AAV-mediated direct in vivo CRISPR screen identifies functional suppressors in glioblastoma

Abstract

A causative understanding of genetic factors that regulate glioblastoma pathogenesis is of central importance. Here we developed an adeno-associated virus-mediated, autochthonous genetic CRISPR screen in glioblastoma. Stereotaxic delivery of a virus library targeting genes commonly mutated in human cancers into the brains of conditional-Cas9 mice resulted in tumors that recapitulate human glioblastoma. Capture sequencing revealed diverse mutational profiles across tumors. The mutation frequencies in mice correlated with those in two independent patient cohorts. Co-mutation analysis identified co-occurring driver combinations such as B2m-Nf1, Mll3-Nf1 and Zc3h13-Rb1, which were subsequently validated using AAV minipools. Distinct from Nf1-mutant tumors, Rb1-mutant tumors are undifferentiated and aberrantly express homeobox gene clusters. The addition of Zc3h13 or Pten mutations altered the gene expression profiles of Rb1 mutants, rendering them more resistant to temozolomide. Our study provides a functional landscape of gliomagenesis suppressors in vivo.

Figure 1. Autochthonous brain tumorigenesis induced by an AAV-mediated CRISPR library

(a) Schematics of direct in vivo AAV-CRISPR GBM screen design. Top panel, AAV-mTSG library design, synthesis and production. Bottom panel, stereotaxic injection of AAV library and subsequent analysis. HPF, hippocampus; LV, lateral ventricle. (b) MRI sections show brain tumors in AAV-mTSG injected mice, but not in matching sections from PBS or AAV-vector injected mice. Arrowheads indicate brain tumors. Scale bar, 5 mm. (c) MRI-based volumetric quantification of time-matched tumor size ± s.e.m. Two-tailed Welch’s t-test, t₁₇ = 2.62, p = 0.018, mTSG vs. vector or PBS (PBS, n = 2 mice; Vector, n = 6; mTSG, n = 18). (d) Kaplan-Meier curves for overall survival (OS) of mice injected with PBS (n = 5), AAV-vector (n = 24) or AAV-mTSG library (n = 56). OS for PBS and vector groups are both 100%, where the curves are dashed and slightly offset for visibility. Log-rank (LR) test, p < 2.20 * 10⁻¹⁶, mTSG vs. vector or PBS.

Figure 2. AAV-mTSG induced brain tumors recapitulate pathological features of GBM

(a) Top panel, representative H&E brain sections from PBS, AAV-vector and AAV-mTSG injected mice. Arrowheads indicate brain tumors. Scale bar = 1 mm. Lower panels, representative images of brain sections from PBS, AAV-vector and AAV-mTSG injected mice stained by Cas9, GFP, GFAP, and Ki67 immunohistochemistry. Cas9 IHC, red arrowheads indicate Cas9-positive cells; GFP IHC, red arrowheads indicate GFP-positive cells; GFAP IHC, representative GFAP-positive astrocytes in PBS, AAV-vector and AAV-mTSG injected mice (blue arrows), as well as representative cancer cells in AAV-mTSG injected mice (red arrowheads); Ki67 IHC, arrowheads indicate representative proliferative cells, which are mostly in tumors (AAV-mTSG) or scattered in tumor-adjacent brain regions (AAV-mTSG). Scale bar = 0.25 mm. (b) Quantification of tumor sizes ± s.e.m. found in H&E brain sections from PBS, AAV-vector and AAV-mTSG injected mice. Two-tailed Welch’s t-test, t₁₀ = 3.97, p = 0.003, mTSG vs. vector or PBS (PBS, n = 3; Vector, n = 7; mTSG, n = 11). (c) Representative higher magnification H&E images showing pathological features of AAV-mTSG induced brain tumors. Clockwise from top left: yellow arrowheads, giant aneuploid cells with pleomorphic nuclei; blue arrows, endothelial cells and angiogenesis; green arrows, hemorrhagic regions; black arrowheads, necrotic regions. Similar features were observed in human GBM patient sections from Yale Glioma tissue bank (Figure S3). Scale bar = 0.5 mm.

Figure 3. Targeted captured sequencing of sgRNA sites in AAV-mTSG induced mouse GBM

(a) Indel variants observed at the genomic region targeted by Mll2 sgRNA 4 in representative PBS, AAV-vector, and AAV-mTSG injected mouse brain samples. (b) Bar plots of variant frequencies in significantly mutated sgRNA target regions from two representative AAV-mTSG injected mouse brain samples. (c) Heatmap of variant frequency across all targeted capture samples (n = 41). Rows denote individual sgRNAs, while columns correspond to samples from mice stereotaxically injected with PBS, AAV-vector, or AAV-mTSG. The liver was considered an off-target organ and thus was used as a background control. Bar plots of the mean variant frequencies for each sgRNA (right panel, orange bars) and each sample (bottom panel, purple bars) are shown. (d) Dot plot of mean variant frequency ± s.e.m., grouped by treatment condition and tissue type. AAV-mTSG injected brains had significantly higher mean variant frequencies (2.087 ± 0.429, n = 25) compared to vector (0.005 ± 0.001, n = 3) or PBS (0.003 ± 0.001, n = 4) injected brains (two-tailed Welch’s t-test, t₂₄ = 4.85 and t₂₄ = 4.86, p = 6.03 * 10⁻⁵ and p = 5.96 * 10⁻⁵ for mTSG vs. vector and mTSG vs. PBS). Comparing brain vs. liver in AAV-mTSG injected mice, mean variant frequencies of brains (2.087 ± 0.429) were significantly higher than livers (0.309 ± 0.261, n = 4) (t_21.48 = 3.54, p = 0.002). (e) Indel size distribution for all filtered variants in each mTSG brain sample (n = 25).

Figure 4. Integrative analysis of functional mutations in driving tumorigenesis

(a) Gene-level mutational landscape of AAV-mTSG induced primary mouse GBM. Center: Tile chart depicting the mutational landscape of primary brain samples from LSL-Cas9 mice injected with the AAV-mTSG library (n = 25), AAV-vector (n = 3) or PBS (n = 4). Genes are grouped and colored according to their functional classifications as noted in the top-right legend. Top: Bar plots of the total number of significantly mutated genes identified in each AAV-mTSG sample. Right: Bar plots of the percentage of GBM samples that were called as significantly mutated for each gene. Left: Heatmap of the numbers of unique significantly mutated sgRNAs (SMSs) for each gene. Bottom: Stacked bar plots describing the type of indels observed in each sample, colored according to the bottom-right legend. (b–c) Comparative cancer genomics in GBM using the TCGA (b) and Yale Glioma (c) datasets. Scatterplot of population-wide mutant frequencies for the genes in the mTSG library, comparing AAV-mTSG treated mouse brain samples to human samples. Representative strong drivers in both species are labeled, with gene names color-coded based on their functional classification (as in a). (b) Mutant frequencies in AAV-mTSG treated mouse brain samples correlated with patients in the TCGA GBM dataset (Pearson correlation R = 0.402, p = 0.002). (c) Mutant frequencies in AAV-mTSG treated mouse brain samples correlated with patients in the Yale Glioma dataset (Yale Glio) (Pearson correlation R = 0.318, p = 0.028).

Figure 5. Co-mutation analysis uncovers synergistic gene pairs in GBM

(a) Upper-left half: heatmap of pairwise mutational co-occurrence rates. Lower-right half: heatmap of -log₁₀ p-values by hypergeometric test for statistical co-occurrence. (b) Scatterplot of the co-occurrence rate of each gene pair, plotted against −log₁₀ p-values. (c) Venn diagrams showing representative strongly co-occurring mutated gene pairs such as Kdm5c and Gata3 (co-occurrence rate = 77.8%, hypergeometric test, p = 6.04 * 10⁻⁶), B2m and Pik3r1 (70.0%, p = 2.28 * 10⁻⁵), as well as Nf1 and Pten (85.7%, p = 7.53 * 10⁻⁸). (d) Upper-left half: heatmap of the pairwise Spearman correlation of variant frequency for each gene, summed across sgRNAs. Lower-right half: heatmap of −log₁₀ p-values to evaluate the statistical significance of the pairwise correlations. (e) Scatterplot of pairwise Spearman correlations plotted against −log₁₀ p-values. (f–g) Scatterplots showing representative strongly correlated gene pairs when comparing variant frequencies summed across sgRNAs, such as Nf1+Pten (f) and Cdkn2a+Ctcf (g). Spearman correlation coefficients are noted on the plot.

Figure 6. Validation of driver combinations

(a) Schematic representation of experiment design. Mixtures of five sgRNAs targeting each gene were cloned as minipools into the astrocyte-specific AAV-CRISPR vector. After packaging, AAV minipools were stereotaxically injected into the lateral ventricle of LSL-Cas9 mice. (b–e) End-point histology (H&E) of representative brain sections from mice treated with AAV sgRNA minipools or relevant controls. In this end-point analysis, mice were euthanized when either macrocephaly or poor body condition score (< 2) was observed, with survival time ranging from 3 to 11 months. Treatments are indicated to the left of each image. Arrowheads indicate the presence of brain tumors. The proportion of tumor-bearing to total mice is indicated in the top right corner of the images. Scale bar = 0.5 mm. (b) Representative histology of brain sections from control mice. No tumors were observed in mice from the vector (0/3), EYFP (0/4) or uninjected (0/2) groups. (c) Representative histology of brain sections from mice treated with various Nf1 minipools, such as Nf1 alone (4/8 mice developed tumors within 11 months), Nf1;Pten (9/9 mice developed tumors within 6 months), Nf1;Mll3 (2/5 mice developed tumors within 6 months), and Nf1;B2m (4/4 mice developed tumors within 11 months). (d) Representative histology of brain sections from mice treated with various Rb1 minipools, such as Rb1 alone (3/3 mice developed tumors within 6 months), Rb1;Zc3h13 (3/3 mice developed tumors within 6 months), and Rb1;Pten (3/3 mice developed tumors within 6 months). (e) Representative histology of brain sections from mice treated with other minipools, such as Mll2 alone (2/10 mice developed tumors within 11 months), Setd2 (1/5 mice developed tumors within 6 months), and Cic (1/5 mice developed tumors within 6 months).

Figure 7. Transcriptional profiling of mouse GBM driver combinations

(a) Schematic of mouse GBM RNA-seq experimental design. Rb1 or Nf1 AAV minipools were stereotaxically injected into the lateral ventricle of LSL-Cas9 mice. Cell lines were derived from mouse GBMs by single-cell isolation. Additional driver mutations were introduced by lentiCRISPR where applicable. Cells were then transcriptionally profiled by RNA-seq (n = 3 samples per condition). (b) Volcano plot comparing gene expression profiles in Rb1 mutant (red) to Nf1 mutant (blue) GBM cells. 616 genes were significantly higher in Rb1 cells, and 982 genes were significantly higher in Nf1 cells (c) Enriched gene ontology categories among Nf1-high genes. (d) Enriched gene ontology categories among Rb1-high genes. (e) Volcano plot comparing Nf1;Mll3 mutant (orange) to Nf1 mutant (blue) GBM cells. 522 genes were significantly higher in Nf1;Mll3 cells, and 175 genes were significantly higher in Nf1 cells. (f) Enriched gene ontology categories among Nf1;Mll3-high genes. (g) Volcano plot comparing Rb1;Zc3h13 mutant (green) to Rb1 mutant (red) GBM cells. 703 genes were significantly higher in Rb1;Zc3h13, and 166 genes were significantly higher in Rb1 cells. (h) Enriched gene ontology categories among Rb1;Zc3h13-high genes. Differentially expressed genes were defined as Benjamini-Hochberg adjusted p < 0.05 and log fold change ≥ 1 or ≤ −1.

Figure 8. Transcriptional profiling of mouse GBM driver combinations in the presence and absence of a chemotherapeutic agent

(a) Schematic of drug treatment RNA-Seq experimental design. Rb1, Rb1;Pten, and Rb1;Zc3h13 GBM cells were treated with either temozolomide (TMZ) or DMSO. Cells were then subjected to phenotypic analysis and RNA-seq. (b–c) Survival fraction ± s.e.m of Rb1 (blue), Rb1;Pten (purple), and Rb1;Zc3h13 (green) cells with 1 mM (b) or 2 mM (c) TMZ treatment. Individual cell replicates are shown (n = 3 for all conditions). (b) Rb1;Pten and Rb1;Zc3h13 cells had significantly higher survival fractions than Rb1 cells with 1 mM TMZ (two-sided t-test,p = 6.20 * 10⁻⁶ and p = 1.94 * 10⁻⁵, respectively). (c) Rb1;Pten and Rb1;Zc3h13 cells had significantly higher survival fractions than Rb1 cells with 2 mM TMZ (two-sided t-test, p = 9.06 * 10⁻⁷ and p = 2.84 * 10⁻⁶, respectively). Individual cell replicates are shown (n = 3 for all conditions). (d) Volcano plot comparing Rb1 cells treated with TMZ (dark blue) or DMSO (blue). 352 genes were significantly higher in TMZ-treated cells (TMZ-induced genes), and 332 genes were significantly higher in DMSO-treated cells (TMZ-reduced genes). (e) Volcano plot comparing Rb1;Pten cells treated with TMZ (pink) or DMSO (purple). 345 genes were significantly higher in TMZ-treated cells, and 313 genes were significantly higher in DMSO-treated cells. (f) Volcano plot comparing Rb1;Zc3h13 cells treated with TMZ (turquoise) or DMSO (green). 703 genes were significantly higher in TMZ-treated cells, and 166 genes were significantly higher in DMSO-treated cells (g) Heatmap of all differentially expressed genes among the TMZ vs. DMSO comparisons. Clustering was performed by average linkage using Pearson correlations. Values are shown in terms of z-scores, scaled by each gene. (h) Venn diagram of TMZ-reduced genes for each tested genotype. While 69 genes were similarly downregulated among all 3 genotypes upon TMZ treatment, the differential expression signatures were nevertheless distinct, suggesting differential responses to TMZ treatment. (i) Venn diagram of TMZ-induced genes for each tested genotype. Though 42 genes were consistently upregulated in all 3 groups upon TMZ treatment, numerous transcriptional differences were nevertheless apparent, suggesting differential responses to TMZ treatment. Differentially expressed genes were defined as Benjamini-Hochberg adjusted p < 0.05 and log fold change ≥ 1 or ≤ −1.