Kinome Profiling of Primary Endometrial Tumors Using Multiplexed Inhibitor Beads and Mass Spectrometry Identifies SRPK1 As Candidate Therapeutic Target

Protein kinases (collectively, termed the kinome) represent one of the most tractable drug targets in the pursuit of new and effective cancer treatments. However, less than 20% of the kinome is currently being explored as primary targets for cancer therapy, leaving the majority of the kinome untargeted for drug therapy. Chemical proteomics approaches such as Multiplexed Inhibitor Beads and Mass Spectrometry (MIB-MS) have been developed that measure the abundance of a significant portion of the kinome, providing a strategy to interrogate kinome landscapes and dynamics. Kinome profiling of cancer cell lines using MIB-MS has been extensively characterized, however, application of this method to measure tissue kinome(s) has not been thoroughly explored. Here, we present a quantitative proteomics workflow specifically designed for kinome profiling of tissues that pairs MIB-MS with a newly designed super-SILAC kinome standard. Using this workflow, we mapped the kinome landscape of endometrial carcinoma (EC) tumors and normal endometrial (NE) tissues and identified several kinases overexpressed in EC tumors, including Serine/Arginine-Rich Splicing Factor kinase, (SRPK1). Immunohistochemical (IHC) analysis of EC tumor TMAs confirmed MIB-MS findings and showed SRPK1 protein levels were highly expressed in endometrioid and uterine serous cancer (USC) histological subtypes. Querying large-scale genomics studies of EC tumors revealed high expression of SRPK1 correlated with poor survival. Inhibition of SRPK1 in USC cells altered mRNA splicing, downregulating several oncogenes including MYC and Survivin resulting in apoptosis. Taken together, we present a SILAC-based MIB-MS kinome profiling platform for measuring kinase abundance in tumor tissues, and demonstrate its application to identify SRPK1 as a plausible kinase drug target for the treatment of EC.

abundance of a significant portion of the kinome, providing a strategy to interrogate kinome landscapes and dynamics. Kinome profiling of cancer cell lines using MIB-MS has been extensively characterized, however, application of this method to measure tissue kinome(s) has not been thoroughly explored. Here, we present a quantitative proteomics workflow specifically designed for kinome profiling of tissues that pairs MIB-MS with a newly designed super-SILAC kinome standard. Using this workflow, we mapped the kinome landscape of endometrial carcinoma (EC) tumors and normal endometrial (NE) tissues and identified several kinases overexpressed in EC tumors, including Serine/Arginine-Rich Splicing Factor kinase, (SRPK1).
Immunohistochemical (IHC) analysis of EC tumor TMAs confirmed MIB-MS findings and showed SRPK1 protein levels were highly expressed in endometrioid and uterine serous cancer (USC) histological subtypes. Querying large-scale genomics studies of EC tumors revealed high expression of SRPK1 correlated with poor survival. Inhibition of SRPK1 in USC cells altered mRNA splicing, downregulating several oncogenes including MYC and Survivin resulting in apoptosis. Taken together, we present a SILAC-based MIB-MS kinome profiling platform for measuring kinase abundance in tumor tissues, and demonstrate its application to identify SRPK1 as a plausible kinase drug target for the treatment of EC.

INTRODUCTION
Protein kinases are a family of ~535 enzymes that, collectively, are termed the kinome (1).
Uncontrolled protein kinase activity has been linked to the development of nearly 25% of all cancers; consequently, protein kinases represent one of the most promising avenues for cancer therapy (2,3). Indeed, >30 kinase-specific inhibitors are currently approved for therapy of various cancer types, with more than 150 kinase inhibitors in Phase 1-3 clinical trials across all diseases (4). However, most of these kinase-specific inhibitors target a relatively small fraction of the human kinome with only about 20% avidly being explored as primary targets for drug therapy (5,6). Thus, the majority of the kinome remains untargeted for cancer therapy and about 50% of the kinome is largely uncharacterized with respect to the function and role of these kinases in cancer, representing the understudied or 'dark' cancer kinome (6)(7)(8). Notably, several CRISPR/cas9 and/or RNAi loss-of-function studies have shown that many dark kinases are essential for cancer cell viability highlighting the therapeutic potential of the dark kinome for the treatment of cancer (9,10).
Identifying kinase-signaling networks that are essential for tumor growth and resistance requires a detailed knowledge of global kinome activity (sum of all 518 human kinases), not simply measuring one or a few kinases in a pathway. To accomplish this, quantitative proteomics strategies including Kinobeads, Kinativ and Multiplexed Inhibitor Beads (MIBs) have been developed that are capable of detecting dynamic changes globally across the kinome at baseline or following treatment with a targeted inhibitor (11)(12)(13). These chemical proteomics techniques couple kinase affinity capture with quantitative mass spectrometry, providing a systems biology platform to profile global kinome signaling at the proteomic level. Several reports have applied kinome enrichment strategies to define kinase signatures of cancer cell lines (14), however, a robust quantitative kinome enrichment strategy for measuring kinase abundance across cohorts of tissues, (i.e., primary tumors) has not been systematically explored. As the usage of patientderived xenografts (PDX) models (15) for drug studies increases, and frozen tissue samples from SRPK1 Identified as Drug Target in Endometrial Cancer clinical trials become more frequently available for analysis, a tumor kinome profiling strategy specifically designed for tissue analysis will be of particular interest.
Here, we present an optimized Quantitative-Multiplexed Inhibitor Beads (Q-MIBs) workflow for profiling the kinome in tissues that incorporates a newly-designed super-SILAC kinome reference standard permitting robust quantitation of ~70% of the human kinome.
Following quality control testing, we applied Q-MIBs to identify candidate kinase targets in endometrial carcinoma (EC), which is the most common gynecologic malignancy in the United States with limited effective targeted therapies (16). Using Q-MIBs, we profiled the kinome of in Perseus software the following manner: normalized MIB-MS s-SILAC ratios were transformed 1/(x) to generate light / heavy ratios, followed by log2/(1/x) transformed. Columns were then filtered based on a valid value of 150, MIB-MS technical replicates averaged, and rows filtered for minimum valid kinases measured (n=>3 kinases). Imputation of missing values was performed as previously described (17), where in the s-SILAC data, a width of 0.3 and the downshift of 0.5, was employed. Principal component analysis (PC1 vs PC2, PC2 vs PC3 and PC1 vs PC3) was then performed to visualize kinome profiles amongst samples. Hierarchical clustering (Euclidean) of s-SILAC ratios was performed and column clusters annotated selecting 4 clusters at a distance threshold of 27.87. Difference in kinase abundance amongst endometrial tumors (n=17) and normal endometrial tissues (n=14) was determined using a two-sample Student's t-test with the following parameters, (S0 0.1, and Side, Both) using p-value < 0.05 using Perseus software.
Analysis of SRPK1 mRNA alterations in endometrial tumors (n=232) from TCGA studies (17) association with survival was performed using Fisher's two-sided p < 0.05. Immunohistochemistry analysis of overexpression of SRPK1 in endometrioid (n=18) and serous (39) endometrial tumors relative to normal endometrial tissues (n=12) was determined by two-sample Student's t-test P < 0.01. For splicing analysis following SPHINX3.1 treatment, RNA-seq reads for both treatment and controls were aligned to the human genome (Hg38) using TopHat (18). Detection of splicing changes between treatment and controls was assessed using MATS algorithm, using the aligned BAM files with default parameters (19). The spliced events were filtered using False Discovery Rate by Benjamini-Hochberg method (FDR < 0.01). Genes significantly altered by SPHINX31 treatment in SPEC-2 cells were determined by two-sample Student's t-test P < 0.001 comparing DMSO vs SPHINX31-treatment and analyzed using g:Profiler and Database for Annotation,

Immunohistochemistry (IHC) and IHC Evaluation-
Immunohistochemical staining was carried out according to standard methods by the FCCC Histopathology Facility. Briefly, 5µm formalin-fixed, paraffin-embedded TMA sections were deparaffinized and hydrated. Sections were then subjected to heat-induced epitope retrieval with 0.01 M citrate buffer (pH 6.0). software. One mismatch was allowed for index sequence identification. RNA sequencing reads for both treatment and controls were aligned to the human genome (hg38) using TopHat. (19).
Alternative splicing analysis was performed as described previously (22). Briefly, for detection of splicing changes between treatment and controls, the MATS algorithm was implemented using the aligned BAM files with default parameters (19). The spliced events were filtered using False Discovery Rate by Benjamini-Hochberg method (FDR < 0.01). Furthermore, the events were sorted based on difference between average of inclusion level for treated and control samples.
Each splicing change was visualized using the IGV program (Integrative Genomics Viewer).
Enrichment analysis for Gene Ontology (GO) terms was assessed using the GOstats program (23).

Bioinformatics Analyses of TCGA Datasets-
Analysis of SRPK1 CNA and mRNA alterations in EC from TCGA studies was performed at the Biostatistics and Bioinformatics Facility, FCCC. SRPK1 expression is associated with survival (Fisher's two-sided p < 0.05).
Overall survival was based on vital status and "days to death" from initial pathologic diagnosis.
Individuals who were still alive at the time of the last follow-up were censored. Survival curves were compared with log-rank tests, and these calculations were done using the R 'survival' package (Therneau, T. M. & Grambsch, P. M. Modeling Survival Data: Extending the Cox Model (Springer-Verlag, 2010). Survival data was obtained from TCGA data repository.

Abundance in Tissues
Kinome profiling strategies have been widely utilized in cancer cell line models, however, application of kinase enrichment strategies to measure kinase abundance in tumor tissues has not been extensively characterized. Here, to quantitate the levels of protein kinases in tissue samples, we designed a super-SILAC (24) kinome standard (SKS) that can be paired with Multiplexed Inhibitor Beads and Mass Spectrometry (12) to measure kinase abundance across samples, which we've termed Quantitative-Multiplexed Inhibitor Beads (Q-MIBs) (Fig. 1). An equal amount of the SKS is spiked into any non-SILAC labeled sample (snap frozen tissues), endogenous kinases purified by MIBs, kinases eluted, digested, and peptides analyzed by LC-MS/MS. Kinase protein levels in tissues are then quantitated by comparing SILAC-labeled peptides from SKS with non-labeled peptides from tissues using MaxQuant Software (25).
Collectively, the Q-MIBs workflow generates kinase signatures that can be immediately actionable through small molecules, as well as identifying perturbations of understudied kinases, which could represent new drug targets for cancer therapy.
To enrich endogenous protein kinases from tissue lysates, we utilized an equal mixture of 50 µL the pan-kinase inhibitor resins Purvalanol B, PP58, VI16832 and CTx-0249885, each inhibitor-resin purifying a distinct fraction of the kinome (Fig. S1A, Data file S1 Fig. 2B-C, and S1B-C, Data file S1). Following testing of various combinations, we identified the mixture of UACC257, MOLT4, COLO205, ACHN and PC3 cell lines provided the most diverse MIB-enriched kinome profile and was therefore selected as our SKS to be used repeatedly to quantify kinase levels across samples ( Fig. 2D-E, Data file S1).
The SKS can be propagated indefinitely, demonstrating high correlations in kinome quantitation across 3 distinct SKS batches prepared months apart, allowing quantitative comparison of kinome signatures from distinct kinome profiling projects (Fig. 2F, Data file S1).
Next, we explored the optimal protein concentration input of the unlabeled sample to achieve maximal SILAC quantitation of the MIB-enriched kinome. The number of kinases measured by s-SILAC was overall similar amongst the various concentration of protein inputs ranging from 236 (1 mg) to 257 (5 mg), however, the number of kinases quantified by at least 3 unique peptides was lower using 1 mg input (n=173) compared to 5 mg protein input (n=201), favoring the use of 5 mg protein inputs if sample amounts are not limiting (Fig. S1D, Data file S1). Various protein input concentrations produce highly correlated Q-MIB signatures, and although reduced numbers of kinases are quantitated using 1 mg relative to 5 mg inputs, those quantitated in both samples, exhibited high correlation (Fig. 2G).
Thus, using the SKS permits the measurement and comparison of low yielding patient specimens such as biopsies with larger tumor sections from surgeries. Reproducibility of the MIB-MS paired with SKS assay to measure kinase abundance was confirmed using a model-based approach (27) for assessing technical reproducibility and outlier detection (Fig. S1E).

Characterizing the Kinome Landscape Measured by Q-MIBs
Next To define the fraction of the kinome measured by Q-MIBs, we kinome profiled a cohort of tumors (n=40) from 8 distinct cancer types, as well as a variety of cancer cell lines (n=19) using MIB-MS paired with the SKS (Fig. 3C, Data file S2). In total, 349 kinases were SILACquantitated with ~60% of the kinases exhibiting MQ scores >100 (Fig. 3D). The average number of kinases measured per sample was 254 (tumors) and 244 (cell lines) with the majority of kinases sequenced identified by 10-24 unique peptides ( Fig. 3E-F, Data file S2). Frequency analysis of individual kinases measured across the 59 Q-MIBs samples showed 127 kinases were measured in every run, while 177 were measured at 90% frequency and ~73% were measured in ≥50% of the Q-MIBs runs (Fig. 3G, Data file S2). A detailed breakdown of individual kinase frequency of measurement by Q-MIBs across the 59 samples can be found in Data file S2.
Characterization of the kinome measured by Q-MIBs revealed significant coverage of kinases with FDA approved drugs (n=43), those currently being investigated in clinical trials (n=45) with several having established kinase driver function (n=33) (5) (Fig. 3H, Data file S2).
It should be noted that most of the kinases quantified by Q-MIBs are largely untargeted for cancer therapies, with many (n=73) having sparse (<50) number of publications, representing the understudied or "dark" cancer kinome (Fig. 3I, Data file S2). For example, differential MIB-binding of understudied kinases PHKG2 and BMP2K was observed amongst tumor types, with pancreatic cancer exhibiting elevated levels of PHKG2, while BMP2K protein levels were the greatest in pancreatic and prostate tumors (Fig. 3J).

Mapping the Kinome of Primary Endometrial Carcinoma Tumors Using Q-MIBs
Endometrial cancer (EC) is the most common gynecologic malignancy in the United States with 60,050 new cases and 10,470 deaths expected in 2020 (28). Most patients are diagnosed with uterine-confined disease (i.e. stage 1) (28) and thus, will have an overall favorable prognosis with 5 year disease-free survival greater than 80% (29). However, there has been a steady increase in the mortality rate for endometrial cancer which has been attributed to higher proportions of patients with advanced stage, higher grade and serous histology (30). Uterine Serous Carcinoma (USC) is one of the most common and lethal forms of endometrial carcinoma, and current treatments have only modestly impacted survival (31). To identify novel kinase therapeutic avenues in EC, we performed Q-MIBs profiling on 17 primary EC tumors (n=14 endometrioid, n=3 serous) and 14 normal endometrial (NE) tissues (Fig. 4A, Fig. S2A and Data file S3).
In total, we measured MIB-binding values for 299 kinases across EC tumors and NE tissues, with an average of 235 and 231 kinases measured per MIB-MS samples in NE tissues and EC tumors, respectively ( Fig. 4B-C). Frequency analysis of kinase measurement across MIB-MS runs showed 126 kinases were measured in every run, while 236 kinases were quantitated in ≥50% of MIB-MS runs. (Fig. 4D, Data file S3). Principal component analysis or unsupervised hierarchical clustering of EC and NE tissue kinome profiles showed EC tumors clustered together and were overall distinct from NE tissues (Fig. 4E-F). Volcano plot analysis of MIB-MS signatures amongst EC tumors and NE tissues revealed many kinases were differentially expressed (Fig. S2B). Kinases commonly elevated (Fig. 4G) or reduced (Fig. 4H) in tumors relative to NE tissues are depicted on kinome trees.

Many kinases overexpressed in EC tumors have inhibitors FDA-approved inhibitors or
drugs currently being evaluated in clinical trials, representing potentially actionable kinases for treatment (Fig. 4I)

SRPK1 Is Overexpressed in Endometrial Tumors and Represents a Prognostic Indicator for Poor Survival
Serine/Arginine-Rich Splicing Factor kinase, (SRPK1) overexpression has been correlated with poor survival of several cancers including colorectal, lung and breast cancers (36).
Overexpression of SRPK1 has been observed in highly aggressive breast cancers, such as triple negative breast cancers (TNBCs) and may be linked to development of resistance to therapy, highlighting SRPK1 as an attractive therapeutic target (37). SRPK1 has been shown to be essential for mRNA alternative splicing in cells and blockade of SRPK1 function alters several oncogenic processes including angiogenesis, migration/invasion, proliferation, cancer stem cell (CSC) phenotype, and sensitivity to chemotherapy (38).
In the present study, Q-MIBs kinome profiling identified SRPK1 to be elevated in EC tumors relative to normal endometrial tissues (Fig. 5A). Immunoblot analysis comparing SRPK1 protein levels amongst EC tumors and paired matched normal tissue confirmed Q-MIBs findings, where the majority of EC tumors overexpressed SRPK1 (Fig. 5B) (Fig. 5E). Notably, SRPK1 mRNA levels were upregulated in 31% of USC tumors (13 out of 42) and 11% of endometrioid tumors (20 out of 186). Levels of SRPK1 mRNA expression were associated with survival in EC, where patients harboring EC tumors with high levels of SRPK1 exhibited reduced overall survival (P <0.05) (Fig. 5F). 3' splice site (3'SS) and 640 retained intron (RI) (Fig. 6B, Data file S5). Pathway analysis of genes exhibiting mis-splicing following SRPK1 inhibition were enriched in cell-cell adhesion, protein transport, cell cycle, protein phosphorylation and mRNA splicing signaling (Fig. 6C).

SRPK1 Inhibition Promotes Alternative Splicing Impairing Survival Signaling in USC Cells
Similarly, BIOCARTA pathway analysis showed an enrichment of alternative spliced targets in RHO pathway, ECM pathway and VEGF pathway signaling (Fig. 6D). Consistent with these findings, inhibition of SRPK1 has been previously shown to block splicing of VEGF 165a to VEGF 165b impacting Vegf signaling and angiogenesis (38).
Analysis of mRNA expression changes following SPHINX31-treatment of SPEC-2 cells revealed SRPK1 inhibition induced significant changes to gene expression, including 1306 gene upregulated and 909 genes downregulated (Fig. 6E, Data file S5). Reactome pathway analysis of genes downregulated by SPHINX31-treatment were enriched in Rho GTPase, cell cycle, DNA replication, transcriptional regulation by TP53 and extracellular matrix organization signaling ( Fig.   6F and S3A). Genes downregulated following SRPK1 inhibition were also enriched in PI3K-AKT and FoxO signaling, consistent with previous reports demonstrating SRPK1 inhibition impaired AKT signaling through disruption with PHLPP interaction (42) (Fig. S3B). Moreover, several genes involved in negative regulation of apoptosis and cell cycle progression were downregulated in response to SRPK1 blockage, including BIRC5 (Survivin), MYC, MYCN and CCND1, all established oncogenic drivers (Fig. 6G). Knockdown of Survivin has been shown to induce apoptosis and mitotic catastrophe in several cancer cell lines (43), while MYC depletion induces cell cycle arrest (44). Immunoblot analysis of SPHINX31-treated cells confirmed several mRNA expression changes including reduced protein levels of established oncogenes, MYC and Survivin (Fig. 6H). Furthermore, an increase in cleaved PARP was observed by immunoblot, demonstrating SRPK1 inhibition induces apoptosis in SPEC-2 cells.
Together, SRPK1 inhibition inhibited cell viability and induced apoptosis distinctly in serum-starved USC cells. Moreover, SPHINX31-treatment induced significant alternative splicing of genes involved in cell motility and cell cycle resulting in the downregulation of several genes essential for growth and survival, including oncogenes MYC and Survivin. Finally, SPHINX31treatment induced apoptosis in USC cells nominating SRPK1 as a plausible drug target for the treatment of EC.

DISCUSSION
The protein kinome represents one of the most promising and actionable classes of drug targets for the treatment of cancer; yet, the majority of the kinome remains untargeted for drug therapy, with many kinases having no established oncogenic function (4,6). The vast majority of kinase publications focus on a small group of well-understood kinases, yet synthetic lethal screens repeatedly identify various untargeted kinases as playing a vital role in cancer cell proliferation and survival (45). Our inability to routinely probe these enzymes has hindered previous attempts to understand how they are regulated and function in cancer. Importantly, emergence of unbiased proteomics strategies such as Kinobeads and MIB-MS has provided methods to capture and quantify these elusive untargeted kinases in normal and in cancer cells (12,46). Notably, MIB-MS has been extensively utilized in the study of cancer cell line models, however, its application in tumor models has not been systematically characterized. Here, we Mis-spliced gene targets following SPHINX31-treatment were enriched in VEGF pathways and angiogenesis, consistent with recent studies showing SRPK1 function has been shown to block splicing of VEGF 165a to VEGF 165b reducing angiogenesis in cells and tumor models (38).
Importantly, Bevacizumab, an antibody against VEGF-A has shown moderate clinical responses in EC, suggesting SRPK1 inhibition could represent a promising anti-angiogenic therapy for EC (16). However, further studies in tumor models will be required to determine the impact of SRPK1   Figure S1: Designing a s-SILAC kinase standard for measuring kinase abundance in tissues. Figure S2: Kinome profiling of EC tumors and NE tissues using Q-MIBs.