Two-tiered atlas of the developing human brain offers new hope in Parkinson’s treatment
Parkinson’s disease robs individuals of their ease and movement by targeting a narrow band of midbrain neurons. For those in the field of neuroscience, the quest to grow these dopaminergic cells in dishes has been a source of both promise and frustration. Results often fluctuate, identities blur, and unwanted cell types infiltrate the culture. Now, a groundbreaking study provides a clearer path forward.
Scientists from Duke-NUS Medical School have developed a two-tier single-cell atlas of the developing human brain and a tool to assess the purity of lab-grown cells. This approach offers a practical evaluation system for models and a roadmap towards more effective therapies in the future.
Two Atlases, One Goal
The research team initially constructed a comprehensive fetal whole-brain atlas, followed by a high-detail map specifically focusing on the midbrain. They integrated two large human datasets, profiling cells from weeks 3 to 14 after conception, and standardized labels and markers across 679,666 cells from 39 donors. Neurons, neural progenitors, and non-neural classes were meticulously defined, with a particular emphasis on transmitter-based groups within neurons, such as dopaminergic, glutamatergic, GABAergic, serotonergic, and cholinergic.
To evaluate the accuracy of cell identities in representing their origin, the researchers employed a k-nearest neighbor classifier. Midbrain dopaminergic neurons scored an impressive 0.92 for region specificity, while forebrain glutamatergic neurons achieved a perfect score of 1.0. Interestingly, some groups exhibited weaker regional ties, like forebrain serotonergic neurons at 0.08, indicating that location alone should not be solely relied upon to determine cell type.
The team also identified 21 midbrain markers for dopaminergic identity, including EN1 and PITX3, as well as less familiar genes like ELOVL3 and MYRIP.
A Sharper Look at the Fetal Midbrain
The midbrain subatlas, comprising 102,335 cells across weeks 3 to 14, provided an independent dataset for the ventral midbrain, expanding coverage. Six progenitor subtypes were mapped to downstream lineages, encompassing dopaminergic, red nucleus, ventral and dorsal GABAergic, glutamatergic, and a mixed dopaminergic or subthalamic nucleus fate.
Trajectory analysis of ventral cells revealed four distinct branches, each leading to specific neuronal types: dopaminergic neurons, red nucleus neurons, ventral GABAergic neurons, and a hybrid dopaminergic or subthalamic lineage. Classical floor plate genes, such as LMX1A and SOX6, were instrumental in tracing dopaminergic precursors, while TH and KCNJ6 marked the maturation of dopaminergic cells.
A notable group emerged, labeled as hDA.STN, with dopaminergic-like characteristics and subthalamic features. These cells could be mistaken for the target cell type in culture. PITX2 marked these subthalamic-related cells, and the team demonstrated that hDA cells enhance axon growth and dopamine metabolism, while hDA.STN cells are more inclined towards protein translation and folding. Signaling cues also differed, with EPHA and Netrin pathways specific to hDA.
BrainSTEM: Two-Step Mapping and Measurement
With both atlases established, the team introduced BrainSTEM, a mapping framework. This innovative approach first projects any lab dataset onto the whole brain and then focuses on midbrain-like cells for precise classification. By doing so, it prevents the common error of misidentifying cultures as 'midbrain' when they are actually more similar to the forebrain or hindbrain.
The authors curated a vast collection of over 1.4 million cells across 50 conditions from 12 published datasets and one in-house organoid protocol. They compared 2D and 3D systems, tracked time points, and built synthetic mixtures to thoroughly test the method.
In real datasets, the initial mapping tier revealed that off-target forebrain and hindbrain cells often constituted over half of samples labeled as midbrain. The in-house organoids and one 2D protocol yielded the highest proportions of true midbrain cells. Midbrain-like cells typically peaked between days 30 and 40, with organoids reaching around 25 to 35 percent midbrain identity at optimal time points. One 2D protocol sometimes exceeded this, but many midbrain-labeled cells were progenitors rather than mature neurons.
Dopaminergic purity was assessed using a gene signature score, with higher scores correlating with more dopaminergic neurons, as evidenced by Spearman correlations exceeding 0.7. Interestingly, some protocols exhibited rising serotonergic signatures by day 45, indicating excess caudalization that pushes cells towards hindbrain fates.
The Whole Brain Filter
After applying the whole-brain filter, seven datasets with sufficient midbrain-like cells progressed to the second tier. Here, the subatlas meticulously parsed fine-grained progenitor and neuron identities, revealing a diverse range of dopaminergic, ventral GABAergic, and glutamatergic lineages across protocols.
To capture ventral identity, the team trained a random forest, generating a ventral score per cell. Cultures rich in dopaminergic cells tended to have higher ventral scores, aligning with floor plate trajectory. The group then compared each protocol's cell mix with in vivo midbrain at specific gestational weeks. Many datasets matched weeks 6 to 8, while an organoid grown to day 120 correlated with weeks 10 to 11, suggesting that longer culture periods nudge cells towards later fetal stages.
Crucially, a one-step direct mapping inflated dopaminergic counts by including forebrain-like cells. The two-step BrainSTEM approach effectively reduced these false positives, enhancing the reliability of readouts.
Implications for Parkinson’s Research
For those working on Parkinson’s disease, the implications are clear. Cells must not only appear dopaminergic but also possess midbrain identity. BrainSTEM highlights the frequent drift of cultures and provides insights into the necessary adjustments. Across various protocols, conditions that included FGF8 and used lower CHIR doses tended to improve dopaminergic output.
The atlas also draws attention to rare, subthalamic-related cells that can skew results if overlooked. It's important to consider the limitations, as the references cover the first trimester, and predictions become less certain in later forebrain stages after week 17. Mature glia are sparse because they appear later in development. Extending the atlas into the second trimester and beyond would capture maturation and interactions tied to adult disease.
Part of a Larger Brain Mapping Initiative
This groundbreaking work aligns with the U.S. National Institutes of Health’s BRAIN Initiative Cell Atlas Network, an international collaboration aimed at creating a comprehensive atlas of human brain cells. The latest collection of papers in Nature and related journals maps how cell types arise, specialize, and regulate gene expression across various species.
Hongkui Zeng from the Allen Institute emphasized the brain's diverse cell types and their collaborative role in generating behaviors, emotions, and cognition. Aparna Bhaduri, a UCLA neuroscientist, highlighted the enigmatic nature of the developing brain, which is challenging to access due to its complexity and rapid changes.
These studies delve into cell development in the neocortex and hypothalamus, among other regions, and even explore the potential of brain tumors to hijack embryonic programs. The team identified new human cell types and revealed slower, prolonged differentiation in the cortex, reflecting the extended human development process.
The ultimate goal is to inform gene and cell therapies and deepen our understanding of conditions like autism, ADHD, schizophrenia, and others. Zeng stated, 'By studying and comparing brain development in humans and animals, we will gain insights into human specialization and the origins of our unique intelligence. Understanding normal brain development in humans and animals will also enable us to study changes in diseased brains.'
Voices from the BrainSTEM Team
Scientists from Duke-NUS Medical School and their collaborators underscored the clinical significance of their findings. Dr. Hilary Toh emphasized the potential of their data-driven blueprint to produce high-yield midbrain dopaminergic neurons that faithfully reflect human biology. Such grafts are crucial for enhancing cell therapy efficacy and minimizing side effects, paving the way for alternative therapies for Parkinson’s disease.
Dr. John Ouyang noted that BrainSTEM's single-cell resolution mapping enables the precise distinction of even subtle off-target cell populations. Assistant Professor Alfred Sun highlighted the significant step forward in brain modeling, emphasizing the rigorous, data-driven approach that will accelerate the development of reliable cell therapies for Parkinson’s disease.
Professor Patrick Tan hailed the study as a redefined benchmark and emphasized the open access to the atlases and toolkit. The research findings are available online in the journal Science Advances.
