A Pan-Cancer Catalogue of Cancer Driver Protein Interaction Interfaces

Abstract

Despite their importance in maintaining the integrity of all cellular pathways, the role of mutations on protein-protein interaction (PPI) interfaces as cancer drivers has not been systematically studied. Here we analyzed the mutation patterns of the PPI interfaces from 10,028 proteins in a pan-cancer cohort of 5,989 tumors from 23 projects of The Cancer Genome Atlas (TCGA) to find interfaces enriched in somatic missense mutations. To that end we use e-Driver, an algorithm to analyze the mutation distribution of specific protein functional regions. We identified 103 PPI interfaces enriched in somatic cancer mutations. 32 of these interfaces are found in proteins coded by known cancer driver genes. The remaining 71 interfaces are found in proteins that have not been previously identified as cancer drivers even that, in most cases, there is an extensive literature suggesting they play an important role in cancer. Finally, we integrate these findings with clinical information to show how tumors apparently driven by the same gene have different behaviors, including patient outcomes, depending on which specific interfaces are mutated.

Fig 1. Using e-Driver to analyze PIK3R1 PPI interfaces.

a) The PDB coordinate set 3HMM contains 2 chains, A (a region of PIK3CA, shown in gray) and B (a region of PIK3R1, shown in brown). Residues from PIK3R1 colored in red make the interaction interface with PIK3CA. b) Using BLAST we then map these residues to the corresponding ENSEMBL protein, define a PPI interface (shown in red). Note that the interface is not continuous in sequence. c) Distribution of mutations in PIK3R1 across all cancer types. d) Using e-Driver we can identify the interface between PIK3R1 and PIK3CA as strongly enriched in missense somatic mutations (p < 2e-6).

Fig 2. Genes with driver interfaces.

X,Y coordinates reflect the q value (FDR) in the Pancancer analysis and their lowest q value in all 23 project-specific analysis, respectively. Gray dashed lines are located at 0.01 FDR for reference. Dots are colored, sized and labeled according to their FDR: genes with an FDR of 1 are colored in white, are smaller, and have no label whereas genes with lower FDRs are redder, bigger, and labeled. Note that there are 8 genes that are not in the plot because their FDR value was too small (FDR < 1e-15): TP53, EGFR, KRAS, HRAS, NRAS, RPSAP58, PIK3CA and UBBP4. Some driver interfaces are also illustrated in the plot, with the chain belonging to the gene with the driver interface colored in orange and the interacting chain colored in blue. The PDB coordinates used are 3IFQ for CTNNB1, 1S78 for ERBB2, 2FLU for NFE2L2, 3B13 for RAC1 and 3FGA for PP2R1A.

Fig 3. Network properties of interface-driver genes.

a) Most genes identified by e-Driver (shown in red) cluster together with other cancer driver genes (shown in light blue). Genes are sized according to their number of interactions (genes with more interactions are bigger and vice versa). Note that genes identified by e-Driver that are also known drivers (shown in purple) tend to have many interactions with other cancer genes. Edges are weighted according to the number of networks where they can be identified, with thicker edges being found in more networks, and vice versa. b) Interface driver genes have similar network properties when compared to known cancer driver genes. Both groups of genes have higher degree (left) and betweenness (right) than the rest of the genes in the genome. c) Known drivers identified by e-Driver (in purple) have higher degree and betweenness than other drivers (known-in blue- or predicted-in red-) or the rest of the genes in the genome (shown in light gray). Statistics in panels (b) and (c) are based on network properties from the network “Vidal merged”. The statistics for the all networks are in Tables AI and AJ in S1 Table.

Fig 4. Protein expression changes induced by mutations in EGFR dimerization interface.

a) Structure showing an EGFR dimer in active conformation (based on[78]). The residues involved in the EGFR-EGFR interaction for the left EGFR molecule are shown in red, the other EGFR protein is shown in blue, the two EGF ligands are shown in green and the lipid bilayer in brown. b) Histogram showing that most glioblastoma mutations (black bars) in EGFR are located in its dimerization interface (red bars). c) Protein expression in different glioblastoma populations. Patients with mutations in EGFR’s dimerization interface (shown in red) have higher levels of EGFR and phosphorylated EGFR proteins than those with other or no EGFR mutations (shown in green and blue respectively).

Fig 5. Mutations in TP53 interface predict patient's outcomes.

a) Our analysis identified the interface between TP53 and SV40 (shown in gray and orange respectively) as strongly enriched in mutations (panel c). b) This interface is the same as the one that TP53 uses to interact with other TP53 molecules (shown in dark blue) and DNA (shown in light blue). d) Breast cancer patients with mutations in this interface (shown in red) have worse outcomes than those with other or no TP53 mutations (shown in blue and red respectively).