Fig 1.
Pipeline overview from 3D mapping to linking cell type identities.
The top module shows the workflow for building 3D transcriptomic cell type atlases from MERFISH data. We produced both unscaled and scaled versions of the densities by (i) integrating various datasets with MERFISH metadata, (ii) calculating cell type densities for each leaf region of the mouse brain, and (iii) projecting these densities back into the average brain space. T-type density estimates can subsequently be integrated with other neuron phenotypes through a probabilistic mapping framework. The bottom module derives from patch-seq data. The reference dataset for the three modalities was used to align the patch-seq datasets on established classification and derive the probability of observing an me-type given a t-type. The
can be combined with the t-type density atlas to produce an met-type density atlas. The workflow integrates all these datasets and literature values and no additional input is required.
Fig 2.
Regional transcriptomic cell counts in the extended 3D atlas format (566, 320, 456).
Sagittal (y = 200) and coronal (x = 300) sections from the A. extended and improved CCFv3 annotation volume (The annotation colors match those in the AIBS reference atlas.), B. regional cell counts with Nissl granularity, C. regional neuron counts (averages), and D. regional non-neuronal counts (averages). Colorbar shows cell numbers in number of cells / mm3 for panel B. E. Sagittal (y = 200) and coronal (x = 300) sections of an excitatory cell type at 4 different hierarchies (class, subclass, supertype, cluster). F. Inhibitory cell type examples at different hierarchical levels. G. Astrocytic cell type examples at different hierarchical levels. H. Oligodendrocytic cell type examples at different hierarchical levels. Colorbars (E-H) show cell numbers for the first cell type in the row. Resolution: 25 μm3 voxel size.
Fig 3.
Stepwise construction of the extended probabilistic map.
A. Principle overview. (i) High-dimensional mRNA expression space is defined by many genes (g1, g2, g3). (ii) Gene selection and dimensionality reduction project the space into a subspace (g2, g3) that captures information relevant for predicting morphological and electrophysiological properties. (iii) This projected gene space approximates the functional feature space (f1, f2) used for label prediction. B. Detailing the gene selection step (Step1) using supervised models. (i) Example embedding of neurons in mRNA space colored by me-type labels. (ii) Candidate genes are evaluated using Random Forest classifiers and regression models trained to predict me-labels. (iii) Feature importance (e.g., Gini importance) is calculated for each gene. (iv) The final set of selected genes is defined by the intersection of best predictors for me-types and genes with low region-predictive power. C. kNN-based mapping of uncovered t-types. (i) Neurons of uncovered t-types are projected into the selected gene subspace (G1∩G2). Distances between cells are computed using Euclidean metrics, and k-nearest neighbors (k = 10) are identified (Step 2). (ii) Probabilities for uncovered t-types are inferred by averaging the probabilities of their nearest neighbors, weighted by distance, yielding the extended probabilistic map (Step 3).
Fig 4.
Comparison of predicted density with literature values.
A. We extracted inhibitory neuron densities from the cerebrum from the unscaled atlas and compared them with inhibitory neuron numbers from the literature [7,38]. B. Inhibitory densities extracted from the cerebellum, and C. the brain stem. D. Comparison of total neuron density values from the scaled atlas with values from the literature [7]. Insets in the top-left corners show sampled areas, colored according to the annotation scheme used in the AIBS reference atlas. For regions with multiple reported literature values, we depict the variability using error bars. Acronyms are listed in S2 Table.
Fig 5.
Probabilistic mapping application on t-type densities and validation of mRNA counts globally.
A. Example of a coronal section of the density of the IN_dend_1_ax_3|bAC me-type, which preferentially maps to basket cells related t-types after applying the probabilistic mapping (left) and a heatmap (right) of obtained densities for a selection of brain regions (rows) and a selection of me-type (columns). For clarity, only a subset of me-types and brain regions are shown (including regions of CA1, DG, VIS, SS). B. mRNA counts obtained by multiplying cell types gene expression profiles with their densities (see Methods) in a selection brain regions and a selection of genes for t-types (left) and me-types (middle) and their relative error (right). C. Semilogarithmic plots of mRNA counts summed over regions (left) or genes (right). The top row shows the relative error, defined as (log₁₀ scale), while the bottom row shows the corresponding summed mRNA counts (log₁₀ scale). Blue points correspond to values derived from transcriptomic t-type data (
), and orange points to values obtained from morphological-electrophysiological me-type estimates (
).
Fig 6.
Brain region cell types composition.
A. UMAP projection of brain regions based on (i) me-type and (ii) t-type density profiles. Each point represents one brain region, positioned according to its high-dimensional composition projected into a two-dimensional UMAP space (UMAP1 and UMAP2 are arbitrary axes that preserve relative similarity, not direct biological variables). Colors indicate k-means cluster assignments, which group regions with similar cellular compositions. (iii) Overlap matrix showing the percentage of brain regions shared between me-type based clusters (rows) and t-type based clusters (columns), with color intensity reflecting overlap magnitude. B. Exemplar region profiles: pie charts show the me-type composition of me-types in six representative brain regions chosen from distinct clusters. Only me-types comprising more than 5% of the brain region are displayed for clarity. Excitatory me-types are shown in shades of red, and inhibitory me-types in shades of blue. A complete color legend for all me-types is provided in the Supplementary Information.
Table 1.
Total cell and neuron counts across the major brain regions, as estimated from MERFISH data. We aggregated the density values and transcriptomic cell type counts across major brain areas, expressed as the number of cells/mm³. See supplementary materials for the detailed table.