Research ArticleCANCER

Mapping a functional cancer genome atlas of tumor suppressors in mouse liver using AAV-CRISPR–mediated direct in vivo screening

See allHide authors and affiliations

Science Advances  28 Feb 2018:
Vol. 4, no. 2, eaao5508
DOI: 10.1126/sciadv.aao5508
  • Fig. 1 AAV-CRISPR mTSG library rapidly induces robust liver tumorigenesis in LSL-Cas9 mice.

    (A) Schematics of the overall design and experimental outline. First, the top MGs were identified from pan-cancer TCGA data sets. After removing known oncogenes and genes without mouse orthologs, a set of 49 most recurrently mutated putative TSGs were chosen (mTSG). Seven additional genes with housekeeping functions were spiked-in, leading to a final set of 56 genes. sgRNAs targeting these genes were then identified computationally, and five were chosen for each gene. Two hundred eighty sgRNAs plus 8 NTC sgRNAs were synthesized, and the sgRNA library (mTSG; 288 sgRNAs) was cloned into an expression vector that also contained Cre recombinase and a Trp53 sgRNA. AAVs carrying the mTSG library were produced and injected into the tail veins of LSL-Cas9 mice. After a specified time period, the mice were subjected to MRI, histology, and MIP capture sequencing for readout and deep variant analysis of molecular landscape of all targeted genes and mutations. (B) MRI of abdomens of mice treated with PBS, vector, or mTSG library. Detectable tumors are circled with green dashed lines. PBS-treated mice (n = 3) did not have any detectable tumors, whereas vector-treated mice (n = 3) occasionally had small nodules. In contrast, mTSG-treated mice (n = 4) often had multiple detectable tumors. (C) Kaplan-Meier survival curves for PBS-treated (purple, n = 10), vector-treated (teal, n = 11), and mTSG-treated (orange, n = 27) mice. No mTSG-treated mice survived longer than 4 months after treatment, whereas all PBS- and vector-treated animals survived the duration of the experiment. Statistical significance was assessed by the log-rank test (P = 1.8 × 10−11). (D) Bright-field images with GFP fluorescence overlay (green) of livers from representative PBS-, vector-, and mTSG-treated mice 4 months after treatment. Large GFP+ tumors are marked with yellow arrowheads. In contrast to PBS- or vector-treated mice, mTSG-treated mice had numerous detectable GFP+ nodules.

  • Fig. 2 Histology analysis of autochthonous tumors generated by AAV-CRISPR mTSG library.

    (A) Hematoxylin and eosin (H&E) staining of liver sections from mice treated with PBS (n = 7), vector (n = 5), or mTSG library (n = 13). Tumor-normal boundaries are demarcated with yellow dashed lines. No tumors were found in PBS samples, whereas small nodules were found, although rare, in vector samples. On the other hand, mTSG-treated livers were replete with tumors [statistics in (B) and (C)]. (B) Dot plot of the total tumor area per mouse (mm2) in liver sections from mice treated with PBS (black, n = 7), vector (gray, n = 5), or mTSG library (purple, n = 13). mTSG-treated mice had a significantly higher total tumor burden than PBS-treated (one-sided Welch’s t test, P = 0.027) or vector-treated mice (P = 0.034). (C) Dot plot of the individual tumor area (mm2) in liver sections from mice treated with PBS (black, n = 7), vector (gray, n = 9), or mTSG library (purple, n = 49). mTSG-treated mice had significantly larger tumors than PBS-treated (one-sided Welch’s t test, P < 0.0001) or vector-treated mice (P = 0.0003). (D) Representative immunohistochemical staining of an LIHC marker, pan-cytokeratin (AE1/AE3), from mice treated with PBS, vector, or mTSG library. The tumors from mTSG-treated samples shown revealed positive staining for AE1/AE3, consistent with LIHC pathology. Certain mTSG tumors were partially positive for cytokeratin, revealing tumor heterogeneity. The tumors from vector-treated samples were relatively small and almost always negative or slightly positive for cytokeratin. Scale bar, 0.5 mm.

  • Fig. 3 Mutational variant-level mutational landscape of mouse AAV-mTSG–induced LIHC.

    (A) Unique variants observed at the genomic region targeted by Setd2 sg1 in representative PBS-, vector-, and mTSG-treated liver samples. The percentage of total reads that correspond to each genotype is indicated on the right in the blue boxes. No indels were found in the PBS- or vector-treated samples, whereas several unique variants were identified in the mTSG-treated sample (mTSG 042). (B) Waterfall plots of two mTSG-treated samples (042 and 066) detailing sum variant frequencies in mutated sgRNA sites (MSs). Individual mice presented with distinct mutational signatures, suggesting that a wide variety of mutations induced by the mTSG library had undergone positive selection. (C) Global heat map detailing the square root of sum variant frequency across all sequenced samples (n = 133) from mTSG-treated (n = 98 samples), vector-treated (n = 21 samples), or PBS-treated mice (n = 14 samples) in terms of sgRNAs. Square root transformation was used to even out the distribution of variant frequencies for visualization. Each row represents one sgRNA, whereas each column represents one sample. Treatment conditions and tissue type are annotated at the top of the heat map: BAT (dark purple), detectable tumor outside liver (light purple), liver (teal), brain (light pink), gastrointestinal (GI; dark pink), lung (brown), and other organs (gray). Bar plots of the mean square root variant frequencies for each sgRNA (right, green bars) and each sample (bottom, purple bars) are also shown. mTSG-treated organs without visible tumors (0.148 ± 0.037 SEM) had significantly lower mean variant frequencies compared to mTSG-treated tumors and livers (BATs, 3.098 ± 0.600; two-sided unpaired t test, P < 0.0001), non-liver tumors (1.919 ± 0.338; P < 0.0001), and livers (1.451 ± 0.203; P < 0.0001). Livers and other organs from vector-treated animals (0.398 ± 0.179 and 0.054 ± 0.004, respectively) and PBS-treated animals (0.140 ± 0.067 and 0.063 ± 0.021, respectively) all had significantly lower variant frequencies than mTSG-treated livers (P < 0.0001 for all comparisons).

  • Fig. 4 Mouse gene-level mutational landscape of LIHC.

    Each row in the figure corresponds to one gene in the mTSG library, whereas each column corresponds to one mTSG-treated liver sample. (Top) Bar plots of the total number of MGs identified in each mTSG-treated liver sample (n = 37). Samples originating from the same mouse are grouped together and denoted with a gray bar underneath. (Middle) Tile chart depicting the mutational landscape of primary liver samples infected with the mTSG library. Genes are grouped and colored according to their functional classifications (DNA repair/replication, epigenetic modifier, cell death/cycle, repressor, immune regulator, ubiquitination, transcription factor, cadherin, ribosome-related, and RNA synthesis/splicing), as noted in the legend in the top-right corner. Colored boxes indicate that the gene was mutated in a given sample, whereas a gray box indicates no significant mutation. Asterisks denote several preselected genes that were generally considered housekeeping genes. (Right) Bar plots of the percentage of liver samples that had a mutation in each of the genes in the mTSG library. Trp53, Setd2, Pik3r1, Cic, B2m, Vhl, Notch1, Cdh1, Rpl22, and Polr2a were the top MGs in each of the 10 functional classifications, respectively. (Bottom) Stacked bar plots describing the type of indels observed in each sample, color-coded according to the legend in the bottom-right corner. Frameshift insertions or deletions comprised the majority of variant reads (median, 59.2% across all samples). (Left) Heat map of the number of mutated sgRNA sites (0 to 5 MSs) for each gene. Multiple mutated sgRNA sites for a given gene are indicative of a strong selective force for loss-of-function mutations in that gene.

  • Fig. 5 Co-mutation analysis of liver samples from mTSG-treated mice reveals potential synergistic combinations of driver mutations.

    (A) Upper-left triangle: Heat map of the co-occurrence rates for each gene pair. To calculate co-occurrence rates, the intersection is defined as the number of double-mutant samples, and the union is defined as the number of samples with a mutation in either of the two genes. The co-occurrence rate was then calculated as the intersection divided by the union. Lower-right triangle: Heat map of −log10 P values by hypergeometric test to evaluate whether specific pairs of genes are statistically significantly co-mutated. (B) Scatterplot of the co-occurrence rates for each gene pair, plotted against −log10 P values by hypergeometric test. Highly co-occurring pairs include Cdkn2a + Pten (co-occurrence rate = 7/10 = 70%; hypergeometric test, P = 2.63 × 10−5), Cdkn2a + Rasa1 (co-occurrence rate = 6/9 = 67%; P = 7.96 × 10−5), Arid2 + Cdkn1b (co-occurrence rate = 11/17 = 65%; P = 9.13 × 10−5), and Kansl1 + B2m (co-occurrence rate = 11/18 = 61%; P = 3.6 × 10−4). (C) Venn diagrams showing the strong co-occurrence of mutations in B2m + Kansl1 (top left), Cdkn2a + Pten (top right), Cdkn2a + Rasa1 (bottom left), and Arid2 + Cdkn1b (bottom right). Numbers shown correspond to the number of mTSG-treated liver samples with a given mutation profile. (D) Upper-left triangle: Heat map of the pairwise Spearman correlation of sum % variant frequency for each gene, summed across sgRNAs. Lower-right triangle: Heat map of −log10 P values by t distribution to evaluate the statistical significance of the pairwise correlations. (E) Scatterplot of pairwise Spearman correlations plotted against −log10 P values. The top four correlated pairs were Cdkn2a + Pten (R = 0.817, P = 6.97 × 10−10), Nf1 + Rasa1 (R = 0.791, P = 5.86 × 10−9), Arid2 + Cdkn1b (R = 0.788, P = 7.16 × 10−9), and Cdkn2a + Rasa1 (R = 0.761, P = 4.45 × 10−8). (F) Scatterplot comparing sum level % variant frequency for Arid2 versus Cdkn1b across all mTSG-treated liver samples. Spearman and Pearson correlation coefficients are noted on the plot (Spearman R = 0.788; Pearson R = 0.746). (G) Heat map of the P values associated with the top mutation pairs that were found to be statistically significant (Benjamini-Hochberg–adjusted, P < 0.05) in both co-occurrence (left) and correlation (right) analyses.

  • Fig. 6 Systematic dissection of variant compositions across individual liver lobes within a single mTSG-treated mouse reveals substantial clonal mixture between lobes.

    (A) Schematic of the experimental workflow for analysis of multiple liver lobes (n = 5) from a single mTSG-treated mouse. (B) Heat map of Spearman rank correlation coefficients among five liver samples from a single mTSG-treated mouse, calculated on the basis of variant frequency for all unique variants present within the five samples. Notably, lobes 1 to 4 are all significantly correlated with lobe 5, with lobe 3 having the strongest correlation to lobe 5. (C) Heat map of variant frequencies for each unique variant identified across the five individual liver lobes after square root transformation. Rows correspond to different liver lobes, whereas columns denote unique variants. Eight clusters were identified based on binary mutation calls and are indicated on the bottom of the heat map. (D) Pie charts depicting the proportional contribution of each cluster to the five liver lobes. In order for a cluster to be considered, at least half of the variants within the cluster must be present in that particular sample. For each lobe, variant frequencies within a cluster were averaged and converted to relative proportions, as shown in the pie charts. The pie charts accurately recapture the correlation analysis in (B) while additionally providing quantitative insight into the shared variants between the five liver lobes. (E) Each box corresponds to one cluster, color-coded as in (C) and (D), showing the top four variants in each cluster. On the basis of whether a variant cluster was present in multiple liver lobes, each box is also classified as either a private or a shared variant cluster. Clusters 1, 2, 3, 5, and 6 are largely unique to individual lobes (private variant clusters), whereas clusters 4, 7, and 8 are present in multiple lobes (shared variant clusters). Cluster 8 was found in four of five lobes and is characterized by mutations in Mll3, Setd2, and Trp53.

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/4/2/eaao5508/DC1

    fig. S1. Representative full-spectrum MRI series of livers from PBS-, vector-, and mTSG-treated mice.

    fig. S2. Additional bright-field images of mTSG-treated livers with GFP overlay.

    fig. S3. Representative full-slide scanning images of mouse liver sections in PBS, vector, and mTSG treatment groups.

    fig. S4. Representative histology and immunohistochemistry images of mouse liver sections in PBS, vector, and mTSG groups.

    fig. S5. MIP capture sequencing statistics and indel size distribution of mTSG livers.

    fig. S6. Mutated sgRNA sites across all liver samples from mice treated with AAV-mTSG library.

    fig. S7. Heat map of gene-level sum variant frequency across all mTSG liver samples.

    fig. S8. Additional co-mutation analysis.

    fig. S9. Heat map of all unique variants across all mTSG liver samples.

    fig. S10. Investigation and comparison of single or combinatorial knockout of screened TSGs in liver tumorigenesis.

    fig. S11. Mutant clonality and clustering analysis.

    table S1. DNA sequences of sgRNA spacers in mTSG library.

    table S2. Raw read counts of mTSG plasmid library.

    table S3. Tumor volume data as measured by MRI.

    table S4. Survival data for PBS-, vector-, or mTSG-treated animals.

    table S5. Tumor area data as measured by tissue histology.

    table S6. Sequence information and annotation for all MIPs used in the study.

    table S7. Metadata for all of the 133 sequenced samples.

    table S8. MIP capture sequencing coverage statistics across all predicted cutting sites of sgRNAs in AAV-mTSG library.

    table S9. Raw indel variant calls of all samples with targeted capture sequencing before filtering.

    table S10. sgRNA-level sum indel frequency table for all samples with targeted capture sequencing.

    table S11. sgRNA-level binary MS calls in livers from mice treated with AAV-mTSG library.

    table S12. Gene-level binary MG calls in livers from mice treated with AAV-mTSG library.

    table S13. Co-occurrence analysis of MG pairs in livers from mice treated with AAV-mTSG library.

    table S14. Correlation analysis of gene-level sum indel frequency in livers from mice treated with AAV-mTSG library.

    table S15. Mutant frequencies for all unique variants present across all mTSG liver samples.

    table S16. Spearman rank correlation matrix for five individual liver lobes within a single mouse.

    table S17. Mutant frequencies for all unique variants present in five individual liver lobes from a single mouse.

  • Supplementary Materials

    This PDF file includes:

    • fig. S1. Representative full-spectrum MRI series of livers from PBS-, vector-, and mTSG-treated mice.
    • fig. S2. Additional bright-field images of mTSG-treated livers with GFP overlay.
    • fig. S3. Representative full-slide scanning images of mouse liver sections in PBS, vector, and mTSG treatment groups.
    • fig. S4. Representative histology and immunohistochemistry images of mouse liver sections in PBS, vector, and mTSG groups.
    • fig. S5. MIP capture sequencing statistics and indel size distribution of mTSG livers.
    • fig. S6. Mutated sgRNA sites across all liver samples from mice treated with AAV-mTSG library.
    • fig. S7. Heat map of gene-level sum variant frequency across all mTSG liver samples.
    • fig. S8. Additional co-mutation analysis.
    • fig. S9. Heat map of all unique variants across all mTSG liver samples.
    • fig. S10. Investigation and comparison of single or combinatorial knockout of screened TSGs in liver tumorigenesis.
    • fig. S11. Mutant clonality and clustering analysis.
    • Legends for tables S1 to S17

    Download PDF

    Other Supplementary Material for this manuscript includes the following:

    • table S1 (Microsoft Excel format). DNA sequences of sgRNA spacers in mTSG library.
    • table S2 (Microsoft Excel format). Raw read counts of mTSG plasmid library.
    • table S3 (Microsoft Excel format). Tumor volume data as measured by MRI.
    • table S4 (Microsoft Excel format). Survival data for PBS-, vector-, or
      mTSG-treated animals.
    • table S5 (Microsoft Excel format). Tumor area data as measured by tissue histology.
    • table S6 (Microsoft Excel format). Sequence information and annotation for all MIPs used in the study.
    • table S7 (Microsoft Excel format). Metadata for all of the 133 sequenced samples.
    • table S8 (Microsoft Excel format). MIP capture sequencing coverage statistics across all predicted cutting sites of sgRNAs in AAV-mTSG library.
    • table S9 (Microsoft Excel format). Raw indel variant calls of all samples with targeted capture sequencing before filtering.
    • table S10 (Microsoft Excel format). sgRNA-level sum indel frequency table for all samples with targeted capture sequencing.
    • table S11 (Microsoft Excel format). sgRNA-level binary MS calls in livers from mice treated with AAV-mTSG library.
    • table S12 (Microsoft Excel format). Gene-level binary MG calls in livers from mice treated with AAV-mTSG library.
    • table S13 (Microsoft Excel format). Co-occurrence analysis of MG pairs in livers from mice treated with AAV-mTSG library.
    • table S14 (Microsoft Excel format). Correlation analysis of gene-level sum indel frequency in livers from mice treated with AAV-mTSG library.
    • table S15 (Microsoft Excel format). Mutant frequencies for all unique variants present across all mTSG liver samples.
    • table S16 (Microsoft Excel format). Spearman rank correlation matrix for five individual liver lobes within a single mouse.
    • table S17 (Microsoft Excel format). Mutant frequencies for all unique variants present in five individual liver lobes from a single mouse.

    Download Tables S1 to S17

    Files in this Data Supplement:

Navigate This Article