Phosphoproteomics identifies determinants of PAK inhibitor sensitivity in AML cells

Background The p21 activated kinases (PAK) are frequently dysregulated in cancer and have central roles in oncogenic signalling, prompting the development of PAK inhibitors (PAKi) as anticancer agents. However, such compounds have not reached clinical use because, at least partially, there is a limited mechanistic understanding of their mode of action. Here, we aimed to characterize functional and molecular responses to PAK inhibitors (PF-3758309, FRAX-486 and IPA-3) in multiple AML models to gain mechanistic insights on how these inhibitors work in this disease and identify determinants of response in patient samples. Methods We used proteomics and phosphoproteomics to profile PAKi impact on protein phosphorylation and expression and kinase activity in P31/Fuj and MV4-11 AML cells, and primary cells from 8 AML patients. We also integrated gene dependency data with phosphoproteomics and proteomics to identify which proteins targeted by PAKi are necessary for the proliferation of AML. We studied the effect PAKi on cell cycle progression, proliferation, differentiation and apoptosis. Finally, we used phosphoproteomics data as input for machine learning models that predicted “ex vivo” response of primary AML cells to PF-3758309 and identify markers of response. Results PF-3758309 was the most effective PAKi in reducing proliferation and inducing apoptosis in AML. PF-3758309 inhibited PAK, AMPK and PKCA activities in all AML cell lines and primary cells tested. This compound reduced the phosphorylation and expression of c-MYC and multiple ribosomal proteins in both cell lines and targeted the FLT3 pathway in FLT3-ITD mutated cells. In primary cells, PF-3758309 reduced STAT5 phosphorylation at Y699 in models with high phosphorylation on this site. Functionally, PF-3758309 reduced cell-growth, induced apoptosis, blocked the cell cycle progression and promoted differentiation in a model-dependent manner. ML modelling accurately classified primary AML samples as sensitive or resistant to PF-3758309 “ex vivo” treatment. Analysis of the models highlighted phosphorylation of PFH2 at Ser705 as a potential biomarker of response to PF-3758309. Conclusions In summary, our data define the proteomic, molecular and functional response of AML cells to PF-3758309 and suggest a route to personalise treatments in AML based on PAK inhibitors.