Abstracts

Machine-learning Enables High-throughput, Low-replicate Reverse Genetic Screen for Novel Anti-seizure Targets in Larval Zebrafish

Abstract number : 1.115
Submission category : 2. Translational Research / 2D. Models
Year : 2023
Submission ID : 764
Source : www.aesnet.org
Presentation date : 12/2/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: Christopher McGraw, MD, PhD – Mass General Hospital, Boston Childrens, HMS

Barbara Robens, PhD – Boston Childrens Hospital; Cristina Baker, B. Sc. – Boston Childrens Hospital; Guoqi Zhang, MD – Boston Childrens Hospital; Christopher Lacoursiere, MS – Boston Childrens Hospital; Annapurna Poduri, MD, MPH – Neurology – Boston Childrens Hospital

Rationale:

New anti-seizure medications (ASMs) with increased efficacy and tolerability are urgently needed, but there have been few systematic efforts to identify new anti-seizure targets. We hypothesize that gene knock-outs that confer resistance to proconvulsants by loss-of-function in living organisms may suggest candidate targets for novel ASM development. Zebrafish are an established model of chemical and genetic seizures. To facilitate anti-seizure screens in zebrafish, we developed a method using machine learning to detect anti-seizure responses using calcium fluorescence. To identify novel anti-seizure targets, we propose a reverse genetic screen using the recently published MIC-Drop approach to deliver multiple sgRNA and Cas9 RNPs in oil droplets from a library targeting presynaptic genes.



Methods:

Calcium fluorescence from unrestrained larval zebrafish expressing a neuronal genetically encoded calcium indicator (elavl3::Gcamp6s) was recorded in 96-well plates using the Hamamatsu FDSS7000EX fluorescent plate reader (“FDSS”). MATLAB was used to extract multiple per-fish and per-event statistics including fish movement and calcium fluorescence changes. An elastic net logistic classifier was trained on N >4000 events from N=63 fish before and after PTZ treatment (15mM) using 70:30 train:test split and 10-fold cross-validation. Reverse genetic screen. Experimentally confirmed “presynaptic” genes with high human CNS expression / enrichment and known zebrafish orthologs are targeted by 4 sgRNA per gene (N=1195 sgRNA total; 310 genes). Injected F0 CRISPant knock-out fish (N=12 per gene) are assessed for morphology; locomotor activity; and seizure-like activity (spontaneous and proconvulsant-induced). Positive hits showing resistance to proconvulsant without other phenotypes are prioritized for further characterization, and genes of interest identified by barcode sequencing.



Results:

The logistic classifier detects seizure-like events with high accuracy (AUC-ROC 0.98, AUC-PRG 0.85). The rate of seizure-like events increases as a dose-response to proconvulsant (PTZ) and reduces as a dose-response to anti-seizure drug treatment duration and dose. Bootstrap simulation (5000 resamples) shows that the anti-seizure response can be detected with N=8 replicates based on robust strictly standardized mean difference (RSSMD) thresholds (brief VPA, RSSMD < = -0.82, TPR 92.2%; prolonged TGB, RSSMD< = -0.76, TPR 91.2%) while maintaining 5% false positive rate. In a test screen, the anti-seizure effect of four out of five known ASMs was detected at a single concentration (200uM) based on N=4-6 replicates.

Translational Research