IQGAP1 and RNA splicing in the context of head and neck via phosphoproteomics

Abstract

IQGAP1 (IQ motif-containing GTPase-activating protein 1) scaffolds several signaling pathways in mammalian cells that are implicated in carcinogenesis, including the RAS and PI3K pathways that involve multiple protein kinases. IQGAP1 has been shown to promote head and neck squamous cell carcinoma (HNSCC); however, the underlying mechanism(s) remains unclear. Here, we report a mass spectrometry-based analysis identifying differences in phosphorylation of cellular proteins in vivo and in vitro, in the presence or absence of IQGAP1. By comparing the esophageal phosphoproteome profiles between Iqgap1^+/+ and Iqgap1^−/− mice, we identified RNA splicing as one of the most altered cellular processes. Serine/arginine-rich splicing factor 6 (SRSF6) was the protein with the most downregulated levels of phosphorylation in Iqgap1^−/− tissue. We confirmed that the absence of IQGAP1 reduced SRSF6 phosphorylation both in vivo and in vitro. We then expanded our analysis to human normal oral keratinocytes. Again, we found factors involved in RNA splicing to be highly altered in the phosphoproteome profile upon genetic disruption of IQGAP1. Both the Clinical Proteomic Tumor Analysis Consortium (CPTAC) and the Cancer Genome Atlas (TCGA) datasets indicate that phosphorylation of splicing-related proteins is important in HNSCC prognosis. The BioGRID repository also suggested multiple interactions between IQGAP1 and splicing-related proteins. Based on these collective observations, we propose that IQGAP1 regulates the phosphorylation of splicing proteins, which potentially affects their splicing activities and, therefore, contributes to HNSCC.

Graphical Abstract

INTRODUCTION

Head and neck squamous cell carcinoma (HNSCC) is one of the most frequent human cancers worldwide.^1,2 Around 50,000 new cases of HNSCC were diagnosed in the United States in 2020, and the current 5-year survival rate is about 60%.^1,2 HSNCC begins in the squamous cells of the mouth and throat regions, and the largest risk factors for HNSCC include smoking, heavy alcohol consumption, and infection with high-risk human papillomavirus (HPV).³ Activation of the epidermal growth factor receptor (EGFR)-Phosphoinositol-3 kinase (PI3K) pathway is often observed in HNSCC.^4,5 Efficient PI3K signaling pathway requires scaffolding of the multiple, enzymatic components of this pathway by IQ motif-containing GTPase-activating protein 1 (IQGAP1).⁶ IQGAP1 is a multifunction protein that participates in many different signaling pathways and is overexpressed in many cancers, including HNSCC.^7-9 Loss of IQGAP1 or blocking the interaction between IQGAP1 and PI3K can inhibit the PI3K signaling pathway.^6,10-12 Recently, we showed that IQGAP1 contributes to the development of HNSCC, at least in part, through PI3K signaling.^11,12 However, the exact underlying mechanisms by which IQGAP1 promotes HNSCC remain largely unknown.

Alterations in protein phosphorylation play a vital role in cellular signaling.^13,14 To study IQGAP1-mediated signaling pathways in the head and neck region in an unbiased and global manner, we leveraged mass spectrometry (MS)-based phosphoproteomics to identify IQGAP1-mediated changes in the phosphoproteome. We began by quantifying differences in protein phosphorylation between Iqgap1^−/− and Iqgap1^+/+ (wild type) mouse tissues. Loss of IQGAP1 led to significant differences in several signaling cascades, such as RAS and RHO mediated signaling, and multiple cellular processes, including cell cycle, multiple metabolic processes, and RNA splicing. While many of these changes were predicted, based upon known functions of IQGAP1, its effect on phosphorylation of RNA splicing factors was unanticipated. To validate this finding in human cells, we turned to MS-based phosphoproteomics studies in tert-immortalized, normal human oral keratinocytes (NOKs) in which we knocked out IQGAP1 using CRISPR/Cas9 technology. Again, we observed significant alterations in the phosphorylation of RNA splicing factors. Given our evidence for a role of IQGAP1 in HNSCC, we interrogated the Clinical Proteomic Tumor Analysis Consortium (CPTAC) and the Cancer Genome Atlas (TCGA) databases and found phosphorylation of splicing-related proteins to be important in HNSCC prognosis, indicating a potential link between IQGAP1 and splicing in the context of this common human cancer.

Prior literature suggests a role of alternative splicing in HNSCC. Multiple studies have attempted to systematically identify alternative splicing events associated with HNSCC prognosis.^15-19 For example, re-clustering TCGA HNSCC RNA-Seq data based on alternative splicing events resulted in four clusters that highly resemble the four molecular subtypes of HNSCC.¹⁷ Additionally, a previous phosphoproteomics and proteomics study comparing normal versus HNSCC cell lines identified RNA splicing as one of the most changed cellular processes in this cancer type.²⁰ The levels of phosphorylated and total levels of splicing factors and their related kinases were found to be upregulated in HNSCC.^20,21 Furthermore, inhibiting SR protein kinase 2 (SRPK2) reduced colony-forming and invasive capacity of HNSCC cell lines.²⁰ Together, these studies implicate alterations in RNA splicing to be playing a role in HNSCC.

In this report, we describe our phosphoproteomics studies in two different systems that implicate IQGAP1 in modulating the phosphorylation of splicing factors and other splicing-related proteins. From our results, we posit that IQGAP1 regulates splicing protein phosphorylation via PI3K signaling, which affects splicing activity and alters cancer-associated splicing events, eventually contributing to HNSCC.

EXPERIMENTAL PROCEDURES

Mouse tissues.

5-week-old Iqgap1^+/+ and Iqgap1^−/− mice of mixed genetic backgrounds (F1 progenies of FVB mice crossed to 129+C57BL/6 mice) were used for mass spectrometry analysis. We included 5 Iqgap1^+/+ (2 males, 3 females) and 6 Iqgap1^−/− (3 males, 3 females) mice. Full-length esophagi were harvested from these mice and embedded in optimal cutting temperature compound (OCT) for preservation. The samples were stored at –80°C until the time of extraction.

Cell culture.

Tert-immortalized NOKs (a gift from Dr. Karl Munger, Tufts University School of Medicine, Boston, MA) were cultured in Keratinocyte Serum-free Medium (KSFM) supplemented with human epidermal growth factor and bovine pituitary extract (Gibco, Thermo Fisher Scientific).²² IQGAP1 knockout NOKs (NOKs^IQGAP1KO) cells were previously generated using CRISPR-Cas9.¹² For MS experiments, NOKs and NOKs^IQGAP1KO cells were grown on 10 cm dishes until they reached 70% confluency, collected, and washed with ice-cold phosphate-buffered saline (PBS). The resulting cell pellets were stored at –80°C until the time of protein extraction. Four replicates of each cell line were analyzed.

Mouse keratinocytes were isolated from the skin of neonate pups. After incubation in phosphate-buffered saline (PBS) containing 10% antibiotics for 2 min, skin pieces were incubated in 0.25% trypsin overnight at 4°C. The epidermis was then separated from the dermis using sterile forceps, minced with a single-edge razor blade, and then stirred for 1 h at 37°C in F-medium (composed of 3 parts Ham's F12 medium to 1 part Dulbecco's modified Eagle's medium and supplemented with the following components: 5% fetal bovine serum (FBS), adenine (24 μg/ml), cholera toxin (8.4 ng/ml), epidermal growth factor (10 ng/ml), hydrocortisone (2.4 μg/ml), insulin (5 μg/ml)) to generate a single-cell suspension. The cells were strained using 0.7-mm membrane (102095-534; VWR) and cultured in F-medium containing 10 μM Rho-kinase inhibitor (Y-27632; Selleck Chemical) in the presence of mitomycin C (M4287; Sigma)-treated 3T3 J2 fibroblasts.

Cell lysis and protein digestion.

Frozen mouse esophageal tissues were carefully extracted from OCT, resuspended in lysis buffer (6 M guanidine hydrochloride, 100 mM Tris pH 8), and probe sonicated (Misonix) until homogenized. After sonication, protein concentration of the lysates was determined by BCA assay (Pierce). Each lysate contained about 2 mg of protein. Aliquots of 300 μg per lysate were subsequently precipitated by adding methanol to 90% and centrifuging at 10,000 x g for 5 minutes. The supernatant was discarded, and the protein pellets were air dried for 5 minutes. After drying, the pellets were resuspended in digestion buffer (8 M urea, 10 mM Tris(2-carboxyethyl)phosphine (TCEP), 40 mM chloroacetamide (CAA), 100 mM Tris) and sonicated in a bath sonicator (Qsonica) for 7.5 minutes at 4 °C. Lysyl endopeptidase (LysC, Wako Chemicals) was added to an estimated 100:1 protein/enzyme ratio, and the samples were incubated at room temperature on a rocker for 4 hours. The samples were then diluted with 100 mM Tris to a final urea concentration of 1.5 M, and trypsin (Promega) was added to an estimated 50:1 protein/enzyme ratio. The samples were incubated on a rocker overnight at room temperature, and the resulting peptides were acidified to pH 2 with trifluoroacetic acid (TFA) and desalted with Strata-X Polymeric solid phase extraction cartridges (Phenomenex). The Strata-X cartridges were prepared by flowing 100% acetonitrile (ACN), followed by 0.2% formic acid. Each acidified sample was then loaded onto the cartridges, and the bound peptides were washed with 0.2% formic acid. The peptides were eluted with 80% ACN/0.2% TFA and subsequently dried with a SpeedVac Vacuum Concentrator (Thermo Scientific).

NOKs were treated identically to the esophageal tissue for the lysis and digestion workflow above except for the following changes: frozen cell pellets were resuspended in lysis buffer and bath sonicated (Qsonica) with a program of 20 seconds on/10 seconds off for 12 minutes, and aliquots of 400 μg per lysate were used for digestion and subsequent steps. For the following sections, the esophagus peptide samples and NOKs peptide samples underwent the same steps, but independently. Points where there were differences between the two workflows are noted.

Isobaric labeling of peptides.

Dried peptides were resuspended in 200 mM triethylammonium bicarbonate (TEAB) and labeled with tandem mass tags (TMT, Thermo Scientific) according to manufacturer instructions. Label efficiency was verified for each of the tags, and equal amounts of labeled peptides from each sample were combined. The resulting mixture (~3.3 mg or ~3.2 mg of peptides in total for esophagus and NOKs, respectively) was desalted and dried as described above.

Phosphopeptide enrichment.

The dried tryptic peptide mixture was dissolved in wash buffer (80% ACN, 6% TFA) and enriched for phosphopeptides using Ti(IV)-immobilized metal affinity chromatography beads (Resyn Biosciences).²³ For every 1 mg of peptides, 50 μL of beads was used. The beads were first washed three times with wash buffer and then incubated with the sample on a vortexer for 60 minutes at room temperature. The supernatant was removed and saved (representing the unbound, unmodified peptides), and the beads (containing the bound phosphopeptides) were washed three times with wash buffer, one time with 80% ACN, one time with 0.5 M glycolic acid/80% ACN, and three more times with 80% ACN. The bound phosphopeptides were finally eluted with 1% NH₄OH/50% ACN. Both peptides and phosphopeptides were desalted and dried as described above.

Peptide and phosphopeptide fractionation.

The entire yield of phosphopeptides (33 μg for esophagus and 32 μg for NOKs, as determined using a NanoDrop spectrophotometer [Thermo Scientific]) and 500 μg of peptides were resuspended in 0.2% formic acid and fractionated by high-pH reversed-phased chromatography across a 4.6 mm × 150 mm, 3.5 μm XBridge BEH C18 column (Waters) with a 1260 Infinity II HPLC system (Agilent). Mobile phase A was 10 mM ammonium formate and mobile phase B was 10 mM ammonium formate in 80% methanol. The peptides and phosphopeptides were separated over a 25-minute gradient in which mobile phase B increased to 35% at 2 minutes, 75% at 8 minutes, and 100% at 14 minutes. The method ended with a 1-minute wash at 100% B and a 10-minute equilibration at 0% B. The flow rate was held at 0.8 mL/minute. The fractions were collected directly in a round-bottom 96-well plate, and detection of eluting fractions was accomplished with variable wavelength detector. Phosphopeptides and unmodified peptides were separated into 8 and 16 final fractions, respectively.

Mass spectrometry data acquisition.

Each peptide and phosphopeptide fraction was dried as described above and resuspended in 0.2% formic acid. The fractions were injected onto a 75 μm i.d. x 360 μm o.d. capillary column with an embedded emitter that was fabricated in-house and packed to 30 cm with 1.7 μm BEH C18 particles.²⁴ The column was installed onto a Dionex UltiMate 3000 nano high performance liquid chromatography (HPLC) system (Thermo Scientific) and held at 50 °C using a heater that was fabricated in-house.²⁵ For the esophagus samples, the peptide and phosphopeptide fractions were loaded in 100% mobile phase A (0.2% formic acid in water) and eluted with increasing % of mobile phase B (0.2% formic acid in 70% ACN) over 90 minutes. For each peptide fraction, 2 μg was injected onto the column, while one-third of the total volume was injected for phosphopeptide fractions. Two injection replicates were analyzed for each phosphopeptide fraction. Eluted peptides and phosphopeptides were analyzed by an Orbitrap Fusion Lumos (Thermo Scientific). Orbitrap survey scans were performed at a resolving power of 60,000 with an automatic gain control (AGC) target of 1 x 10⁶, maximum injection time of 50 ms, and scan range of 300-1350 m/z. The instrument was operated in Top Speed mode with cycle times of 1 s. Precursor ions were isolated in the quadrupole with an isolation window of 1.3 m/z and fragmented in the higher energy collisional dissociation (HCD) cell with a collision energy of 35%. MS/MS scans were performed in the Orbitrap with a resolving power of 60,000. Charge states 2-4 were included, AGC target was 5 x 10⁴, scan range was 100-1200 m/z, and maximum injection time was 118 ms.

The differences in the NOKs acquisition workflow are as follows: separations occurred over 120 minutes, eluted peptides and phosphopeptides were analyzed by an Orbitrap Eclipse (Thermo Scientific), precursor ions were isolated in the quadrupole with an isolation window of 1.5 m/z, charge states 2-6 were included, AGC target was 1 x 10⁵, and dynamic scan range began at 100 m/z.

Mass spectrometry data analysis.

Raw files were processed with MaxQuant (version 1.5.2.8) and searched against a database of reviewed mouse proteins plus isoforms (downloaded from UniProt on 6 April 2018) or reviewed human proteins plus isoforms (downloaded from UniProt on 20 January 2022).^26,27 Cysteine carbamidomethylation was set as a fixed modification, and methionine oxidation and acetylation (protein N-terminus) were set as variable modifications. For the phosphopeptide fractions, phospho(STY) was added as a variable modification. Default searching parameters were used, and TMT reporter ion isotopic impurities were accounted for based on manufacturer specifications for the product lot. Searches were performed using a 1% false discovery rate (FDR).

The “proteinGroups.txt” and “Phospho(STY) Sites.txt” output files from MaxQuant were filtered, analyzed, and visualized in Perseus (version 1.6.0.7).²⁸ Reverse sequences, sequences only identified by site, and contaminants were removed, and a localization probability cutoff of 0.75 was applied for the phosphorylation analyses. The data were log₂-transformed, and proteins or sites were removed if they were not observed in all channels. No normalization of channels was performed. Statistical comparisons between groups were performed with two-tailed Student’s t-tests, and resultant p-values were corrected for multiple hypotheses with the Benjamini-Hochberg algorithm (yielding q-values). Specific proteins and phosphosites in the text were considered significantly up- or downregulated at q < 0.05 and log₂ fold change > ∣0.5∣. The log₂ fold change cutoff was chosen based on the observation that most protein or phosphosite changes were within a log₂ fold change of ±1 due to TMT ratio compression and tissue heterogeneity.²⁹ Proteomics and phosphoproteomics output files from MaxQuant for esophagus are given in Supplemental File 1 and for NOKs in Supplemental File 2.

Hierarchical clustering was performed in Perseus using Euclidean distance, average linkage, and k-means processing. Sequence motif enrichment was also performed in Perseus for phosphosites with q < 0.05 and log₂ fold change > ∣0.5∣ using a Fisher’s exact test. For relative enrichment, “protein” was selected. Cluster analysis was performed with the Database for Annotation, Visualization, and Integrated Discovery (DAVID, 2021 Update) for proteins with differentially regulated phosphosites (p < 0.01, log₂ fold change > ∣0.5∣).³⁰ GO terms, KEGG pathways, and Reactome pathways were identified. Protein-protein interaction clustering was performed with Search Tool for the Retrieval of Interacting Genes/Proteins (STRING). Proteins with phosphosites q < 0.05 were included. Default parameters were used, except for a high confidence score (0.700); textmining, experiments, and databases for interaction sources; a maximum of 5 first shell interactors; and MCL clustering. Interactions with over 2 proteins were kept. KEGG pathways and UniProt annotation terms were extracted from the STRING analysis for the NOKs.

Raw file availability.

MS raw files were deposited to the MassIVE database with the identifier MSV000087770.

Protein lysate preparation for immunoblotting.

For tissues, frozen skin and esophagus samples were minced into small pieces on ice using razor blades, homogenized in 300 μL RIPA buffer (25mM Tris pH 8, 150mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X-100) with protease and phosphatase inhibitors using a plastic pestle (Axygen), and incubated on an orbital shaker at 4 °C for 20 minutes. The homogenates were centrifuged at 14,000 rpm for 20 minutes at 4 °C, and the supernatants (protein lysates) were collected. For tissue culture cells, cells were harvested at sub-confluency and lysed with RIPA lysis buffer as described above. Protein concentrations were determined using the Bradford Protein Assay (Bio-Rad). Equivalent amounts of protein were resolved on precast Mini-PROTEAN TGX 7.5% gels (Bio-Rad) and transferred to nitrocellulose membranes. The membranes were blocked with 5% nonfat dry milk in Tris-buffered saline with 0.1% Tween 20 (TBST). The primary antibodies that were used are summarized in Table S1. Horseradish peroxidase-conjugated secondary antibodies (1:10,000, Jackson ImmunoResearch) and chemiluminescent substrates (Clarity ECL Substrates; Bio-Rad) were used for visualization on a Bio-Rad ChemiDoc Imaging System.

RESULTS AND DISCUSSION

To comprehensively dissect which pathways are mediated by IQGAP1 in head and neck epithelium in vivo, we performed phosphoproteomics using esophageal tissue lysates from Iqgap1^+/+ and Iqgap1^−/− mice (Figure 1). Additionally, we used the unbound peptides from the enrichment flow-through to perform proteomics to complement the phosphoproteomics results. To obtain quantitative information for proteins and phosphosites, we used TMT isobaric labeling. We chose to study esophagus over tongue—another easily obtainable source of murine head and neck tissue—for two major reasons. First, the percentage of epithelium is higher in esophagus than tongue, which makes it easier to study the epithelial signaling changes. Second, we observed more significant differences in cancer formation upon carcinogen treatment in the esophagus than in the tongue between Iqgap1^+/+ and Iqgap1^−/− mice.¹¹ This observation indicates that IQGAP1 may play a more significant role in promoting tumors in the esophagus and therefore have larger impact on esophageal cell signaling. To prepare the samples, we collected whole-length esophagus from five Iqgap1^+/+ mice (3 females, 2 males) and six Iqgap1^−/− mice (3 females, 3 males), which were embedded in OCT prior to their preparation for MS analysis.

Figure 1. Mass spectrometry workflow for analyzing Iqgap1^+/+ and Iqgap1^−/− mouse esophagus.

Proteins were extracted from the tissue and digested into peptides with trypsin. Peptides from each tissue sample were labeled with a different TMT channel. Channels were tested for label efficiency and then mixed. Phosphopeptides were enriched from the mixture, while the unmodified peptides were retained from the flow-through. Phosphopeptides and peptides were fractionated by high-pH reversed-phase chromatography. Each fraction was ionized and analyzed by the mass spectrometer. “KO” refers to knockout tissue (Iqgap1^−/−), and “WT” refers to wild type tissue (Iqgap1^+/+).

Overall, we quantified 5,830 proteins and 10,725 sites of phosphorylation (corresponding to 3,136 unique phosphoproteins). Before performing our analyses, we investigated the quality of our proteomics and phosphoproteomics data by assessing peptide and phosphopeptide matching scores (Figure S1A-B), coefficients of variation (Figure S1C-D), and dendrograms for the proteome and phosphoproteome (Figure S1E-F). The high median matching scores, tight coefficients of variation, and expected genotype clustering all indicated that the data were of high quality. We then performed Student’s t-tests, which yielded 984 and 1,751 significantly changing proteins and phosphosites, respectively, between Iqgap1^+/+ and Iqgap1^−/− tissue lysates (p-value < 0.05). To obtain more robust changes for further biological interrogation, we applied a stringent p-value correction (Benjamini-Hochberg q-value < 0.05) and a fold change cutoff (log₂ > ∣0.5∣), which narrowed the number of significantly changing proteins and phosphosites to 25 and 114, respectively (Figure 2A-B).

Figure 2. Changes in the proteome and phosphoproteome between Iqgap1^+/+ and Iqgap1^−/− mouse esophagus.

Volcano plots show quantified proteins (A) and quantified and localized phosphosites (B). Plotted q-values were corrected with the Benjamini-Hochberg algorithm. Highlighted up- (orange) or downregulated (green) proteins or phosphosites had q < 0.05 and log₂ fold change > ∣0.5∣ between Iqgap1^−/− and Iqgap1^+/+ mice. “FC” refers to fold change.

As expected, IQGAP1 was the most downregulated protein, approaching a 4-fold decrease in expression levels in Iqgap1^−/− tissues compared to Iqgap1^+/+ tissues. Olfactomedin-like 1 protein (OLFML1) was the most upregulated protein in the absence of IQGAP1, with a more than 7-fold increase. OLFML1 plays a role in HIPPO signaling by negatively regulating YAP1-associated transcription.³¹ The change in OLFML1 is in line with a report that IQGAP1 promotes YAP1-associated transcription; however, more evidence is needed to confirm such an association.³² Because fewer overall changes were detected in the proteomics dataset and we were primarily interested in deciphering the cellular signaling pathways that involve kinases, we focused on the changes in abundance of phosphorylation rather than of proteins. The proteomics dataset was used to confirm that changes in phosphorylation did not occur simply due to a significant change in the total protein amount.³³

Two sites (S314 and S316) on SR-rich splicing factor 6 (SRSF6) were the most downregulated of all phosphorylation events in the absence of IQGAP1 (MS/MS shown in Figure S2A; log₂ fold change = −2.22; q-value = 0.02). The proteomics data indicated that SRSF6 was not altered in its protein levels in the absence of IQGAP1 (log₂ fold change = −0.06; q-value = 0.83). SRSF6 belongs to the SR protein family of splicing factors that regulate the assembly of spliceosomes.^34-37 Compared to other members in this family, such as SRSF1 and SRSF2, SRSF6 has been relatively understudied, especially in the context of cancer. A few studies have suggested that SRSF6 acts as an oncogene.^38-40 SRSF6 is overexpressed in colon cancer, lung cancer, and, interestingly, HNSCC.^38,41

The subsequently most downregulated phosphosite was S591 on Protein Niban 1 (FAM129A), a positive regulator of protein translation (MS/MS shown in Figure S2B).⁴² Protein translation is one of the cellular processes controlled by PI3K/AKT/mTOR signaling, which is known to be modulated by IQGAP1. In contrast, the phosphorylation of S273 in Na(+)/H(+) exchange regulatory cofactor (NHE-RF1, or SLC9A3R1) was the most upregulated site in the absence of IQGAP1, with a more than 6-fold increase (MS/MS shown in Figure S2C). NHE-RF1/SLC9A3R1 is an actin-binding protein that may also play a role in PI3K/PTEN/AKT as well as Wnt signaling, which is also mediated by IQGAP1.^6,9,43,44 Altogether, our analyses identified many proteins with significantly altered levels of phosphorylation due to the loss of IQGAP1. Several of these proteins are associated with IQGAP1 in various cellular processes, which help to pinpoint the IQGAP1-mediated pathways in the head and neck, including ones known to contribute to HNSCC.

To analyze the phosphoproteomics data more globally, we first examined enriched linear sequence motifs among the significantly changing sites. The top 10 enriched motifs include substrate motifs of DNA-dependent protein kinase, Glycogen synthase kinase-3 (GSK3), β-adrenergic receptor kinase, MAPKAPK2 kinase, Casein kinase 1, Pyruvate dehydrogenase kinase (PDK), Calmodulin-dependent protein kinase II, Checkpoint kinase 1, Growth-associated histone H1 kinase, and 14-3-3 domain-binding motif (Figure 3A). Most of these kinases are associated with IQGAP1 or IQGAP1-mediated signaling. For example, DNA-dependent protein kinase and PDK are responsible for AKT phosphorylation, a critical step in PI3K/AKT/mTOR signaling.^6,45,46 Furthermore, GSK3 casein kinase 1 is involved in Wnt signaling, and both 14-3-3 proteins and calmodulin-dependent protein kinase are associated with CDC42, one of the most well-characterized IQGAP1-binding partners.^47-51 Interestingly, some of these kinases, including GSK3, Checkpoint kinase 1, MAPKAPK2, and 14-3-3 proteins, are linked to RNA splicing and processing, which builds upon the observation that SRSF6 had the most downregulated levels of phosphorylation with the loss of IQGAP1.^52-56

Figure 3. Pathway analysis of the altered phosphoproteome with the loss of IQGAP1 in mouse esophagus.

A) Top 15 linear sequence motifs that were enriched with the loss of IQGAP1. The p-values were FDR-corrected. B) Top 10 general clusters from DAVID for the proteins with significantly changing phosphosites with the loss of IQGAP1. C) STRING network for the proteins with significantly changing phosphosites with the loss of IQGAP1 (6 clusters are shown).

Next, we performed a cluster analysis of the proteins with significantly changing phosphosites using DAVID. The significantly altered phosphosites were enriched according to pathways from GO terms, KEGG pathways, and Reactome pathways. The top 10 general clusters were kinase activity, Rho GTPase signaling, metabolism, splicing, cell communication, cardiac regulation, muscle contraction, cell cycle, motility, and cell adhesion (Figure 3B). Therefore, the loss of IQGAP1 affects the dynamics of many cellular processes. Interestingly, splicing was one of the most changed pathways when IQGAP1 is absent, which is in line with the results from motif analysis and the downregulation of SR protein phosphorylation.

It is also worth noting that a high number of proteins altered in their phosphorylation in the absence of IQGAP1 are associated with metabolism. Investigations into the role of IQGAP1 in metabolism are limited. Most of the existing studies suggest that IQGAP1 positively regulates metabolic processes. For example, one study showed that Iqgap1^−/− mice have decreased transcription of genes related to gluconeogenesis and fatty acid synthesis, and another showed that IQGAP1 interacts with and mediates the activation of NRF2, one of the master regulators of cell metabolism.^57-59 IQGAP1 also participates in the exocytosis and endocytosis of insulin.^60,61 Suppressing IQGAP1 lowers insulin secretion from β-cells.⁶⁰ The IQGAP1-Rab27a interaction is also implicated in insulin endocytosis.⁶¹ Increased metabolic activity may be due to increased cell cycle activity, which was another top cluster. IQGAP1 has been described previously to promote cell proliferation.^62-68 Nevertheless, our previous studies did not observe increased levels of proliferation in Iqgap1^−/− mice compared to Iqgap1^+/+ mice, reflected by the numbers of BrdU-positive cells.¹² Overall, our phosphoproteomics data showed that loss of IQGAP1 affects the dynamics of multiple cellular processes.

To complement the linear sequence motif and pathway analyses, we used the STRING database to identify known interactions among the proteins with significantly changing phosphosites. After applying our filtering criteria, interactions were found among 34 proteins, with a protein-protein interaction enrichment p-value of 5.37E-10, indicating that the interactions identified are significant. These 34 proteins were separated into 6 interaction clusters (Figure 3C). These interaction clusters include proteins involved in cytoskeleton regulation (cluster 1), stress response (cluster 2), spliceosome regulation (cluster 3), cell cycle (cluster 4), metabolism (cluster 5), and cell adhesion (cluster 6). Some of these clusters are known to be related to IQGAP1, such as cytoskeleton regulation and mitotic signal transduction. Novel, again, was an alignment with spliceosome regulation (cluster 3). Altogether, these global analyses of the phosphoproteomics data strongly implicate that IQGAP1 plays a role in regulating RNA splicing. We decided to focus further investigation into the link between IQGAP1 and splicing. First, we confirmed the phosphoproteomics results for SRSF6 by western blotting. We collected both esophagus and skin tissues from Iqgap1^+/+ and Iqgap1^−/− mice, prepared protein lysates, and examined phospho-SRSF6 levels (Figure S3A-B). We also pursued in vitro studies by isolating primary keratinocytes from neonates of Iqgap1^+/+ and Iqgap1^−/− mice and examining phospho-SRSF6 levels (Figure S3C; Supplemental File 3).¹² Each of these western blots confirmed a reduction in phospho-SRSF6 upon loss of IQGAP1.

Next, to extend our analyses to human cells, we conducted a second unbiased phosphoproteomics/proteomics experiment using NOKs, a tert-immortalized human oral keratinocyte cell line that has been extensively used for human head and neck cancer studies. We had previously generated NOKs knocked out for IQGAP1 (NOKs^IQGAP1KO) using CRISPR/Cas9 to study the role of IQGAP1 in head and neck carcinogenesis mediated by human papillomavirus.¹² We quantified 8,345 proteins and 16,626 sites of phosphorylation (corresponding to 4,573 unique phosphoproteins) in the NOKs. Of these, 173 proteins and 453 phosphosites were significantly changing between NOKs and NOKs^IQGAP1KO (q-value < 0.05 and log₂ > ∣0.5∣). As expected, IQGAP1 levels were largely reduced in the NOKs^IQGAP1KO cells when compared to wild type NOKs (over an 8-fold decrease). We used STRING to analyze the proteins with significantly changing phosphosites. Again, the KEGG pathway and UniProt annotation outputs from STRING revealed splicing to be highly implicated: spliceosome, alternative splicing, and mRNA splicing (Figure 4A-B). Other pathways and processes identified were related to ErbB signaling pathway, Rap1 signaling pathway, MAPK signaling pathway, PI3K-Akt signaling pathway, cytoskeleton, and cell cycle, which are all cellular processes regulated by IQGAP1. Additionally, many related post-translational modification (PTM) terms were identified in the NOKs upon loss of IQGAP1: phosphoprotein, methylation, Ubl conjugation, and acetylation.

Figure 4. Alterations in splicing in the NOKs with the loss of IQGAP1.

A) Top 11 KEGG pathways for the proteins with significantly changing phosphosites. B) Top 10 UniProt annotation terms for the proteins with significantly changing phosphosites. C) STRING network for the splicing factors and splicing-related proteins with significantly changing phosphosites. D) Sequence windows and direction of change for the proteins shown in the STRING network.

Focusing in on splicing, we found several splicing factors and related proteins with significantly changed phosphosites (around 2-fold increases or decreases) in the NOKs when IQGAP1 was knocked out: RNA-binding motif protein X-linked (RBMX, S102), Matrin-3 (MATR3, S206), Heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1, S6), Pre-mRNA processing factor homolog B (PRPF40B, S853), Apoptotic chromatin condensation inducer 1 (ACIN1, S230), SR repetitive matrix 1 (SRRM1, S421), SR repetitive matrix 2 (SRRM2, S1424), Coiled-coil domain containing 94 (CCDC94, S179), SR-rich splicing factor 4 (SRSF4, S431), SR-rich splicing factor 5 (SRSF5, S117), and SR-rich splicing factor 9 (SRSF9, S204). Annotated MS/MS spectra of each of the phosphopeptides containing these sites are shown in Figure S4. None of these proteins were significantly changed in their total protein levels (q > 0.05), which indicates that the phosphorylation changes are not due to changing protein levels. When inputting these proteins separately into STRING, we found that they all clustered together (Figure 4C). Most of these sites had decreased levels of phosphorylation with the loss of IQGAP1, except for SRRM1 and SRRM2, which were increased (Figure 4D). Although S314 and S316 in SRSF6 were the most downregulated phosphosites in esophagus (Figure 2B), they were unchanged in the NOKs, which perhaps points to species-specific differences among splicing proteins and their interactions with IQGAP1.

Several of the splicing factors and related proteins altered in NOKs^IQGAP1KO have been previously implicated in HNSCC. A recent phosphoproteomics investigation of HNSCC cell lines found many of the same proteins identified in our dataset to be hypo- or hyperphosphorylated.²⁰ Another investigation of HNSCC cell lines and tissues found MATR3, an RNA-binding protein and regulator of alternative splicing, to be upregulated in HNSCC.^69,70 Additionally, SRSF5, a splicing factor involved in constitutive splicing that can modulate the selection of alternative splice sites, has been shown to be upregulated in HNSCC tumors.⁷¹ Lower expression of SRSF9, another splicing factor similar to SRSF5, and RBMX, a protein implicated in alternative splicing that interacts with SRSF10, correlates with improved survival in HNSCC patients.^72-74 Searching the CPTAC HNSCC data also revealed that many of these splicing-related proteins have significantly altered phosphorylation levels between HNSCC tumors and normal tissue.⁷⁵ Two of the significantly changing phosphosites in our NOKs dataset (S853 in PRPF40B and S1424 in SRRM2) were significantly changed in the same direction between tumor and normal tissue as in the NOKs and NOKs^IQGAP1KO (without significant change in the proteins).

Beyond HNSCC, these splicing-related proteins, particularly splicing factors, have been implicated in many other cancers.⁷⁶ For example, many tumors overexpress SRSF9, such as those in glioblastoma, squamous cell lung carcinoma, and malignant melanoma.⁷⁷ Overexpression of SRSF9 (along with SRRM1) has also been associated with clinical parameters of aggressiveness (e.g., metastasis) in prostate cancer.⁷⁸ SRSF5 is overexpressed in lung cancer and colorectal cancer tissues.^79,80 PRPF40B, a splicing processing factor, is underexpressed in acute myeloid leukemia.⁸¹ RBMX and its related proteins are involved in multiple cancers, such as bladder cancer and hepatocellular cancer.^82-84 HNRNPA1, a multipurpose RNA-binding protein and splicing factor, is also highly associated with cancer and can regulate many alternative splicing events.⁸⁵ For example, HNRNPA1 is an oncogenic driver in glioblastoma.⁸⁶ Therefore, the changes in phosphorylation of these proteins in the NOKs with the loss of IQGAP1 are reasonable not only within the context of HNSCC but also within the larger context of cancer.

Our phosphoproteomics results from two different mammalian species provide evidence for a potential link between IQGAP1 and the phosphorylation of splicing factors and splicing-related proteins. These findings correlated with the CPTAC HNSCC tumor phosphoproteomics data. However, we were also interested in whether gene expression datasets indicate a role for splicing factor phosphorylation in HNSCC. To test for this, we examined whether splicing factor phosphorylation affects HNSCC prognosis using the TCGA RNA-Seq HNSCC dataset.⁸⁷ While we cannot directly quantify splicing factor phosphorylation in gene expression data, levels of SR protein kinases (SRPKs) can serve as partial proxies.^88,89 SRPKs regulate pre-mRNA splicing by phosphorylating spliceosome components and SR proteins, including SRSF4, SRSF5, SRSF6, and SRSF9.⁸⁹ SRPK2 was overexpressed in HNSCC cell lines, and inhibiting SRPK2 reduced the colony-forming and invasive ability of HNSCC cells.²⁰ Interestingly, the AKT-SRPK-SR axis links SR protein phosphorylation with PI3K signaling, which may be the manner by which IQGAP1 regulates splicing factor phosphorylation.^90,91 Inhibiting PI3K activity reduced SR phosphorylation.^90-92 In a recent report studying protein-protein interactions in HNSCC, multiple SR proteins were associated with proteins in the PI3K signaling pathway (EGFR, HER2, PI3K, PTEN) in three different cell lines, which further strengthens the hypothesis that PI3K signaling can regulate SR proteins in HNSCC.⁹³

To test for an association between SRPKs and HNSCC patient survival, we performed a Kaplan-Meier survival analysis on the TCGA RNA-Seq dataset regarding SRPK levels. For these analyses, the expression cutoff for each gene was determined by the FPKM (fragments per kilobase of transcript per million mapped reads) value that yielded the maximum difference regarding survival between the two groups at the lowest log-rank p-value. The results showed that neither SRPK1 nor SRPK2 alone was significantly associated with 5-year HNSCC patient survival (Figure 5A-B; log-rank test: SRPK1, p = 0.06; SRPK2, p = 0.17). However, when considering levels of both SPRK1 and SPRK2 together, patients with low levels of both SPRK1 and SPRK2 showed significantly improved 5-year survival rates compared to those with high levels of SRPK1 and SRPK2 (Figure 5C; log-rank test: p = 0.03). From these analyses, we can speculate that SRPKs may play a role in HNSCC, and by inference, the phosphorylation of SR protein, including splicing factors.

Figure 5. Kaplan-Meier survival analysis for SRPK and HNSCC prognosis correlation.

This analysis was performed on the TCGA RNA-Seq dataset. A) Correlation between SRPK1 levels and 5-year survival of HNSCC patients (log-rank test p-value: 0.06). B) Correlation between SRPK2 levels and 5-year survival of HNSCC patients (log-rank test p-value: 0.17). C) Correlation between the combination of SRPK1 and SPRK2 levels and 5-year survival of HNSCC patients (log-rank test p-value: 0.03).

Lastly, we utilized the Biological General Repository for Interaction Datasets (BioGRID) to investigate whether there are reported interactions between IQGAP1 and splicing-related proteins that were curated from high-throughput screens and focused studies.^94,95 IQGAP1 has a total of 576 unique interactions in the repository. Of these, we found 24 interactions with splicing-related proteins (Figure S5). These interacting proteins are RNA helicases, RNA binding proteins, heterogeneous nuclear ribonucleoproteins (hnRNPs), regulators of mRNA processing and splicing, splicing factors, kinases of splicing factors, components of the spliceosome, and exporters of mRNA. Interestingly, of the specific proteins with significantly changing phosphosites in esophagus and NOKs that were mentioned above, five of them have known interactions with IQGAP1 in BioGRID (SRSF6, SRRM2, MATR3, ACIN1, and HNRNPA1). These BioGRID interactions support not only our discussions of specific phosphosites but also the hypothesis of IQGAP1’s association with splicing. Additionally, the variety of splicing-related activities carried out by these interacting proteins indicates that IQGAP1 may have a widespread role in regulating or influencing the phosphorylation of splicing.

One specific example of the interactions between IQGAP1 and splicing proteins is a study by Birladeanu et al. that links hnRNPs, another large family of splicing factors, to IQGAP1 in the context of gastric cancer.⁹⁶ Their study showed that IQGAP1 regulates alternative splicing through hnRNPs. In a gastric cancer cell line, they found that IQGAP1 can interact with spliceosome component proteins, including hnRNPs and SR proteins. By comparing cells with and without IQGAP1 using RNA-Seq, they identified many IQGAP1-dependent alternative splicing events, most of which related to cell cycle and cell division. We tested the 12 specific IQGAP1-dependent alternative splicing events validated in their study using RT-PCR. However, we did not observe significant differences in the alternative splicing patterns in these genes when comparing NOKs and NOKs^IQGAP1KO (data not shown), which indicates that these alternative splicing events are regulated by IQGAP1 either only in the gastric region or specifically in the context of cancers. This study also suggested that IQGAP1 mediates the SUMOylation status of hnRNPM, which controls its subnuclear localization and regulates the splicing activities of hnRNPM. These findings align with our hypothesis that IQGAP1 can regulate splicing activity by controlling the PTMs of splicing factors and related proteins.

Among the IQGAP1-dependent alternative splicing events identified in Birladeanu et al., the Fanconi Anemia complementation group I (FANCI) was one of the most frequently occurring events.⁹⁶ The FANCI protein is part of the Fanconi Anemia (FA) complex for DNA repair. FA deficiency has been indicated to be associated with HNSCC for a long time.^97,98 The loss of FA leads to increased invasion in HNSCC cell lines, at least in part, through the RAC1 pathway, which can also be mediated by IQGAP1.^49,99-101 Relatedly, we found one phosphorylation site in the FANCJ protein (S1237) in the NOKs dataset that was significantly upregulated by 1.7-fold with the loss of IQGAP1. Another interesting gene with IQGAP1-related alternative splicing events was anillin (ANLN). ANLN is an actin-binding protein that is involved in cell division and migration.¹⁰² Proteomics analysis showed that ANLN is overexpressed in HNSCC cells.²⁰ Overexpression of ANLN also correlates with poorer prognosis in HNSCC patients.^103,104 Knocking down ANLN in HNSCC cell lines decreases HNSCC cell invasion and increases cell apoptosis.^103,104 Interestingly, ANLN interacts with several splicing factors, including SRSF4, SRSF6, and SRSF9.^94,95 In our NOKs dataset, ANLN was significantly increased with the loss of IQGAP1 (up 1.8-fold). Additionally, the alternative spliced isoforms of ANLN contribute to HNSCC by regulating cell proliferation, migration, and invasion, which are activities regulated by IQGAP1.^104,105 Therefore, these recent studies raise the possibility that both ANLN and FANCI could be alternatively spliced genes regulated by IQGAP1 and contribute to HNSCC.

CONCLUSIONS

Here, we describe two global MS-based analyses investigating IQGAP1-dependent signaling in the context of head and neck cancer. By comparing the phosphoproteome profiles of esophagus from Iqgap1^+/+ and Iqgap1^−/− mice as well as NOKs and NOKs^IQGAP1KO cells, we detail the link between IQGAP1 and RNA splicing. Additional analysis of the CPTAC and TCGA HNSCC datasets and known IQGAP1 interactions in the BioGRID database provided further context to suggest that the association between IQGAP1 and RNA splicing is likely. Collectively, these results suggest that IQGAP1 is associated with alternative splicing by regulating the phosphorylation of splicing-related proteins, possibly via PI3K signaling, which may affect splicing events and in turn contribute to the development of HNSCC. Our results not only hint at the underlying mechanisms of how IQGAP1 contributes to HNSCC but also help to open a new area in investigating the function of IQGAP1, an oncogene involved in many types of cancer, as splicing regulation and involvement is one of the understudied activities of IQGAP1. With a better understanding of how IQGAP1 functions in different cancers, we can develop new and more precise targeted therapies in the future.