Intelligence in a Dish: Developing Computational Phenotypes for Human Cortical Organoid Models of Neurodevelopmental Disorders

Human cortical organoids are generated from stem cells, forming a brain-like tissue in the laboratory dish. Prior work has shown that these organoids follow the developmental milestones of the foetal brain and recapitulate the human brain’s cell type diversity, cortical layering and neuronal network activity. When grown from patient-specific induced pluripotent stem cells (iPSC), cortical organoids offer a unique in vitro model for mechanistic and therapeutic studies of neurodevelopmental disorders. In addition, organoids hold promise for personalised screening of different drugs and drug combinations. However, such approaches depend critically on the availability of in vitro cellular-scale phenotypes of cognitive processing, which are currently lacking.

In this project we propose a new class of neuroinformatic phenotypes which directly measure the computational properties of human cortical organoids, thereby promising to unlock their therapeutic potential. In particular, we piloted this approach in a human organoid model of Rett Syndrome - a severe neurodevelopmental disease marked by a regression in cognitive and behavioural functions that becomes apparent in infants during the first year of life. In over 95% of cases Rett Syndrome is caused by mutations in a single gene, called MECP2. However, despite the known genetic cause, there is currently no treatment which is able to prevent or slow the cognitive decline in development.

In this project, we generated human cortical organoids from patient-derived iPSCs. To test their computational properties, we recorded neuronal communication using an array of  microelectrodes in the organoids.  With these microelectrode arrays (MEAs), we could deliver input stimuli and read output activity of the organoid. We then used this system to perform simple cognitive tasks, training a basic machine-learning layer on the organoids outputs. We observed above-chance average performance - confirming that the tissue was processing and storing task-relevant information. We also observed a trend for better performance in the control than the MeCP2-deficient organoids. This is in line with our hypotheses that (i) the neuronal network architecture and dynamics in healthy tissue are able to support better computational performance than brain tissue in Rett Syndrome, and (ii) that the organoid-MEA framework we developed is able to capture and quantify these differences in task-performance. Taken together, these results provide a first step towards neuroinformatic phenotyping of biological tissue. 

The CIND grant was instrumental in pulling together a multidisciplinary team with the required experience and track-record in areas as diverse as organoid biology, neurodevelopmental disorders, and computational neuroscience. It also allowed us to develop a promising proof of principle, and we believe we are now in an excellent position to apply for larger grants.

by Prof. Petra Vertes