The OncoPPi network of cancer-focused protein-protein interactions to inform biological insights and therapeutic strategies

Abstract

As genomics advances reveal the cancer gene landscape, a daunting task is to understand how these genes contribute to dysregulated oncogenic pathways. Integration of cancer genes into networks offers opportunities to reveal protein-protein interactions (PPIs) with functional and therapeutic significance. Here, we report the generation of a cancer-focused PPI network, termed OncoPPi, and identification of >260 cancer-associated PPIs not in other large-scale interactomes. PPI hubs reveal new regulatory mechanisms for cancer genes like MYC, STK11, RASSF1 and CDK4. As example, the NSD3 (WHSC1L1)-MYC interaction suggests a new mechanism for NSD3/BRD4 chromatin complex regulation of MYC-driven tumours. Association of undruggable tumour suppressors with drug targets informs therapeutic options. Based on OncoPPi-derived STK11-CDK4 connectivity, we observe enhanced sensitivity of STK11-silenced lung cancer cells to the FDA-approved CDK4 inhibitor palbociclib. OncoPPi is a focused PPI resource that links cancer genes into a signalling network for discovery of PPI targets and network-implicated tumour vulnerabilities for therapeutic interrogation.

Figure 1. Design and workflow of the high throughput PPI screening platform leading to OncoPPi v1.

(a) Lung cancer genomics information was utilized to construct the OncoPPi expression vector library (see Supplementary Table 1) for pairwise TR-FRET-based high-throughput screening in H1299 lung cancer cells. (b) Analysis of high throughput PPI datasets included monitoring expression of each fusion protein construct using the fluorescence of the Venus-fusion protein and a FRET-based GST biosensor for the GST-fusion protein. Positive PPIs were defined based on FOC values, P values, and q-values calculated using the Benjamini–Hochberg procedure. See also Supplementary Table 2. (c) Definition of OncoPPi. Venn diagram representation of the OncoPPi network as a defined set of HC-PPIs plus previously reported interactions validated in these studies (Supplementary Fig. 2, Supplementary Table 2). (d) Schematic heatmap representation of OncoPPi expansion of the PPI landscape for lung cancer-associated genes, including membrane proteins, transcriptional regulators, adaptor proteins, kinases and others. Blue are previously described PPIs, magenta and yellow are experimentally determined OncoPPi and SS-PPI sets, respectively. See also Supplementary Fig. 2.

Figure 2. OncoPPi network architecture.

(a) A connectivity map of the OncoPPi network involving 83 lung cancer-associated proteins linked via 397 interactions (Supplementary Cytoscape file available for visualization and detailed analysis, see Supplementary Data 3). Major hubs highlighted in green. PPIs with mutual exclusivity of genomic alterations in LUAD are indicated with blue lines. (b) Analysis of network topology, including degree and BC reveals major PPI hubs. (c) Heatmap showing GO annotations of cellular localization for OncoPPi network genes. (d) Bar graph showing the number of OncoPPi PPIs supported by predicted domain–domain interactions. Domain–domain pairs in OncoPPi using Pfam domain annotation are listed on the x axis, the number of PPIs in OncoPPi on the y axis. Examples of co-crystallized CDK4/CyclinD, ARNT/HIF1α, and a homology model of MST1/RASSF1 constructed with the Swiss-Model server (swissmodel.expasy.org) based on MST1/RASSF5 crystal structure are shown to illustrate the interactions between different structural domains. (e) Venn diagram showing the distribution of PPIs in the OncoPPi network supported by cellular co-localization (Co-loc) data and/or structural domain–domain interactions (DDI). See also Supplementary Fig. 3.

Figure 3. Major hub proteins in the OncoPPi network.

(a) Hub and spoke diagrams for the 41 proteins that form at least five heterodimers in the OncoPPi network are shown. The red, blue and green sectors inside the nodes represent the percent of LUAD cases (based on LUAD TCGA provisional dataset) with gene amplifications, deletions or mutations, respectively. PPIs identified as physical interactions in public IntAct, BioGrid, String or GeneMania databases are indicated with dashed lines (Supplementary Data 2). Newly discovered PPIs are indicated with solid lines. Functional connectivity between interacting partners was evaluated with mutual exclusivity analysis of genomic alterations (mutual exclusivity) and FUSION analysis (Supplementary Data 2). PPIs positive in mutual exclusivity or FUSION analyses are highlighted with blue lines. The proteins are identified by the Human Genome Organization Gene Nomenclature Committee approved symbols. (b) Confirmation of CDK4 and (c) RASSF1 PPIs with GST-pull down assay in HEK293T cells. Both Venus (vector) and GST tags were used as negative controls.

Figure 4. OncoPPi-suggested regulatory mechanism for the MYC oncogene.

(a) Direct interaction of NSD3-s with MYC. GST-MYC and Venus-Flag-tagged NSD3-s were expressed in HEK293T cells. Tb-conjugated anti-GST-antibodies were incubated with cell lysates to detect GST-MYC. The TR-FRET signal is expressed as the FRET ratio (520 nm/486 nm × 10⁴) Representative results of three independent experiments are shown. The error bars show the mean±s.d. of three replicates. (b) NSD3–s/MYC interaction by GST-affinity pull-down assay. GST-MYC was captured by glutathione-resin to probe the presence of Flag-NSD3-s with western blotting. (c,d) Endogenous interaction of NSD3-s with MYC. The NSD3-s/MYC complex was co-immunoprecipitated with an anti-MYC antibody from lung cancer (c) H1299 and (d) H1944 cells with anti-IgG as control. (e) Schematic diagram of truncated NSD3 constructs. The MYC-binding fragments are indicated in grey. (f) GST-affinity pull-down assay with MYC and NSD3-s fragments. (g) NSD3-s stabilizes MYC. Immunoblot showing MYC and tubulin levels in HEK293T cells at different time points after inhibition of protein synthesis with cyclohexamide with or without co-expressed NSD3-s. (h) Graph of MYC protein levels at indicated time points based on densitometric analysis of results in (g). 100% corresponds to the total MYC detected at the 0 time point. MYC levels are normalized to tubulin protein levels. (i) NSD3-s activates MYC transcriptional activity. HEK293T cells were co-transfected with Venus-NSD3-s and either wild-type or mutant E-box luciferase reporter. Relative luciferase activity was measured, normalized to internal Renilla luciferase control. Representative results of three independent experiments are shown. The error bars show the mean±s.d. of three replicates. (j) NSD3-s interacts with MYC, NSD3-s and BRD4. GST-NSD3-s was co-transfected with FLAG-tagged constructs for MYC, NSD3-s and BRD4 into HEK293T cells, followed by affinity chromatography with glutathione-conjugated beads, SDS-PAGE and immunoblotting with anti-Flag antibodies. (k) Proposed working model. BRD4 utilizes its ET domain to regulate MYC through a transcription-independent mechanism via the BRD4-NSD3-MYC pathway, in addition to the well-established BRD4-pTEFb-mediated pathway via the C-terminal fragment of BRD4. Both BRD4 and NSD3-s interact with modified histones.

Figure 5. Physical and functional interaction of tumour suppressor STK11 with CDK4.

(a) The OncoPPi network links STK11 to CDK4 and palbociclib. (b) Mutual exclusivity map for the STK11 and CDK4 pathways in LUAD patient samples. Missense and truncation mutations are shown in green and black, respectively. DNA amplification is shown in magenta, deletions are shown in blue. Alterations of the PRKAA1 gene coding alpha 1 catalytic subunit of AMPK are shown. (c) Heatmap with expression signatures for FUSION reporter genes with individual knockdowns of STK11, AMPK, mTOR, CDK4, CDKN2A and CDKN2B based on Pearson correlation values from highest (blue) to lowest (magenta). (d) Interaction of STK11 with CDK4 and CCND2 using Renilla luciferase-PCA. Representative results of three independent experiments are shown. The error bars show the mean±s.d. of three replicates. (e,f) Endogenous STK11 was co-immunoprecipitated with CDK4 in lung cancer (e) H1299 and (f) H1792 cells. (g) Effect of STK11 status on CDK4 and STK11 activity as shown by pRB and pAMPK levels in isogenic H1299 lung cancer cells (STK11 wild-type and knockdown). (h) Effect of CDK4 inhibition by palbociclib on pAMPK status in isogenic H1792 cells with wild-type (STK11-WT) and STK11 knockdown (STK11-KD). (i) CDK4 inhibition disrupts the interaction of CDK4 with STK11. HEK293T cells transfected with indicated plasmids were treated with palbociclib at the indicated concentrations. GST-CDK4 pull-down assay was carried out and protein expression was examined by western blot with the indicated antibodies. (j) Silencing of STK11 (H1792-408 cells) enhanced lung cancer cell response to phenformin and to (k) palbociclib in a cell viability assay. (l) Overexpression of STK11 in STK11 null cells (H157) reduced sensitivity of H157 cells to palbociclib. Western blots are representative of three independent experiments. Cell viability is expressed as % of control (mean±s.d.).