We have obtained hippocampal organoids from rodent neural stem cells (NSCs), devised a minimally invasive closed-skull neurosurgery technique and formulated an injectable fluid pre-tissue combining NSCs and a novel alginate-based scaffold. Although organoids could not achieve robust electrical functionality, observations supported the possible risk of epileptiform behavior. Three-dimensional cellular constructs from primary hippocampal cells exhibited different behavior in the absence and presence of alginate. Hippocampal spheroids (no alginate) unveiled an intrinsic epileptogenicity, whereas alginate-based hippocampal hydrogels did not. The spontaneous epileptogenicity of hippocampal spheroids pictured a tangible risk from a translational perspective. Therefore, while in vivo research proceeded toward in situ tissue formation, in vitro studies retained spheroids (hereafter, model graft) to further characterize their interaction with an established ex vivo model of MTLE (hereafter, model host).
In the absence of artificial counterparts, the functional outcome of bidirectional closed-loop stimulation between model host and model graft proved influenced by the electrical phenotype of the latter and corroborated the requirement of a controlled dialogue. In this direction, we obtained a self-governing biohybrid construct, wherein the model graft is functionally coupled with a neuromorphic computing system (NCS). For this, we evaluated a hybrid CMOS-neuron/memristive-synapse (MXBAR) system, MoS2-based charge-trap memory devices, memristor computation algorithms and memristorless designs, ultimately retaining the latter.
An analog CMOS-based NCS featuring online learning was tested for closed-loop stimulation in epileptic rodents, demonstrating decrease of seizure severity and modulation of epileptogenic network dynamics. This was the final achievement of the in vivo work since at last we could not reach hippocampal regeneration despite the diverse strategies deployed.
A digital pretrained NCS was validated for closed-loop stimulation of hippocampal spheroids, yielding a biohybrid construct. As additional layer of supervision, we devised a multi-agent AI system, wherein agent0 operates seizure forecasting, while agent1 adaptively tunes the NCS during its interaction with the model graft. Agent0 was first validated for its capability of adaptive electrical stimulation to prevent ictal activity in the model host. Subsequently, AI and NCS converged to the validation of the AI-tuned biohybrid construct.
At last, we deployed the fully fledged biohybrid system in vitro. This work evidenced the behavior of the NCS as primary driver of the graft-host interplay, while the model graft coupled to it served as biological probe and biohybrid modulator of the ongoing dynamics of the model host. The AI system proved its character of super partes teacher and coordinator of the biohybrid interplay through dynamic NCS tuning based on compound seizure forecasts from graft and host during their bidirectional interaction, guiding the biohybrid restoration of seizure-prone brain networks through dynamic NCS tuning.
