Fig 1.
Functional genomic data generated and analysis workflow.
Flow-chart including transcriptome, proteome and exome data generated and integrative analysis of these data sets. 12 cell lines (not drug-treated) were characterized using exome sequencing (1), gene expression arrays (2), and reverse phase protein arrays (3). Variant calling and enriched gene variants were identified by cross referencing our results with mutated genes associated with cancer identified by Lawrence et al [20]. Cell lines were exposed to PLX4720 and their responses were assayed with gene expression arrays and reverse phase protein arrays. Protein response to treatment was correlated with cytotoxic effects of PLX4720 treatment to identify proteins that might be mitigating the cytotoxic response. Cytotoxicity (Cytotox) groups were identified by clustering the cytotoxicity data. Differential gene expression responses to treatment within each cytotoxicity group were identified and underwent both gene set and pathway enrichment analysis via MSigDB and Pathway Express respectively. Pathway enrichment analysis revealed ErbB signaling as a key response to treatment and gene set enrichment analysis revealed a number of transcription factors that are enriched which putatively regulate ErbB signaling pathway genes.
Fig 2.
Clustering and PCA analysis of basal gene expression reveals MITF expression and gene regulation separates melanoma cell lines.
We performed a one-way analysis of variance (ANOVA) and applied a 0.1% false discovery rate (FDR) cutoff to identify genes with significantly varying expression levels, and the genes (y-axis) and cell lines (x-axis) were organized by unsupervised hierarchical clustering. Clustering the genes using the correlation distance and average linkage yielded 91 gene clusters with the number of genes in each cluster ranging from 1 to 73 when applying a cluster height cutoff of 0.41. Hierarchical clustering of samples (x-axis) by genes (y-axis) of gene expression for significantly varying genes across twelve melanoma cell lines (A). The largest cluster (cluster IV), which contains MITF along with three others that contain cell line specific signatures (clusters I, II and III) are highlighted as they were the only clusters with greater than 20 genes. Cluster IV genes were analyzed using Ingenuity Pathway Analysis revealing the cluster contains MITF as a regulatory hub and its target genes (B). Principal component plot of basal expression across the twelve cell lines (C) separates cells according to MITF expression along the first principal component (i.e., decreasing MITF expression going from left to right) and Bliss score across the second principal component.
Fig 3.
Clustering by cytotoxicity reveals differential responses to PLX4720 treatment.
Clustering of cell lines (x-axis) according to cytotoxic responses to PLX4720, lapatinib, and combination treatment at three doses for each individual drug and a three by three dose response for the combination (y-axis) (A). Five groups were identified: CGA (most left, 3 lines), CGB (near left, 1 line), CGC (center, 4 lines), CGD (near right, 2 lines), and CGE (far right, 2 lines). Pathway enrichment analysis of differential gene expression to PLX4720 treatment identified ERBB signaling as being broadly activated in CGC (B). Refer to S6 Table for fold change values used to generate pathway diagram.
Fig 4.
Exome analysis reveals trend between zygosity of V600E allele in the BRAF locus and sensitivity to PLX4720.
Ranked IC50 values for BRAFV600E melanomas from our panel and the CCLE. For lines in our panel that were also in the Cancer Cell Line Encyclopedia [26] (CCLE), we used the IC50 values from the CCLE. Lines from our panel are denoted with “*”. Zygosity at the BRAF locus was assayed using IGV visualization of BAM files. Lines found to be homozygous for the V600E allele were colored blue, lines found to be heterozygous for the V600E allele were colored red, and lines with only wild type BRAF were colored green (A). Box plots of IC50 values for BRAFV600E/V600E (Homozygous), BRAFWT/V600E (Heterzygous), and BRAFWT/WT (Wild Type), melanoma cell lines (B). Exome sequencing identifies potential causal variants. Variants present in our lines were compiled using exome sequencing and cross referenced with genes identified as being mutated across multiple cancer types [20]. BRAF zygosity, genes uniquely mutated in each drug cytotoxicity group, and the genes uniquely mutated in each line are listed (C).
Fig 5.
Transcription factors associated with ErbB pathway genes upregulated in combination sensitive cell lines treated with PLX4720.
Unsupervised clustering of gene expression response (log2 fold change) to PLX4720 treatment (y-axis) of the ErbB signaling pathway genes of each cytotoxicity group (left heat map). Transcription factors (x-axis) that have binding motifs in the promoters of ERBB signaling pathway genes (black squares in binary matrix on right). Genes were selected for plotting based on passing the FDR corrected P Value threshold of 1% or better in at least one cytotoxicity group. Genes that demonstrated a FDR corrected P Value of 1% or better are denoted with an “*”.
Fig 6.
Reverse phase protein array reveals downregulation of RTK signaling pathways and a compensatory response of ErbB4 to PLX4720 treatment.
Heat map of protein phosphorylation changes at 5% FDR across the four cytotoxicity groups. (A). Epitopes were selected for plotting based on passing the FDR corrected P Value threshold of 1% or better in at least one cytotoxicity group. Epitopes that demonstrated a FDR corrected P Value of 1% or better are denoted with an “*”. Correlating the response to treatment from the protein array with the cytotoxicity identified ErbB4 total protein fold changes as being the most anti correlated (B).