BrainSTEM: A multi-resolution fetal brain atlas to assess the fidelity of human midbrain cultures

Many midbrain dopaminergic neuron (mDA) differentiation protocols aimed at Parkinson’s disease (PD) modeling and cell replacement therapy have been developed. However, comprehensive evaluations of the transcriptomic fidelity of these protocols at the single-cell level against a common in vivo reference have been lacking. To this end, we constructed an integrated human fetal whole-brain atlas and a midbrain subatlas to use as a standard of comparison. From the whole-brain atlas, we observed distinct brain-region-specific gene expression in most neural cell types, emphasizing the need to first evaluate in vitro protocols at the whole-brain level to identify midbrain-associated cells. These cells are then mapped to the midbrain subatlas for more refined neuronal subtype specification and trajectory analysis specific to the midbrain. We surveyed all publicly available single-cell datasets of human midbrain culture models and performed the two-tier mapping. Using this biologically-driven multi-resolution mapping strategy which we termed BrainSTEM (brain Single-cell Two tiEr Mapping), we confirmed the presence of multiple midbrain cell types (‘on-target’), but also a substantial proportion of cells associated with non-midbrain regions and subtypes (‘off-target’). This leads to an overall ‘inflation’ of mDA presence, stemming from non-midbrain-associated cells, across all published protocols. BrainSTEM thus offers an unbiased framework for understanding the current state of midbrain models and aids the improvement of midbrain differentiation protocols for PD studies.