Ascertaining cells' synaptic connections and RNA expression simultaneously with barcoded rabies virus libraries

Abstract

Brain function depends on synaptic connections between specific neuron types, yet systematic descriptions of synaptic networks and their molecular properties are not readily available. Here, we introduce SBARRO (Synaptic Barcode Analysis by Retrograde Rabies ReadOut), a method that uses single-cell RNA sequencing to reveal directional, monosynaptic relationships based on the paths of a barcoded rabies virus from its "starter" postsynaptic cell to that cell's presynaptic partners. Thousands of these partner relationships can be ascertained in a single experiment, alongside genome-wide RNAs. We use SBARRO to describe synaptic networks formed by diverse mouse brain cell types in vitro, finding that different cell types have presynaptic networks with differences in average size and cell type composition. Patterns of RNA expression suggest that functioning synapses are critical for rabies virus uptake. By tracking individual rabies clones across cells, SBARRO offers new opportunities to map the synaptic organization of neural circuits.

Fig. 1. Single-virion RNA tracking enabled by libraries of rabies virus particles encapsidating millions of uniquely barcoded genomes.

a Monosynaptic SBARRO. Shapes indicate neuron type. A TVA-expressing starter cell (“S”, magenta border) complemented in trans with rabies virus glycoprotein (G; orange text/fill) are selectively transduced by EnvA-RVdG-EGFP_VBC (green text/fill) in which G is replaced by barcoded EGFP. G-complemented clonal particles spread a single retrograde synapse into presynaptic cells (“P”, black border). Uninfected tan cells make no starter cell synapses. Single-cell RNA profiles (dotted lines) inform (1) synaptic groupings (from viral barcode (VBC) sharing, green); (2) Starter or presynaptic status (from TVA mRNAs, magenta) and (3) host cell type (by capturing thousands of cellular mRNAs, black). b Monosynaptic relationships are inferred through unitary clonal infectivity paths (uCIPs). uCIPs are defined by VBCs sufficiently rare in the infecting rabies library to seed single founder infections in starter cells (magenta border). Example library-abundant (orange) or library-rare (blue) VBCs. Abundant VBCs infect multiple starter cells and thus conflate presynaptic networks one and two. Rare VBCs define a uCIP which corresponds to monosynaptic network 2. c Viral particles are distinguished by a 20 bp VBC in the 3′ UTR of EGFP. SBARRO libraries contain millions of uniquely barcoded genomes. d Schematic of VBC diversity during rabies packaging. e Sequencing-based VBC quantification using unique molecular identifiers (UMIs). f–h VBC diversity metrics (color-coded as in d). f Unique VBCs at each packaging stage after down-sampling to 0.67 million UMI counts (left) or with 8.2 million counts to more thoroughly characterize the final EnvA pseudotyped library (right). The dotted “theoretical limit” indicates a library in which every sampled VBC is unique. g Cumulative distribution of VBCs binned by “VBC abundance group” (AG) across packaging stages (0.67 M counts/stage). Counts for all VBCs sampled once belong to AG = 1; sampled twice belong to AG = 2, etc. h The relationship between unique VBCs and total VBC draws from the EnvA-RVdG-EGFP_VBC library (8.2 M counts; blue line) and after removing the 88 most abundant VBCs (dashed blue line). The dotted line shows maximum diversity in which every VBC draw is unique. Source data are provided as a Source Data file.

Fig. 2. Properties of barcoded library infection revealed through single-cell RNA profiling from mouse brain cultures lacking cell-to-cell viral spread.

a Experimental schematic. EnvA-RVdG-EGFP_VBC library transduced starter cells (+TVA) from which the rabies virus could replicate but not spread to other cells (-G). RNA profiles were captured (n = 60.8 K cells), including from rabies-infected cells (n = 17.2 K), allowing a direct analysis between how viral genomic barcodes are expressed in many single starter cells. b Representative image of dissociated mouse brain cell cultures (14 days in vitro) expressing TVA (magenta) and EGFP (green). TVA expression was induced by a cocktail of two high-titer rAAVs (CAG-Flex-TVA-mCherry: serotype 2–9, 2.2 × 10¹⁰ genomes/well; Syn1-EBFP-Cre: serotype 2–1, 6 × 10⁹ genomes/well). Cultures consist of TVA-/EGFP-, TVA+/EGFP- and TVA+/EGFP+ cells. c Inference of founder VBC sequences and accurate UMI-based counts from single cell RNA profiles in light of subsequent barcode mutations. A dendrograms illustrating VBC sequence relationships (top) and UMI counts (below) before (left) and after (right) “within-cell VBC collapse” for a single example RNA profile (Methods). The mean (red dotted line) and two standard deviations (pink shading) expected from the distribution of edit distances across 20 base pair barcode sequences. d UMAP embedding of 60,816 scRNA profiles color-coded by molecular identity (left) or VBC ascertainment status (right) following LIGER analysis (Methods). A subset of infected scRNA profiles (n = 2635) could not be definitively identified (light green). e, f Comparison of single cell VBC properties ascertained from RNA profiles of TVA+ neurons (n = 4222) or glia (n = 914; ***p < 2.2e−16, two-sided Kolmogorov-Smirnov Test). Data from only the 1:10 EnvA-RVdG-EGFP_VBC dilution (MOI ~ 1.5) are shown (Methods and Supplementary Fig. 3e). e Cumulative distribution of unique VBCs. f Total VBC UMIs (left) or % EGFP mRNA (right). g Critically evaluate the performance of the EnvA-RVdG-EGFP_VBC through a corpus of 17.2 K starter cell RNA profiles. Left, for ascertained VBCs in the library (94%), the relationship between library abundance and the number of independent starter cell infections. Right, for library-absent VBCs (6%), the number of independent infections. VBCs observed in more starter cell RNA profiles than expected based on quantitative library abundance or library absence were flagged for exclusion (Methods). Source data are provided as a Source Data file.

Fig. 3. Massively parallel inference of cell-type-specific synaptic connectivity using SBARRO.

a Experimental schematic. The EnvA-RVdG-EGFP_VBC library transduced starter cells (+TVA/+G) from which individual virion clonally replicate and undergo monosynaptic spread into presynaptic cells. scRNA-seq libraries were prepared from either (1) SBARRO EGFP+ cells (n = 22 culture wells from n = 3 mouse preparations; n = 130.5 K scRNA profiles) or (2) preparation-matched control cells (n = 147.6 K scRNA profiles). b Sagittal brain schematic color-coded by region from which cells were co-cultured. c UMAP embedding of scRNA profiles color-coded and labeled by coarse molecular identity (left, Supplementary Fig. 6a) or SBARRO/control status (right) following LIGER analysis. d An example SBARRO network inferred through shared expression of the VBC assigned the name “immatureness_22”, a barcoded genome with <1% chance infecting >1 starter cell. The “immatureness_22” network consists of an SPN starter cell and heterogenous collection of putative presynaptic cells (color-coded as in c). e Horizontal bar plot representing the cell types of the “immatureness_22” presynaptic network . f Horizontal bar plots for n = 9865 SBARRO networks with ≥2 cells (left) and the largest 100 networks (right). Of all networks, n = 365 networks (3.7%) included starter cell assignments. g–i Properties of presynaptic networks. g Fractional cell-type compositions from RV uninfected control wells (left) and presynaptic networks stratified by starter cell type (right). A subset of cell types observed in controls (dotted vertical line) were never or rarely observed in presynaptic networks and excluded from network analysis. Presynaptic cell type composition varies by starter cell type (*p <0.05; **p < 0.01; ***p <0.001, Chi-Square Test). The number of aggregated networks and total presynaptic cells are shown (networks/total cells). Network data include n = 365 networks identified across n = 22 culture wells derived n = 3 primary cell culture replicates. h Inferred presynaptic network sizes differ by starter cell type (**p < 0.05, two-sided Wilcoxon Test). i UMAP embedding color-coded and labeled by glutamatergic neuron and interneuron subtype (Supplementary Fig. 6a, c). j Inferred presynaptic network sizes by starter cell subtype (**p < 0.01; two-sided Wilcoxon Test). Boxes define the interquartile range and whiskers delineate 1.5 times this range. Source data are provided as a Source Data file.

Fig. 4. Postsynaptic RNA levels associated with rabies-based inferences of presynaptic network size.

a Schematic of postsynaptic starter cells with small (brown triangle) or large (green triangle) numbers of presynaptic partner cells. b Histogram of inferred presynaptic network sizes for n = 144 starter cell RNA profiles belonging to one of four major cell types (glutamatergic neurons, interneurons, SPNs or astrocytes). c, d Differential expression testing identifies Arpp21 upregulated in Pvalb interneurons (n = 9 small versus n = 7 large RNA profiles) and Cdh13 as Tfap2d Glutamatergic neurons (n = 5 small versus n = 3 large RNA profiles) starter cells with large presynaptic networks. Left, Volcano plots illustrating results from differential expression testing of starter cell subtypes in which UMI counts were aggregated by inferred presynaptic network size category (Fisher’s Exact Test; Methods). Genes passing corrected p value thresholds (p corrected < 0.05, blue dots) were further tested for differences in single-cell scaled expression (two-sided Wilcoxon Test; Methods) and those that pass this additional test (p uncorrected < 0.05, red dots) are labeled. Right, normalized expression levels (^*p < 0.05. Arpp21, p = 0.044; Cdh13, p = 0.026. Two-sided Wilcoxon Test). Boxes define the interquartile range and whiskers delineate 1.5 times this range. Source data are provided as a Source Data file.

Fig. 5. The developmental emergence of rabies transmission co-occurs with the maturation of synaptic function.

a UMAP embedding of scRNA profiles (n = 32,503) along a trajectory of development from immature neural precursor cells (NPCs) to mature spiny projection neurons (SPNs). Left, color-coded by rabies virus infected (i.e. SBARRO; n = 8837 profiles) or uninfected control cells from paired cultures (n = 23,666 RNA profiles); Middle, pseudotime; Right, pseudotime bins (n = 10). b Example expression plots for four developmentally regulated genes. c For each pseudotime bin, the percentage of scRNA profiles corresponding to rabies virus infected SBARRO cells over the total number of all cells. d RNA levels across pseudotime bins. Of four described rabies receptors, Ncam1 (top) has the only appreciable expression. Ncam1 expression precedes the major developmental increase in rabies transmission. The subset of genes (n = 55 of 171) in the “synaptic vesicle” SynGO category (bottom) whose RNA levels correlate with rabies virus infectivity (n = 3309 genes total). e Gene Ontology Biological Pathways (GOBP) and Synaptic Gene Ontology (SynGO) analyses for rabies-infectivity correlated genes (n = 3309; Methods). Left, analyses were conducted on the correlated gene list (red) as well as two sets of control genes (each with 1000 replicates of n = 3309 genes). In the “Expression-matched” set (dark gray), genes were selected from the mature SPN metacell (pseudotime bin = 10) in a manner that matched expression levels of the correlated genes. In the “Random” set (light gray), genes were selected at random from those for which expressed RNA was detected. Statistical testing was performed using one-sided Fisher’s exact test with false discovery rate correction using default parameters for GOBP and SynGO. Categories with adjusted p-values < 0.05 were considered enriched. Uncorrected p-value distributions for n = 3 GO-BP categories (middle) and n = 4 SynGO categories (right) for which the rabies-infectivity correlated gene set was statistically enriched. Source data are provided as a Source Data file.