HERMES is a complex polyhedral project, requiring the parallel development of biological and artificial components, and their convergence in the functional biohybrid. The latter was achieved during this third reporting period.
Bioengineered brain tissue graft – We have established a reproducible protocol to obtain bioengineered organoids with hippocampal signature from rodent neural stem cells (NSCs), and developed a bioactive biopolymer-based extracellular matrix (ECM) for in vivo co-injection with NSCs, to support their in vivo differentiation in hippocampal neurons. In this context, we have also established the key turning point for fluid pre-tissue grafting to avoid excessive stemness (hence uncontrolled proliferation), and provide a favorable microenviroment for NSCs differentiation. Further, we have established the minimally invasive closed-skull neurosurgery technique to ablate the sclerotic hippocampal tissue in epileptic rodents along the lines of hippocampal ablation in TLE patients. In parallel, we have developed spheroids and alginate hydrogels of primary hippocampal cells as hippocampal tissue replicas recapitulating the final NCS differentiation in mature grafts. These replicas demonstrated robust and reliable electrical activity and offer the opportunity of testing the biohybrid interplay in two distinct biological scenarios. Namely, the spheroids revealed their ictogenic potential, substantiating the basic assumption of HERMES, i.e. the risk of a purely biological brain regeneration; on the other hand, hydrogels never generate ictal activity, suggesting an anti-ictogenic potential.
Neuromorphic engineering – The neuromorphic computing system (NCS) will guide the functional graft-host integration. To achieve it, we have designed and fabricated the CMOS-neuron chip, the analogue front-end, and the memristor array. For the latter, we have selected materials based on their specific suitability to address the required biologically plausible plasticity. These components have been integrated in a desktop system for in vitro use, for which we have implemented custom communication hardware to interface the NCS with a commercial microelectrode array (MEA) recording system. In addition to the memristor-based NCS (here, MXBAR), we have developed an emulator based on a microcontroller and implementing a model of the MXBAR. We have also developed a new algorithm for memristor-based seizure prediction, termed 'memristor transform'. In addition to the MXBAR system originally foreseen by the project, we have developed new MoS2-based memristor devices for a memristor-only NCS, as well as two memristorless NCS designs (one fully analog CMOS-based, the other fully digital). All these systems have been succesfully tested with MEA recording of epileptiform patterns generated by brain slices or hippocampal spheroids, achieving the recognition of different epileptiform events and, in some instances, seizure prediction.
Aiming at the in vivo setting, we have finalized the flexible probe design and the implantation technique, and we have tested the implanted probes via benchmark tests and in vivo recording and stimulation.
Artificial Intelligence – We have developed the computational models and signal processing tools aiding the AI algorithms design. For the latter, we have achieved a two-AI agent design, wherein agent 1 pursues seizure prediction (prediction agent) and agent 2 monitors and trains the NCS (reinforcement learning control agent).
