Simultaneous CRISPR screening and spatial transcriptomics reveal intracellular, intercellular, and functional transcriptional circuits

Abstract

Pooled optical screens have enabled the study of cellular interactions, morphology, or dynamics at massive scale, but they have not yet leveraged the power of highly plexed single-cell resolved transcriptomic readouts to inform molecular pathways. Here, we present a combination of imaging spatial transcriptomics with parallel optical detection of in situ amplified guide RNAs (Perturb-FISH). Perturb-FISH recovers intracellular effects that are consistent with single-cell RNA-sequencing-based readouts of perturbation effects (Perturb-seq) in a screen of lipopolysaccharide response in cultured monocytes, and it uncovers intercellular and density-dependent regulation of the innate immune response. Similarly, in three-dimensional xenograft models, Perturb-FISH identifies tumor-immune interactions altered by genetic knockout. When paired with a functional readout in a separate screen of autism spectrum disorder risk genes in human-induced pluripotent stem cell (hIPSC) astrocytes, Perturb-FISH shows common calcium activity phenotypes and their associated genetic interactions and dysregulated molecular pathways. Perturb-FISH is thus a general method for studying the genetic and molecular associations of spatial and functional biology at single-cell resolution.

Keywords: multimodal screening; pooled CRISPR screen; pooled optical profiling; single-cell perturbations; spatial transcriptomics.

Figure 1.. Perturb-FISH allows recording both gRNAs and transcriptome in cells in their spatial context

(A) Perturb-FISH workflow. (Ai) Guides are packaged into lentiviral particles using a modified version of lentiguide-puro that contains a T7 promoter between the end of the U6 promoter and the beginning of the guide. (Aii) The lentivirus is used to insert the guide sequence in the genome of the cell, resulting in genome editing. (Aiii) T7 polymerase locally generates many copies of the gRNA in fixed cells. (Aiv) DNA-encoding probes anneal on both the target mRNA and the amplified gRNA. (Av) Fluorescent readouts anneal on the encoding probes and are imaged. The fluorophores are cleaved, and this step is repeated. (Avi) After sequentially imaging the gRNAs, rounds of hybridization/imaging/cleavage continue to image the transcriptome in the same cells. (B) Representative images of Perturb-FISH: amplified gRNA generates a bright spot in the nuclei of cells, and the identity of gRNAs is encoded in the sequence of images in which they fluoresce (left). The transcriptome is read out the same way with MERFISH (right). T7 transcription yields higher signal amplification than the tiling of an mRNA with 30 probes, as visible by the larger size of the spots they generate. Scale bar, 10 μm.

Figure 2.. Overview of experimental design and analysis

(A) Overview of matched Perturb-seq and Perturb-FISH screens for genetic regulators of LPS response in THP1-cells. (B) We compare perturbation effect sizes from the two screens to evaluate Perturb-FISH in terms of consistency, recapitulation of global regulatory structure, and power (Figure 3). (C) We first characterize the effects of local cell density on perturbations effects, then we recover effects from perturbed cells onto their unperturbed control neighbors. We validate these results using both qPCR and RNAscope (Figure 4). (D) We demonstrate the use of Perturb-FISH in conjunction with live imaging characterization of calcium phenotypes, in a screen of ASD risk genes (Figure 5). (E) We verify that Perturb-FISH can be applied in 3D tissues, in a screen of NF-κB genes in tumors.

Figure 3.. Perturb-FISH robustness and power analysis

(A) Scatterplots of significant effect sizes (q < 0.1) determined in Perturb-FISH sample 1 (x axis) and the same effect sizes as determined in sample 2 (y axis). Effects represent log-fold changes (LFCs; natural log base) in expression relative to control cells. Individual plots depict significant effects across all perturbations or for individual perturbations MAP3K7, IRAK1, and TRAF6. (B) Scatterplots of significant effect sizes (q < 0.1) determined in Perturb-seq (x axis) and combined analysis of both Perturb-FISH replicates (y axis). Effects represent LFCs(natural log base) in expression relative to control cells. Individual plots depict significant effects across all perturbations or for individual perturbations MAP3K7, IRAK1, and TRAF6. (C) Heatmap of LFC effect sizes (color bar) found in Perturb-FISH (upper left triangle of each square) and the same effects in Perturb-seq results (lower right triangle). Rows and columns are clustered based on unweighted pair group method with arithmetic mean (UPGMA) clustering of Perturb-FISH data. (D) Held-out validation accuracy (y axis; Pearson correlation, left; AUPRC, right) of effects learned in downsampled Perturb-FISH data using increasing numbers of cells (x axis). Orange line shows approximate correlation (left) or AUPRC (right) obtained with 50 cells in Perturb-seq, estimated using previously published data. See also Figure S1.

Figure 4.. Perturb-FISH analysis of density-dependent and intercellular effects

(A) Violin plots show the expression level (y axis) of TNF (top) and IL1A (bottom) in cells with different numbers of neighbors (x axis). (B) Heatmap of LFC effect sizes (color bar) found by Perturb-FISH in cells with two or fewer neighbors (upper left triangle of each square) or three or more neighbors (lower right triangle). Effects represent log-fold changes (LFCs; natural log base) in expression relative to control cells at matching density. Black squares highlight effects that are significant (q < 0.1) in at least one density and different by at least 0.4 between densities. (C) Comparison of the density dependency of the effects of perturbing NFKB1, TRAF6, TAB2, or IKBKB on TNF, ETV3, or IL1A expression. Bars show the difference between effects at high density and effects at low density, when effects are measured with Perturb-FISH (in orange) or with a qPCR (blue). Effects are expressed as LFCs (natural log base). (D) RNAscope images showing TNF transcripts (red) in NFKB1 KO cells (blue) and unperturbed cells (black) at both low density (left) and high density (right). (E) Heatmap of LFC effect sizes (color bar) of gene knockouts (x axis) on the expression of genes (y axis) in neighboring unperturbed cells. Effects are expressed as LFCs (natural log base). Circle size shows significance, with q values from 0.05 to 0.5 s. See also Figure S2.

Figure 5.. Perturb-FISH analysis of the role of ASD risk genes in generating different calcium activity phenotypes in astrocytes

(A) Traces showing the variation in cytoplasmic calcium (y axis) concentration in example cells (line) across time (x axis), following ATP stimulation. Variation in calcium concentration shown as difference in fluorescence. Cells are six groups using k-means clustering on features extracted from the traces: large peaks; inactive cells; cells with large early transients; cells with a small peak; cells with a step; and finally, cells with a delayed response. (B) Heatmap of LFC effect sizes (color bar) found by Perturb-FISH in iPSC-derived astrocytes. Perturbations on the y axis; genes (x axis) are aligned with and labeled on (D). (C) Heatmap of LFC perturbation enrichment (y axis, aligned with B, color bar) per calcium activity signature (x axis). (D) Heatmap showing genes with differential expression (x axis, aligned with B, color bar) and their Z scored expression (scale bar) in six clusters of calcium activity. See also Figures S3 and S4.

Figure 6.. Perturb-FISH screen of NF-κB pathway in a tumor xenograft

(A) Heatmap of intrinsic LFC effect sizes (color bar) of perturbations (x axis) on gene expression (y axis) in tumor cells. (B) Top, spatial map of cell-type identity in the tumor. Infiltrated T cells and three groups of tumor cells are identified using only their transcriptomes. Bottom, example violin plots for six genes that represent the signatures of the four identified clusters. (C) Spatial map of identified CRISPR-KO perturbations. (D) Heatmap of LFC effect sizes (color bar) of perturbations in tumor cells (y axis) on gene expression in T cells (y axis). See also Figure S5.