Integration of single-cell transcriptomic atlases of the adult human midbrain

Abstract

The cellular composition of the human brain has shown to be more heterogeneous than previously understood. The ventral midbrain is no exception. Advances in single-cell technologies has enable the creation of high-resolution transcriptomic datasets. However, current atlases lack representation of rare cell types, obscuring their distinction from noise in subsequent analyses. In this thesis, we present an integrated midbrain reference atlas that combines data from two midbrain atlases, three substantia nigra (SN) datasets and one ventral tegmental area (VTA) atlas. The additional data enhances the sensitivity of the dataset, allowing for more refined subtyping of the (ventral) midbrain. Additionally, we optimized the integration pipeline with harmonized realignments of the original snRNA reads, benchmarks of integration methods such as scVI, Harmony and Scanorama using batch correction metrics, and hyperparameter tuning on scVI. By investigating the hierarchical relationships between previously identified subtypes, we found agreements across datasets. Notably, we provided evidence for VTA-related combinatorial subtypes that express both gamma-aminobutyric acid (GABA) and dopamine neurotransmitters. Our findings support the existence of specialized subtypes that can be efficiently discerned through region specific atlases. We anticipate this reference genome to serve as a cornerstone for future research on the human midbrain and to expand over time with the inclusion of additional atlases.