Spatially Resolved in vivo CRISPR Screen Sequencing via Perturb-DBiT

Abstract

Perturb-seq enabled the profiling of transcriptional effects of genetic perturbations in single cells but lacks the ability to examine the impact on tissue environments. We present Perturb-DBiT for simultaneous co-sequencing of spatial transcriptome and guide RNAs (gRNAs) on the same tissue section for in vivo CRISPR screen with genome-scale gRNA libraries, offering a comprehensive understanding of how genetic modifications affect cellular behavior and tissue architecture. This platform supports a variety of delivery vectors, gRNA library sizes, and tissue preparations, along with two distinct gRNA capture methods, making it adaptable to a wide range of experimental setups. In applying Perturb-DBiT, we conducted un-biased knockouts of tens of genes or at genome-wide scale across three cancer models. We mapped all gRNAs in individual colonies and corresponding transcriptomes in a human cancer metastatic colonization model, revealing clonal dynamics and cooperation. We also examined the effect of genetic perturbation on the tumor immune microenvironment in an immune-competent syngeneic model, uncovering differential and synergistic perturbations in promoting immune infiltration or suppression in tumors. Perturb-DBiT allows for simultaneously evaluating the impact of each knockout on tumor initiation, development, metastasis, histopathology, and immune landscape. Ultimately, it not only broadens the scope of genetic inquiry, but also lays the groundwork for developing targeted therapeutic strategies.

Keywords: Spatial omics; clonal dynamics; genetic perturbation; genome-wide CRIPR library; in vivo CRISPR screen; tumor colonialization; tumor immune microenvironment; whole transcriptome.

Figure 1:. Perturb-DBiT overview and technical performance

(A) Schematic of Perturb-DBiT and its applications. (B) Violin plot of detected gene/UMI counts per spatial pixel from gene expression data across 7 samples utilizing Perturb-DBiT co-profiling. (C) Table of fourteen samples utilizing Perturb-DBiT co-profiling and Perturb-DBiT sgRNA capture in different tissue types, resolutions, viral delivery methods, embeddings, and sgRNA library sizes. (D) Bubble plot depicting Perturb-DBiT sgRNA capture diversity. (E) Table depicting reproducibility of Perturb-DBiT direct capture across five adjacent samples.

Figure 2:. Perturb-DBiT demonstrates robust performance and reveals tumor clonal heterogeneity in small and medium-sized CRISPR-screening libraries

(A) Top: Schematic of the dual sgRNA CRISPR KO retroviral vector used for Perturb-DBiT with the small (2 and 86 sgRNA) libraries. Bottom: Schematic of the AAV-CRISPR vector containing two sgRNA expressing cassettes used for Perturb-DBiT with the medium (288 sgRNA library). (B) Small library guide detection (86 sgRNAs) in Cas9-expressing tumor cells in a tumor tissue using a library of 38 epigenetic regulators and 10 non-targeting controls (NTC). Top: Top sgRNA hit spatial plot and adjacent H&E image corresponding to the region of interest. Bottom: Top sgRNAs are presented with points atop a 2D density map (sgRNA1 and sgRNA2 overlap = blue-to-gold). (C) Detection of a medium-sized guide library (288 sgRNAs) in a murine model of autochthonous liver cancer. Left: Liver tissue stained with hematoxylin and eosin (H&E), demonstrating the ROI used for Perturb-DBiT. Right: spatial clustering of gene expression data overlaid on brightfield image, labeled with pathologist annotations. (D) Left: Spatial visualization of the top gene perturbations, depicted by color-coded 2D-density maps atop an autochthonous liver tumor ROI. Right: Bar plot of the top 10 gene perturbations detected by 50-micron Perturb-DBiT. The area and total counts of the gene perturbations are presented by bars and blue line, respectively. (E) Left: UMAP dimensional reduction embedding of gene expression, leading to distinct spatial clusters. Right: Distribution of detected gene/UMI counts per spatial pixel from gene expression data. (F) Heat map of differentially expressed (DE) genes between GEX pixel clusters. The top 15 expressed DE genes are shown for each cluster with z-score values, arranged by hierarchical clustering of genes, pixels, and clusters. (G) Left: Recapitulation analysis comparing sgRNA capture accuracy between Perturb-DBiT and the traditional pooled sequencing method (MIP). Right: Receiver-operator curve comparing the sgRNA detection by 50-micron Perturb-DBiT vs the traditional pooled sequencing method (MIP).

Figure 3.. Genome-wide Perturb-DBiT mapping reveals high-resolution tumor clonality by integrating with histology

(A) Spatial mapping of tissue sections collected from HT29 lung metastatic colonization model transduced with genome-wide Brunello sgRNA library. Left: H&E staining of an adjacent section with region of interest (ROI) indicated with blue square. Right: combined spatial distribution of top 10 sgRNAs. (B) Spatial colocalization of top enriched sgRNA pairs. (C) Unsupervised clustering of the combined exonic and intronic expression matrix revealed 20 transcriptomic clusters. (D) Distribution of detected gene/UMI counts per spatial pixel from reads mapped to exonic or intronic region. (E) Spatial mapping of the lung section using 20 μm device covering 100,000 pixels. Left top: H&E of an adjacent section and the UMAP clustering analysis. Left bottom: H&E of the enlarged ROI region. Right: Spatial UMAP showing 13 transcriptomic clusters. (F) Spatial distribution of cluster 1 overlapped with the H&E image, and the top genes defining this cluster. (G) Super-resolved spatial clustering of the top 16 sgRNA profile and representative sgRNA expression enhanced with iStar. These sgRNAs exhibit cluster-specific imputed expression. See also Figures S3.

Figure 4.. Perturb-DBiT reveals distinct tumor suppressor/promoter programs

(A) Colony formation assay results of top enriched sgRNA from HT29 lung metastatic colonization model. (B) Hazard Ratio plot of top enriched sgRNA from HT29 lung metastatic colonization model. Squares surrounded by colored boxes represent significant differences in hazard ratios. (C) UMAP visualization of pixels based on the dimensional reduction of Mixscape perturbation scores (pUMAP). Non-perturbed (NT) pixels were included in the analysis to serve as a control/reference point. Additionally, pixels were clustered (shown by color and outline) then named by the major perturbations of the cluster (>= 20% of clustered pixels). One cluster with NT and 21 minor perturbations (< 20% of cluster) was named “NT+21”. (D) Violin plot of the expression of 5 selected oncogenes, compared between pUMAP clusters. (E) Monocle3 pseudotime analysis of sgRNA detected from HT29 lung metastatic colonization model. (F) Top: Pseudotime analysis of top 5 enriched sgRNAs detected by Perturb-DBiT. Bottom: spatial mapping of top 5 enriched sgRNAs detected by Perturb-DBiT, overlaid on the brightfield image of the ROI. (G, H) Volcano plots of the transcriptomic DEG analysis for top enriched sgRNA perturbations and the top 8 most significant DE genes were labeled for selected analyses and gene ontology analysis (H).

Figure 5.. Genome-wide Perturb-DBiT highlights genes that modulate the structural features of tumors and TME

(B) H&E stained image of E0771 syngeneic lung fresh-frozen tissue section with labeling of four anatomically distinct tumor regions and adjacent lymph nodes. (C) Unsupervised clustering of the gene expression matrix revealing 6 distinct clusters overlain on brightfield image of tissue section (right). (D) Spatial intensity maps of four selected genes Orc2, Cdh13, S13gal4, Pvt1 overlain on brightfield image of tissues section. (E) Spatial intensity plot of top 6 sgRNA hits revealed by Perturb-DBiT overlain on brightfield image of tissue section. (F) Individual spatial intensity plots of top 6 sgRNA hits. (G) Left: ligand-receptor interactions within each of the 6 clusters elucidated by unsupervised clustering of gene expression data. Right: spatial maps of four significant ligand-receptor interaction analyses. (H) CODEX staining (26 marker panel) is performed on the adjacent tissue section. Left: Pan B cell markers (Cd19, IgM, and IgD), tumor-specific markers (GFP, Ki67, and PD-L1) and vasculature (Cd31). Middle panel, top: background of GFP-positive (tumor) cells and T cells (Cd3+, Cd4+ helper T cells and Cd3+ Cd8a+ cytotoxic T cells). Middle panel, bottom: Tumor cells showing positivity for Pd-l1. Top right: zoomed in image with green circle representing infiltrating region of Cd8+ Tcells and the tumor. Bottom right: zoomed in image with yellow line representing the border of spatial transcriptomics cluster 1 and cluster 2 within the tumor region. (I) UMAP clustering of the CODEX data revealed 17 distinct protein clusters (J) Heatmap showing the top differentially expressed proteins for each cluster