Transcriptional and morphological profiling of parvalbumin interneuron subpopulations in the mouse hippocampus

Abstract

The diversity reflected by >100 different neural cell types fundamentally contributes to brain function and a central idea is that neuronal identity can be inferred from genetic information. Recent large-scale transcriptomic assays seem to confirm this hypothesis, but a lack of morphological information has limited the identification of several known cell types. In this study, we used single-cell RNA-seq in morphologically identified parvalbumin interneurons (PV-INs), and studied their transcriptomic states in the morphological, physiological, and developmental domains. Overall, we find high transcriptomic similarity among PV-INs, with few genes showing divergent expression between morphologically different types. Furthermore, PV-INs show a uniform synaptic cell adhesion molecule (CAM) profile, suggesting that CAM expression in mature PV cells does not reflect wiring specificity after development. Together, our results suggest that while PV-INs differ in anatomy and in vivo activity, their continuous transcriptomic and homogenous biophysical landscapes are not predictive of these distinct identities.

Fig. 1. Morpho-transcriptomic profiling of PV-INs.

a Experimental pipeline: tdTomato+ PV-INs were recorded in acute brain slices, analyzed for their morphological and electrophysiological properties, and processed for single-cell RNA-seq. b Morphological reconstructions show different morphological PV types (red: vertical axo-axonic cell, vAAC, orange: vertical bistratified cell, vBIC, yellow: horizontal bistratified cell, hBIC, blue: vertical basket cell, vBC, dark blue: horizontal basket cell, hBC; these color code applies to all figures). c Violin plots show expression of top ten genes that are enriched in PV versus SST-OLM (upper plots) and in SST-OLM versus PV cells (lower plots). d Plot shows transcriptomic data-based dimension reduction using UMAP. Symbols correspond to different PV and the SST-OLM types. Insert shows UMAP for only PV-INs. e Bar plot shows classification accuracy for random-forest-based PV versus SST-OLM, BIC versus non-BIC, and for BC versus non-BC discriminations.

Fig. 2. Transcriptomic properties of PV-INs.

a Plot shows unsupervised, transcriptomic data-based dimension reduction using nbt-SNE. Single cells are labeled with circles, which are colored according to morphological classification. Background colors refer to transcriptomic classification using proMMT. b Confusion matrix between proMMT-based transcriptomic types and morphological types. Numbers represent how many cells belong to both a given morphological type (rows) and a given transcriptomic type (columns). c Heatmap of all genes that were differentially expressed (FDR < 0.05 and fold change>2), in at least one pairwise comparison of transcriptomic types using edgeR. Cells are grouped by transcriptomic types. d Plot shows the mapping of PV-INs of this study onto the CA1-IN dataset. Cell-type labels, e.g., Pvalb.Tac1, are also imported from the original CA1-IN study. Gene selection for mapping was performed using the method described in Kobak et al.. e Quantification of mapping efficacy using six different gene selection methods (see “Methods” for details).

Fig. 3. Electrophysiological properties of PV-INs.

a Plots show electrophysiological properties measured in the five morphological PV types. Circles represent single cells. Data represent mean ± S.E.M. b All panels show dimension reduction on electrophysiological data using UMAP. Circles represent single cells and are colored to show the two dendro-morphological (horizontal and vertical), three axo-morphological (AAC, BIC, and BC), four proMMT transcriptomic (Pthlh.2900055J20Rik.Npy, Pthlh.2900055J20Rik.Pcp4, Pthlh.Snca, and Synpr), and five morphological (vAAC, vBIC, hBIC, vBC, and hBC) PV types.

Fig. 4. Transcriptomic correlates of morphological PV types.

a–c Heat maps (upper plots) of differentially expressed genes (using edgeR, fold difference>2, FDR < 0.05) between the five morphological PV types (panel a), three axo-morphological types (panel b), and two dendro-morphological types (panel c). Cells are ordered according to proMMT types. Genes that also appeared in proMMT comparisons (Fig. 2c) are highlighted in red. PCA plots (lower plots) were made using the differentially expressed genes. d PCA plot of cells, colored by the five morphological PV types, using an extended set (with a cutoff of FDR < 0.15) of n = 52 differentially expressed genes from panels a to c. e Comparison of expression rate of genes in PV-INs from this current versus the CA1-IN study, displayed as a heatmap. Black line marks the loess regression fit. Black points label the 52 differentially expressed genes (with a cutoff of FDR < 0.15) from a to c. f UMAP-based embedding of PV-INs from the CA1-IN study, and mapping the PV-INs of this current study onto the UMAP embedding using the extended set of n = 52 differentially expressed genes (with a cutoff of FDR < 0.15) from panel a to c. Three clusters of mapped cells are shown (labeled as 1, 2, and 3) as determined by K-means clustering. Symbols refer to the following transcriptomic subtypes, as described in the original study: rightward triangle Pvalb.Tac1.Akr1c18, leftward triangle Pvalb.Clql1.Cpne5, upward triangle Pvalb.C1ql1.Npy, downward triangle Pvalb.Clql1.Pvalb, plus sign Pvalb.Tac1.Nr4a2, open circle Pvalb.Tac1.Sst, and star Pvalb.Tac1.Syt2. The dashed line separates Pvalb.Tac1 and Pvalb.C1ql types.

Fig. 5. Analysis of CAM expression in morphological PV types.

a Heatmap showing all CAMs expressed in at least three PV-INs, grouped by morphological types. Genes are ordered via hierarchical clustering. b Violin plots show the top ten most significantly differentially expressed CAMs (using edgeR, fold difference >2), between morphological PV types (top panel), and between pooled PV-INs and SST-OLM cells (bottom panel). For comparison, relevant types from the CA1-IN study are shown on the right. c Similarity matrices between the five morphological (top left), three axo-morphological (top right), and two dendro-morphological PV types (bottom left). Similarity scores were measured based on only CAM expression and using the average of Pearson’s correlation coefficients of cells. d Heatmap shows an expression of genes that were previously described to be (1) ubiquitous in CA1 pyramidal cells (PYR), fast-spiking interneurons (FS; presumed PV-INs), and regular-spiking interneurons (RS; presumed CCK-INs); or specifically expressed in (2) FS and RS cells; (3) RS cells; (4) FS cells; or in (5) PYR cells.

Fig. 6. Morphological and electrophysiological analysis of vBC type PV-INs during circuit maturation.

a Morphological reconstructions show vBC type PV-INs at different ages. The first image shows Sholl analysis for dendritic branching based on concentric 3D spheres (plot shows circles for clarity). b Plot shows the number of intersections at different distances from the soma of n = 8 < P21 and n = 16 > P21 vBC. Circles denote individual cells, which represent a randomly chosen subset (~50%) of all <P21 and >P21 vBC cells included in the whole study. None of the statistical comparisons (two-sided Welch’s t test) revealed a FDR smaller than 0.255. c Plots show the total dendritic (left) and axonal (right) length of <P21 and >P21 vBC-type cells. P values are shown on top and were determined using two-sided Welch’s t test. Data represent mean ± S.E.M. d Scatter plots and linear regression fits of axon and dendrite length against age. Neither value was shown to correlate with age (lowest P value and highest r² for fit were 0.272 and 0.209, respectively). e Plots show electrophysiological properties of n = 19 < P21 versus n = 31 > P21 vBC type PV-INs. Attenuation (FDR = 1.7 × 10⁻⁵, two-sided Welch’s t test), AP half-width (FDR = 0.008) and AP amplitude (FDR = 0.05) show significant changes. Data represent mean ± S.E.M. f UMAP-based dimension reduction of electrophysiological properties of vBC-type cells. All cells represent the vBC type and are colored by age.

Fig. 7. Age-dependent transcriptomic changes and the onset of hemoglobin mRNA expression in PV-INs.

a Expression levels of genes showing a statistically significant (FDR < 0.10) transition using Gini impurity (see “Methods”). Genes (rows) ordered by the day of transition, cells (columns) are ordered according to age. All cells represent vertical basket (vBC) type. b UMAP-based on genes from panel a. All cells represent vBC type and are colored by age. c Plot of consistency with which age can be predicted using Random Forest Classifier based on genes from panel a (see “Methods”), fitted with a loess curve. d Heatmap shows the expression of hemoglobin mRNA and related genes in vBC-type cells (columns), which are sorted by age left to right. e Normalized loess fits of hemoglobin mRNA expression levels versus age in vBC-type cells. f Violin plots show expression of hemoglobin mRNA and related genes from panel d only considering >P21 PV-INs and SST-OLM cells. g Plots show average Hba-a1, Hba-a2, Hbb-bs, and Hbb-bt expression in each morphological PV type with single-nucleotide resolution. Exon and intron lengths are shown according to their original scale.