A Cross-Linking-Aided Immunoprecipitation/Mass Spectrometry Workflow Reveals Extensive Intracellular Trafficking in Time-Resolved, Signal-Dependent Epidermal Growth Factor Receptor Proteome

Abstract

Ligand binding to the cell surface receptors initiates signaling cascades that are commonly transduced through a protein–protein interaction (PPI) network to activate a plethora of response pathways. However, tools to capture the membrane PPI network are lacking. Here, we describe a cross-linking-aided mass spectrometry workflow for isolation and identification of signal-dependent epidermal growth factor receptor (EGFR) proteome. We performed protein cross-linking in cell culture at various time points following EGF treatment, followed by immunoprecipitation of endogenous EGFR and analysis of the associated proteins by quantitative mass spectrometry. We identified 140 proteins with high confidence during a 2 h time course by data-dependent acquisition and further validated the results by parallel reaction monitoring. A large proportion of proteins in the EGFR proteome function in endocytosis and intracellular protein transport. The EGFR proteome was highly dynamic with distinct temporal behavior; 10 proteins that appeared in all time points constitute the core proteome. Functional characterization showed that loss of the FYVE domain-containing proteins altered the EGFR intracellular distribution but had a minor effect on EGFR proteome or signaling. Thus, our results suggest that the EGFR proteome include functional regulators that influence EGFR signaling and bystanders that are captured as the components of endocytic vesicles. The high-resolution spatiotemporal information of these molecules facilitates the delineation of many pathways that could determine the strength and duration of the signaling, as well as the location and destination of the receptor.

Graphical Abstract

INTRODUCTION

Protein–protein interaction (PPI) networks determine many fundamental biological functions in cells, and mapping PPI networks through a proteome-wide approach has been an active research field for deep understanding of many important cellular processes.^1–4 As an emerging powerful strategy, affinity enrichment with or without genetic manipulation followed by mass spectrometry (AP-MS) has been utilized to generate protein complex data.^5–9 While most efforts have been focused on soluble protein complexes at steady states, PPI networks of membrane proteins, such as transmembrane receptors and receptor tyrosine kinases that function in response to extracellular stimuli and initiate dynamic intracellular signaling, remain to be explored.

Transmembrane receptors typically contain at least one, often multiple transmembrane domains that facilitate the anchoring of the receptor to the plasma membrane. This property presents a challenge for isolating membrane-bound protein complexes as high detergent-containing buffers used for solubilizing membrane-bound proteins could disrupt protein–protein interactions, particularly the weak and transient interactions. Chemical cross-linking results in covalent bonds that freeze the interactions and therefore is considered a useful strategy to preserve weak and transient protein–protein interactions during detergent extraction and stringent wash conditions.

As one of the most extensively characterized cell surface receptors, epidermal growth factor receptor (EGFR, also known as ERBB1 or HER1) plays pivotal roles in many essential biological processes, including embryonic development and tissue homeostasis by promoting cell proliferation, differentiation, and survival.^10–12 Without ligand engagement, EGFR exists as a membrane-bound monomer; upon ligand stimulation, EGFR undergoes conformational change that facilitates its dimerization and internalization and induces its autophosphorylation and phosphorylation of a plethora of substrates to activate downstream signaling pathways. EGFR-activated pathways, including PI3K/AKT/mTOR, RAS/ MAPK, JAK/STAT3, and JNK,^10,12 are critical for cell growth and survival. Moreover, hyperactivation of EGFR by gene amplification, protein overexpression, or constitutive kinase activation mutations has been found in many types of human malignancies, including lung, breast, brain, head and neck, and colon cancers.^13–16 EGFR monoclonal antibodies and small-molecule inhibitors have been successfully used in clinical treatment of lung, colon, and head and neck cancers with oncogenic EGFR.^16–18

The EGFR-mediated PPI network has been extensively studied, and as a result, more than 150 proteins with biochemical evidence and nearly 1000 proteins through high-throughput screening have been reported as EGFR-binding partners in the literature.^6,19–23 It has been well established that activated EGFR undergoes internalization and endocytic trafficking through multiple intracellular vesicle compartments and is eventually degraded in the lysosome or recycled back to the plasma membrane.^24–26 Therefore, it is likely that the large number of interacting proteins reported in the literature represents a collection of proteins that associate with EGFR as the receptor undergoes the dynamic translocation through various intracellular vesicles. Although the endocytic trafficking was initially viewed as a means of signal attenuation following RTK activation, more recent studies showed that signal transduction continues to be active during receptor trafficking,^24,27,28 which raises the question of what roles the trafficking proteins play in determining the final destination, protein stability, and signaling capacity of EGFR. However, the involvement of multiple intracellular compartments and the dynamic nature of the signaling complexes make it challenging to quantitatively map the spatiotemporal network.

Here, we describe a cross-linking-aided immunoprecipitation/mass spectrometry (MS) workflow for isolation and identification of the signal-dependent EGFR proteome. Employing formaldehyde as a cross-linking reagent, we performed immunoprecipitation (IP) of endogenous EGFR-associated complexes upon EGF treatment and analyzed the associated proteins by quantitative mass spectrometry. We present a comprehensive time-resolved and signal-dependent EGFR proteome composed of both signaling proteins that are stably associated with EGFR and proteins that are more transiently associated with EGFR during its traffic.

MATERIALS AND METHODS

Antibodies and Reagents

The antibodies used in this study were β-actin (#A2066, Sigma-Aldrich), EGFR (Ab13, Thermo Fisher Scientific), p-EGFR-(Y1068) (#4267, Cell Signaling Technology), p-ERK1/2-(T202/Y204) (#9101, Cell Signaling Technology), and ERK (#4695, Cell Signaling Technology). EGF and formaldehyde were purchased from Sigma-Aldrich. Small interfering RNAs (siRNAs) used in this study were ordered from Sigma-Aldrich, and the product information is listed in Table S7. The specificity of anti-EGFR antibody was tested and is shown in Figure S9.

Cell Culture and Treatment

HeLa cells (1 × 10⁷/IP) were grown in DMEM supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin (Sigma-Aldrich) and incubated at 37 °C with 5% CO₂. The cells that were 70% confluent were serum-starved for 16 h and then stimulated with 50 ng/mL EGF for indicated times. siRNA was transfected into HeLa cells with Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer’s instruction.

Cross-Linking and Cell Cytosolic Fraction Preparation

Cells were cross-linked by adding formaldehyde to the media to a final concentration of 1% and incubated at 37 °C for 8 min; the cross-linking was quenched by adding glycine to a final concentration of 0.2 M. Cells were collected and washed twice with ice-cold PBS. The soluble fraction was extracted by resuspending the pellet containing ~1 × 10⁷ cells in 100 μL of LB1 buffer [50 mM HEPES-KOH (pH 7.5), 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, and 0.25% Triton X-100 with 10 mM β-mercaptoethanol, phosphatase inhibitor, and protease inhibitor cocktails] for 30 min at 4 °C followed by sonication (Sonics Ultrasonic Processors VCX; 15 s, 5 s on, 10 s off, 25%). After centrifuging for 20 min at 66,000g, the supernatant was used for the IP experiments.

Immunoprecipitation

For each IP experiment, 1 mg of protein lysate was incubated with 5 μg of antibody for 2 h at 4 °C and cleared by ultracentrifugation (100,000g, 15 min). The supernatant was then incubated with 30 μL of 50% protein A-Sepharose slurry (#17-0780-01, GE Healthcare, Piscataway, NJ) for 1 h at 4 °C. The bead-bound complexes were washed four times with NETN buffer [20 mM Tris (pH 7.5), 1 mM EDTA, 0.5% NP-40, and 150 mM NaCl] and eluted with 20 μL of 2× Laemmli buffer and heated at 95 °C for 10 min.

Mass Spectrometry

IP samples were resolved on a NuPAGE 10% Bis-Tris gel (Life Technologies, WG1201BX10) in 1× MOPS running buffer; the gel was cut into three molecular weight regions plus the IgG heavy and light chain bands. Each band was in-gel-digested overnight with 100 ng of trypsin, which cleaves peptide chains at the C-terminus of lysine or arginine (MS grade, GenDepot) in 20 μL of 50 mM NH₄HCO₃ at 37 °C. Peptides were extracted with 350 μL of 100% acetonitrile and 20 μL of 2% formic acid and dried in a Savant SpeedVac. Digested peptides were dissolved in 10 μL of loading solution (5% methanol containing 0.1% formic acid) and subjected to LC–MS/MS assay as described previously.²⁹ Briefly, digested peptides were analyzed by a nano-HPLC 1000 system (Thermo Fisher Scientific) coupled to an LTQ Orbitrap Elite (Elite) or an Ultimate 3000 UHPLC coupled to an Orbitrap Fusion Tribrid (Fusion, Thermo Fisher Scientific) mass spectrometer. Samples were enriched on an in-housed 2 cm × 100 μm i.d. trap column with 3 μm Reprosil-Pur Basic C18 beads (Dr. Maisch HPLC GmbH, Germany); the trap column was then switched in-line with an in-housed 5 cm × 150 μm capillary column packed with 1.9 μm Reprosil-Pur Basic C18 beads. For the Elite instrument, 75 min discontinuous gradient of 2–24% acetonitrile and 0.1% formic acid at a flow rate of 800 nL/min was applied to the column and then electrosprayed into the mass spectrometer. The instrument was operated under the control of Xcalibur software ver. 2.2 (Thermo Fisher Scientific) in data-dependent mode, acquiring fragmentation spectra of the top 25 strongest ions. The parent MS spectrum was acquired in the Orbitrap with a full MS range of 375–1300 m/z in the resolution of 240,000 and an AGC target of 3 × 10⁶. The CID fragmented MS/MS spectrum was acquired in an ion trap with rapid scan mode with an AGC target of 3 × 10⁵. Dynamic exclusion was applied as 1 repeat count, 10 s of repeat duration, 40 s of exclusion duration, 1.5 m/z of high and low exclusion mass, 1 expiration count, and expiration S/N threshold as 2.0. For the Fusion instrument, the flow rate for HPLC was 850 nL/min with 45 min of discontinuous gradient of 4–26% acetonitrile and 0.1% formic acid. The MS spectrum was acquired under the control of Xcalibur version 4.0 in data-dependent mode of the top 35 strongest ions. The parent ion acquisition range was 400–1300 m/z in the resolution of 120,000 with 50 ms maximum injection time at an AGC target of 5 × 10⁵. CID fragmented MS2 scan was acquired in an ion trap with rapid scan mode with a maximum injection time of50 ms at an AGC target of 1 × 10⁴. Dynamic exclusion was applied as 1 repeat count, 20 s of repeat duration, and 25 ppm of high and low exclusion mass.

Protein Identification and Label-Free Quantification

The MS/MS spectra were searched against target-decoy Human refseq database (release 2015_06, containing 73,637 entries) in Proteome Discoverer 1.4 interface (Thermo Fisher Scientific) with Mascot algorithm (Mascot 2.4, Matrix Science). Dynamic modifications of acetylation of N-term and oxidation of methionine were allowed. The precursor mass tolerance was confined within 20 ppm with a fragment mass tolerance of 0.5 Da, and a maximum of two missed cleavages was allowed. Assigned peptides were filtered with 1% false discovery rate (FDR) using Percolator³⁰ validation based on the q value. Calculated area under the curve ofpeptides was used to calculate iBAQ for protein abundance based on the previous publication.³¹

Peptides were grouped into gene products as previously described.³¹ Briefly, peptides were mapped to gene products (collapsing all protein isoforms that originate from the same genetic locus). Gene products are quantified by iBAQ,³² with modification. For peptides that map to more than one gene product, the peptide peak area is distributed in a weighted procedure based on unique peptide ratios. For each peptide shared across multiple gene products, the peptide peak area assigned to gene product A is defined as

$$s_A=\frac{\Sigma u_A}{\Sigma u_A+\Sigma u_B+\dots }$$

where u_{i, A} is one of each unique-to-gene peptide for gene product A, and (A, B, …) are all gene products to which shared peptide s maps. For cases where no unique-to-gene peptides exist, shared peptide quantities are divided evenly across all gene products.

For normalization across experiments, the fraction of total (iFOT) was calculated, as defined for a given protein p by

$$p_{\text{iFOT}}=\frac{p_{\text{iBAQ}}}{{\Sigma }_i{p}_{\text{iBAQ}}}$$

Parallel Reaction Monitoring

Parallel reaction monitoring (PRM) measurements of 106 selected proteins were carried out using an Orbitrap Fusion Tribrid mass spectrometer (Thermo Fisher Scientific). The selected proteins are separated into two groups based on relative amount (iFOT), 47 abundant and 59 low abundant proteins. Up to three strongest unique peptides for each protein were selected for quantification based on full MS intensity. Peptides for PRM of each protein are listed on Tables S3 and S4. In-gel-digested samples from cross-linked IPs were analyzed at each machine run with 4–24% of acetonitrile gradient for 45 min. All other nano-HPLC conditions were the same as those used for protein identification. Precursor ions (full mass) were scanned with a resolution of 120,000, an AGC target of 2.0 × 10⁵, and 50 ms maximum injection time. Target precursor ions were isolated by Quadrupole with an isolation width of 2 m/z filtering the predicted elution time with a 6 min time window. Product ions (MS2) were generated by CID at 35% collision energy and scanned at 350–1400 m/z with an AGC target of 5.0 × 10⁴ and maximum 35 ms injection time in Orbitrap with 30,000 resolutions. The raw spectral files were converted to mgf format by PD1.4 and then imported to Skyline together with the raw data files. The sum of the area under the curve (AUC) of the top three fragment ions was used for quantification after being normalized to the AUC sum of the top three EGFR peptides of corresponding samples. No labeled internal standard was used. The results as calculated by Skyline were manually inspected within the program.

Experimental Design

For identifying the EGF treatment-dependent temporal EGFR proteome, four independent replications of time series were performed with HeLa cells treated with 50 ng/mL EGF for 0, 2, 10, 30, and 120 min. EGFR-associated proteins were identified as those that showed an iFOT ratio (treated/nontreated) of more than 2-fold increase and appeared in more than three experiments. The PRM experiment was done twice using two affinity-purified samples from two independent preparations. The intensity of the tested peptides in the PRM for the same protein was averaged between values of the two replicate experiments. GO term enrichment analysis was performed with the Gene Ontology Consortium (http://geneontology.org/). Interactive network analysis was performed using the Search Tool for the Retrieval of Interacting Genes (STRING) database (https://string-db.org).

Statistical Analysis

We used the R (R version 3.3.2) statistical programming language for statistical analyses and visualization. Spearman’s rank correlation coefficient was used to measure the degree of correlation between independent biological replicates at each time point. The corrplot³³ function was used for visualizing the correlation matrix. K-means clustering analysis³⁴ was performed to partition EGFR proteome components into K-groups based on their temporal response kinetics. The K-means function from basic R stats package was executed with the initial centers equal to 6 (k = 6) and other parameters by default. The optimal number of clusters was determined using the elbow method.³⁵

The corresponding original expression profile was required to transform into Z-score as an input before plotting; the expression pattern generated by the average of FOT in each cluster depicted its typical characteristic. The dot blots of the spatiotemporal profiles were generated by importing R ggplot2 package,³⁶ and the input data were normalized using the highest value of FOT for each protein before calling the ggplot function. All heatmaps were achieved by the R statistical software using pheatmap³⁷ package based on Z-score transformation data.

Western Blot Analysis

Whole-cell lysates were prepared with RIPA buffer with protease inhibitors and phosphatase inhibitors. Thirty micrograms of lysates was resolved on 8% SDS-PAGE and transferred onto nitrocellulose membranes. After blocking with 5% milk in TBST for 1 h, the membranes were incubated with appropriate primary-antibodies in 5% milk in TBST overnight at 4 °C. The membranes were incubated for 2 h with horseradish peroxidase-conjugated secondary antibodies. Signals of target protein bands were detected using chemiluminescent detection reagents on X-ray films or a CCD camera (Tanon 5200 Luminescent Imaging Workstation).

Immunofluorescence

Cells grown on poly-d-lysine (Sigma-Aldrich, St. Louis, MO)-coated coverslips were washed twice with PBS and fixed with 4% paraformaldehyde for 15 min at room temperature. Coverslips were rinsed three times with PBS and then permeabilized with 0.1% Triton X-100-containing PBS for 10 min on ice. Cells were blocked with 5% bovine serum albumin for 30 min and then incubated with the primary antibody overnight at 4 °C. After briefly washing with PBS, the coverslips were incubated with FITC- or Texas Red-conjugated secondary antibodies for 30 min at room temperature. The images were visualized and captured by a Leica TCS SP5 confocal microscope.

RESULTS

Establishing a Streamlined Procedure for Isolating Membrane-Bound Protein Complexes

Formaldehyde-mediated cross-linking is a well-established method for preserving protein–protein and protein–DNA interactions as covalent bonds formed between interacting partners are resistant to detergent extraction and subsequent stringent wash for removing nonspecifically bound proteins. The method has been widely used in chromatin immunoprecipitation (ChIP), tissue fixation, and protein–protein interaction studies.^20,38,39 While there are many stronger cross-linking reagents, we chose formaldehyde for this work as it contains the shortest carbon chain and would be reactive only to proteins that are in close proximity to the bait protein.

We first evaluated the effect of formaldehyde concentration on capturing EGFR-associated complexes. Serum-starved HeLa cells were stimulated with 50 ng/mL EGF for 10 min or left untreated; they were then subjected to 0, 0.1, 0.3, 0.5, and 1% formaldehyde treatment followed by IP with an anti-EGFR antibody (AB13, a cocktail of monoclonal antibodies that specifically recognize epitopes of two separate extracellular domains).⁴⁰ The IP samples were separated on SDS-PAGE followed by in-gel digestion, and the resulting tryptic peptides were analyzed with MS (Figure 1A). On average, 700–1400 proteins (with ≥2 high quality and unique peptides) were identified from each sample, which included those that were known to associate with EGFR and those that were nonspecifically associated but repeatedly appeared in each IP experiment. To focus on EGF-induced association, we restricted the data analysis on proteins whose abundance was increased upon EGF stimulation. The effect of the formaldehyde was apparent as its addition led to a >2-fold increase in abundance of 400–700 proteins compared with no cross-linking (Table S1).

Figure 1.

Signal-dependent membrane proteome captured by a formaldehyde-aided cross-linking method. (A) Schematic representation of the workflow to capture and analyze the EGF-dependent EGFR proteome. (B) HeLa cells starved overnight were treated with EGF for 5 min and subjected to cross-linking with various concentrations of formaldehyde. Proteins that were associated with EGFR in an EGF-inducible manner (>3-fold increase) were selected; the normalized iFOT ratios (Z-score) of fold of increase (EGF+/−) of a partial list of the proteins were plotted in the heatmap.

In the absence ofcross-linking, we detected 30 proteins whose abundance was increased by at least 2-fold upon EGF stimulation (Figure 1B and Table S1). While the majority of these proteins were known EGFR-binding proteins, close examination revealed that many of them contain structural domains, such as SH2, SH3, and PH domains, which mediate protein–protein interactions.⁴¹ In samples treated with 0.1% formaldehyde, we found 63 proteins that satisfied the criteria, in which 33 of them were not detected without cross-linking. Among them are ERBB2, a heterodimeric partner of EGFR, and STAM, SNX1, and SNX2, which are involved in EGFR signaling and receptor endocytic trafficking.^42–44 Increasing the formaldehyde concentration to 0.3, 0.5, and 1% led to the identification of 83, 92, and 89 proteins, respectively. The comparable number of proteins detected at 0.5 and 1% formaldehyde suggested that further increase in formaldehyde concentration over 1% was unlikely to gain significant benefit; on the contrary, over-cross-linking could lead to the loss of native peptides and thereby compromise the MS detection. Taken together, our data suggest that, although using 0.5% formaldehyde could identify a few more proteins, using 1% formaldehyde produced more reproducible results to capture the weak and transient interacting proteins in an extended EGFR proteome.

Charting a Temporal EGFR Proteome Stimulated by EGF

As a prototypical RTK, EGFR governs extensive signaling networks.^6,26 While a landmark study by Zheng and colleagues has elegantly determined the temporal regulation of EGF signaling networks defined by the scaffold protein SHC1,⁴¹ the study used a stably expressed, doubly tagged SHC1 and was focused mainly on the immediate impact of EGF stimulation and strong interactions. Proteins that participate in the later events of EGFR following EGF stimulation, such as receptor trafficking, degradation, and recycling, have not yet been systematically examined. To obtain an extended EGFR proteome induced by EGF, we performed cross-linking with formaldehyde at various times (0, 2, 10, 30, and 120 min) upon EGF stimulation and analyzed the endogenous EGFR-associated proteome. These time points were selected to obtain a global view of the dynamic EGFR proteome when the receptor sequentially underwent internalization, endocytosis, endosome sorting, recycling, and degradation. Multiple biological replicates were performed to identify proteins that were significantly increased in EGFR IP samples. In total, we analyzed 20 IP experiments, including 5 time points and 4 independent biological repeats of the complete time course (Table S2) that showed high degree of correlations (Figure S1).

We defined an EGF-induced, EGFR-associated protein as a gene product (GP) that was detected with two or more unique and strict peptides (ion score > 20) and showed a greater than 2-fold increase upon EGF treatment. To be a significant identification, a protein had to be detected in at least three out of four replicates at one time point or in two out of the four replicates in two consecutive time points (detected total of four times).

To focus on relatively abundant proteins, we further required that the iFOT value was greater than 1⁴⁵ in at least one time point. These criteria eliminated most of the commonly abundant, nonspecific binding proteins. We further removed proteins that failed to pass the E_cutoff filters described by Malovannaya et al.⁹ in our database and those appeared frequently in the CRAPome⁴⁶ as nonspecific proteins. FYCO1, which did not satisfy the above criteria but was confirmed by PRM, was also included as being a member of the FYVE family proteins (see below for more details). In total, we found 55, 100, 88, and 65 proteins at 2 min, 10 min, 30 min, and 2 h, respectively, resulting in a total of 140 proteins excluding EGFR (Figure 2A and Table S2). The heatmap patterns of the four replicates shown in Figure 2A demonstrated the high degree of reproducibility of the assignment. We therefore designated the collection of these 140 proteins as the EGF-induced EGFR proteome. Of the 140 proteins, only 10 (CBL, UBASH3B, PIK3R2, ERBB2, GRB2, SHC1, VAV2, UBB, ANKFY1, and VPS16) were found in all time points in at least three of the four replicates, suggesting that they may constitute a core EGFR proteome. The PANTHER GO slim biological process analysis software showed that the EGFR proteome was enriched mainly in protein endocytosis processes, including vesicle-mediated transport, receptor-mediated endocytosis, intracellular protein transport, and transmembrane receptor protein tyrosine kinase signaling pathway (Figure 2B), consistent with the known processes that EGFR is involved in. Using the STRING database, we found apparent interaction modules based on previously published experimental evidence, including the core signaling module, adaptor protein complexes, retromer, and tethering complexes (Figure S2), and that distinct modules were enriched at each time point (Figure 2C). At the 2 min time point, a protein module containing VAV2, GAB2, PIK3CB, PRKCD, CD2AP, INPPL1, ARAF, RAF1, and PIK3C2B was detected (Figure 2C, shown in green) with the EGFR core complex (Figure 2C, shown in red). This module is known to be components of the FcγR-mediated phagocytosis pathway (KEGG pathway: cfa04666). Appearance of this module at the 2 min time point suggests that the initiation of intracellular translocation of the EGFR complex might be regulated by a similar mechanism as that of phagocytosis. The FcγR-mediated phagocytosis protein module is absent at 10 min, while two distinct new modules emerged in the EGFR core complex (Figure 2C). One is known to be related with SNARE interactions in vesicle transport (KEGG pathway: hsa04130), containing AP1B1, VAMP8, AP1G1, AP1M1, AP1S1, STX7, VAMP7, STX8, and VTI1B; the other plays a role in receptor transport from endosome to lysosome in the GO term (GO: 0008333), containing VPS45, ZFYVE20, VPS16, VPS18, VPS39, VPS33A, TGFBRAP1, and VPS8. These two modules that appeared at 10 min are also copresent at 30 and 120 min despite losing some components at 120 min. Together, these results demonstrate that the cross-linking IP combined with the high-quality MS measurement provides a useful tool to identify new components in the EGFR signaling pathway.

Figure 2.

EGF-dependent EGFR proteome. (A) Heatmaps of four replicates of EGFR proteome obtained in HeLa cells treated with EGF for the indicated times. The tables on the right summarize the frequencies of each protein scored as significant observations in the four replicates at each time point. (B)GOterm analysis of significantly enriched proteins in the EGFR proteome. The −log q values of the significance were plotted with the values of each GO term marked. (C) Previously known protein–protein interaction modules at each time point obtained from the STRING database. Each module is indicated by a different color.

The abundance of the EGFR proteome components varied in a wide range, with their iFOT values varying as high as 1500fold. From the chemical reaction point of view, proteins should be in the right amount/concentration to be functional. We ranked the top 20 most abundant proteins at each time point and plotted their fold of induction (y axis) and abundance (circle size) (Figure 3). The top three most abundant proteins identified at each time points were UBB, GRB2, followed by either SHC1 (at 2 min, 10 min, and 2 h) or SNX2 (30 min) (Figure 3). GRB2 and SHC1 are well-characterized EGFR-binding partners that are phosphorylated by activated EGFR and play key roles in the signal transduction.^41,47 Ubiquitin had similar abundance as that of GRB2 in the EGFR proteome; this is consistent with the notion that ubiquitination plays an important regulatory role in EGFR signaling, sorting, degradation, and recycling.^21,48–53 Intriguingly, SNX2 was the third most abundant protein at 30 min time point; furthermore, additional components of the SNX-BAR complex, including SNX3, 6, 27, and other retromer components (VPS26A and 35) that facilitate cargo retrieval, were also among the top 20 highly abundant components of the EGFR proteome (Figure 3). Together, these results highlighted the critical roles of ubiquitination and trafficking in EGFR-mediated signaling transduction.

Figure 3.

Relative abundance of the top 20 EGFR-associated proteins at each time point (mean of four replicates). The areas of the circles indicate the abundance in FOTs of the top 20 most abundant EGFR-associated proteins obtained in EGF(+) IPs. The y axis indicate the iFOT ratios of proteins identified in EGF(+) versus EGF(−) in the log2 scale, which are arranged from low to high along the x axis.

Quantification by PRM Reveals Distinct Dynamics of EGFR Association

We next performed PRM measurements on selected components to obtain more accurate quantification of the EGFR proteome. For each selected protein, we picked one to three peptides based on the peptide recovery results obtained from the data-dependent acquisition (DDA) runs. We loaded 10 or 30% of the independently purified samples to ensure linear response and repeated the measurement two times to ensure reproducibility. Based on the protein abundance from the discovery runs, target proteins were separated into two groups, namely, strong and weak interacting proteins (Tables S3 and S4); they were run separately to get enough dwelling time for each PRM target peptide. One hundred out of 106 tested interacting proteins (94%) identified in the DDA mode were further confirmed by PRM (Figure 4A, Figure S3, and Tables S3, S4, and S9). The median correlation coefficient among five paired measurements was 0.91, indicating a high overall concordance between PRM and DDA measurements and providing another layer of confidence for our streamlined label-free quantification procedure.

Figure 4.

Distinct kinetics of response revealed by the PRM measurements. (A) Comparison of response kinetics between PRM and DDA. (B) Line graphs of PRM results of representative EGFR proteome components that display different association kinetics. Each time point is normalized to the maximal FOT value of the entire time course.

The PRM measurements also revealed that EGFR proteome components responded to EGF stimulation with distinct kinetics. For instance, increased binding of GRB2 and SHC1 to EGFR was detected as early as 30 s, peaked at 2 min, and maintained with high intensity until 2 h (>60% of the maximal binding); in contrast, SOS1 and MAP4K3, which were also maximally detected at 2 min, were quickly dissociated as their amounts were decreased to <5% of the maxima by 30 min (Figure 4B). Interestingly, class I PI3K subunits PIK3R2 and PIK3CB exhibited almost identical binding kinetics and peaked at 2 min, but the class II PI3K subunit PIK3C2B clearly peaked at 10 min (Figure 4B). It has been well known that PIK3C2B differs from the class I kinase with the presence of a C-terminal C2 domain but lacks a regulatory subunit and shows different substrate preference.^54,55 The distinct response kinetics suggests that these kinases might play different roles in EGFR signal transduction.

Distinct Response Kinetics to EGF Stimulation Classifies the EGFR Proteome to Six Clusters

We next carried out K-means clustering to classify EGFR proteome components based on their temporal response kinetics, suggesting the distinct association/dissociation kinetics of its members with a line chart by invoking the line function. As shown in Figure 5 and Table S5, the EGFR proteome could be classified into six clusters, which were correlated roughly with the timing of their maximal binding. The proteins in clusters 1 and 2 responded most rapidly to EGF stimulation with a peak time within 2 min and included all but one component (VPS16) of the core EGFR proteome. Many proteins in these two clusters participate in the transduction of the EGFR-activated signals as the substrates of the EGFR kinase. Clusters 1 and 2 differ in their dissociation kinetics: cluster 1 members maintained their associations with EGFR throughout the 2 h time frame and included well-characterized EGFR-associated proteins, such as ERRB2, GRB2, and SHC1; in contrast, the cluster 2 proteins associated more transiently with EGFR, and their intensities rapidly declined. Cluster 2 included scaffold proteins (GAB1 and GAB2), protein kinases (MAP4K3 and MAP4K5), lipid kinases (PIK3R2), phosphatases (INPPL1), and GTPase cofactors (VAV2), which play important roles in receptor internalization and rapid activation of the signaling cascade. Most of the cluster 3 and 4 proteins were maximally bound to EGFR at ~10 or 30 min but exhibited different dissociation kinetics. Clusters 3 and 4 included membrane-binding proteins, such as adaptor complexes 1 and 2 and proteins that function in cytoskeleton reorganization. Cluster 5 proteins appeared maximally at 30 min, followed by immediate dissociation. Cluster 6 proteins exhibited a common trend of continued increase until at least 2 h. This group contains many proteins that attenuate EGFR signal activity. For instance, ERRFI (ERBB receptor feedback inhibitor 1) is a cytosolic protein that promotes EGFR degradation.^56,57 LAMTOR1 binds to the lysosomal membrane and has been reported to mediate mTOR signaling at the lysosome.⁵⁸ The appearance of the vacuolar ATPase (V-ATPase), a multisubunit enzyme that mediates acidification of eukaryotic intracellular organelles (ATP6V0D1 and ATP6V1B2) and autophagy (TMEM59), indicated the arrival of EGFR to the lysosome at 2 h time point. This timing is also consistent with the apparent reduction of EGFR intensity.

Figure 5.

The EGFR proteome is classified into six clusters based on the distinct association/dissociation kinetics of its members. Clusters 1 and 2 show the highest association at 2 min but display different dissociation kinetics, clusters 3 and 4 show the highest association at 10 min but display different dissociation kinetics, cluster 4 shows the highest association at 30 min, and cluster 6 shows increased association in 2 h.

The Dynamic EGFR Proteome Reveals a High-Resolution Spatiotemporal Track of EGFR Trafficking

While previous proteomics studies have established the critical role of intracellular trafficking in EGFR signal transduction, most of them were focused on limited time scales or on addressing specific biological questions, which confined their application in the global proteomic context. Our quantitative characterization on a more detailed time course provided a high-resolution spatiotemporal view of the EGFR proteome (Figure 6). For instance, the AP2 complex is localized to the plasma membrane and facilitates clathrin-dependent endocytosis of the receptor,^59–61 whereas the AP1 complex, which is also associated with clathrin-coated vesicles, is located mainly on the endosomes as well as the trans-Golgi network (TGN).⁶² While more than 10 subunits of AP2 and AP1 were reproducibly detected, 4 of the 5 AP2 components were significantly enriched as early as 2 min, and their intensities were mostly declined by 30 min; in contrast, all AP1 subunits were not significantly detected until 10 and 30 min, and their association with EGFR was still maintained at high levels at 30 min (Figure 6). These observations are consistent with the notion that AP2 is mainly involved in endocytosis initiation, whereas AP1 functions during endocytosis, as well as on the endosome membranes and in the TGN.

Figure 6.

Response kinetics of proteins involved in EGFR traffic. The size of the circle indicates the mean FOT value of four biological replicates of EGFR IPs.

EGFR is sorted at the early endosome membrane to enter either the degradation pathway or the recycling pathway. Key components of the retromer complex (four core components, VPS26A/VPS26B, VPS29, and VPS35, and two accessary factors, SNX3 and SNX27), which mediates cargo retrieval at the endosomal membranes,⁶³ were found in the EGFR proteome at 10 min and peaked at 30 min (Figure 6). This timing also coincided with the detection of three WASH complex components: WASH1, KIAA0196, and KIAA1033. The SNX-BAR proteins participate in the retrograde export pathway and return cargos to the trans-Golgi network (TGN).^63,64 Our data support the association of EGFR with the TGN with the identification of multiple TGN-associated complexes, including SNX1, SNX2, SNX5, and SNX6, as well as the WASH complex.

The ESCRT complexes catalyze the membrane fission event that leads to the formation of intraluminal vesicle in the multivesicle body that is destined for lysosomal degradation.^48,65–67 Of the four ESCRT complexes (0, I, II, and III), onlyESCRT-0 complex subunits (HGS and STAM and, in some instances, STAM2)⁶⁸ were found in the EGFR proteome (Figure 6). Noticeably, these subunits are responsible for the recognition of the ubiquitinated cargos. The appearance of the ESCRT-0 complex only in the EGFR proteome seemed to indicate that only the ESCRT-0 complex may be directly involved in EGFR binding, while ESCRT-I/II/III may play indirect roles.

Tethering factors are proteins or protein complexes that establish long-range interaction between the transport carriers (TCs), and the acceptor compartment before contacts between v-SNARE and t-SNARE occurs and promotes SNARE complex assembly for membrane fusion.⁶⁹ We found single-protein tethering factors, including long coiled-coil proteins, such as EEA1, ZFYVE20, and RABEP1, which bind the endosomal membrane, as well as multisubunit tethering complexes such as HOPS, CORVET, and GARP (Golgi-associated retrograde protein).^70,71 CORVET and HOPS are tethering complexes, connecting the proteins that are essential for early-to-late endosome transition to lysosome biogenesis. VPS16, 18, 11, 33A, and 33B are core subunits shared by both CORVET and HOPS. Interestingly, the CORVET-specific subunit VPS8 appeared at 5–10 min, whereas the HOPS-specific subunit VPS39 appeared at 30 min and 2 h, timings that were consistent with their association with early and late endosomes/lysosomes, respectively. Two subunits of the GARP, VPS51 and VPS52, were found and peaked at 30 min. Taken together, these findings demonstrate the remarkable spatial and temporal resolution of our cross-linking method, which enabled us to track the progress of the receptor trafficking to various vesicle compartments (Figure 7).

Figure 7.

Schematic representation of EGFR spatiotemporal translocation upon EGF stimulation to various intracellular vesicles. The time points are not drawn to scale due to space limitation.

ZFYVE Proteins Facilitate EGFR Intracellular Trafficking

We noticed that multiple FYVE domain-containing proteins were induced by EGF treatment and enriched by cross-linking. The FYVE domain is responsible for specifically binding to phosphatidylinositol 3-phosphate (PI3P) in lipid bilayers and involved in regulating membrane trafficking.^72–75 At least nine proteins (ANKFY1, ZFYVE20, ZFYVE9, EEA1, HGS, ZFYVE16, RUFY1, RUFY2, and FYCO1) were reproducibly detected with the presence of as low as 0.1% FA (Figure 1B). While some of the FYVE domain-containing proteins have been reported previously,^76–78 their roles as a protein class in EGFR trafficking have not been systematically examined. We found that the association of the nine FYVE domain-containing proteins with EGFR varied in timing and abundance (Figures 4A and 6). ANKFY1 was detected as early as 2 min; HGS, EEA1, ZFYVE16, ZFYVE9, and ZFYVE20 appeared and peaked at 5–30 min and then dissociated from EGFR by 2 h; RUFY1, RUFY2, and FYCO1 started to appear at 10–30 min and maintained their association at high levels until 2 h. The distinct kinetics of different FYVE domain-containing proteins could indicate the distinct functional roles that they play during EGFR endocytosis endosomal sorting and signaling at different cellular compartments, that is, from the plasma membrane to various endosomal vesicles.

As the first FYVE protein that appeared in the EGFR proteome upon EGF treatment, ANKFY1 was also detected at all time points and was one of the top 20 most abundant components. ANKFY1 contains a coiled-coil structure, a BTB/ POZ domain at its N-terminus, ankyrin repeats in the middle, and a FYVE-finger motif at its C-terminus. ANKFY1 has been reported to localize to early endosome and interact with VPS26/ 35 in sorting,⁷⁹ but its role in EGFR signaling is not clear. We knocked down ANKFY1 with two siRNA sequences and first examined its impact on the EGFR proteome. We used PRM-MS to verify the ANKFY1 knockdown efficiency. We selected two tryptic peptides from ANKFY1 and picked the top three most abundant fragment ions for PRM plotting. As shown in Figure 8A, ANKFY1 levels were reduced to less than 20% of those in control knockdown cells. However, IP-MS experiments showed no significant alternation in either the composition or the abundance of the EGFR proteome at 2 and 30 min following EGF stimulation (Table S6). We then examined its effect on EGFR-mediated signaling by Western blotting. We found that, although loss of ANKFY1 did not significantly affect the rate of EGFR degradation, it led to the attenuation of the EGFR Y1068 phosphorylation at the 30 min time point (Figure 8B and Figure S4). Moreover, phosphorylation of ERK1/2, a downstream effector in the EGFR signaling, showed a moderate but consistent increase of ANKFY1 knockdown cells (Figure 8B and Figure S4). Hence, these data suggested that ANKFY1 may play a role in EGFR signaling but has no major impact on EGFR degradation or its associated proteome.

Figure 8.

The FYVE domain proteins play functional roles in EGFR traffic. (A) Parallel reaction monitoring for two selected ANKFY1 peptides shows the siANKFY1 knockdown efficiency. NC, nonspecific control; KO1, siRNA sequence #1; KO2, siRNA sequence #2. (B) Knocking down ANKFY1 by siRNA affects EGFR autophosphorylation and phosphorylation of its downstream effector ERK. (C) Effect of knocking down FYVE domain proteins on EGFR intracellular distribution.

Likewise, we found that, when the other four FYVE proteins were effectively reduced by siRNA knockdown (Figure S5), no obvious change in EGFR degradation was observed (Figure S6). We then examined the EGFR intracellular distribution by immunofluorescence. In starved HeLa cells, EGFR exhibited a diffused pattern. Upon EGF treatment, EGFR was first accumulated on the plasma membrane and then formed bright foci that were colocalized with EEA1 (an endosome marker, Figure S7) at 10–30 min and VPS11 (a late endosome and lysosome marker, Figure S8); its intensity gradually decreased by 2 h. A common phenotype observed in cells depleted of FYVE genes was the persistent presence of bright EGFR foci in the cytoplasm even at 2 h after EGF treatment when these foci were largely diffused in control cells (Figure 8C), suggesting that reduced level of FYVE proteins indeed affected the EGFR trafficking. Together, our data implied that FYVE domain-containing proteins might be involved in EGFR signaling by affecting its intracellular trafficking.

DISCUSSION

As a prototype of RTK and a therapeutic target in a variety of cancers, EGFR signaling pathways and the EGFR-associated interactome have been extensively studied. These studies include both function-based characterizations focusing on one or a few interacting partners as well as large-scale studies that examine the EGFR-associated proteomic landscape.^26,41,80 The cross-linking approach that “freezes” the interaction by forming covalent bonds also allows the stringent wash conditions that effectively removes contaminating proteins. As a result, nearly 1000 proteins have been reported as putative EGFR-interacting proteins in the literature. Our current study employed a well-established, formaldehyde-mediated cross-linking strategy to stabilize transient and weak interactions, which examined the dynamic EGFR proteome by label-free quantitative MS during the 2 h time frame following EGF stimulation. This work provides a high-resolution spatiotemporal view of the EGFR proteome, which includes proteins that directly and indirectly associated with the receptor.

Previously, the membrane yeast two-hybrid (MYTH) and the mammalian membrane two-hybrid (MaMTH) have been utilized to identify protein–protein interactions with high throughput.^81–83 Recently, several proximity labeling methods based on ascorbate peroxidase (APEX)^84,85 or biotin, including BioID^86,87 and TurboID,⁸⁸ have also been successfully applied to identify protein–protein interactions with high sensitivity and specificity. However, all these strategies require exogenous expression of the labeled bait proteins as well as labeled prey proteins, which inevitably increase the complexity and workload and likely result in artificial protein–protein associations. In the present work, we targeted endogenous EGFR using specific anti-EGFR antibody without engineering exogenous genes and achieved proximity-dependent interaction by cross-linking with formaldehyde. Therefore, the chemical cross-linking-based IP of endogenous proteins herein is an efficient and economic method to evaluate dynamic interactomes of RTK in the natural context.

We compare the EGFR binding proteins in this study to two previous publications (Table S8). Foerster et al.⁶ reported 183 EGFR-binding proteins from a human epidermoid carcinoma cell line A431 treated with EGF for 30 min; in another study by Tong et al.,⁸⁹ the authors investigated the effect of temperature on EGFR activation and reported 323 EGFR-binding proteins at 37 °C following 15 min of EGF treatment in the HEK-EGFR T-Rex cells. Of the 140 proteins in our study, only 31 were found in all three studies, and only 40 were common in the two previous reports. The 31 common proteins appeared to be the core binding proteins shown in Figure 2C. Most of the time-dependent binding modules discovered for this study, including the FcγR-mediated phagocytosis pathway (KEGG pathway: cfa04666), the snare interactions in vesicle transport (KEGG pathway: hsa04130), and the receptor transportation from endosome to lysosome (GO: 0008333), were not reported by the two previous studies. This low level of overlap may be explained by, among others, the difference in treatment duration, the targeting for either endogenous protein or ectopic expressed Flag-tagged EGFR, and the different antibodies used in each study. It could also be explained partially by the different cell lines used or the status of EGFR. For example, an EGFR-mutant interactome of 263 proteins with a 14-protein core network critical for the viability of lung cancer has been reported. Of the proteins that are specifically associated with mutant cell survival, MK12, CD11A, CD11B, ARHG5, and GLU2B are not identified reproducibly in HeLa.¹⁹

Different concentrations of EGF could lead to different outcomes.⁹⁰ For instance, EGFR could take clathrin-mediated or non-clathrin-mediated endocytic pathways at low or high EGF concentrations, respectively.⁹¹ In the human body, EGF levels vary from low (1–5 ng/mL) in plasma, serum, and saliva to medium (5–50 ng/mL) in tears, follicular fluid, sperm, and seminal plasma to high (50–500 ng/mL) in bile, urine, milk, and prostate fluid.^92–94 To activate EGFR, 10, 25, 50, and 100 ng/ mL EGF concentrations have all been used in EGFR-related studies. However, while using 100 ng/mL EGF could ensure maximum activation of EGFR with the fastest rates, it might result in loss of some critical interactions due to the fast progression of EGF/EGFR and rapid dissociations between EGFRand its binding partners.^26,41 Taken together, we chose 50 ng/mL EGF for our investigation to ensure the capture of a more complete EGFR proteome.

Our results showed that, among the 140 EGFR-associated proteins identified, 10 proteins (CBL, UBASH3B, PIK3R2, ERBB2, GRB2, SHC1, VAV2, UBB, ANKFY1, and VPS16) were reproducibly detected with high confidence at all time points, suggesting that they constitute the core EGFRproteome and play key roles in EGFR-mediated signal transduction and intracellular trafficking. Consistent with previous findings that ubiquitination is a critical regulatory mechanism for EGFR signal transduction,^48,52,53,95 we found that ubiquitin (UBB) is one of the most abundant components of the EGFR proteome, likely because of the receptor is mono- and polyubiquitinated at multiple positions. Moreover, an E3 ligase (CBL) and an Ub binding domain-containing protein (UBASH3B) are also part of the core EGFR proteome that are present at all time points examined. Therefore, our results further highlight the importance of ubiquitination in the EGFR signaling.

In addition to several well-characterized tyrosine and lipid kinases, we also found two germinal center-related kinases, MAP4K3 and MAP4K5, which have been previously shown to interact with GRB2.⁴⁷ MAP4K3, also known as happyhour in Drosophila melanogaster,⁹⁶ is a component of the TORC1 signaling complex that modulates cell growth and viability; it has also been shown as an inhibitor in the EGF signaling pathway that regulates ethanol sensitivity in Drosophila.^97,98 MAP4K5 (also known as GCKR) interacts with the SH3 domain of CRK and CRKL and is required for the SAPK activation.^99,100 The detection of both MAP4K3 and MAP4K5 as early as 2 min suggests that these two kinases play previously underappreciated roles in the upstream of the EGF signaling.

The high-quality quantification of the large number of EGFR-associated proteins also allowed us to classify them based on their temporal behavior and infer the possible functions of poorly characterized proteins. For instance, our data identified nine FYVE domain-containing proteins throughout the 2 h time course. Since the FYVE domain is associated with PIPs at various membrane structures, the timing of the FYVE protein appearance together with the copresence of the RAB proteins could implicate their functions. FYCO1, RUFY1, and RUFY2 all appear relatively late upon EGFR stimulation, and their levels rise gradually and are maintained longer than the other FYVE domain proteins. This distinct temporal behavior suggests that they are most likely involved in receptor degradation and recycling.

We showed that ANKFY1, one of the core EGFR proteome proteins, plays a role in EGFR signaling as siRNA knockdown of ANKFY1 resulted in moderate attenuation of pEGFR and moderate enhancement of pERK. However, neither the rate of EGFR degradation nor its associated proteome was appreciably affected. Likewise, although all the FYVE domain proteins examined affect EGFR subcellular distribution, the rate of EGFR degradation or signaling was also minimally affected. It is possible that, since receptor signaling can occur in various vesicle compartments, the receptor redistribution has a small overall effect on the signaling. The functional redundancy of the multiple FYVE proteins could also compensate for the loss of a single protein.

While a subset ofthe EGFR proteome modulates the receptor function, many proteins in this collection likely serve as “bystanders”, which are simply present as part of the endocytic vesicles that are captured when the receptor reaches these compartments via intracellular trafficking. This observation is consistent with a previous study showing the minor roles of trafficking in EGFR-mediated transcriptional response.¹⁰¹ It is important to note that the appearance of these bystanders at specific times during receptor endocytic trafficking could mark the location of the receptor and implicate the processes that it undergoes, providing the spatial information without the need of subcellular fractionation.

The temporal regulation of EGF signaling is complex and regulated in multiple mechanisms. Notably, several core components that are associated with EGFR throughout the time course are scaffold/adaptor proteins (such as SHC1 and GRB2) that contain protein–protein interaction domains facilitating the recruitment of subcomplexes in a signal-dependent manner. As the activated receptor is autophosphory-lated, phosphorylation and dephosphorylation play key roles in temporal regulation. Enzymes such as kinases (such as MAP4K3 and MAP4K5) and phosphatases (PTPN11) associate brieflyas they may be required only transiently for modification. Also, the feedback that terminates the signals together with activation could determine the duration of the association. Durations of the passenger protein association likely depend on the time for the receptor to translocate through the intracellular vesicle where they reside. However, these logical speculations need to be validated by additional experimental evidence. Furthermore, unlike PTPN11, two other tyrosine phosphatases, PTPN9 and PTPN23, remain associated with the receptor for an extended time. Follow-up investigation on their recruitment and their substrates is needed to reveal their specific roles in the signaling network.

Taken together, our study identified an EGF-induced spatiotemporal EGFR proteome that contains both functional regulators and bystanders. The high-resolution spatiotemporal information of these molecules provides a rich resource to generate testable hypotheses to delineate the many pathways that determine the strength and duration of the signal, as well as the location and fate of the receptor.

ABBREVIATIONS

AP-MS

affinity purification followed by mass spectrometry

AUC

area under the curve

DDA

data-dependent acquisition

EGF

epidermal growth factor

EGFR

epidermal growth factor receptor

FDR

false discovery rate

iBAQ

intensity-based absolute quantification

iFOT

iBAQ-based fraction of total

IP

immunoprecipitation

MS

mass spectrometry

PD1.4

Proteome Discoverer 1.4 interface

PPI

protein–protein interaction

PRM

parallel reaction monitoring

RTK

receptor tyrosine kinase

siRNA

small interfering RNA

STRING

Search Tool for the Retrieval of Interacting Genes

TGN

trans-Golgi network