Figure 1.
Analysis at the functional region level allows us to gain novel insights from pharmacogenomics data.
(a, b) Mapping of the different ERBB3 functions to specific regions of the protein. Each functional relationship can be associated to a specific domain or intrinsically disordered region in ERBB3. For example, EGF receptor domains (red boxes in (b)) mediate physical interactions between ERBB3 and EGFR and NRG1 (red edges in (a)). (c) Methods focusing at the whole-protein level can not find any association between ERBB3 mutations and the activity of PF2341066. (d) Mutations altering specifically the N-terminal EGF receptor are associated to lower drug activity. (e) Mutations affecting another PFR in ERBB3, its kinase domain, and that, thus, are mainly affecting other functional regions, are not associated to any changes in drug activity. (f), Venn diagram showing the different thresholds that we have established in order to minimize false positives. We only kept PFRs with (I) p<0.001 when compared to cell lines with no mutation in the protein, (II) p<0.05 when compared to cell lines with mutations in other regions of the same protein and (III) with p>0.01 at the protein level.
Figure 2.
Perturbations of different regions in the same protein can have different drug effects.
(a) Missense mutations in different PFRs of MSH6 lead to increased sensitivity towards three different drugs: AEW541, RAF5 and Lapatinib. The protein level analysis on the other hand reveals a potential association with Erlotinib (shown in blue). This highlights the complementarity between protein and PFR-centric approaches.
Figure 3.
Using complimentary datasets to validate some of the predictions by e-Drug.
(a) Missense mutations in PIK3CA can have opposite effects in terms of AEW541 activity depending on the PFR affected. Mutations in the p85-binding and PIK accessory domains are associated with lower and higher drug activities respectively (upper panel). By integrating our analysis with proteomics data from TCPA we have been able to propose a mechanism for that. It appears that IRS1 protein expression is lower in cells with p85-binding mutations, but higher in those with PIK mutations (second panel). Moreover, Akt1 phosphorylation levels are higher in cell lines with p85-binding domain mutations (two lower panels). (b) Proposed mechanisms for the two PFR-AEW541 associations. AEW541 inhibits the kinase domain of IGF1R (upper blue protein). In those cell lines with mutations in the PIK domain of PIK3CA (shown in blue PIK3CA's structure), there is a gain of interaction between this protein and IRS1 (I). This will likely increase the signaling through IGF1R (II), explaining why cell lines with mutations in this domain are more sensitive to the inhibition of this receptor. On the other hand, cell lines with mutations in the p85-binding domain (shown in red in PIK3CA's structure) have lower IRS1 expression and higher AKT1 phosphorylation levels. Together, this suggests that PIK3CA is active in this cell lines independently of its interaction with extracellular receptors, signaling directly downstream towards AKT1 (III). This would explain why these cells are resistant to AEW541.
Figure 4.
PFR perturbations identified using data from cell lines predict the survival of patients treated with Irinotecan.
(a) Proteins with PFR associated to Irinotecan resistance can not be used to successfully stratify cancer patients treated with this drug, as there are no differences between patients with mutations in such proteins (gray) and those without them (black) (b) Specific PFR in these proteins do predict the outcome of cancer patients. Patients with mutations altering the PFRs found using CCLE (red) have worse outcomes that those with mutations in other regions of the same protein (green) or no mutations (black).
Figure 5.
Enrichment map of the proteins associated with differential drug activity at both, whole-protein and individual region levels.
We performed a gene-set enrichment analysis by comparing Gene Ontology (GO) annotations of the 316 proteins associated with different drugs at both levels of resolution (whole-protein and individual PFRs) against the whole human genome. All the GO terms identified here showed an enrichment in the biomarker group, and most of them relate to pathways and functions associated with carcinogenesis, metastasis, and drug resistance, such as regulation of cell proliferation, kinase activity, cell migration, cell adhesion, MAPK cascade, or response to hypoxia. In the figure, GO terms are connected when they are related according to the gene ontology.