An in vivo and in vitro spatiotemporal profile of human midbrain development

Abstract

The dopaminergic system has key roles in human physiology and is implicated in a broad range of neurological and neuropsychiatric conditions that are increasingly investigated using induced pluripotent stem cell-derived midbrain models. To determine similarities of such models to human systems, here we undertake single-cell and spatial profiling of first and second trimester fetal midbrain and compare it to in vitro midbrain models. Histological examination reveals that, by the second trimester, fetal midbrain tissue exhibits structural complexity comparable to that of adults. At the molecular level, single-cell profiling uncovers differences in cellular composition across models, with brain organoids most closely resembling late first trimester tissue - an observation supported by meta-integration of existing midbrain datasets. By reconstructing developmental trajectories of neuronal and astrocytic lineages, we map gene expression dynamics associated with maturation. Importantly, integration of spatial transcriptomics provides critical context for aligning organoid models, revealing that their spatial organization and intercellular signaling resemble the architecture and microenvironment of the second trimester midbrain. Ultimately, we leverage our findings to study Dopamine Transporter Deficiency Syndrome progression in patient-derived midbrain organoids, validating their relevance. Understanding the extent of human tissue recapitulation in midbrain laboratory models is essential to justify their use as biological proxies.

Fig. 1. Morphological changes of the ventral human midbrain during the second trimester.

A Experimental design indicating age of fetal samples analysed. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/f8y0e0d. B–D Immunofluorescence analysis of fetal midbrain tissue coronal sections collected at 12 (B), 19 (C) and 22 (D) PCW for TH and GFAP (upper panel). Nuclei are stained with DAPI. Scale bars upper panel = 300 μm (B) and 500 μm (C, D). Graphical representation of the ventral midbrain coronal section and TH positive cells radial and tangential migration (middle panel). Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/yci1tqt. Higher magnification of radial and tangential migrating TH+ and GFAP+ neural cells at 12 (B), 19 (C) and 22 (D) PCW (lower panel). Dotted lines demarcate zoomed radial (1) and tangential (2) neuronal migration regions. Cerebral aqueduct (aq), dorsal (D), ventral (V), red nucleus (RN), substantia nigra. Scale bars lower panel= 50 μm.

Fig. 2. In vitro derived 2D mDA neurons and 3D MLO models cellular composition.

A Experimental design for the in vitro system analysis. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/culevfq. B–I Representative images of 2D mDA neuronal culture and MLO at 40 days of differentiation. Cultures were stained for the midbrain-related and neuronal proteins FOXA2, LMX1A, CORIN, OTX2, TH, MAP2, and proliferative marker Ki67. J–N Representative images of 2D mDA neuronal culture and MLOs at 70 days of differentiation showing expression of TH, MAP2, AADC and GABA. O–R Immunofluorescence analysis of MLOs at 120 days of differentiations. MLOs were stained for TH, NEUN, AADC, GABA and synaptophysin (SYP). Nuclei are staining with DAPI. Scale bars = 100 μm (left panel) and 50 μm (right panels). S Bright field image visualizing melanin aggregates. Scale bar = 2 mm. T Fontana-Masson staining and haematoxylin and eosin (H&E) stain of MLO at 120 days of differentiation showing melanin precipitation. Scale bar = 50 μm. Necrotic core (nc), Matrigel embedding (me).

Fig. 3. Single-cell RNA-seq profiling of in vitro and in vivo fetal samples.

A Scheme of the in vitro (2D, 3D) and fetal samples profiled with droplet-based scRNA-seq. Only control samples were considered for all models: 2 iPSC lines from pediatric controls were differentiated to a midbrain dopaminergic fate, either as a 2D culture (days 40, 70) or 3D organoids (days 40, 70, 120). As for fetal samples, the midbrain sections of four donors were profiled at single timepoints: 10 and 12 PCW for the first trimester, and 16 and 20 PCW for the second trimester. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/pt104zy. B UMAP visualization based on gene expression colored by in-vitro (2D and 3D) or in-vivo dopaminergic models (fetal tissue). C UMAP visualization (as in b) with cells grouped in 10 broad categories (low-resolution annotation), encompassing the 24 cell types defined by the high-resolution annotation. The largest cell clusters correspond either to neuronal cell population (top-left) or to radial glia/floor-plate progenitors (bottom-right). D Upper: cell type composition per model and time point. Lower: number of cells profiled in each corresponding group to estimate cell type composition. E, F Differential abundance (DA) analysis comparing cell types between 3D vs 2D in-vitro models (E), and between fetal samples and the 3D model (F). Each point represents a neighborhood (group of cells connected by an edge in the KNN graph), with coloured points indicating significant DA (SpatialFDR < 0.1). Only in (E) the sampling timepoint is included as a covariate. G The mitochondrial respirasome is commonly upregulated in 3D versus 2D models across neuronal populations (hDA1a, hDA1b and hDA2) indicating a prominent role of mitochondria metabolism in the maturation of dopaminergic neurons. Differential gene expression between 3D and 2D cells was calculated within the same cell type (neighborhood group), accounting for timepoints as a covariate (adj P.val<0.05, FC > 1.75, hypergeometric test for over-representation analysis implemented in gprofiler2 (gost)). H Fetal hDA2 cells show an upregulation of synaptic-related pathways when compared with the 3D hDA2 cells (only the top-5 biological processes of the gene ontology enrichment are shown, pval<0.05, minSize=25, hypergeometric test for over-representation analysis implemented in the gprofiler2 R package). I Force-directed graph (FDG) representation of inferred trajectories for astrocytes (top) and neurons (bottom). Cells highlighted in yellow belong to the inferred trajectory, and the arrow indicates the direction of increasing pseudotime. All cells shown in B, C and I were profiled based on experimental design available in Supp Fig. 10A. J PC1 loadings of highly variable genes defined along the neuronal lineage. Genes whose expression correlates with neuronal maturation or a progenitor-like state are shown in green and red, respectively. K Smoothed expression profiles of a curated set of neuronal marker genes (same as in J), across five evenly spaced pseudotime bins. Asterisks indicate significant start-to-end gene expression changes. L Immunofluorescence analysis of fetal midbrain at 19 and 22 PCW and MLOs at 70 and 120 days of differentiation for TH and the potassium channel GIRK2. Arrowheads indicate colocalization of GIRK2 and TH either in the soma or in the neurites. Nuclei are stained with DAPI. Scale bar= 50 μm.

Fig. 4. Meta-integration of multiple dopaminergic scRNA-seq datasets and analysis of pseudotemporal dynamics.

A UMAP visualization based on the integrated gene expression (MNN correction) of our dataset with in-vitro 3D dopaminergic models (Fiorenzano et al. 2021, Zagare et al. 2022), in-vivo fetal data (La Manno et al. 2016, Birtele et al. 2022 and Braun et al. 2023), and adult post-mortem samples (Zagare et al. 2022). The final integrated dataset contains 194,521 cells. B Cell type composition per dataset and time point, expressed in percentage. For comparability between datasets, a unified cell type annotation (17 identities) has been defined (legend on 3rd row). Abbrev: floor plate progenitors (FPP), intermediate progenitor cells (IPC), oligodendrocytes (ODC), oligodendrocyte progenitor cells (OPC) and vascular leptomeningeal cells (VLMC). C, D UMAP visualization from (a) colored by either cell types (unified annotation) or just dopaminergic neurons, respectively. E UMAP visualization from (a) colored by dataset and timepoint. Only cells collected from this work and fetal samples are highlighted. F Transcriptomic similarity (Pearson’s correlation of MNN corrected gene expression, n = 2000 features) between fetal gestational timepoints and in vitro 3D dopaminergic differentiation timepoints. Abbrev: gestational week (GW), post-conceptional week (PCW). G Ridge density plot of pseudotime distribution for neurons, colored per dataset and displayed per timepoint (each row represents a different timepoint). Differences in pseudotime distribution were statistically assessed across six within-dataset comparisons (This work: 2D-day 40 vs 2D-day 70; 3D-day40 vs 3D-day120; fetal-PCW10 vs fetal-PCW20; Birtele: fetal-PCW6 vs fetal-PCW11; Braun: fetal-PCW8 vs fetal-PCW14; and Fiorenzano: 3D-day15 vs 3D-day120). Significance was evaluated using a two-sided Wilcoxon test (all comparisons showed p < 2.22·10⁻¹⁶, indicated by ****). H Ridge density plot of the pseudotime distribution for dopaminergic neurons subtypes annotated in this study. The plot is colored by subtype and displayed per timepoint (each row represents a different timepoint), shown only when enough cells ( > 50) are available per combination. Differences in pseudotime distribution were statistically assessed across three comparisons (hDA1b: 2D-day40 vs 2d-day70; hDA1b: 3D-day40 vs 3d-day120; and hDA2: 2D-day40 vs 2D-day70). Significance was evaluated using a two-sided Wilcoxon test (all comparisons displayed p < 2.22·10⁻¹⁶, indicated by ****). I, J Split violin plots showing activation scores (aggregated gene expression) for the five neuron-specific pseudotime-dependent gene modules, comparing fetal neurons from this study with those from Birtele et al. 2022 (I) or from Braun et al. 2023 (J), respectively. All comparisons displayed p < 10⁻³ (***), except for Module 12 in (I) with p = 0.418. Significance was assessed using a two-sided one-way permutation-test (10,000 label permutations per test, mean p-value for 100 resamplings reported). Violin plots display the full data distribution per group, with embedded boxplots showing the median (center line in black), the interquartile range (box, 25th–75th percentiles) and whiskers extending to 1.5x interquartile range. K Activation score distribution at single-cell level for the neuron-specific module 12, colored by dataset and displayed per timepoint (each row represents a different timepoint). The plot shows mean ± standard deviation across individual neurons (n = 33,504), as defined by the unified annotation.

Fig. 5. Spatial distribution of midbrain cell types across development.

A Experimental design for the spatial transcriptomics. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/f0bwfhn. B Haematoxylin and eosin (H&E) stain of human fetal samples at 7, 11 and 17 PCW and MLO samples at 40, 70 and 120 days of differentiation. Scale bars = 300 µm (upper panel); 500 µm (lower panel). Ventricular zone (VZ), cerebral aqueduct (aq), dorsal (D), ventral (V), necrotic core (nc). C Abundance of hRgl1 and hDA2 cells in tissues and organoids, as estimated by the Cell2Location deconvolution algorithm. Expression of two marker genes for each cell type are displayed below. D Manually annotated maturation paths on tissues and organoids. E Moving averages of the expression of genes that are differentially expressed with respect to maturation paths, calculated over a window of 50 spots along the path of maturation. F Progenitor and post-mitotic cell populations in tissues and organoids, defined by clusters annotated according to anatomical features and marker genes (displayed below). G  Approximations of spatial locations of RNA molecules associated with midbrain patterning and specification.

Fig. 6. Spatially defined maturation of dopaminergic neurons at 17 PCW.

A Immunofluorescence analysis of 17 PCW human ventral midbrain for TH. Squares indicate different anatomical regions of TH expression across the ventral midbrain, including the midline and lateral areas. Nuclei are stained with DAPI. Scale bars = 400 µm (left panel); 50 µm (right panel). B Regions associated with TH expression in 17 PCW tissue, overlaid over expression of markers of ventral midbrain dopaminergic subtypes CALB1 and ALDH1A1. C Expression of genes associated different TH regions. D  Expression of genes associated with dopaminergic neurons maturation and specification shown across spots positive for TH, CALB1 or ALDH1A1, split by the TH regions shown in (B). Each y axis starts at 0.

Fig. 7. Fetal and MLO midbrain systems share spatial cell-cell communication pathways and spatial characteristics.

A Expression in 11 PCW tissue of the PTN-PTPRZ1 ligand-receptor pair, identified by COMMOT to be highly active and shown by Moran’s I to be spatially correlated. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/3akidui. B Expression in organoids of the PTN-PTPRZ1 ligand receptor pair. C Immunofluorescence analysis of 11 PCW human fetal coronal section for PTPRZ1 and SOX2. Nuclei are stained with DAPI. Scale bar = 100 µm. Cerebral aqueduct (aq). Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/gfxtdrp. D Schematic of similarity quantification approach: cell types of neighboring spots are counted and used to compute similarity between samples. E Quantification of spatial similarity between tissues and organoids. F Temporal alignment of organoids and tissues, where the timepoints of the tissues are averaged and weighted according to the organoid’s spatial similarity to each one.

Fig. 8. MLOs derived from DTDS iPSCs allow investigation of disease progression in dopaminergic neurons.

A Experimental design for the analysis of DTDS-derived MLOs at different maturation time points (40, 70, 120 days of differentiation) to investigate progressive neurodegeneration. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/wvn03ts. B Immunofluorescence analysis for the neuronal markers TH and MAP2 at d40 and d70 of differentiation in Control 1, Control 2, Patient 1, Patient 2 and CRISPR isogenic-derived MLOs. Nuclei were counterstained with DAPI. Scale bar= 50 µm. C, D Quantification of TH positive and TH/MAP2 double-positive neurons for controls and patient lines (n = 4 for every line). Error bars indicate SEM. DTDS lines were independently compared to controls using a two-tailed Student’s t-test for all analyses. Statistically significant differences indicate comparison of Control 1 or 2 to Patient 1 or 2 and Patient 2 to CRISPR. E, Experimental design for the differentially expressed genes (DEGs) analysis in mDA clusters. Created in BioRender. BUDINGER, D. (2025) https://BioRender.com/oo6rdkc. F, G Gene ontology enrichment analysis (GO) on cell compartments based on the DEGs in neuronal lineage and hRgl1 clusters, respectively, in patients compared to controls. H, I List of genes contributing to GO enrichment in hDA1 and hRgl1 clusters, respectively. Error bars indicate SEM. DTDS lines were independently compared to controls using a two-tailed Student’s t-test for all analyses. Statistically significant differences indicate comparison of Control 1 or 2 to Patient 1 or 2 and Patient 2 to CRISPR. Necrotic core (nc), Matrigel embedding (me). Source data are provided as a Source Data file.