based labeling, particularly for intermediate forward bouts.
  • The HMM framework successfully distinguishes between individual fish based on their bout sequences, with a high accuracy rate (14 out of 18 fish correctly identified).
  • A temporal re-scaling factor of approximately 0.44 was found to align the sojourn times of behavioral and neural states, indicating that neural dynamics are roughly 2.3 times faster than behavioral bout initiation.
  • Generated synthetic trajectories from the neural HMM closely matched the statistical properties (e.g., bout angle distributions, Mean Square Reorientation) of real behavioral data across different temperatures.

Limitations and Future Work

  • Lack of Direct Multimodal Data: The study acknowledges the absence of simultaneously recorded neural and behavioral data from freely swimming larvae. This limits direct bout-by-bout comparison and necessitates indirect methods for validating the correspondence between neural and behavioral states.
  • ARTR Neuron Interactions: While the HMM models the overall ARTR activity, it assumes independence among neurons within a state. Future work could incorporate recurrent interactions to better model the ARTR circuit's internal dynamics.
  • Experimental Artifacts: The neuronal data was collected from immobilized larvae, which might introduce biases (e.g., phototaxis responses) that could affect the interpretation of neural states.
  • Generalizability: While the HMM framework is powerful, its application to other species or more complex behaviors would require further validation.

The authors suggest that their work provides a foundation for future studies that could leverage more advanced HMM variants or integrate with energy-based models to gain deeper insights into neural network mechanisms. The availability of their code and tutorials aims to facilitate broader adoption of HMMs in behavioral and neuroscience research.

Summary updated

Abstract

Understanding how collective neuronal activity in the brain orchestrates behavior is a central question in integrative neuroscience. Addressing this question requires models that can offer a unified interpretation of multimodal data. In this study, we jointly examine video-recordings of zebrafish larvae freely exploring their environment and calcium imaging of the Anterior Rhombencephalic Turning Region (ARTR) circuit, which is known to control swimming orientation, recorded in vivo under tethered conditions. We show that both behavioral and neural data can be accurately modeled using a Hidden Markov Model (HMM) with three hidden states. In the context of behavior, the hidden states correspond to leftward, rightward, and forward swimming. The HMM robustly captures the key statistical features of the swimming motion, including bout-type persistence and its dependence on bath temperature, while also revealing inter-individual phenotypic variability. For neural data, the three states correspond to left- and right-lateral activation of the ARTR circuit, known to govern the selection of left vs. right reorientation, and a balanced state, which likely corresponds to the behavioral forward state. To further unify the two analysis, we exploit the generative nature of the HMM, using the neural sequences to generate synthetic trajectories whose statistical properties are similar to the behavioral data. Overall, this work demonstrates how state-space models can be used to link neuronal and behavioral data, providing insights into the mechanisms of self-generated action.
BibTeX
@misc{dommangetkott2024linking,
  author = {Mattéo Dommanget-Kott and Jorge Fernández-de-Cossío and Monica Coraggioso and Volker Bormuth and Rémi Monasson and Georges Debrégeas and Simona Cocco},
  title = {{Linking Brain and Behavior States in Zebrafish Larvae Locomotion using Hidden Markov Models}},
  year = {2024},
  doi = {10.1101/2024.11.22.624881},
  howpublished = {bioRxiv},
}
Figure