Elucidating synergistic dependencies in lung adenocarcinoma by proteome-wide signaling-network analysis

Abstract

To understand drug combination effect, it is necessary to decipher the interactions between drug targets-many of which are signaling molecules. Previously, such signaling pathway models are largely based on the compilation of literature data from heterogeneous cellular contexts. Indeed, de novo reconstruction of signaling interactions from large-scale molecular profiling is still lagging, compared to similar efforts in transcriptional and protein-protein interaction networks. To address this challenge, we introduce a novel algorithm for the systematic inference of protein kinase pathways, and applied it to published mass spectrometry-based phosphotyrosine profile data from 250 lung adenocarcinoma (LUAD) samples. The resulting network includes 43 TKs and 415 inferred, LUAD-specific substrates, which were validated at >60% accuracy by SILAC assays, including "novel' substrates of the EGFR and c-MET TKs, which play a critical oncogenic role in lung cancer. This systematic, data-driven model supported drug response prediction on an individual sample basis, including accurate prediction and validation of synergistic EGFR and c-MET inhibitor activity in cells lacking mutations in either gene, thus contributing to current precision oncology efforts.

Fig 1. Framework for the reverse engineering of TK signaling networks from phosphoproteomic profiles.

(A) Workflow to identify pharmacological synergistic dependencies based on context-specific signaling network analysis. (B) Top panel shows the schematic diagram of a TK→S interaction. The non-phosphorylated kinase is inactive in terms of phosphorylating a substrate, while the active isoform successfully phosphorylates the substrate. Bottom panel shows schematic diagram showing the correlation between TK phosphorylation and that of its potential substrates. The first two rows in the heatmap show proteins representing candidate TK substrates (C) Illustration of the pARACNe framework including 6 steps. Step-1 depicts peptides collection from primary lung cancer tissue and cell lines for whole phosphrtyrosine proteomics quantification. Step-2 depicts inferences of TK→S interactions using Mutual Information by Step-3 of the iterative quantile discretization methods and Step-4 Naïve-Bayes estimator. Step-5 and 6 depict network pruning and bootstrapping to construct final network. (C) Framework of pARACNe from LC-MS/MS data normalization, Iterative Quantile Discretization (IQD) process, Mutual Information (MI) calculation, Data Processing Inequality (DPI) process, bootstrapping to network consolidation.

Fig 2. Predicted TK-TK network and validation of EGFR and c-MET prediction.

(A) pARACNe-inferred densely inter-connected TK-TK network, with red nodes representing candidate TKs involved in auto-phosphorylation, where the phospho-state of a tyrosine is correlated with the phospho-state of a different tyrosine on the same TK protein. (B) pARACNe-inferred EGFR and (C) c-MET substrate overlap with SILAC-based and Database reported substrates, respectively.

Fig 3. Inference of master regulator and combination.

(A). Schema of Master Regulator analysis in lung cancer using pVIPER. (B) shows prioritized Master Regulators and (MR) (C) shows prioritized Master Regulator Pairs as significantly activated (red circle) or de-activated (blue) molecules in different lung cancer cell lines (by column). In panel (B) and (C), pVIPER infers, for each MR / MR pairs in each cell line, the enrichment of the MR’s target peptides’ in the differentially phosphorylated peptides compared to the average of all normal samples. The size and color represent the scale of enrichment. Red color represents an enrichment of substrates hyper-phosphorylation by a Master Regulator or Master Regulator Pairs. Blue color represents that of hypo-phosphorylation. A white small circle refers to no significant enrichment; red (blue) large circle represents substrates of a kinase to be significantly hyper (hypo)-phosphorylated in the corresponding cell line when compared with the average from all normal samples.

Fig 4. Experimental validation of EGFR and c-MET combination by colony formation assay.

(A) Colony formation assay schema shows the image of long-term EGFR and c-MET double inhibition effects in HCC78 cell line with different treatments. (B) shows long-term colony formation data for 14 cell lines with different EGFR, BRAF and KRAS genomic mutation status.

Fig 5. MTT Assay validation of EGFR and c-MET combination.

(A). MTT assay experimental schema. (B) MTT assay of HCC78 cell line shows synergistic effects of Crizotinib and Erlotinib treatment. (C) shows short-term effects of EGFR and c-MET inhibitors’ combination index in 11 cell lines include 2 control cell lines (red).

Fig 6. Master regulating peptides in primary lung cancer samples.

EGFR and c-MET co-regulate in three scenarios (A) when their common substrates are hyperphosphorylated, the patient responds to combination treatment well; (B) when most EGFR substrates are hyper-phosphorylated, the patient responds to EGFR inhibitor; (C) when substrates of both EGFR and c-MET are mostly hypophosphorylated, the patient does not respond. (D) shows the Master Regulator and Master Regulator Pairs regulating hyper/hypo-phosphorylation of their network substrates in each primary samples. In panel (A-C), Red node represent hyper-phosphorylated, yellow node represent hypo-phosphorylated and gray represents un-differentiated phosphorylated proteins compared to average normal samples.