Machine-learning Approach Reveals Behavioral Fingerprints in Epileptic Zebrafish
Abstract number :
3.282
Submission category :
3. Neurophysiology / 3F. Animal Studies
Year :
2024
Submission ID :
196
Source :
www.aesnet.org
Presentation date :
12/9/2024 12:00:00 AM
Published date :
Authors :
Presenting Author: Anjelica Vance, MS, BS – University of California San Francisco
Paige Whyte-Fagundes, PhD, MSc – University of California San Francisco
Scott Baraban, PhD – UCSF
Rationale: Epilepsy research and therapy development is traditionally based on recognizing distinct convulsive behaviors in preclinical animal models. This is commonly achieved in rodents using labor-intensive and highly subjective scoring of observed behaviors. Zebrafish (Danio rerio) are a valuable alternative preclinical epilepsy model, with a high degree of genetic homology to humans (Howe et al., 2013). To model a spectrum of genetic epilepsies, we recently generated 40 mutant zebrafish lines using CRISPR technology (Griffin, et al. Commun. Biol. 2021). Zebrafish models have proven extremely valuable in translational epilepsy research and facilitated ‘aquarium-to-bedside’ drug discoveries. To progress this model further, we explored the transformative potential of machine-learning to phenotype zebrafish representing seven rare genetic epilepsies e.g., ARX, GABRB3, PCDH19, PNPO, SCN1A, SCN8A, and SYNGAP1B.
Methods: Adult heterozygote F8+ generation CRISPR-Cas9 edited zebrafish (arxa, gabrb3, pcdh19, pnpo, scn1a, scn8a, syngap1b) were crossed to generate larvae. Larvae were plated individually in Falcon 96 well-plates and habituated for 30 min. Wells contained 150 µl embryo media or media containing 15 mM pentylenetetrazole (PTZ). A Ramona Optics Multi-Camera Array Microscope (MCAM™) with 24 high-resolution CMOS cameras was used for video acquisition at 160 frames per second (10 min). All larvae were genotyped post hoc.
Results: Eight-point skeletal tracking revealed consistent changes in high-speed ( > 100mm/s) swim movements and tail angle metrics for PTZ exposed zebrafish but not age-matched controls at 3, 5 and 7 days post-fertilization, dpf (n = 1632 fish). Custom supervised machine-learning (ML) algorithms also identified inter-eye distance as a reliable unbiased metric for quantification of convulsive seizure episodes. Next, data was acquired in a blinded manner from WT, heterozygote and homozygote larvae for all 7 CRISPR-generated zebrafish lines at 5 dpf. Metrics including total activity, distance moved, tail and head angles, and inter-eye distance were measured for >1000 larvae. ML algorithms revealed a spectrum of behaviors ranging from hypoactivity, evident through ataxia, to hyperactivity and convulsive episodes, which was recorded through excessive and often high velocity swimming and tail angle deflections.
Conclusions: Our results establish a new ‘gold standard’ for unbiased identification of complex seizure behaviors in larval zebrafish. This approach significantly advances deep computational phenotyping of zebrafish models for epilepsy, autism and other neurological disorders at a scale and precision not previously possible.
Funding: NIH/NINDS grants R01-NS096976 and R21-NS138525 (to S.C.B.)
Neurophysiology