Research ResourceSystems Biology

Time-resolved dissection of early phosphoproteome and ensuing proteome changes in response to TGF-β

See allHide authors and affiliations

Sci. Signal.  22 Jul 2014:
Vol. 7, Issue 335, pp. rs5
DOI: 10.1126/scisignal.2004856

Abstract

Transforming growth factor–β (TGF-β) signaling promotes cell motility by inducing epithelial-to-mesenchymal transitions (EMTs) in normal physiology and development, as well as in pathological conditions, such as cancer. We performed a time-resolved analysis of the proteomic and phosphoproteomic changes of cultured human keratinocytes undergoing EMT and cell cycle arrest in response to stimulation with TGF-β. We quantified significant changes in 2079 proteins and 2892 phosphorylation sites regulated by TGF-β. We identified several proteins known to be involved in TGF-β–induced cellular processes, such as the cytostatic response, extracellular matrix remodeling, and epithelial dedifferentiation. In addition, we identified proteins involved in other cellular functions, such as vesicle trafficking, that were not previously associated with TGF-β signaling. Although many TGF-β responses are mediated by phosphorylation of the transcriptional regulators of the SMAD family by the TGF-β receptor complex, we observed rapid kinetics of changes in protein phosphorylation, indicating that many responses were mediated through SMAD-independent TGF-β signaling. Combined analysis of changes in protein abundance and phosphorylation and knowledge of protein interactions and transcriptional regulation provided a comprehensive representation of the dynamic signaling events underlying TGF-β–induced changes in cell behavior. Our data suggest that in epithelial cells stimulated with TGF-β, early signaling is a mixture of both pro- and antiproliferative signals, whereas later signaling primarily inhibits proliferation.

INTRODUCTION

Transforming growth factor–βs (TGF-βs) belong to a superfamily of 33 cytokines in mammals, including TGF-βs, bone morphogenetic proteins, activins, inhibins, and nodals. TGF-β signaling influences various biological processes, including cell proliferation, differentiation, morphogenesis, tissue homeostasis, and regeneration. TGF-β family proteins are implicated in diseases such as cancer and autoimmune disorders (13). Moreover, cellular responses to TGF-βs not only are diverse but also depend on cellular context; for example, TGF-β can promote either differentiation or maintenance of pluripotency in embryonic stem cells (4).

TGF-β signaling plays an important role in the epithelial-to-mesenchymal transition (EMT), a process wherein epithelial cells lose their cell junctions and apical basal polarity, reorganize their cytoskeleton, secrete extracellular matrix (ECM) proteins, and thereby transdifferentiate into motile mesenchymal cells (5). EMT is required for normal embryonic development and for tissue remodeling and wound healing. Inappropriate reactivation of EMT contributes to the progression of various human pathologies, particularly those associated with tissue fibrosis and cancer cell invasion and metastasis (6, 7). TGF-β induces the formation of actin stress fibers and the production of ECM proteins, including plasminogen activator inhibitor 1 (PAI1) and fibronectin 1 (FN1) (810). Both normal epithelial cells and cancer cells respond to TGF-β by undergoing EMT, resulting in migration or invasion, whereas TGF-β inhibits the proliferation of epithelial cells, but not cancer cells (11, 12). In several cultured epithelial cell lines, including the human keratinocyte–derived HaCaT cells, stimulation with TGF-β inhibits proliferation and induces a phenotypic switch that resembles the EMT involved in wound healing, known as type II EMT (13).

TGF-βs bind to the type II transmembrane receptor (TβRII), which then recruits the type I transmembrane receptor (TβRI) and phosphorylates its juxtamembrane region (2). Activated TβRI binds to and phosphorylates the C terminus of SMAD2 or SMAD3 (14). The adaptor protein SARA (SMAD anchor for receptor activation) facilitates the interaction between TβRI and SMAD2 or SMAD3 (15). The heteromeric complex formed by activated SMAD2 or SMAD3 and the regulatory co-SMAD (SMAD4) localizes to the nucleus, where it binds to the promoters of target genes (16, 17). Transcriptional activation is regulated by DNA binding coactivators AP1 (activating protein 1) and ATF2 (activating transcription factor 2) (18) and co-repressors TGIF, Ski, and SnoN (19, 20). Known TGF-β and SMAD4 target genes include CDKN1A encoding the cyclin-dependent kinase (CDK) inhibitor p21CIP1 and CDKN2B encoding the CDK4 inhibitor p15INK4B, which promote cell cycle arrest (21, 22), as well as FN1 and PAI1, which are involved in remodeling of the ECM (2325). Increasing evidence supports a role for SMAD-independent mechanisms in TβR-induced signaling, which is typically mediated by mitogen-activated protein kinase (MAPK) cascades, involving p38, c-Jun N-terminal kinases, or extracellular signal–regulated kinase 1 (ERK1) and ERK2 (ERK1/2) (23, 26, 27).

Despite that TGF-β signaling has been studied since the 1980s, the proteins that elicit the many diverse downstream effects have not been extensively characterized at the systems level. Recent studies using proteomics approaches have identified relatively few TGF-β–regulated proteins (28, 29), likely due to technical limitations. In addition, proteome-wide analysis of phosphorylation events in TGF-β signaling has resulted in the identification of relatively few sites (30, 31). Here, to obtain systems-level insight into the mechanisms involved in the TGF-β pathway, we used high-resolution mass spectrometry (MS) to investigate temporal changes in the abundance and phosphorylation of proteins in HaCaT cells stimulated with TGF-β. Our extensive proteomic analysis identified many proteins known to be involved in TGF-β signaling as well as many proteins that were previously uncharacterized with respect to TGF-β signaling, including ECM and cytosolic proteins, transcription factors, and substrates of TGF-β receptors and downstream kinases. The combined analysis of proteomic and phosphoproteomic data enabled visualization of the interplay of transcription factors, kinases, and other molecular pathways driving cytostasis, EMT, and other cellular processes induced by TGF-β.

RESULTS

Proteomic analysis of TGF-β–induced signaling across time

To identify mechanisms of TGF-β signaling at the systems level, we performed quantitative MS analysis of protein lysates from HaCaT cells stimulated with TGF-β for various times. We found that exposing HaCaT cells to TGF-β for 3 days induced G1-phase cell cycle arrest (Fig. 1A), as well as internalization of E-cadherin and formation of actin stress fibers (Fig. 1B), indicative of a mesenchymal phenotype. We exposed HaCaT cells to TGF-β and harvested lysates at 0, 6, 12, 24, 36, and 48 hours in biological quadruplicate and performed single-run MS analyses (fig. S1). To account for the autocrine secretion of TGF-β and other growth factors, which is proportional to cell density (32, 33), we collected unstimulated control HaCaT cells at each time point for protein normalization.

Fig. 1 Phenotypic and MS analyses of HaCaT cells stimulated with TGF-β.

(A) Graph of the distribution of cells labeled with 5-bromo-2′-deoxyuridine (BrdU) (y axis) and propidium iodide (x axis) and analyzed by flow cytometry. (B) Representative confocal images of cells stained for E-cadherin or F-actin (stress fibers). For (B) and (C), cells were left unstimulated or stimulated with TGF-β for 40 hours. Data are representative of n = 3 independent experiments. (C) Representative spectra and peptide intensity heatmap used to identify and quantify a peptide corresponding to FN1. The upper left panel shows multiple different precursor ions (indicated as different colors) eluted across time in a single LC-MS experiment. The elution time and mass-to-charge ratio (m/z) of an FN1 peptide is indicated by the black box. The upper right panel shows a representative MS2 spectrum of the fragmented FN1 peptide used for identification. The middle panels show heatmaps of the intensity of precursor peptides across LC elution times for multiple LC-MS experiments corresponding to samples of HaCaT cells exposed to TGF-β for the indicated times. For quantification, intensities were determined as the intensity maximum over the retention time profile. The bottom panels show MS1 intensity of the precursor peptide at the peak maximum.

Combined analysis of the spectra from all samples resulted in the identification of 7491 proteins at a false discovery rate (FDR) of 1% (6149 ± 302 proteins per replicate across 44 samples) (table S1). To improve overall depth of coverage, peptides from a fifth replicate per time point were fractionated by strong cation exchange (SCX), but this analysis did not lead to an appreciable increase in number of identified proteins (<100 additional proteins per time point) [table S1 and “protein groups table” in the data in the Proteomics Identifications (PRIDE) repository ID# PXD000496]. Using the label-free quantification algorithm in MaxQuant (34), we measured the abundance of 6113 proteins in at least one pairwise comparison (Fig. 1C and table S1). The abundance of quantified proteins spanned greater than seven orders of magnitude (table S1), demonstrating the sensitivity of our approach. Analysis of quadruplicate samples showed excellent reproducibility with a Pearson’s correlation coefficient of at least 0.96 for all pairwise comparisons at a given time point (fig. S2). The correlations between samples of unstimulated and TGF-β–stimulated cells were also high but gradually diminished with increasing length of stimulation times (fig. S2), indicating a systematic and reproducible change in the abundance of proteins.

Statistical and bioinformatic analyses of TGF-β–regulated proteins

We calculated the fold change for the abundances of the 6113 proteins in each sample at each time point in TGF-β–stimulated relative to unstimulated cells (table S2). The average abundances of 2079 proteins showed significant differences in at least one time point [analysis of variance (ANOVA), 5% FDR] (table S2).

TGF-β signaling induces a cytostatic response (22, 35); therefore, we analyzed proteins involved in the regulation of cell cycle. Consistent with the fact that the genes encoding p21CIP1 (CDKN1A) and p15INK4B (CDKN2B) are TGF-β and SMAD targets (22, 35), we observed increased abundance of these proteins in cells stimulated with TGF-β (table S2). Antigen Ki-67 (encoded by MKI67) is a prototypic cell cycle–related nuclear protein commonly used as a marker for cell proliferation, including in cancer cells (36). The abundance of Ki-67 decreased by 2.6 orders of magnitude in cells stimulated with TGF-β for 48 hours compared to unstimulated cells (Fig. 2A), and this was the largest decrease in abundance of any protein measured (table S2). In addition, other cell cycle–related proteins including kinesin family member 4A (KIF4A), anillin (ANLN), and minichromosome maintenance complex component 6 (MCM6) showed a marked time-dependent decrease in abundance of TGF-β–stimulated cells (Fig. 2A). Regulation of the abundance and activation of CDKs is an important mechanism of the antiproliferative effects of TGF-β (37). We found that the abundance of CDK1 remained constant until after 12 hours of stimulation with TGF-β and then gradually decreased to about 10% of its initial value (Fig. 2B), showing that CDK1 was regulated by TGF-β signaling in HaCaT cells and confirming that our approach was not limited to highly abundant proteins.

Fig. 2 Functional annotation analysis of proteins with TGF-β–regulated abundance.

(A) Label-free quantification of the abundance of proteins for the indicated gene names, which showed the largest increase or decrease in cells exposed to TGF-β at 48 hours. The color indicates the cellular function. Data are means ± SEM of biological quadruplicates. (B) Quantification of the abundance of CDK1 and ITGB6 in cells exposed to TGF-β. Circles indicate values from individual replicates. Bars indicate means ± SEM. (C) Heatmap of z score– and log2-transformed ratios of the average abundance of proteins that showed significant differences across time of exposure to TGF-β. Proteins were grouped using unsupervised hierarchical clustering. Examples of significantly enriched functional annotations for the individual clusters are shown. 2% FDR, Fisher’s exact test. (D) Heatmap of z score–transformed position score for the indicated KEGG pathway annotations, which were significantly enriched at the indicated time points. Annotations were grouped using unsupervised hierarchical clustering of position scores.

In contrast to cell cycle–related proteins, we predicted that stimulation with TGF-β would increase the abundance of proteins associated with the EMT-like phenotypic switch. We found increased abundance of proteins involved in this process, including smooth muscle actin (ACTA2), integrins (ITGA2, ITGA3, ITGA6, ITGB1, ITGB4, ITGB5, ITGB6, ITGA8, and ITGAV), SERPINE1 (also known as PAI1), FN1, transgelin (TAGLN), and transglutaminase 2 (TGM2) (8, 25, 38, 39) (table S2). FN1, TAGLN, and TGM2 were among the top five proteins that increased in abundance in TGF-β–stimulated cells (Fig. 2A). FN1 production and deposition in the ECM is a hallmark of TGF-β–induced EMT (25), and FN1 had the largest increase in abundance of all proteins measured in TGF-β–stimulated cells, increasing by more than 3.5 orders of magnitude by 48 hours (table S2 and Fig. 2A).

We also identified proteins with TGF-β–regulated abundances that were involved in other cellular functions. For example, integrin β6 (ITGB6) is an ECM-interacting plasma membrane protein that also binds and activates latent, inactive TGF-β (43). The abundance of ITGB6 was significantly increased in TGF-β–stimulated cells (Fig. 2B), suggesting that it may be part of a positive feed-forward loop. Moreover, the Rho-interacting kinase Citron (CIT) and the previously uncharacterized protein proline-rich protein 9 (PRR9), neither of which has been implicated in TGF-β signaling, were among the most highly regulated proteins in TGF-β–stimulated cells (Fig. 2A).

To gain a broader understanding of the biological processes and pathways involved in TGF-β–induced responses, we performed bioinformatics analyses. We used unsupervised hierarchical clustering to group proteins with abundances that were significantly altered in TGF-β–stimulated cells and found that these proteins clustered into two major groups corresponding to proteins that generally increased (cluster one) or decreased (cluster two) in abundance (Fig. 2C). We performed an enrichment analysis of functional annotations using KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways and GO (Gene Ontology) biological processes. Cluster 1 was significantly enriched for annotations for processes that may be involved in the TGF-β–induced EMT (44), including ECM-receptor interaction, focal adhesion, actin cytoskeleton, and other processes (tables S3 and S4). In addition, we also found enrichment for annotations related to vesicle-mediated transport that are not generally associated with TGF-β signaling, including protein glycosylation, vesicle organization, and vesicle coating (tables S3 and S4). Cluster 2 was significantly enriched for different types of functional annotations for processes that may be influenced by TGF-β signaling (45, 46), including cell cycle process, DNA replication, mRNA processing, spliceosome, and regulation of transcription (table S4).

In an orthogonal approach, we assessed whether annotations for specific biological processes were enriched in TGF-β–regulated proteins at each time point. We used a recently developed algorithm that relies on both the number and abundance of proteins to determine statistical enrichment for a given annotation (47). We identified 49 KEGG pathway annotations that were significantly enriched in at least one time point (table S5). Using unsupervised hierarchical clustering of annotations, we found that proteins that increased in abundance with exposure to TGF-β were associated with functional annotations consistent with EMT-like cell behaviors, and those that decreased in abundance were associated with annotations related to cell cycle inhibition, nucleic acid metabolism, transcription, and the regulation of DNA replication and repair (Fig. 2D). This approach also facilitated a more in-depth analysis of vesicle-mediated transport, identifying proteins annotated with SNARE [soluble NSF (N-ethylmaleimide–sensitive factor) attachment protein receptor] interactions in vesicular transport, lysosome, endocytosis, and phagosome, implicating these processes as potential targets of TGF-β signaling. Furthermore, sphingolipid metabolism and mammalian target of rapamycin (mTOR) signaling pathway annotations were specifically enriched with proteins that increased in abundance after 36 hours of stimulation with TGF-β (Fig. 2D). We also observed functional enrichment of unexpected KEGG pathway annotations, such as toxoplasmosis. However, inspection of the identity of proteins that contributed to these enrichments revealed that many were components of the ECM, proteins encoded by genes that directly respond to TGF-β signaling, and kinases involved in SMAD-independent TGF-β signaling (table S5), suggesting that the identification of these processes may be an artifact of the generalized nature of protein annotations.

Functional validation of TGF-β–regulated proteins

To validate our proteomics study, we analyzed changes in the abundance of several proteins by Western blot. We selected proteins with the highest fold increase in TGF-β–stimulated cells that were also not previously known to be affected in TGF-β signaling and had known functional or structural domains. We found that the immunoreactivity for adhesion molecule with immunoglobulin-like domain 2 (AMIGO2), inositol polyphosphate-4-phosphatase type II (INPP4b), programmed cell death protein 4 (PDCD4), transmembrane protein 2 (TMEM2), CDK17, and OCIA domain–containing protein 2 (OCIAD2) increased in a time-dependent manner in cells exposed to TGF-β (Fig. 3A).

Fig. 3 Functional validation of TGF-β–induced changes in protein abundance.

(A) Western blot with the indicated antibodies in lysates of HaCaT cells exposed to vehicle control or TGF-β for the indicated times. Data are representative of n = 3 biological replicates. (B and C) HaCaT cells stably expressing nontargeting control shRNA or shRNAs targeting PRSS23 and exposed to vehicle control or TGF-β for the indicated times. (B) Graph of the abundance of PRSS23 mRNA shown as a ratio relative to the abundance of actin. Data are means ± SD of biological triplicates. (C) Western blot with the indicated antibodies. Data are representative of n = 3 biological replicates.

The serine protease PRSS23 was increased in TGF-β–stimulated cells (table S2) and was recently shown to be involved the endothelial-to-mesenchymal transition (48). Using reverse transcription quantitative polymerase chain reaction (RT-qPCR), we found that the abundance of PRSS23 mRNA was increased in TGF-β–stimulated cells (Fig. 3B). PRSS23 contains a signal peptide and, thus, may be secreted and play a role in ECM remodeling. To assess the possible function of PRSS23 in the context of TGF-β signaling, we used three independent short hairpin RNAs (shRNAs), which reduced the abundance of PRSS23 mRNA (Fig. 3B), and found that depletion of PRSS23 reduced the ability of TGF-β to increase the abundance of PAI1 but had no effect on the phosphorylation of SMAD2 (Fig. 3C and fig. S3). We also assessed whether PRSS23 was required for molecular changes that occur later in EMT in HaCaT cells exposed to TGF-β for 48 hours, and found that PRSS23 knockdown did not consistently affect stress fiber formation, fibronectin production, or changes in the abundance of E-cadherin and N-cadherin (fig. S3). Thus, PRSS23 is likely a transcriptional target of TGF-β signaling that may affect ECM proteolysis and likely acts redundantly with other TGF-β–regulated proteins to induce complete EMT.

Analysis of upstream transcription regulators induced by TGF-β treatment

To identify factors that are either activated or inhibited in response to TGF-β and act upstream of the observed changes in protein abundance, we analyzed the TGF-β–regulated proteins using the “upstream regulator” analysis module in Ingenuity Pathway Analysis (IPA). We identified 253 significantly enriched upstream regulators, including 50 “transcription regulators,” which were primarily DNA-associated transcription factors, as well as numerous kinases, microRNAs, translational regulators, cytokines, and growth factors, including TGF-β itself (table S6). We filtered for transcription regulators that were predicted to be activated or inhibited by IPA analysis at least at one time point. We clustered the resulting 40 transcription regulators on the basis of the absolute value of the IPA activation score at any time point to identify temporally regulated factors (table S6 and Fig. 4A). SMAD2, SMAD3, and SMAD4 had a sustained activation profile, whereas the activity of other activated regulators, such as SNAIL2, appeared to be delayed in comparison (Fig. 4A). We identified several transcription factors, some of which have previously been implicated in TGF-β signaling, including p53 and SMARCA4 (40, 49), and others that are likely not key components of SMAD-dependent TGF-β–induced signaling, such as vitamin D receptor (VDR), TATA box binding protein–associated factor 4 (TAF4), SAM pointed domain–containing ETS transcription factor (SPDEF), CUT-like homeobox 1 (CUX1), T-box 2 (TBX2), and specificity protein 1 (SP1) (41, 42, 5053) (Fig. 4A). Finally, our analysis revealed previously uncharacterized transcriptional regulators of TGF-β signaling, including myotrophin (MTPN) and huntingtin (HTT) (Fig. 4A). This analysis correctly predicted key direct and indirect transcriptional regulators of TGF-β signaling, including SMAD2, SMAD3, SNAIL2, SMAD7, and c-MYC (18), solely from the temporal changes in the abundance of their targets (Fig. 4, A and B).

Fig. 4 Upstream regulator analysis of proteins with TGF-β–regulated abundance.

(A) Heatmap of the activation scores for upstream transcriptional regulators predicted to be activated (olive green) or inhibited (brown) by exposure to TGF-β in at least one time point. (B) Quantification of MS data showing TGF-β–induced changes in the abundance of proteins corresponding to targets for the indicated transcriptional regulators. Proteins in red correspond to those that were predicted to increase abundance when the transcriptional regulator is activated, and proteins in blue correspond to those that were predicted to decrease abundance when the transcription is activated. The proteins corresponding to the targets of TBX2 and SMAD7 showed the opposite of the predicted response (proteins that were predicted to increase in abundance with activation of the transcriptional regulator were decreased in the MS data and those that were predicted to decrease with activation of the transcriptional regulator were increased in the MS data), suggesting that exposure of cells to TGF-β inhibited these transcriptional regulators.

Phosphoproteomic analysis of early signaling events

Molecular events that occur immediately after activation of the TGF-β receptor are mediated in part by protein phosphorylation. Thus, we analyzed proteome-wide changes in protein phosphorylation in HaCaT cells stimulated with TGF-β for 0, 5, 10, 15, or 20 min using a “double-triple” SILAC (stable isotopes labeling by amino acids in cell culture) approach (54) (fig. S1). We assumed that changes in protein abundance would be minimal at these time points (55), and thus, any changes in the abundance of phosphopeptides could be attributed to changes in protein phosphorylation. Combined analysis of the spectra from all samples in biological triplicate identified 22,388 phosphopeptides corresponding to 5345 proteins (PRIDE repository ID# PXD000496). For 15,696 phosphosites, we identified the specific phosphorylated amino acid with a localization probability >0.75 (also referred to as class I sites) with an average localization probability >0.99 (table S7). To assess the coverage of the measured phosphoproteome, we compared the TGF-β–dependent class I sites identified here to all phosphosites in the PhosphoSitePlus database (56) and found 11,595 matching sites, indicating that we identified 4101 previously uncharacterized phosphosites. We compared our data to a previously published analysis of TGF-β–regulated phosphorylated proteins in colorectal cancer cells (30) and found that 9 of the 13 sites identified in that study were also present in our data (table S7). Of the 15,696 phosphosites, we quantified 14,010 sites (either singly or multiply phosphorylated) in at least six measurements in TGF-β–stimulated samples (table S7) and used these for further analyses. We identified 2892 phosphosites with significantly altered abundance in TGF-β–stimulated cells in at least one time point (ANOVA, 1% FDR) (table S7). We identified phosphorylation of activating residues on the C terminus of SMAD2 (Ser465 and Ser467) (table S7), which is a direct substrate of TβR1 (57, 58). In addition, exposure to TGF-β increased phosphorylation of other known TGF-β–regulated sites, including phosphorylation of CDK1, CDK2, or CDK3 (30, 59), eukaryotic initiation factor (EIF)–4E binding protein 1 (EIF4EBP1) (60), p21 protein (Cdc42Rac)–activated kinase 2 (PAK2) (61) (Fig. 5A), and PAK1 (62), as well as the mTOR pathway–associated proteins, including ribosomal protein S6 (RPS6) kinase α-1 (RPS6KA1), EIF4B, EIF4EBP1, EIF4G1, EIF5B, and ribosomal protein S3 (RPS3) (63, 64) (table S7). We did not identify phosphopeptides corresponding to the known phosphorylation of the GS (glycine-serine-rich) domain of TβR1, which has also not been found in any previous large-scale MS study, which may be because tryptic digestion of this protein is predicted to generate a large hydrophobic peptide that is difficult to detect by MS.

Fig. 5 Analysis of phosphoproteomic changes in response to TGF-β.

(A) Quantification of the abundance of phosphosites for the indicated residues and proteins. Circles indicate values from individual replicates. Bars indicate means ± SEM. (B) Western blot with the indicated antibodies in lysates of HaCaT cells used for phosphoproteomic MS and exposed to vehicle control or TGF-β for the indicated times. (C) Heatmap of z score– and log2-transformed ratios of the average abundance of phosphosites that showed significant differences across time of exposure to TGF-β. Phosphosites were grouped using unsupervised hierarchical clustering. (D) Sequence logo graphs of examples of significantly enriched kinase substrate motifs for individual clusters of phosphosites shown in (C). β-Adrenergic receptor kinase (β-ARK) and AMP-activated protein kinase (AMPK) motifs were enriched in cluster 1. Polo-like PBD binding kinase (PLK-like) and GSK3 motifs were enriched in cluster 2. ERK1 and ERK2 kinase (ERK1/2) and CDK1, CDK2, CDK4, and CDK6 kinase (CDK) motifs were enriched in cluster 3. Details of enrichment scores are given in table S8.

We performed validation of TGF-β–induced phosphorylation of proteins by Western blot. Consistent with the MS data, we found that phosphorylation of Tyr15 of CDK1, CDK2, or CDK3 and Ser807 and Ser811 of retinoblastoma 1 (RB1), which are proteins involved in cell cycle control (65), was increased in HaCaT cells exposed to TGF-β (Fig. 5B). Likewise, phosphorylation of Ser142, Ser143, or both of these residues of zyxin, which localizes to focal adhesions (66), was also increased in HaCaT cells exposed to TGF-β (Fig. 5B).

We analyzed TGF-β–dependent phosphorylated proteins for the presence of phosphorylation motifs annotated in the Human Protein Reference Database (HPRD) (67). Using a Fisher’s exact text, we found significant enrichment of kinase substrate motifs in proteins corresponding to the 2892 TGF-β–regulated phosphosites compared to proteins corresponding to all 22,388 phosphosites identified in our screen (table S8). Among these, ERK1/2, CDKs, AKT, ataxia-telangiectasia mutated (ATM), and glycogen synthase kinase 3 (GSK3) substrate motifs showed strong enrichment (table S8, P < 1 × 10−10).

To better understand the temporal regulation of phosphorylation in cells exposed to TGF-β, we performed unsupervised hierarchical clustering of the 2892 TGF-β–regulated phosphosites and identified three major clusters (Fig. 5C). The phosphosites in cluster 1 (538 phosphosites) had abundances that initially increased in TGF-β–stimulated relative to unstimulated cells and then decreased with increasing time of TGF-β stimulation, those in cluster 2 (782 phosphosites) showed an oscillating pattern of changes in abundance, and those in cluster 3 (1386 phosphosites) had abundances that initially decreased and then increased with longer times of TGF-β stimulation (Fig. 5C). Because the data are based on only four time points, it is not clear whether the oscillating pattern of phosphosites in cluster 2 represents a biological response or experimental variability.

We explored the functional implications of TGF-β–stimulated phosphorylation by performing enrichment analysis of functional annotations for GO biological process and KEGG pathways on the proteins corresponding to the 2892 TGF-β–regulated phosphosites. These proteins were enriched for cell cycle, regulation of actin cytoskeleton, spliceosome, DNA replication, mTOR signaling pathway, adherens or tight junction, and focal adhesion, which are known signaling responses to TGF-β (table S8) (1, 63, 68, 69). Additional annotations corresponded to functions not generally associated with TGF-β signaling, but similar to those identified in our analysis of changes in protein abundance in TGF-β–stimulated cells, such as DNA mismatch repair (tables S3 to S5).

Το identify potential differences in kinase substrate motifs among clusters of TGF-β–regulated phosphosites, we performed enrichment analysis using annotations from HPRD and created sequence logos in the iceLogo software (70). Phosphosites in cluster 1 were significantly enriched for adenosine monophosphate (AMP)–activated protein kinase and β-adrenergic receptor kinase consensus motifs (Fig. 5D and table S8), suggesting that early events that occur downstream of TβR1 activation may trigger activation of these kinases. However, we also observed enrichment of casein kinase 2 (CK2) consensus motifs (P < 1 × 10−16) in cluster 1 phosphosites (table S8), and the β-adrenergic receptor kinase motif ([D/E][pS/pT]XXX) contains a subset of CK2 motifs ([D/E]pS[D/E]X[D/E]) (67), suggesting that some of these phosphosites may be target by one or both these kinases. Phosphosites in cluster 2 were significantly enriched for polo-like kinase 1 (PLK1) polo box domain (PBD) binding kinase motifs and GSK3 motifs (Fig. 5D and table S8), consistent with the known roles of phosphatidylinositide 3-kinase, AKT, and GSK3 downstream of TGF-β (27). Phosphosites in cluster 3 showed increased abundance after 15 min and were significantly enriched for motifs recognized by proline-directed kinases, such as MAPKs (Fig. 5D and table S8), consistent with activation of these kinases by a SMAD-independent mechanism (26). Phosphosites in cluster 3 were also significantly enriched for CDK motifs, consistent with similar phosphorylation profiles for the majority of known CDK substrates (fig. S4).

Pathway-specific view of TGF-β signaling in cell cycle

One of the major cellular effects of TGF-β stimulation in nontransformed cells is cell cycle arrest in G1 (35), and the annotation cell cycle was enriched in both the proteomic and phosphoproteomic analyses (tables S3 to S5, S8, and S9). We found that 83 of 98 proteins annotated in KEGG pathways with cell cycle had altered abundance or phosphorylation in TGF-β–stimulated cells (tables S2 and S7) (modeled in Fig. 6). Moreover, peptides corresponding to 48 of the 83 proteins associated with cell cycle were among those with the largest changes in abundance in TGF-β–stimulated cells at the phosphoproteome (24 proteins) or proteome level (31 proteins) or both (7 proteins). TGF-β induced early changes in phosphorylation that primarily occurred in the network around CDKs and their downstream effectors (Fig. 6 and fig. S4), whereas changes in the abundance of proteins occurred throughout the cell cycle pathway (Fig. 6). A number of proteins driving the cell cycle [CDKs, RB1, and PCNA (proliferating cell nuclear antigen)] were reduced in abundance, whereas cell cycle inhibitors (p15INK4B and p21CIP1) were increased in abundance after 24 hours of exposure of cells to TGF-β (Fig. 6), which would be predicted to lead to cell cycle arrest. The temporal profiles of phosphorylation induced by activation of TGF-β signaling varied among individual proteins involved in cell cycle. For example, phosphorylation of Tyr15 of CDK1, CDK2, or CDK3, which inhibits its kinase activity (71), or phosphorylation of Ser64 of S-phase kinase–associated protein 2 (SKP2), which protects CDKs from degradation (72), increased over time in cells stimulated with TGF-β (Fig. 6). In contrast, phosphorylation of Ser1068 and Ser1112 on retinoblastoma-like 2 (RBL2, also known as p130), which inhibits its interaction with E2F transcription factor (73), was reduced after 15 min in cells stimulated with TGF-β (Fig. 6). Whereas phosphorylation of some proteins suggested increased cell proliferation (such as RB1 phosphorylation), that of others (activation of SMAD2 or inhibition of CDK1, CDK2, or CDK3) suggested inhibition of cell proliferation. We found a twofold decrease in the phosphorylation of Ser120 of checkpoint kinase 2 (CHK2) (Fig. 6 and table S7), suggesting TGF-β–induced activation of G1 checkpoint and cell cycle arrest (74, 75).

Fig. 6 Network diagram of cell cycle–related proteins affected by TGF-β.

Proteins are indicated by rectangles. White indicates that the protein was not identified, and gray indicates that the protein was identified but not significantly regulated in our MS data. Regulated proteins are shown with varying hues of purple indicating the relative abundance. Regulated phosphosites from the phosphoproteome analysis are shown with varying hues of orange indicating the relative abundance. e, gene expression; +p, phosphorylation; −p, dephosphorylation; +u, ubiquitylation; O, expression as annotated in the KEGG database.

Mechanistic network of the regulated proteome and phosphoproteome

Functional enrichment analysis of all proteins that showed altered abundance or phosphorylation in TGF-β–stimulated cells revealed annotations consistent with changes in cell cycle and cell motility similar to those seen when the data sets were analyzed individually. Using the BioCarta pathways database (http://www.biocarta.com) (76), we found enrichment of VDR signaling, ATM signaling pathway, SWI-SNF chromatin remodeling, and cell cycle G1 to S checkpoint in the combined data (table S9).

Recent evidence shows that, in addition to SMAD-dependent, there are also SMAD-independent TGF-β signaling pathways (7779) that crosstalk to other signaling pathways (80). To establish a systems-level perspective of TGF-β–induced EMT and cell cycle arrest, we combined our MS data with knowledge of kinase-substrate and protein-protein interactions. We constructed a model using 100 proteins with the largest change in abundance in cells stimulated with TGF-β; 271 proteins corresponding to phosphosites that were significantly altered in abundance in cells stimulated with TGF-β over the time course, which included 40 phosphosites that are used as indicators of protein activation or inhibition in other studies [from PhosphoSitePlus (56)]; and SMAD2, SMAD3, SMAD7, and SNAIL2, which were identified in our analysis of upstream regulators of TGF-β–induced changes in protein abundance. We input this list into STRING (81), identifying a network of 150 proteins that were highly interconnected, annotated kinase-substrate interactions using information from PhosphoSitePlus (56), and rendered the network using Cytoscape (Fig. 7 and fig. S5).

Fig. 7 Protein interaction network of proteins with TGF-β–induced changes in abundance or phosphorylation.

Selected proteins or groups of proteins from the complete network shown in fig. S5. Proteins are indicated by rectangular nodes. Gray indicates that the protein was identified but not significantly regulated in our MS data. Regulated proteins are shown with varying hues of purple indicating the relative abundance. Regulated phosphosites from the phosphoproteome analyses are shown with varying hues of orange indicating the relative abundance. Red arrows indicate kinase-substrate interactions, blue borders indicate transcription factors, and red borders indicate kinases.

We focused on the network architecture and identified several functionally linked proteins within the network, including members of the mTOR signaling pathway (RPS6KA1, EIF4B, EIF4EBP1, EIF4G1, EIF5B, and RPS3), small guanosine triphosphate–binding proteins (RAPGEF1, RACGAP1, and ARHGEF12), nucleoporins, the MCM complex, ECM adhesion components, and proteins regulating the actin cytoskeleton. The temporal changes in the abundance of these proteins in cells exposed to TGF-β and cellular location of the individual proteins in the network indicated that TGF-β likely promotes the accumulation of ECM proteins, an essential process for cell migration after EMT (44).

Crosstalk among proteins that localize to the plasma membrane may be important for TGF-β signaling. We observed that multiple proteins, including integrins (ITGAV, ITGB4, and ITGB6) and tight junction proteins (TJP1 and TJP2), showed changes in phosphorylation in TGF-β–stimulated cells, suggesting potential crosstalk between TβR1 and other receptors. Other membrane-associated proteins that changed in abundance in TGF-β–stimulated cells included those that concentrate at areas of cell-cell and cell-matrix contacts such as FN1 and integrins (39, 43) and scaffolding proteins, such as talin1 (TLN1), breast cancer anti–estrogen receptor protein 1 (BCAR1), neurofibromin 2 (NF2), and β-spectrin 2 (SPTBN1), which link plasma membrane to actin cytoskeleton and focal adhesions (8286), suggesting mechanisms that may promote cell migration. The protein tyrosine kinases SRC and EGFR (epidermal growth factor receptor) contribute to noncanonical TGF-β signaling in mesenchymal or dedifferentiated epithelial cells (4) and showed altered phosphorylation in TGF-β–stimulated cells.

Protein kinases and transcription factors are key regulators in signaling systems. Our network analysis suggested involvement of putative key transcription factors and kinases downstream of TGF-β signaling. Several transcription factors including SMADs and SMARCA4, mediator complex subunit 1 (MED1), β-catenin (CTNNB1), RB1, and p53 had altered phosphorylation in TGF-β–stimulated cells. The kinases CDK1, CDK2, SRC, protein kinase Cδ (PRKCD), PAK1, PAK2, Rho-associated coiled-coil containing protein kinase 2 (ROCK2), and mitogen-activated protein 4 kinase 4 (MAP4K4) were the primary regulated kinases identified in TGF-β–stimulated cells. These proteins were central to the network, connecting phosphorylation signaling to proteomic changes associated with TGF-β–induced cellular responses.

DISCUSSION

Here, we analyzed time-resolved, proteome-wide changes in protein phosphorylation and abundance in response to TGF-β stimulation of HaCaT keratinocytes. Our data overlapped with TGF-β–induced phosphosites found in a previous published proteome study in colorectal cancer cells (30) and with many protein and phosphorylation changes known to occur in HaCaT cells in response to activation of TGF-β signaling, such as those associated with cell cycle inhibition, ECM remodeling, and EMT (4, 10, 44). We also discovered many previously uncharacterized TGF-β–regulated protein and phosphorylation changes. We validated that TGF-β increased the abundance or phosphorylation of several proteins by Western blot. In addition, we provided evidence that TGF-β increased the expression of PRSS23 and that PRSS23 may be required for TGF-β–induced remodeling of ECM (Fig. 3). A recent study showed that PRSS23 is essential for endothelial-to-mesenchymal transition in human aortic endothelial cells and that PRSS23 knockdown did not affect the phosphorylation of SMAD2 but decreased the transcription of SNAI1 (48), the expression of which lies downstream of SMAD4 in cells exposed to TGF-β (87). Thus, our data are consistent with previous studies and may be a valuable resource for future studies of TGF-β signaling in multiple cell types.

Investigation of upstream transcription regulators and integrated network analysis of proteomic and phosphoproteomic data suggested that phosphorylation of several transcription factors and cofactors could contribute to TGF-β–induced changes in gene expression (Figs. 5 and 7). SMAD2 and SMAD3 were activated and c-MYC was inhibited throughout the time course as identified by upstream regulator analysis in IPA (Fig. 4 and table S6). We detected additional transcriptional regulators that may act independently or in conjunction with the SMAD complex to promote increased affinity or selectivity for subsets of target genes. For example, the vitamin D3 receptor (VDR), a ligand-dependent transcription factor that is activated upon binding to vitamin D or its analogs, was found to be activated in the upstream regulators analysis, and the control of gene expression by VDR was the most enriched category in the functional enrichment analysis of combined proteome and phosphoproteome data sets (Fig. 4A and table S9). VDR is implicated in the regulation of proliferation (87), differentiation (88), and EMT in keratinocytes (89) and in a negative feedback loop in regulation of SMAD signaling in multiple cell types (50). The abundance of the ankyrin repeat–containing protein MTPN was increased in TGF-β–stimulated cells, and MTPN was found to be activated in the upstream regulator analysis (Fig. 4 and table S6). The role of MTPN in SMAD-dependent TGF-β signaling is unknown, but MTPN overexpression leads to myocardial hypertrophy and, in a mouse model, increases TGF-β and FN1 expression (90). Thus, MTPN and TGF-β could be part of a positive feedback loop.

Epithelial cells, including HaCaT keratinocytes, respond to TGF-β by inhibition of cell cycle and activation of cell motility through EMT. We found evidence that vesicular transport may be an important target of TGF-β–regulated proteins involved in cell motility. TGF-β–regulated proteins were enriched for the annotations: endocytosis, SNARE interactions, and lysosomes (Fig. 2D and tables S3 to S5 and S9). This may represent feed-forward or negative regulatory loops that are activated upon TGF-β signaling to modulate signaling capacity. For example, caveolin-dependent internalization of activated TGF-β receptor complexes leads to receptor degradation and terminates signaling (91). Moreover, during EMT, tight junction proteins are internalized by clathrin-mediated endocytosis, which, together with recycling of focal adhesion proteins, such as integrins, is necessary for migration (92, 93). We observed that TGF-β stimulation increased the abundance of ECM proteins, including fibronectin, collagens, and laminins, as well as proteins in the secretory pathway, which may reflect the increased demand for secretion of ECM during EMT and migration (94).

TGF-β signaling also leads to cell cycle arrest in epithelial cells, and our MS analysis identified changes consistent with this effect. We found several known TGF-β–induced changes in the phosphorylation or abundance of cell cycle proteins, including p21CIP1 and p15INK4B (21, 53) and CDK1, CDK2, or CDK3 (30, 95). However, changes in substrate phosphorylation suggested that CDKs may also be activated by TGF-β. For example, TGF-β induced phosphorylation of Ser375 of the phosphatase CDC25B, which promotes its ability to dephosphorylate and thereby activate CDK1 (96). Likewise, TGF-β stimulation increased phosphorylation of known CDK substrates (97), including SKP2, DNA ligase 1 (LIG1), high mobility group protein HMGA1, nuclear casein kinase and cyclin-dependent kinase substrate (NUCKS), and RB1 (Figs. 6 and 7 and figs. S4 and S5). Moreover, we found that the abundance of phosphosites corresponding to CDK substrates peaked around 15 to 20 min after exposure to TGF-β (Fig. 5C, cluster 3, and fig. S4).

Direct phosphorylation of the tumor suppressor RB1 by CDK1 or CDK2 prevents its ability to inhibit cell cycle. TGF-β signaling can induce cell cycle arrest by inhibiting the activity of CDK1 and CDK2 toward RB1 (98). However, in some cell types where TGF-β promotes proliferation, it activates CDK1 and CDK2, leading to increased RB1 phosphorylation (99). We found that TGF-β stimulation for 15 to 20 min increased phosphorylation of RB1 at multiple sites (Figs. 5B, 6, and 7). Thus, very early responses to TGF-β may have both stimulatory and inhibitory effects on the cell cycle, whereas subsequent responses may primarily inhibit cell cycle. Consistent with this model, we found that exposing cells to TGF-β for as little as 15 to 20 min increased phosphorylation of Thr70 of the translational repressor EIF4EBP1, which prevents its binding to the EIF4E of the translational initiation complex and thereby relieves repression of protein synthesis required for cell growth and proliferation (100, 101) (Fig. 7). In contrast, exposing cells to TGF-β for longer than 24 hours results in increased abundance of EIF4EBP1 and a corresponding increase in its binding to EIF4E, thereby leading to inhibition of cell proliferation (102). Therefore, early TGF-β signaling is composed of phosphorylation events that inhibit proliferation (CDK inhibitory phosphorylation, SMAD activation) as well as those that promote proliferation (EIF4EBP1, RB1, and CDK substrate phosphorylation) (Fig. 5B), whereas extended stimulation with TGF-β (longer than 24 hours) induces changes in protein abundance that override the proliferative signaling, leading to cell cycle arrest. Therefore, imbalances in the relative activation of SMAD-dependent and SMAD-independent TGF-β signaling may underlie the protumorigenic effects of TGF-β (4).

Our MS data suggested that biological responses to TGF-β result from extensive crosstalk with other signaling pathways. For example, we found that TGF-β induced phosphorylation of AKT consistent with its activation (103), as well as phosphorylation of its downstream substrates, including GSK3α (104) (fig. S5). Phosphorylation by AKT inhibits GSK3α (105), which phosphorylates the transcriptional coactivator β-catenin, promoting its degradation (106). We found that TGF-β stimulation increased phosphorylation of β-catenin, not on the site downstream of GSK3 (consistent with GSK3 inhibition by AKT) but rather on the site phosphorylated by PAK1 (107), a kinase that was activated in response to TGF-β. Phosphorylation of this site on PAK1 increases the stability of β-catenin (107). Thus, the observed increased abundance of β-catenin in TGF-β–stimulated HaCaT cells (fig. S5) could be explained by the opposite regulation of upstream kinases acting on both these sites. This example illustrates that the combination of a systems-level analysis with knowledge of specific phosphorylated sites can enable a deeper understanding of the mechanisms of pathway crosstalk. In conclusion, our study provides a time-resolved, proteome-wide understanding of the changes in phosphorylation and protein abundance in TGF-β signaling, which suggests that temporal regulation of different proteins may be a mechanism mediating context-dependent effects of TGF-β in keratinocytes and other systems.

MATERIALS AND METHODS

Workflow for proteome sample preparation

HaCaT cells were grown in a 5-cm dish to 60% confluence and preincubated with Dulbecco’s modified Eagle’s medium (DMEM; PAA Laboratories) containing 2% fetal bovine serum (FBS) for 12 hours and treated with TGF-β1 (5 ng/ml; PeproTech) for 0, 6, 12, 24, 36, and 48 hours (fig. S1). In the experimental time window that we used, 2% FBS prevents the apoptotic response to TGF-β (108). Corresponding untreated controls were also collected at all time points. Cell lysates were collected in biological quadruplicates. Cells were lysed in buffer containing 6 M guanidium chloride, 40 mM chloroacetamide (CAA), 10 mM tris(2-carboxyethyl)phosphine (TCEP) (Sigma), and 100 mM tris-HCl (pH 8.5) (109). The lysate was sonicated and incubated for 5 min at 95°C, protein concentration was quantified, and samples were processed by in-Stage tip protocol (110). Briefly, 20 µg of cell lysate was reduced, alkylated with CAA during lysis, and digested in a one-step procedure overnight at 37°C in buffer containing 10% acetonitrile, 25 mM tris-HCl (pH 8.5), and 0.03 µg of trypsin (Promega). The peptides were eluted from the SCX in-Stage tips (Empore, 3M-2251) with 5% ammonia and 80% acetonitrile, vacuum-dried, and resuspended in 0.2% formic acid. A fifth replicate per time point was processed as above, except that tryptic peptides were separated into three fractions using SCX by sequential elution with 150 mM ammonium acetate, 300 mM ammonium acetate, and 5% (v/v) ammonium hydroxide as described (110).

Workflow for phosphoproteome sample preparation

For phosphoproteomic studies, we used a double-triple SILAC-based approach as described (111). Briefly, HaCaT cells were grown in DMEM supplemented with 10% dialyzed FBS (Invitrogen), 1% streptomycin (10 mg/ml)/penicillin (10,000 U/ml) (PAA Laboratories), 1% l-glutamine (200 mM) (PAA Laboratories), 1% sodium pyruvate (100 mM) (Invitrogen), and either unlabeled l-arginine (Arg0) at 42 mg/liter and l-lysine (Lys0) at 71 mg/liter or equimolar amounts of the isotopic variants l-[U-13C6,14N4]arginine and l-[2H4]lysine (Arg6, Lys4), or l-[U-13C6,15N4]arginine and l-[U-13C6,15N2]lysine (Arg10, Lys8) (Cambridge Isotope Laboratories). After five cell doublings on culture dishes, cells were >99% labeled with the isotopes. SILAC-labeled HaCaT cells were incubated in serum-free medium for 6 hours. To achieve a five-point time course experiment with three SILAC labels, we divided the conditions into two groups. In one group, “light”-labeled cells were unstimulated (0 min), “medium”-labeled cells were stimulated with TGF-β1 (5 ng/ml; PeproTech) for 5 min, and “heavy”-labeled cells were stimulated with TGF-β1 for 15 min. In the second group, “light”-labeled cells were unstimulated (0 min), “medium”-labeled cells were stimulated with TGF-β1 for 10 min, and “heavy”-labeled cells were stimulated with TGF-β1 for 20 min. The combined data from the two groups with the common 0 time point after MS analysis were used to generate the five-point time course (fig. S1). The entire experiment was done in biological triplicate. SILAC labels were exchanged among conditions in the third replicate. Treated samples were not corrected for solvent (phosphate-buffered saline) addition, which may weakly activate the ERK pathway. For phosphoproteome analysis, samples were prepared as described earlier (112). Briefly, cells from 20-cm dishes at 80% confluence were lysed in buffer containing 100 mM tris (pH 7.6), 4% SDS, 100 mM dithiothreitol supplemented with protease and phosphatase inhibitors; sonicated; and centrifuged for 15 min at 14,000 rpm. Cell lysates from light, medium, and heavy conditions were mixed in a 1:1:1 ratio, and 10 mg of the combined sample was digested with trypsin according to the filter-aided sample preparation method (113). The resulting peptides were fractionated by SCX and subjected to phosphopeptide enrichment with TiO2 beads (MZ-Analysentechnik) (111, 114).

MS analyses

The peptides or phosphopeptides were desalted on StageTips (115) and separated by nanoscale high-pressure liquid chromatography on a reversed-phase column (packed in-house with 1.8-µm C18 Reprosil-AQ Pur reversed-phase beads) (Dr. Maisch GmbH) over 270 min (proteome analysis) or 120 to 240 min (SCX-fractionated phosphopeptides). The peptides eluting at the tip were electrosprayed and analyzed by tandem MS on a Q Exactive (116) (Thermo Fisher Scientific) with higher-energy collisional dissociation–based fragmentation. The instrument was set to alternate between full precursor scans (MS1) and up to 5 or 10 scans of fragmented ions (MS2).

Data processing and analysis

Spectra were analyzed using MaxQuant (117), versions 1.3.10.15 and 1.3.10.18, and Andromeda (118). The MS2 spectra were searched against the UniProt FASTA database version 2/25/2012 (81,213 entries) (119). Enzyme specificity was set to trypsin, allowing for cleavage N-terminal to proline and between aspartic acid and proline. The search included carbamidomethylation of cysteine as a fixed modification and N-acetylation and oxidation of methionine (and phosphorylation of serine, threonine, or tyrosine for phosphoproteins) as variable modifications. Up to two missed cleavages were allowed for protease digestion, and enzyme specificity was set to trypsin, defined as C-terminal to arginine and lysine excluding proline. The “identify” module in MaxQuant was used to filter (1% FDR) identifications at the peptide and protein level. The identity of precursor peptides present in MS1 but not selected for fragmentation and identification by MS2 in a given run was obtained by transferring peptide identifications based on accurate mass and retention times across liquid chromatography (LC)–MS runs where possible using MaxQuant (117, 120). Protein identifications were collapsed to the minimal number that contained the set of identified peptides.

Proteome quantification was performed in MaxQuant using the XIC-based label-free quantification (LFQ) algorithm (34). In MaxQuant, a quantification event was reported only when isotope pattern could be detected and was consistent in terms of charge state of peptide. For quantification, intensities were determined as the intensity maximum over the retention time profile. Intensities of different isotopic peaks in an isotope pattern were summed up for further analysis. Protein abundance in TGF-β–stimulated cells at each time point was normalized by subtracting log scale values of the corresponding untreated control. For samples at time point 0, we calculated the fold change for each of the four replicates by dividing its abundance by the average abundance for two randomly selected replicates. The normalized abundances of proteins that were quantified in at least 50% of all replicates from six time points were analyzed by ANOVA. Replicates were grouped by time point, and the ANOVA was performed with a 1% permutation–based FDR (121).

For phosphopeptide identification, minimum score and minimum δ score thresholds of 40 and 17 were used with Andromeda, respectively. Phosphopeptide quantification was performed as described (111). The phosphoproteome data were normalized within a biological replicate using z-score transformation of the log2 ratios (time point/control) across SILAC experiments. Phosphosites quantified in at least 50% of all replicates from five time points were analyzed by ANOVA. The MS data were deposited at the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) in the PRIDE partner repository with the data set identifier PXD000496 (see also table S10).

Downstream bioinformatics analyses

Bioinformatic analyses were performed using Perseus in MaxQuant (99). Unsupervised hierarchical clustering was performed on the z score–transformed ratios described above. Proteins were annotated with GO biological process terms (122), KEGG pathway terms (123), kinase substrate motifs from HPRD (67), and BioCarta (76). Statistical analysis of the enrichment of proteins with functional annotations was evaluated using a Fisher’s exact test or Expression Analysis Systematic Explorer (EASE) (124). An enrichment score of greater than 1 indicates enrichment, and a score from 0 to 1 indicates negative enrichment.

For the orthogonal analysis of functional annotations in the proteome data, we used a one-dimensional (1D) annotation test and clustering approach in Perseus that is based on a recently published algorithm (47). The 1D annotation matrix algorithm ranked the normalized protein abundance and tested if proteins corresponding to every annotation term tend to be ranked higher or lower than the ranking of all proteins in the data set. Statistical significance was determined with a 2D version of the nonparametric Mann-Whitney test and a Benjamini-Hochberg FDR threshold of 0.05. For significantly different annotations, we calculated a position score (between −1 and 1) that specified where the mean of the normalized abundances for proteins belonging to each annotation was located relative to the mean of all normalized protein abundances at a particular time point. A position score near 1 indicated that proteins with that annotation were concentrated at the high end of the distribution of normalized protein abundances. Position scores of KEGG biological pathways (123) calculated for each of the six time points were used to create the annotation matrix and analyzed by unsupervised hierarchical clustering.

Upstream regulator analysis was performed using IPA (125) on the proteome data set. The analysis was used to determine the P value associated with the overlapping fraction of known target genes of each transcriptional regulator that was represented among proteins with altered abundance in cells stimulated with TGF-β. In addition, the algorithm compared the known effect (transcriptional activation or repression) of a transcriptional regulator on its target genes to the observed changes in protein abundance to assign an activation z score. We considered proteins with a z score >2 as activated and those with a z score <−2 as inhibited.

Kinase motif analysis

The kinase motif information was obtained from the HPRD (67). A Fisher’s exact test was used to analyze enrichment of selected motifs in TGF-β–regulated phosphosites compared to all phosphosites identified in our MS analysis at a 2% FDR. For cluster-specific enrichment, a similar test was performed using a 4% FDR. The sequence of phosphosites belonging to a particular cluster was analyzed using iceLogo (70).

Protein interaction network modeling

The top 100 proteins (maximum fold change) and 271 TGF-β–regulated phosphosites were submitted to the STRING (81) to identify and filter protein-protein interactions with a confidence score of 0.9. The resulting network consisted of 150 proteins that were highly interconnected. These proteins and their interactions were rendered in Cytoscape (126), and abundance and phosphorylation levels along the temporal axis are depicted. Kinase-substrate interactions from PhosphoSitePlus (56) were mapped onto the existing protein interaction network.

Short hairpin RNA

Lentivirus conditioned media from shRNA constructs [Sigma-Aldrich, MISSION shRNA; shPRSS23#1 TRCN0000047039, shPRSS23#2 TRCN0000047040, shPRSS23#3 TRCN0000047042, nontargeting control (SHC002)] were produced as described (127). Cells were incubated with lentivirus-containing media for 12 hours in the presence of polybrene (5 ng/ml; Sigma-Aldrich). Infected cells were selected with puromycin (1 μg/ml; Sigma-Aldrich) for 48 hours, preincubated overnight in DMEM containing 5% FBS, and treated with TGF-β3 (5 ng/ml; provided by K. K. Iwata, OSI Pharmaceuticals, New York).

Reverse transcription quantitative polymerase chain reaction

Total RNA was isolated with a NucleoSpin RNA II kit (Macherey-Nagel), reverse-transcribed with a RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific), and analyzed with primers for PRSS23 (forward 5′-GAGCCGAAGCCAAATTAGAA-3′, reverse 5′-AGGATGTAGATGCCCACCTG-3′), β-actin (forward 5′-AATGTCGCGGAGGACTTTGATTGC-3′, reverse 5′-AGGATGGCAAGGGACTTCCTGTAA-3′), and PAI1 (forward 5′-CACAAATCAGACGGCAGCACT-3′, reverse 5′-CATCGGGCGTGGTGAACTC-3′). Reactions were carried out using SYBR Green PCR Master Mix (Roche) and analyzed on a StepOne Plus real-time PCR system (Applied Biosystems).

Western blotting

Fractions of the lysates prepared for MS were used for Western blot with the following antibodies: INPP4b (Cell Signaling Technology, #4039), PDCD4 (Cell Signaling Technology, #9535), AMIGO2 (R&D Systems, #MAB20801), CDK17 (Prestige Antibodies, #HPA015325),TMEM2 (Abcam, #ab98348), OCIAD2 (Sigma-Aldrich, #SAB3500119), β-actin (Cell Signaling Technology, #5441), phosphorylated Ser465 and Ser467 of Smad2 (Cell Signaling Technology, #3108), phosphorylated Ser809 and Ser811 of RB (Cell Signaling Technology, #8516), phosphorylated Ser143 of zyxin (Cell Signaling Technology, #8467), phosphorylated Tyr15 of CDK1, CDK2, or CDK3 (Cell Signaling Technology, #9111), and GAPDH (glyceraldehyde-3-phosphate dehydrogenase) (Millipore, #MAB 374). For the PRSS23 knockdown experiments, cells were incubated with DMEM containing 5% FBS for 16 hours; stimulated with TGF-β3 (5 ng/ml; provided by K. K. Iwata, OSI Pharmaceuticals) for 0.75, 6, or 24 hours; lysed [0.5% Triton X-100, 50 mM tris-Cl (pH 7.4), 150 mM NaCl, Complete Protease Inhibitor Cocktail (Roche)]; and processed for Western blots with antibodies against PAI1 (BD Biosciences, #612024), phosphorylated SMAD2, and β-actin.

Confocal microscopy

E-cadherin and F-actin were visualized as described (127). Briefly, cells cultured on coverslips were preincubated with DMEM containing 2% FBS for 16 hours, stimulated with or without TGF-β3 (5 ng/ml; provided by K. K. Iwata, OSI Pharmaceuticals) for 48 hours, fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and stained with an antibody against E-cadherin (BD Biosciences, 610181). F-actin was visualized with rhodamine-phalloidin (Invitrogen, R415). Confocal pictures were taken with a Leica TCS SP8 confocal microscope.

Flow cytometry

Cells were incubated in DMEM containing 5% FBS for 16 hours, stimulated with or without TGF-β3 (5 ng/ml; provided by K. K. Iwata, OSI Pharmaceuticals) for 40 hours, labeled with 20 μM BrdU (Roche) for 2 hours, harvested by trypsin, and fixed in 70% ethanol. Fixed cells were incubated with ribonuclease A (50 μg/ml) for 30 min, incubated in a DNA denaturation buffer (5 M HCl, 0.5% Triton X-100) for 20 min, stained with an antibody against BrdU conjugated to fluorescein isothiocyanate (Boehringer Mannheim) and propidium iodide (Sigma-Aldrich), and analyzed on a BD LSR II flow cytometer (BD Biosciences).

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/7/335/rs5/DC1

Fig. S1. Schematic illustration of the MS experimental design.

Fig. S2. Evaluation of the reproducibility of MS replicates.

Fig. S3. Functional assessment of PRSS23 in TGF-β–induced EMT.

Fig. S4. Heatmap of CDK-specific phosphorylation sites regulated by TGF-β.

Fig. S5. Protein interaction network of proteins with TGF-β–induced changes in abundance or phosphorylation.

Table S1. Protein groups identified in proteome MS analysis.

Table S2. Protein groups quantified from proteome MS analysis.

Table S3. Functional annotation enrichment analysis of proteins with TGF-β–regulated abundance.

Table S4. Cluster-specific functional annotation enrichment analysis of proteins with TGF-β–regulated abundance.

Table S5. 1D annotation enrichment analysis for KEGG pathways and time point–specific quantification of corresponding pathway proteins.

Table S6. Upstream regulator analysis of proteins with TGF-β–regulated abundance.

Table S7. Phosphosites quantified from MS analysis.

Table S8. Functional annotation enrichment analysis of proteins with TGF-β–regulated phosphorylation.

Table S9. Functional annotation enrichment analysis of proteins with TGF-β–regulated abundance or phosphorylation or both.

Table S10. Explanation for proteins and phosphopeptides tables from MaxQuant.

REFERENCES AND NOTES

Acknowledgments: We thank J. Cox, N. Neuhauser, and T. Viturawong at the Max Planck Institute of Biochemistry, Martinsried, for helpful discussions. L. Zhang and T. van Laar at the Department of Molecular Cell Biology, Leiden, assisted with shRNA, immunoblotting, and imaging work. TGF-β3 was provided by K. K. Iwata (OSI Pharmaceuticals). M. Wierer, M. Raeschle, S. Humphrey, H. Schiller, and J. Liu critically commented on the manuscript. Funding: This work was partially supported by the European Union 7th Framework project PROSPECTS (Proteomics Specification in Time and Space, grant HEALTH-F4-2008-201645) at the MPI, Martinsried. Author contributions: R.C.J.D., A.M.K., C.C., P.t.D., M.M., and K.S. designed and planned the experiments. R.C.J.D., K.S., and N.N. prepared and measured samples by MS. R.C.J.D. and K.S. analyzed the data. R.C.J.D., A.M.K., and M.v.D. performed shRNA, RT-qPCR, blotting, and fluorescence-activated cell sorting experiments. R.C.J.D., A.M.K., P.t.D., M.M., and K.S. interpreted the data and wrote the paper. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The MS raw data and associated tables can be downloaded from the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) in the PRIDE partner repository with the data set identifier PXD000496.
View Abstract

Navigate This Article