High-content image-based pooled screens reveal regulators of synaptogenesis

Abstract

Synapse formation is a fundamental process that shapes the connectivity and function of the nervous system, but the mechanisms regulating synaptogenesis are incompletely understood. Moreover, the interplay of these mechanisms at distinct synapse types remains to be defined. Using a scalable optical pooled screening platform, we investigated the process of synapse induction to uncover modulators of a prototypical synapse-organizing adhesion molecule, neuroligin-1. Analysis of over two million single-cell phenotypic profiles identified 102 candidate regulators of neuroligin-1 that are linked to cell adhesion, cytoskeletal dynamics, and signaling. Among these, we show that the phosphatase PTEN and the dystrophin-associated glycoprotein DAG1 promote neuroligin's roles in inducing presynaptic assembly, with DAG1 selectively regulating inhibitory synapses. This work establishes a scalable high-content screening approach for cell-cell interactions that enables systematic studies of the molecular interactions guiding synaptogenesis.

Keywords: CP: Neuroscience; CRISPR/Cas9; cell adhesion molecules; cell non-autonomous; cell-cell interactions; functional genomics; high-content screening; neuroligin; optical pooled screening; synapse; synaptogenesis.

Figure 1.. Image-based pooled screening enables high-throughput discovery of synaptogenic regulators

(A) Workflow of optical pooled CRISPR knockout screen. (B) Example field of view from the screen. Insets show the stains used. Arrows indicate presynaptic specializations induced by NLGN1. Scale bar, 50 μm. (C) Images of in situ sequencing (Laplacian-of-Gaussian filtered) for the same field of view as the insets in (B). Scale bar, 50 μm. (D) Histogram of the number of cells analyzed for each highly expressed gene target (n = 1,843,247 cells in the primary screen). (E) Example images of Bassoon (red) and HA-NLGN1 (gray) staining from the screen show decreased induction of synaptogenesis by HEK293 cells with CDH2 or NLGN1 knockouts. Scale bar, 25 μm. (F) Boxplots show the number of Bassoon puncta per HEK293 cell in the primary screen (center line, median; box, upper and lower quartiles, Q1–Q3; whiskers, 1.5× interquartile range [IQR]) (non-targeting sgRNAs, n = 26,920 cells; CDH2 knockout, n = 3,899 cells; NLGN1 knockout, n = 1,648 cells). Statistical significance was measured by a one-way ANOVA test, followed by Tukey’s test for multiple comparisons (***p < 0.001).

Figure 2.. Multidimensional phenotypes underlying synaptogenesis

(A) Scores plot of principal-component analysis (PCA) performed on phenotypic profiles of 644 gene knockouts and 50 non-targeting sgRNAs. (B) Top 15 loadings of PC1 include puncta area, count, and intensity features for Bassoon, VGLUT1, and VGAT stains. Bolded features were summed into a synaptogenesis score for each cell. (C) Two-dimensional PHATE visualization of phenotypic profiles of all gene knockouts. Colors indicate Leiden clusters, with non-targeting sgRNAs in black. (D) PHATE plot shown in (C) with genes colored by the HA-NLGN1 median cell intensity (left) and synaptogenesis score (right). (E) Receiver operating characteristic (ROC) curve of random forest classifier (trained on 70% of the data) predicting cluster labels in (C) using the HA-NLGN1 median cell intensity and component features of the synaptogenesis score. (F) Boxplot of feature importances, calculated from 100 permutations, for the random forest model in (E) (center line, median; box, Q1–Q3; whiskers, 1.5× IQR). (G) Scatterplot comparing HA-NLGN1 median cell intensities and synaptogenesis scores (Z scored) for all genes, Pearson correlation coefficient r = 0.24. (H) Illustrative images show gene knockouts labeled in (G) with increased expression of HA-NLGN1 and modestly increased induction of presynaptic specializations. Scale bar, 25 μm.

Figure 3.. Identification of positive and negative regulators of NLGN-1-induced synapse formation

(A) Volcano plot of synaptogenesis scores (Z scored), highlighting genes whose knockouts resulted in decreased (blue; positive controls in navy) or increased (red) presynaptic specializations (false discovery rate [FDR] < 0.1, dashed line). Orange dots represent 50 random samples of 6 non-targeting sgRNAs. (B) Illustrative images of cells with gene knockouts showing decreased induction of presynaptic specializations. Scale bar, 25 μm. (C) Empirical cumulative distribution functions for the normalized Bassoon puncta area of single cells with CRKL, PFN1, PTEN, CDH2, and NLGN1 knockouts (shades of blue), compared to cells with non-targeting control sgRNAs (gray). Circles indicate cells shown in (B). (D) Illustrative images of cells with gene knockouts showing increased heterologous synapse formation. Scale bar, 25 μm. (E) Empirical cumulative distribution functions for the normalized Bassoon puncta area of single cells with CADM1, NSF, and SFPQ knockouts (shades of red) compared to cells with non-targeting sgRNAs (gray). Circles indicate cells shown in (D). (F) Scatterplot comparing VGLUT1 and VGAT puncta scores (Z scored sum of normalized area and puncta count) for all genes, Pearson’s r = 0.49. (G) Example images of cells with gene knockouts showing a preferential decrease in excitatory (VGLUT1) (left) or inhibitory (VGAT) (right) synapse formation. Scale bar, 25 μm. (H) Heatmap of the synaptogenesis score and its component features for top-scoring genes from the primary screen (FDR < 0.05 and Z score > 1.5 for at least two features). Color indicates the corresponding feature Z score, and circle size indicates the feature FDR. Gene order was obtained from hierarchical clustering.

Figure 4.. Validation of screen hits in heterologous synapse formation co-culture assays

(A) Workflow of arrayed co-cultures with clonal knockout HEK293 cells. (B–D) Quantification of presynaptic specializations induced by knockout HEK293 cells transfected with HA-tagged BFP, wild-type NLGN1, or NLGN1 AChE mutant with inactivating K578A/V579A substitutions. Bar plots of the number of (B) Bassoon, (C) VGLUT1, and (D) VGAT puncta per cell show means ± SEM from 4–8 independent experiments (numbers of cells analyzed are listed in Figure S5F). Statistical significance was calculated by a one-way ANOVA test, followed by Tukey’s test for multiple comparisons (ns = p ≥ 0.05 and ***p < 0.001).

Figure 5.. DAG1 specifically modulates inhibitory synapses in primary neurons

(A) Example images showing DIV 12 hippocampal neurons with shRNA-mediated knockdowns stained for Homer (red) and VGLUT1 (blue) (Laplacian-of-Gaussian filtered). Scale bar, 5 μm. (B and C) Quantification of excitatory synaptic puncta upon Cdh2, Dag1, and Pten knockdowns. Boxplots of the number of (B) Homer and (C) VGLUT1 puncta per μm of dendrite (center line, median; box, Q1–Q3; whiskers, 1.5× IQR). Values are normalized to the mean of non-targeting shRNAs. Data were obtained from n = 102–142 cells per condition, across 2 shRNAs per gene from 4 independent cultures. Statistical significance was calculated using the Kruskal-Wallis test, followed by Dunn’s test with Benjamini-Hochberg correction (ns = p ≥ 0.05 and ***p < 0.001). (D) Example images showing DIV 12 hippocampal neurons with shRNA-mediated knockdowns stained for gephyrin (red) and VGAT (blue) (Laplacian-of-Gaussian filtered). Scale bar, 5 μm. (E and F) Quantification of inhibitory synaptic puncta upon Cdh2, Dag1, and Pten knockdowns. Boxplots of the number of (E) gephyrin and (F) VGAT puncta per μm of dendrite (center line, median; box, Q1–Q3; whiskers, 1.5× IQR). Values are normalized to the mean of non-targeting shRNAs. Data were obtained from n = 133–184 cells per condition, across 2 shRNAs per gene from 4 independent cultures. Statistical significance was calculated using the Kruskal-Wallis test, followed by Dunn’s test with Benjamini-Hochberg correction (ns = p ≥ 0.05, *p < 0.05, and ***p < 0.001).

Figure 6.. Role of DAG1 and PTEN on synaptic organization and vesicle recycling

(A) Example images showing DIV 14 hippocampal neurons with shRNA-mediated knockdowns stained for surface GluA receptors (red) and presynaptic Bassoon (blue). Scale bar, 5 μm. (B and C) Quantification of immunostaining of surface GluA receptors. Boxplots of (B) the number of Bassoon puncta per μm of dendrite and (C) the mean GluA signal intensity in Bassoon puncta (center line, median; box, Q1–Q3; whiskers, 1.5× IQR). Values are normalized to the mean of non-targeting shRNAs. Data were obtained from 2 shRNAs per gene and 3–5 independent cultures (numbers of cells analyzed are listed in Figure S6J). Statistical significance was calculated using the Kruskal-Wallis test, followed by Dunn’s test with Benjamini-Hochberg correction (ns = p ≥ 0.05, **p < 0.01, and ***p < 0.001). (D) Example images showing DIV 14 hippocampal neurons with shRNA-induced knockdowns after live antibody-labeled synaptotagmin I (SytI) uptake (red) and fixed-cell staining for synaptophysin (blue). Scale bar, 5 μm. (E and F) Quantification of SytI uptake. Boxplots of (E) the number of synaptophysin puncta per μm of dendrite and (F) the mean SytI signal intensity in synaptophysin puncta (center line, median; box, Q1–Q3; whiskers, 1.5× IQR). Values are normalized to the mean of non-targeting shRNAs. Data were obtained from 2 shRNAs per gene and 4–6 independent cultures (numbers of cells analyzed are listed in Figure S6N). Statistical significance was calculated using the Kruskal-Wallis test and Dunn’s test with Benjamini-Hochberg correction (ns = p ≥ 0.05, **p < 0.01, and ***p < 0.001).