CRISPRi-based screens in iAssembloids to elucidate neuron-glia interactions

Abstract

The sheer complexity of the brain has complicated our ability to understand the cellular and molecular mechanisms underlying its function in health and disease. Genome-wide association studies have uncovered genetic variants associated with specific neurological phenotypes and diseases. In addition, single-cell transcriptomics have provided molecular descriptions of specific brain cell types and the changes they undergo during disease. Although these approaches provide a giant leap forward towards understanding how genetic variation can lead to functional changes in the brain, they do not establish molecular mechanisms. To address this need, we developed a 3D co-culture system termed iAssembloids (induced multi-lineage assembloids) that enables the rapid generation of homogenous neuron-glia spheroids. We characterize these iAssembloids with immunohistochemistry and single-cell transcriptomics and combine them with large-scale CRISPRi-based screens. In our first application, we ask how glial and neuronal cells interact to control neuronal death and survival. Our CRISPRi-based screens identified that GSK3β inhibits the protective NRF2-mediated oxidative stress response in the presence of reactive oxygen species elicited by high neuronal activity, which was not previously found in 2D monoculture neuron screens. We also apply the platform to investigate the role of APOE- ε4, a risk variant for Alzheimer's Disease, in its effect on neuronal survival. We find that APOE- ε4-expressing astrocytes may promote more neuronal activity as compared to APOE- ε3-expressing astrocytes. This platform expands the toolbox for the unbiased identification of mechanisms of cell-cell interactions in brain health and disease.

Keywords: APOE; CRISPR interference; CROP-seq; GSK3B; NFE2L2; essential genes; functional genomics; neuron-glia co-culture; neuronal activity; oxidative stress; single-nucleus RNA sequencing.

Figure 1.. Integrating hiPSC-derived neurons, astrocytes, and microglia into three-dimensional cultures (iAssembloids)

(A) hiPSCs expressing CRISPRi machinery and inducible NGN2 are pre-differentiated and seeded with hiPSC-derived astrocytes at a 3:1 neurons:astrocyte ratio. After 1 week, hiPSC-derived microglia are seeded at one-third of the number of astrocytes. (B) Brightfield images of iAssembloids 30 min, 1 day, 1 week and 2 weeks post seeding. Scale bar = 800 μm. (C) Neurons (gray), astrocytes (green) and microglia (red) expressing different fluorescent proteins were seeded to form iAssembloids. Left image: maximum-intensity projection, other images: individual images from the horizontal sample images (z-stack) generated from confocal microscopy. Arrows denote neurons co-localized with microglia (closed arrowheads) as well as neuronal extensions across the culture (open arrowheads). Scale bar = 50 μm. Images were taken 14 days post seeding into AggreWell 800 plates. (D) Maximum intensity projections of iAssembloids stained with antibodies against neuronal markers NEUN and TUJ1. Scale bar = 50 μm. Images were taken 14 days post seeding into AggreWell^™ 800 plates. (E) Maximum intensity projections of iAssembloids stained with antibodies against microglia marker IBA1. Three different morphologies are presented. Scale bar = 10 μm. Cells with processes outlined in white. (F) Maximum intensity projections of iAssembloids stained with antibodies against the microglial marker IBA1 and the astrocyte marker S100β. Scale bars = 50 μm. Images were taken 14 days post seeding into AggreWell^™ 800 plates. Arrows denote microglia projections.

Figure 2.. Neurons cultured in iAssembloids are more functionally mature than monocultured neurons

(A) Single-nucleus RNA sequencing was performed on iAssembloids in culture for 1 day, 2 weeks and 4 weeks. In total, 43,182 nuclei passed quality control and are represented in this dataset. Cell types were assigned to clusters using cell type-specific markers such as RBFOX3 and MAPT for neurons, SLC1A3 (GLAST) for Astrocytes, and CSF1R and SPI1 for microglia. The heatmap is subsampled at 5,000 cells for visualization. UMAP is labeled by cell type and sample. (B) Single-nucleus RNA sequencing of neurons in iAssembloids and 2D monoculture neurons were mapped onto excitatory neurons from fetal brain single-cell sequencing data. The proportion of cells that mapped onto represented gestational weeks are reported. (C) Analysis of selected differentially expressed genes comparing iAssembloid and monocultured neuronal culture systems shows that expression of glutamate receptor subunits, glutamate receptor interactors and various ion channel subunits is significantly higher in iAssembloids versus monoculture. (D) Example raster plot from multi-electrode array (MEA, Axion) analysis of spikes of neuronal activity in monocultured neurons versus iAssembloids. (E) Bar graphs represent cumulative spike data over a 15-minute time span from 2D monoculture and 3D cultures with various proportions of glia. Bars represent the median and the error bars represent the 95% confidence interval. Each dot represents a separate well in the MEA plate.

Figure 3.. CRISPRi-based screens in iAssembloids reveals genes affecting neuronal survival not found in screens in 2D monoculture.

(A) Schematic of screen design. (B) Volcano plot of knockdown phenotypes from the survival screen from day 14 iAssembloids. Hit genes (FDR < 0.05) are labeled in either light blue (knockdown decreases survival) or light pink (knockdown increases survival). Selected hit genes were color-coded based on curated functional categories. (C) Scatterplot of gene scores from our previous survival screen in 2D monocultured neurons (x-axis) compared to the screen in iAssembloid (y-axis). Both screens used the H1 sgRNA library targeting the “druggable genome.” Selected genes with consistent phenotypes in both screens are labeled in black. Other genes of interest are color-coded based on functional category as in panel B. (D) Heatmap representing gene scores for neuronal survival from screens in 2D monoculture screens^,, and from iAssembloids (this study, screens were conducted with the H1 sgRNA library and an sgRNA library targeting neurodegeneration-related genes). Neurodegeneration library hits were compared to gene scores from a genome-wide 2D monoculture screen. Genes are grouped by functional categories as in panels B and C. Genes selected for secondary screens are highlighted by yellow boxes on the left.

Figure 4.. CROP-seq screen reveals that GSK3B knockdown induces neuronal expression of NRF2 target genes in iAssembloids, but not 2D monoculture

(A) Schematic of CROP-seq experimental design. (B) Heatmap of combined significantly expressed genes (p<0.05) from CROP-seq in neurons cultured in iAssembloids. Rows represent genes that were knocked down in the cells, columns are differentially expressed genes in cells with knockdowns compared to controls. Highlighted are upregulated genes that are putative NRF2 targets (yellow) and STAT3 targets (green). (C) Differentially expressed genes comparing GSK3B knockdown versus cells containing non-targeting controls in iAssembloids were determined. Red dots represent upregulated genes below the p<0.05 cutoff whereas pink dots represent genes below the p<0.1 cutoff. Dark blue represents downregulated genes meeting p<0.05 cutoff whereas light blue represents downregulated genes below the p<0.1 cutoff. The target gene, GSK3B, is highlighted in purple and genes that have been putatively shown to be regulated by NRF2 are highlighted in yellow. P values were determined using the Wald test. (D) Differentially expressed genes comparing GSK3B knockdown versus cells containing non-targeting controls in 2D neuronal monoculture were determined as specified in C.

Figure 5.. GSK3β activity prevents NRF2 from protecting neurons from oxidative stress

(A) Western blot validation of GSK3β knockdown in 3D Neurons. NTC: non-targeting control sgRNA. 3D cultures generated from NTC cells from results for 2 independent non-targeting control (NTC) sgRNAs and 2 independent sgRNAs targeting GSK3B are shown. (B) Validation of the effect of GSK3Β knockdown on cell survival over 14 days comparing the ratio of surviving cells expressing a non-targeting control sgRNA and GFP versus the proportion of cells expressing an sgRNA targeting GSK3Β and BFP. Experiment was performed with two different NTC sgRNAs as well as two GSK3Β sgRNAs with a total of four different combinations of NTC and GSK3Β sgRNAs. Medians and 95% confidence intervals are represented. ANOVA followed by Tukey’s multiple comparison test was conducted to determine significance. (C) Immunohistochemistry staining for NRF2 (red) in iAssembloids. Neuronal nuclei express BFP (blue). Scale bars = 50 μm. Images taken at 14 days post seeding of iAssembloids into AggreWell^™ 800 plates. Nuclei were identified using BFP in images and the integrated intensity for NRF2 within nuclei was quantified. Images from three different iAssembloids expressing NTC sgRNA 1 (n = 71 cells) and GSK3Β KD sgRNA 1 (n = 94 cells) were taken. Mean with standard deviation is represented. P values were determined using Student’s t-test. (D) Levels of reactive oxygen species (ROS, via CellROX^™ orange staining) determined via flow cytometry of neurons expressing a non-targeting control sgRNA or an sgRNA knocking down GSK3Β in 3D monoculture. NTC sgRNA 1 and GSK3Β KD sgRNA 1 were used in this experiment. Neurons with those sgRNAs were co-cultured and sgRNA identity was assigned based on expression of a BFP (GSK3Β sgRNA 1) or mClover (NTC sgRNA 1) marker. (E) Western blot quantification comparing level of phospho-S9 GSK3β to total GSK3β in 3D monoculture neurons only, iAssembloids, and 2D monoculture neurons. N=4 independent wells. bars represent mean, error bars represent standard error of the mean. ANOVA followed by Tukey’s multiple comparison test was conducted to determine significance. (F) Western blot quantification of pS9-GSK3β levels in 2D monocultured neurons after induction of oxidative stress with rotenone (200 nM for 24 hours). N = 6 independent wells are shown for each the vehicle and rotenone conditions. Bars represent mean, error bars represent standard error of the mean. P values were determined using Student’s t-test. (G) ROS levels (CellROX^™ orange staining) comparing 2D monoculture neurons versus 3D monoculture neurons. N=3 wells per sample. Bars represent means of the median fluorescence level and error bars represent standard error of the mean. P values were determined using Student’s t-test. (H,I) ROS levels CellROX^™ orange staining). comparing 3D monoculture neurons with neuron-astrocyte co-cultures of variable ratios. Bars represent normalized mean ROS levels compared to 3D monoculture neurons. N= 3 independent wells. Error bars represent standard error of the mean. ANOVA followed by Tukey’s multiple comparison test was conducted to determine significance. (J, K) Peroxidated lipid levels of GSK3Β KD cells as compared to NTC. Cells were stained with Liperfluo and median fluorescence levels were measured. Levels were then normalized to the NTC. J, example trace. K, quantification of 5 replicates (5 cell culture wells), bars represent mean, error bars represent standard error of the mean. P values were determined using Student’s t-test. (L) Reactive oxygen species (ROS) levels of 3D neurons treated with tetrodotoxin (TTX, 1.5 μM, duration of 1 week). Neurons are stained with CellROX orange and fluorescence levels were read out using flow cytometry. median fluorescence was normalized to the fluorescence level of the vehicle condition to account for different levels of background fluorescence across experiments. Dots represent n = 4 individual wells. Error bars represent standard error of the mean. P values were determined using Student’s t-test. (M) Peroxidated lipid levels (Liperfluo stain) in 3D monocultured neurons treated with vehicle, tetrodotoxin (1.5 μM, duration of 1 week), and ferrostatin (10 μM, duration of 1 week). Bars represent mean, error bars represent standard error of the mean. Dots represent cells from n = 3 independent wells. ANOVA followed by Tukey’s multiple comparison test was conducted to determine significance. (N) Viability of 3D monocultured neurons after 14 days in culture with and without tetrodotoxin (TTX, 1.5 μM, duration of 1 week) and ferrostatin (10 μM, duration of 1 week) treatment. Cells were stained with trypan blue, and the percentage of trypan blue negative cells were obtained. Bars represent mean, error bars represent. Error bars represent standard error of the mean. N=7 wells for vehicle, 8 wells for ferrostatin and 6 wells for TTX. ANOVA followed by Tukey’s multiple comparison test was conducted to determine significance.

Figure 6.. Proposed model of GSK3β function in oxidative stress and neuronal death

(A) In 2D monocultured neurons, there is not sufficient neuronal activity to drive oxidative stress. Therefore, GSK3β activity is low and neuronal death is not triggered. (B) In 3D monocultured neurons, neuronal activity is high. This triggers an increase in ROS and oxidative stress and leads to an increase in GSK3β activity. Increase of GSK3β activity blocks NRF2 from translocating to nucleus to mount a protective oxidative stress response. In the absence of the protective response, neurons undergo ferroptosis. (C) In 3D monocultured neurons with GSK3B knockdown, neuronal activity and oxidative stress is triggered, but in the absence of GSK3β , NRF2 can translocate to the nucleus, preventing ferroptosis. (D) In iAssembloids, the addition of glial cells reduces neuronal activity and ROS, possibly by regulation of glutamate levels via glutamate reuptake, which could lessen GSK3β activity and protect cells against ferroptosis.

Figure 7.. CRISPRi-based screen for neuronal survival in APOE-ε3 versus APOE-ε4 astrocyte 3D co-cultures

(A) Schematic of experimental design. Screens for neuronal survival were conducted in APOE-ε3 neurons expressing dCas9-KRAB and dox-inducible CRISPRi that were 3D co-cultured with APOE-ε3 vs APOE-ε4 astrocytes. (B) Hit genes that were consistent between the two screens (WTC11 and KOLF2.1) and had the strongest difference in gene score between APOE-ε3 and APOE-ε4 astrocyte co-cultures were selected and annotated. Heatmap represents gene scores. (C) Top 10 enriched GO biological processes differentially expressed genes in neurons 3D-co-cultured with APOE-ε3 versus APOE4-ε4 astrocytes, based on snRNA-seq. Orange bars represent terms related to metabolic processes and red bars represent terms related to neuronal function. P-values are calculated by EASE Score, a Modified Fisher Exact P-value. (D) Violin plot of expression of genes selected from related GO terms across different co-culture conditions from the snRNA-seq dataset represented in C. (E) Validation of screening results of CACNB4 KD. Neurons expressing non-targeting control guides and GFP were seeded at 1 to 1 ratio with neurons expressing sgRNAs targeting CACNB4 and BFP. Neurons were then cocultured with APOE-ε3 astrocytes (n=5 wells for sgRNA 1, n=4 wells for sgRNA 2) vs APOE-ε4 astrocytes (n=4 wells for sgRNA 1, n=4 wells for sgRNA 2) and the ratio of BFP+ to GFP+ cells were assessed after 2 weeks. P-value from Student’s t-test. Error bars represent standard error of the mean . (F) Multi-electrode array analysis of neurons co-cultured with APOE-ε3 vs APOE4-ε4 astrocytes. Spike numbers are represented. n=4 wells for each condition. P-value from Student’s t-test. Error bars represent standard error of the mean . (G) Median CellRox Orange fluorescence levels for neurons co-cultured with APOE-ε3 vs APOE4-ε4 astrocytes. n=3 wells for each condition. P-value from Student’s t-test. sError bars represent standard error of the mean .