The doxorubicin-induced cell motility network is under the control of the ceramide activated protein phosphatase 1 alpha

Abstract

We have recently reported that a specific pool of ceramide, located in the plasma membrane, mediated the effects of sublethal doses of the chemotherapeutic compound doxorubicin on enhancing cancer cell migration. We identified neutral sphingomyelinase 2 (nSMase2) as the enzyme responsible to generate this bioactive pool of ceramide. In this work, we explored the role of members of the protein phosphatases 1 family (PP1), and we identified protein phosphatase 1 alpha isoform (PP1 alpha) as the specific PP1 isoform to mediate this phenotype. Using a bioinformatics approach, we build a functional interaction network based on phosphoproteomics data on plasma membrane ceramide. This led to the identification of several ceramide-PP1 alpha downstream substrates. Studies on phospho mutants of ezrin (T567) and Scrib (S1378/S1508) demonstrated that their dephosphorylation is sufficient to enhance cell migration. In summary, we identified a mechanism where reduced doses of doxorubicin result in dysregulation of cytoskeletal proteins and enhanced cell migration. This mechanism could explain the reported effects of doxorubicin worsening cancer metastasis in animal models.

INTRODUCTION

Although classical chemotherapy remains a mainstay in treating many forms of cancer, cancer cells escaping the toxic effects of chemotherapy might develop an aggressive metastatic phenotype due to effects on the cancer cells themselves. Cancer cells surviving chemotherapy, especially doxorubicin, have been reported to stimulate the tumor microenvironment forming pre-metastatic niches in the primary tumor¹. We have recently shown that sublethal doxorubicin increases cell migration directly in cervical cancer cells².

Mechanistically, we have reported that doxorubicin treatment resulted in transcriptional upregulation of neutral sphingomyelinase 2 (nSMase2), which catalyzes the hydrolysis of plasma membrane sphingomyelin to ceramide. We found that ceramide in the plasma membrane was necessary and sufficient to induce loss of cell adhesion and gain in cell migration in HeLa cells². However, the mechanisms by which plasma membrane ceramide mediated the effects of doxorubicin on these biologies was not determined.

Ceramide mediates a growing number of biologies, including cell death, cell cycle arrest, and senescence (a detailed list of ceramide biologies can be found in^{3, 4}). Although a few proteins have been shown to be regulated by direct interaction with ceramide (Ceramide Binding Proteins, CBP, listed in^{5, 6}), there are no direct connections of CBPs to the list of known biologies. Our group has extensively studied the activation of protein phosphatase -1 (PP1s) by ceramide in vitro and cultured cells^7–10. Although doxorubicin-induced plasma membrane ceramide could exert its effects on migration through more than one mechanism, we explored the potential role of the PP1 family of proteins in this context.

The main result from this work finds that of the three isoforms of the PP1 family, the alpha isoform (PP1 alpha) mediated the effects of plasma membrane ceramide (PM-ceramide) in response to chemotherapy treatment. Using a phosphoproteomics approach, we implicated a group of proteins in the cell adhesion-cell migration network under the regulation of plasma membrane ceramide and PP1 alpha. From these proteins, we validated ezrin and scrib proteins and their specific phospho-sites as key for the regulation of doxorubicin on cell adhesion and cell migration.

MATERIAL AND METHODS

Cell Culture

Human cervical carcinoma HeLa cells (ATCC, Manassas, VA, US) were cultured in Dulbecco's Modified Eagle Medium (DMEM) high glucose from Thermo Fisher (Waltham, MS, US) supplemented with 10% of fetal bovine serum (FBS) inactivated by heat (Thermo Fisher, Waltham, MS, US). Authentication of cell line was performed by ATCC Cell Line Authentication Service with 100% match with HeLa in the ATCC reference Database Profile. Mycoplasma was tested on a monthly base. Cells were cultivated in a 37⁰ C, in a 5% CO2 atmosphere, in a humidified incubator.

Recombinant protein expression and purification

Bacterial sphingomyelinase (bSMase) from Bacillus cereus and bacterial ceramidase (pCDase) from Pseudomonas aeruginosa were expressed in Escherichia coli BL21 DE3, and recombinant proteins were purified in a His-Trap affinity column followed by size exclusion chromatography (Superdex 200 HiScale 26/40) using a Äkta FPLC system from Global Life Science Solutions USA LLC, (Marloborough, MA, US) as previously described¹¹.

Reverse-Transcription PCR

HeLa cells were plated in 60 mm diameter cell culture dishes at 350K cells/ dish and allowed to attach for 16h in complete DMEM media, high glucose-10% FBS. Media were changed to serum free media (DMEM, high glucose) 1h prior to treatment, in 5ml/dish. Cells were treated with 1ul of vehicle (control, DMSO) or 1ul with doxorubicin (to reach 0.8uM or 1 uM). Messenger RNA was extracted using RNeasy kit (Qiagen; Venlo, Netherlands). The concentration of RNA was measured with a NanoDrop ND-2000 (ThermoFisher), and 1 microgram was used to synthesize complementary DNA with Superscript III kit (Invitrogen). Real-time PCR was performed using Taqman assay on an ABI 7500 RT-system and following product recommendations. Briefly, reaction mix was prepared in 20ul final volume (10ul of supermix, 4.67ul water, 0.33ul 60x Taqman assay, 5ul template cDNA). PCR reaction was started with 2 minutes at 50C, 10 minutes at 95C and 40 cycles of 15 seconds 95C, 60 seconds at 60C. All samples were analyzed in triplicates. As a reference, actin probes were used. Ct values were converted to mean normalized expression using the delta delta Ct method. Taqman assay probes were purchased from Life Technologies (a brand from Thermo Fisher Scientific, Waltham, Massachusetts, US). Taqman assay numbers were: actin: Hs99999903_m1, nSMase2: Hs00920354_m1. PPP1CA [pp1 alpha]: Hs00267568_m1. PPP1CB [pp1 beta]: Hs01027793_m1. PPP1CC [pp1 gamma]: Hs00160351_m1.

Interference RNA and plasmid DNA transfection

For messenger RNA silencing experiments, HeLA cells were seeded in 100 mm diameter cell culture dishes in complete media (DMEM, high glucose, 10% FBS) for 16h. Fresh complete media were added 1h prior transfection with lipofectamine RNAiMAX (Life Technologies, Chicago, IL, US) containing the corresponding siRNA, and following the manufacturer’s instructions. Fresh media were added 6h after transfection. After 48h, cells were incubated in serum free media and used for siRNA validation by western blot and for experiments. Small interfering DNA for PP1 alpha was sc-SC-36299 (Santa Cruz Biotechnology, Dallas, TX), PP1 beta: SC-36295, and PP1 gamma: sc-36297. Key results for PP1 alpha siRNA were reproduced using another siRNA sequence (Hs_PPP1CA_10, Qiagen; Venlo, Netherlands) and validated results plotted in supplemental Figure S1. For expression experiments, nSMase2-V5, CFP/EGFP-Ezrin, GFP-SCRIBB, dsRed-EPS8, EGFP-PAK2, GFP-CDH3 were transfected using XtremeGENE 9 using the standard protocol from SIGMA (St. Louis, MO). For validation of the knockdown by western blot, the following antibodies were used: Santa Cruz Biotechnologies (Dallas, TX) anti-PP1 alpha goat ref 4G3, anti-PP1 gamma ref SC6109, Cell signaling Technology (Danvers, MA) anti- Phospho-Myosin Light Chain 2 (Thr18/Ser19) Antibody #3674, and Phospho-Ezrin (Thr567)/Radixin (Thr564)/Moesin (Thr558) Antibody #3141 and Ezrin Antibody #3145. SIGMA (St. Louis, MO) Monoclonal Anti-β-Actin antibody produced in mouse.

Single mutagenesis and fluorescent tags

All constructs were cloned in pEGFP (or dsRed for EPS8), plasmids containing a CMV promotor. DNA for human ezrin (EZR) wt and T567A mutant were previously reported. Mutagenesis from Threonine or Serine to Alanine residues was performed using the Quickchange Site-Directed Mutagenesis Kit from Stratagene (La Jolla, CA, US) following the kit instructions. Primers used for mutagenesis were: SCRIB, primer 5’ CCTAAGCGCGTGGCCCTGGTGGGTGCTG 3’ for S1378A and primer 5’ GCAGCCAGGATGAAGGCATTGGAACAGGACGCTCTCC 3’ for S1508A. PKA2 S197A, primer 5’ CGAAATCAATTTACACACGGGCTGTAATTGACCCTGTTCC 3’. EPS8 T317A, primer 5’ CCTGGAGAGGGCGTTTTAGCACTGAGGGCAAAACC 3’ and CDH3 T682A, primer 5’ CGTGGCACCAGCCATCATCCCGGCACCCATGTACCG 3’.

Wound healing assay

The cell culture Wound Healing assay or Scratch assay was adapted from Liang et al.¹² as previously reported². Cell culture 12-well plates were coated with 5ug/ml of human fibronectin-1 for 16h and washed with PBS at room temperature three times. HeLa cells kept in subconfluent culture in DMEM-10% FBS (complete media) were trypsinized and plated at 200K cells/well in the coated 12-well plates. Next day, cells were observed under the microscope to be sure there was a confluent layer of cells. Cells were washed with serum free media, and then DMEM-2% FBS media was added to the cells. Scratches were performed using a new 200 pipette tip for each scratch. Images at time zero were taken using an EVOS XL-Core imaging system using a 20x phase contrast objective. Cells were kept in humidified incubators at 37C, 5% CO2 for 16h, and images of cells were re-taken. Image analysis was performed as previously indicated using Python-OpenCV software (python.org, OpenCV.org) recording scratched and uncovered surface areas². For each condition, 5 independent experiments were used. Results plotting and statistical analysis (ANOVA multi comparison test, and Tukey post hoc test) were performed using Python-NumPy-pandas.

Adhesion assay

Cell-matrix adhesion assay was based on Piccolo protocol¹³ and reported previously². Cells were plated in 10 cm dishes at 1 million cells per each treatment condition in DMEM-high glucose, 10% FBS. Next day, cells were attached, and culture media were changed to DMEM-high glucose without FBS, for 16h, and then treatments were added for the required time. Cells were washed in DMEM-high glucose, FBS-free media and detached in 10mM EDTA in PBS for 20 minutes and collected with the help of a cell lifter. Cells were collected and centrifuged at 600 x g for 5 minutes and resuspended at 100K cells/ ml in DMEM-high glucose and FBS-free. Cell culture 12 well plates or 35 mm confocal dishes were previously coated with fibronectin -1 for 16h at 5ug/ml, and cells were added at 100K/cells/ml and incubated at 37C for 60 minutes. Culture dishes were washed 3 times with PBS, and cells were fixed with warmed 4% paraformaldehyde in PBS (w/v). Cells were permeabilized in 0.1% Triton X-100 (v/v) in PBS and stained using rhodamine-phalloidin and DAPI. Attached cells were visualized under a laser scanning confocal microscope Leica TCS SP8 at 20x magnification. Images were processed for automatic cell counting using Python-Bioformats-OpenCV.

Immunofluorescence staining

Cells seeded in 35 mm diameter uncoated glass 1.5 bottom dishes (MatTek, Ashland MA, US) were fixed with fresh prepared 4% paraformaldehyde (pre-warmed at 40C) for 15 min, followed by 2 washed with phosphate buffered saline (PBS) and permeabilized with 0.1% Triton X-100 in PBS for 20 min. Permeabilized samples were blocked with 2% bovine serum albumin in PBS (w/v) and washed with PBS.

Plasma membrane ceramide quantification

Ceramide levels were quantified in the plasma membrane as previously reported². Briefly, HeLa cells (control or upon treatment) were washed with serum free media and treated with recombinant bacterial ceramidase from Pseudomonas aeruginosa (pCDase) 500mU or PBS (no pCDase) for 10 min. Plasma membrane ceramide was calculated as: (Sph from pCDase ) – (Sph from no pCDase).

LC-MS sphingosine quantification

Organic extraction solvent iso-propanol:water:ethyl acetate (30:10:60; v:v:v) was added directly to the cells, followed by 50 ul of 50pmol/ml C17-sphingosine internal standard (AVANTI Polar Lipids Inc, Alabaster, AL) for mass spectrometry. Lipids were extracted following the protocol described by Bielawski et at¹⁴ and injected into an Agilent LC-MS (Infinity 1260 - 6120b). Column Spectra C8SR 150x3mm, 3um particle size (Peeke Scientific, Redwood City, CA). Injection volume: 5ul; Flow: 0.5ml/min; Mobile phase A: Fisher Water Optima LC/MS, 1mM ammonium formate, 0.2% formic acid; Mobile phase B: Fisher Methanol Optima, 1mM ammonium formate, 0.2% formic acid. Buffer pre-heated 50 °C; Gradient: 0-1 min 80%B, 1-7.0 min 99%B, 7-16 min 99%B. Parameters were set as Peak width 0.02 min, Step size 0.2, Fragmentor 220 V, positive polarity, Drying gas 10L/min, Nebulizer 30 psi, Drying gas temperature 350C and Capillary voltage (Vcap) +5000V. Transition mass for sphingosine and c17-sphingosine were (m/z) 282.2 and 268.0 respectively. Chromatograms were visualized, and area under the chromatographic peaks were integrated using Mass Hunter software (Agilent, Santa Clara, CA)

Proteomic analysis

Experiments were performed with 2 biological replicates with technical repeats of MS analysis. Cells were washed with prewarmed PBS and Hanks balanced salt solution before the addition of SILAC media. HeLa cells were fed with SILAC media for at least 7 doubling passages. Cells fed with ¹³C₆ ¹⁵N₄ L-Argenine + ¹³C₆ L-Lysine were marked as ‘Heavy’, and cells fed with unlabeled amino acids as ‘Light’. Cells growing in ‘Heavy’ and ‘Light’ media were expanded at subconfluence density to get 20 million cells for each group. Two and a half million cells of each group were plated in 15 cm diameter cell culture dishes, and after 24 h cells were treated with siRNA using Lipofectamine RNAiMAX reagent (Life Technologies) according to the manufacturer’s protocol. For negative control RNAi experiments 20nM of ‘AllStars Negative Control siRNA’ (Qiagen; Venlo, Netherlands) was used. For PP1 alpha knockdown the SC-36299 siRNA from SantaCruz Technology (Dallas, TX) was used. Fresh media were added after 16h of transfection, and at 48h cells were treated with PBS (vehicle) or bSMase 10mU/ ml for 5 min. Cells were collected in 2.5% SDS, and protein was precipitated in chloroform and methanol. Precipitated proteins were submitted to the Stony Brook Proteomics Core for bottom-up proteomics analysis. Protein samples were dissolved in 8 M urea, 100 mM ammonium bicarbonate, sonicated for 30 seconds and subjected to reduction with 5 mM DTT and alkylation in 10 mM iodoacetamide. Peptides were diluted to 2 M urea and digested with trypsin at 37 °C overnight. Peptides were desalted on reverse-phase resin (HLB, Waters), lyophilized overnight and resuspended in 0.1% TFA, 50% acetonitrile (ACN), 1M lactic acid and incubated with TiO₂ beads (10um; GL-BioScience) for one hour. Peptides not binding to TiO₂ beads with collected as a non-phosphorylated, total protein fraction, lyophilized and desalted. Peptides bound to TiO₂ were washed with 0.1% TFA, 50% ACN and eluted with 50mM KH₂PO₄ pH 10.5, neutralized with 5% formic acid, 50% ACN, lyophilized and desalted. Phosphopeptides and total peptides also were analyzed by three hour gradient nano reverse phase LC-MS/MS using a Thermo orbital trap Q-Exactive HF. MS/MS spectra were collected on twenty peptides per ~2s cycle using a resolution of 60,000 for MS and 30,000 for MS/MS. Protein abundance and peptide phosphorylation site abundance were established by protein database searching using the ProteomeDiscoverer v2.3, followed by statistical analysis using JMP12. Three missed tryptic cleavages were allowed and posttranslational modifications considered included cysteine derivatization, phosphorylation (S,T,Y), deamidation (N,Q), water loss (S,T), oxidation (M) and SILAC labels (R,K). Database searches used the human UniProt FASTA database (70391 sequences including common contaminants). Raw data from triplicate experiments was filtered for >2 peptide spectra matches (PSM), digital data (Peptides with inappropriately high H:L ratios in the highest 1% of the distribution) were eliminated and only peptides observed in the two experiments were kept. Peptides were considered deferentially phosphorylated if they had an absolute natural log fold-change >1.6 and a false discovery rate-corrected P<0.05.

Databases

Protein-protein interaction network:

Reactome database was download from Reactome.org: ‘Functional interactions (FIs) derived from Reactome, and other pathway and interaction databases‘ version 2019. Only experimentally confirmed interactions were considered, and all in silico ‘predicted’ interactions were eliminated. STRING database was downloaded from string-db.org: ‘9606.protein.links.v11.0.txt’. Low quality interactions (Score <700) were removed). Reactome and String interactome were then merged and used to build a network using the selected phosphoproteomic hits as seeds¹⁵.

Gene Ontology database

(geneontology.org) data were browsed using AmiGO2 and filtered to extract proteins related to cell adhesion and cell migration. The Biological Process (BP) ‘Cell adhesion’ (GO:0007155) was filtered for ‘Homo sapiens’ (2895 annotations) and ‘experimental evidence’ (1025), from which 968 corresponding to protein level and reduced to 574 unique genes. The BP ‘Cell migration’ (GO:0016477) contained 30948 annotations. The list was filtered for ‘Homo sapiens’, ‘experimental evidence’ (1363 annotations) resulting in 737 unique genes.

Software

Fiji (ImageJ) was used to quantify scratch area in wound healing assays. For batch quantification, cell counting in adhesion assay and cell tracking Python/OpenCV was used as previously reported². Python (python.org) modules NumPy, Pandas, Matplotlib and Seaborn were used for data processing and data representation. Cytoscape was used for network representation¹⁶. NetworkX was used to calculate connectivity and filter Network from low connected nodes¹⁵. Bibliographic references were managed using Mendeley desktop v 1.19.4 software (www.mendeley.com).

Editing and data processing for this manuscript have used free open source software whenever possible, including Debian Linux (debian.org), Libreoffice (libreoffice.org), Cytoscape (cytoscape.org), Anaconda (anaconda.com), OpenCV (opencv.org), Bio-formats [OME] (openmicroscopy.org).

RESULTS

Knock-down of PP1 alpha, but not PP1 beta or PP1 gamma, blocks the effects of PM-ceramide on cell adhesion

As mentioned in the Introduction, in our very recent work, we showed that generation of ceramide at the plasma membrane resulted in loss of cell adhesion and increase of cell migration. However, the mechanism by which ceramide mediated these effects was not addressed. Ceramide has been shown to mediate multiple biologies, and many proteins have been shown to mediate ceramide effects. Among them protein phosphatases 1 (PP1) have been shown by our group and others to couple to ceramide in vitro and in vivo. Therefore, we decided to test if PP1 or any or its isoforms were involved in the PM-ceramide dependent phenotype of cell adhesion and cell migration. As previously reported, low doses of bSMase applied within a window of 10 min generates in HeLa cells ceramide localized at the plasma membrane which is not transported to other main membranes, and it is sufficient to initiate signaling causing dephosphorylation of a set of cytoskeletal proteins within the cell adhesion/migration connectivity network^{2, 11}. PP1 alpha, beta, and gamma were individually knocked down in HeLa cells for 48h. Loss of PP1 isoforms was monitored by western blotting (Figure 1A) for PP1 alpha and gamma using specific antibodies for these isoforms. Unfortunately, we could not find antibodies specific for PP1 beta, and all PP1 beta tested antibodies also recognizes the alpha isoform. Previously, we used phospho-myosin light protein (MLC) signal as read out of knock down of PP1 beta, and the same antibody was used in this work to validate PP1 beta knockdown¹⁰. Furthermore, we validated knockdown of mRNA by RT-PCR as well (Figure 1B). In both approaches knock down of each isoform had little effect on the levels of the other phosphatases. At the protein level, knock down of beta isoform showed some increase of the gamma phosphatase. Once the conditions for knock down were established, cells were treated with bSMase for 5 min, and evaluated for cell adhesion on fibronectin 1 (Figure 1C). Cells exposed to bSMase showed a loss in the cell adhesion phenotype as previously reported. Knock down of PP1 beta or gamma did not show any effect on cell adhesion phenotype per se, and their knock down did not modulate the loss of cell adhesion caused by bSMase. On the other hand, knock down of PP1 alpha completely blocked the effect of bSMase on cell adhesion. These results implicate PP1 alpha as the mediator of the action of bSMase/ceramide on cell adhesion.

Figure 1. Effect of knocking down PP1 family members on plasma membrane ceramide-induced loss of cell adhesion.

(A) HeLa cells were treated with siRNA for PP1 alpha, beta, and gamma, and successful knockdown was validated by western blot and (B) RT-PCR. (C) Control HeLa cells (Mock, blue color) and HeLa cells knocked down for PP1 alpha (red color), beta (green), and gamma (yellow) were evaluated for cell adhesion on fibronectin 1 (solid lines) and loss of cell adhesion upon treatment with bSMase (dotted line). Statistics one-way ANOVA Tukey post hoc vs control (** p-value 0.01; *** p-value 0.001).

PP1 alpha mediates the effects on cell adhesion and migration induced by overexpression of nSMase2

Next, it became important to determine if an endogenously induced ceramide at the plasma membrane also launched this PP1 alpha pathway. Neutral sphingomyelinase 2 has been shown to localize to the plasma membrane in epithelial cancer cells^{9, 17, 18}, act on plasma membrane sphingomyelin, respond to signaling events, and recapitulate the effects of bSMase on loss of cell adhesion². The identification of PP1 alpha as the phosphatase responding to PM-ceramide suggested the hypothesis that it would also respond to overexpression of nSMase2 on the effect of cell adhesion. To test that, HeLa cells were transfected with plasmid DNA V5-tagged nSMase2. After 24h expression, plasma membrane localization of nSMase2 was confirmed by immunofluorecence against V5 tag (Figure 2A), and generation of PM-ceramide was evaluated using recombinant pCDase and LC-MS/MS (Figure 2B). In this assay, exogenous pCDase has access only to ceramide in the outer leaflet of the plasma membrane, and in this case pCDase would generate sphingosine. The results in Figure 2B show that pCDase generated a robust sphingosine signal in response to overexpression of nSMase2. Next, control and nSMase-V5 expressing cells were knocked down for PP1 alpha for 48h, and cells were evaluated for loss of cell-matrix adhesion on FN1 (Figure 2C, representative images are shown in 2D). Cells overexpressing nSMase2 showed loss of cell adhesion when compared to control cells. On the other hand, cells knocked down for PP1 alpha did not show any effect compared to control group. However, the loss of cell adhesion induced by overexpression of nSMase2 did not occurred in cells knocked down for PP1 alpha. These results suggest that PP1 alpha mediates the effects of nSMase2 generated ceramide and the signaling events leading to loss of adhesion.

Figure 2. The depletion of PP1 alpha reverted the effects of nSMase2 on cell adhesion and cell migration.

(A) HeLa cells were transfected with V5-nSMase2, and transfection and localization of nSMase2 at the plasma membrane were validated using an anti V5 peptide antibody and visualized at the 488nm - green channel on a confocal microscope. (B) Control HeLa cells and cells expressing V5-nSMase2 were evaluated for generation of PM-ceramide using bacterial ceramidase (pCD) as detailed in Material and Methods. (C) The effects of control HeLa cells and cells knocked down for PP1 alpha were evaluated for the loss of cell adhesion on fibronectin 1 upon nSMase2 overexpression. (D) Representative images of the cell adhesion experiment. (E) The effects of control HeLa cells and cells knocked down for PP1 alpha were evaluated for the gain of cell migration upon overexpression of V5-nSMase2, (F) representative images are shown. Statistics two-way ANOVA Tukey post hoc vs control (** p-value 0.01; *** p-value 0.001. *** p-value 0.0001; * symbol refers comparison within the same group, and # between groups).

Cell-matrix adhesion plays many roles in cellular, tissue, and organ functionality and integrity. At the cellular level, cell adhesion defines cell morphology, and is an essential part of the cell migration mechanism. We previously related ceramide-induced loss of cell adhesion to an increase of cell migration². Since knocking down PP1 had highly significant effects in reverting the loss of cell adhesion, we evaluated if this observation should also be translated to effects on cell migration. HeLa cells expressing nSMase2 resulted in an increase in cell migration (Figure 2E, representative images are shown in 2F). Knockdown of PP1 alpha blocked the increase in cell migration, and it also blocked basal migration in untransfected cells.

All together, these results identify PP1 alpha as a key target to mediate the effects of PM-ceramide on cell adhesion and migration.

PP1 alpha mediates the effects on cell adhesion and migration induced by chemotherapy

The next mechanistic step was to evaluate if PP1 alpha also mediated the loss of adhesion triggered by doxorubicin, which has been shown to induce the expression of nSMase2 in several epithelial cancer cell lines^2,19, and we have already shown that doxorubicin suppresses cell adhesion and increases cell migration in an nSMase2-dependent manner. Here, PP1 alpha was knocked down for 48h followed by treatment with doxorubicin. Induction of nSMase2 was evaluated at mRNA (Figure 3A) and protein level (Figure 3B), as well as cell adhesion (Figure 3C, representative images are shown in Figure 3D). Doxorubicin decreased cell adhesion to fibronectin whereas loss of PP1 alpha prevented this action of doxorubicin.

Figure 3. The depletion of PP1 alpha reverted the effects of doxorubicin on cell adhesion and cell migration.

(A) Induction of nSMase2 in HeLa cells by doxorubicin was validated using RT-PCR and (B) western blotting. (C) The effects of control HeLa cells and cells knocked down for PP1 alpha were evaluated for the loss of cell adhesion on fibronectin 1 upon 0.6uM of doxorubicin treatment. (D) Representative images of the cell adhesion experiment are shown. (E) The effects of control HeLa cells and cells knocked down for PP1 alpha were evaluated for the gain of cell migration upon doxorubicin treatment, and (F) representative images are shown. Statistics two-way ANOVA Tukey post hoc vs control (** p-value 0.01; *** p-value 0.001. *** p-value 0.0001; * symbol refers comparison within the same group, # between groups).

To investigate if PP1 alpha also was involved in cell motility downstream of the dox-nSMase2-PM-ceramide axis, PP1 alpha was knocked down for 48h, and cell migration was evaluated using the scratch wound assay upon 24h exposed to sublethal doses of doxorubicin. Knock down PP1 alpha reduced migration induced by sublethal doses of doxorubicin (Figure 3E, representative images in Figure 3F).

These results strongly suggest that PP1 alpha is a key player that mediates the effects of the doxorubicin-nSMase2-PM-ceramide axis to mediate loss of cell adhesion and increase in cell migration.

Network analysis of phosphoproteomics identifies cell adhesion and cell migration programs that respond to PM-ceramide

The above results identified PP1alpha under PM-ceramide control. This control was necessary to mediate loss of cell adhesion and increased cell migration during external stimulation with doxorubicin. The fact that a specific protein phosphatase was key to mediate these events strongly suggested that protein dephosphorylation played a fundamental role in these biologies, and it encouraged us to search which proteins were dephosphorylated downstream of PP1 alpha regulation. In order to pursue that goal, control cells (wt) and cells knocked down for PP1 alpha were analyzed by phosphoproteomics (Figure 4A, see Methods). Around 8000 phosphopeptides were identified (Figure 4B) from which 562 phosphopeptides were dephosphorylated, 437 phosphopeptides were phosphorylated and 185 peptides contained both phosphorylated and dephosphorylated residues in response to PM-ceramide in the wt sample (Figure 4C). All three groups of proteins (dephosphorylated, phosphorylated and with both changes) were selected to build a protein connectivity network (supplemental Figure S2A), and in order to identify downstream biologies, the network was clustered using the Markov Cluster Algorithm^{20, 21}. The resulting clusters were biologically annotated using the Reactome Pathways and the Gene Ontology Biological Functions. These clusters were: 1. Cell adhesion/migration, 2. MAPK signaling, 3. Wnt signaling, 4. Lipid metabolism, 5. PI3K/Akt signaling , 6. Growth factor receptor signaling , 7. Cell cycle, 8. DNA damage and 9. Others including small dispersed clusters (supplemental Figure S2B).

Figure 4. Construction of the ceramide functional network and dependency on PP1 alpha.

(A) Experimental design. HeLa cells were fed with ‘Heavy’ and ‘Light’ amino acids, as detailed in Material and Methods. Each group, ‘Light’ or ‘Heavy’, was treated with scrambled control siRNA (scr) and siRNA for PP1 alpha (PP1-KD) for 48h. Scrambled siRNA and PP1 alpha knocked down cells in the ‘Light’ group were treated with vehicle (PBS). The ‘Heavy’ group was treated with bSMase for 5 min and collected with 5% SDS. ‘Light’ and ‘Heavy’ lysates from scramble siRNA treated cells were merged in the same tube, and lysates from cells from the ‘Heavy’ group were merged in a second tube and analyzed for phosphoproteomics. (B) Histogram plot showing the distribution of the natural logarithm of the ratio ‘Heavy’ / ‘Light’ for the control grup treated with scramble siRNA (Scr, in blue color) and from the PP1 alpha knocked down group (PP1, orange). (C) Venn diagram of unique proteins containing identified dephosphorylated peptides (light green) and phosphorylated peptides (dark green). The intersection represents unique proteins containing both identified phosphorylated and dephosphorylated residues. (D) Functional network analysis of the proteins dephosphorylated (light green nodes) and phosphorylated (dark green nodes) by plasma membrane ceramide. Proteins with phosphosites regulated by PP1 alpha are shown as nodes with a red ring.

To identify key proteins in cell adhesion and cell migration downstream of PP1 alpha, proteins from cluster #1 “cell adhesion/migration” were compared to the proteomics results where PP1 alpha was knocked down. Proteins dephosphorylated in wt samples but not dephosphorylated in PP1 alpha knocked down samples were classified as PP1 alpha-dependent. Figure 4D depicts a subnetwork where only proteins from cluster #1 are shown, and PP1 alpha-dependent dephosphorylations are highlighted with red borders. This latter subset of dephosphorylated proteins therefore appears to be critical for mediating effects of PM ceramide on cell adhesion and migration through a PP1 alpha-dependent mechanism (since inhibiting or silencing PP1 alpha prevents those cell responses).

Identification and validation of specific proteins and their identified phospho-sites as regulators of protein function

The above results suggested that specific proteins are important for mediating the effects of PM ceramide on cell adhesion and migration, through their dephosphorylation. Therefore, it could be hypothesized that mimicking dephosphorylation on these PP1 alpha dependent phosphosites should reproduce, at least partially, the PM-ceramide phenotype. For this approach, we focused on a subset of proteins that: 1. underwent dephosphorylation upon PM-ceramide treatment, 2. were controlled by PP1 alpha, and 3. are already related to both cell adhesion and cell migration. This analysis is summarized in Table 1 which shows proteins from cluster #1 (column 1). Proteins from column 1 known to regulate cell adhesion and migration according to the Gene Ontology Consortium (http://geneontology.org/, see Methods section) are listed in column 2. Proteins from column 2 regulated by PP1 alpha are shown in column 3. From the last column, proteins EZR, SCRIB, PAK2, EPS8, and CDH3 were selected as a representative sample from this group (highlighted in column 3). Serine/threonine mutations to alanine generate phospho-incompetent versions with minimal structural changes. The majority of the identified proteins are reported as structural proteins whose active form defines the subcellular localization. Thus, the selected proteins were mutated in the identified phospho-amino acid to alanine as (HUGO gene symbols are shown): EZR T567A, SCRIB S1378A+S1508A, PAK2 S197A, EPS8 T317A, and CDH3 T682A, and then they were tagged with a fluorescent protein tag and expressed in HeLa cells. The subcellular localization was visualized using confocal microscopy. The subcellular localization of wt forms corresponded to the localization reported in the literature and in The Human Protein Atlas (proteinatlas.org)^{22, 23}, which validated the functionality of these constructs. However, some of the mutant forms showed differential localization when compared to the wt forms (Figure 5A).

Table 1.

Proteins (shown as HUGO gene symbol nomenclature) responding to PM-ceramide resulting in dephosphorylation were screened for genes with cell adhesion annotation (first column). These were further selected for cell migration (second column) and for genes identified as regulated by PP1 alpha in the proteomics experiment. Genes selected for further experimental validation are highlighted in yellow.

Cluster #1	Cluster #1 - (Adh+ Migr)	Regulated by PP1 alpha
ACTG1	ACTG1	APC
AFDN	AFDN	CDH3
APC	APC	CLASP1
BAIAP2L1	CDH1	DOCK5
CDH1	CLASP1	EPS8
CDH24	CLN3	EZR
CDH3	CSF1	FLNA
CLASP1	CTNNA2	ITGA5
CLN3	DCHS1	MYH9
CORO1A	DOCK5	PAK2
CSF1	DOCK7	RNF20
CTNNA2	EZR	SCRIB
CTNND1	FAM83H	SLC9A3R1
DCHS1	FBXO5	STRAP
DOCK5	FGFR1OP
DOCK7	FLNA
DSCAML1	HMOX1
DSP	IGF1R
EPS8	ITGA11
EZR	ITGA5
FAM83H	JUP
FBXO5	KANK2
FGFR1OP	LIMCH1
FLNA	MAPK1
GSK3B	MAPK14
HMOX1	MISP
HSPD1	MYH9
IGF1R	PAK2
ITGA11	PHLDB2
ITGA5	PKN2
JUP	PKN3
KANK2	PRKCA
LIMCH1	PXN
MAPK1	RNF20
MAPK14	RTN4
MISP	SCRIB
MYH9	SHTN1
MYL12A	SLC9A3R1
PAK2	SLIT2
PAK4	SLK
PDLIM5	SP100
PHLDB2	STK10
PKN2	STRAP
PKN3	TBXA2R
PPP1R12A	TJP1
PRKAR1A	TTBK2
PRKCA	ZMYND8
PXN
RNF20
RPS3
RTN4
SCRIB
SHTN1
SLC9A3R1
SLIT2
SLK
SP100
SPECC1L
STK10
STRAP
STXBP3
TBXA2R
TJP1
TJP2
TMEM8B
TMOD3
TRIOBP
TTBK2
VCL
VWF
ZMYND8

Figure 5. Effects on subcellular localization of PP1 alpha - ceramide phospho-incompetent substrates.

Five proteins were selected to experimentally validate a role in cell adhesion under the regulation of the identified phosphosite. To mimic the effect on ceramide dephosphorylation on the identified phosphosites, identified sites were mutated to alanine (mutant). (A) The subcellular localization of wild type and mutant forms were compared for each protein (Scale bar = 20 micrometers) and (B) quantified as percentage of cells with localization at the plasma membrane (for ezr, scrib, EPS8), with filopodia for PAK2 and localization at cell-cell contacts for CDH3. Statistics T-test p-value * 0.05; **0.01; *** 0.001.

EZR T567 is already a well-known key regulatory site, which allows EZR (and other members of the ERM family) to bind PIP2 in the plasma membrane. The dephosphorylated form remains cytosolic²⁴. The EZR wt and T567A mutant expressed in HeLa confirmed this result (Figure 5B). EZR T567 was the only protein from the selected list with available antibodies against the identified phospho-site, and we used western blot to confirm the proteomics results (supplemental Figure S3). SCRIB wt was found to be distributed into the cytoplasm and plasma membrane whereas the S1378A+S1508A mutant was mainly cytoplasmic losing the plasma membrane localization. PAK2 also showed a dramatic change in localization. The wt form was localized mainly at the plasma membrane and some was cytosolic. Mutant PAK2 S197A induced strong plasma membrane protrusions and was solely localized at the plasma membrane. EPS8 was found to distribute in discrete regions of the plasma membrane and cytoplasm, mutant EPS8 T317A did not altered this pattern. Protein CDH3 wt and its mutant T682A did not show any apparent change in subcellular localization, both wt and mutant were localized at cell-cell interaction contacts. These results suggested that the phospho-status of the identified residues downstream of PP1 alpha play a regulatory role in the function of at least some of these proteins and that PM-ceramide potentially modulates these functions.

Identification of ceramide activated -PP1 alpha substrates key players in cell adhesion and cell migration

In order to evaluate if the phospho-proteomics substrates identified downstream of PP1 alpha are involved in regulating cell adhesion and cell migration, cells expressing wt and mutant forms of these proteins were analyzed for these cell functions.

HeLa cells were transfected with wt and mutant forms for EZR, SCRIB, PAK2, EPS8 and CDH3 for 24h. Similar transfection efficiency among samples was evaluated by flow cytometry. Cells were detached from the culture dish and replated on fibronectin 1-coated dishes for 60 min, and the number of attached cells was quantified using a colorimetric assay. Unfortunately, cells overexpressing PAK2 wt and mutant failed to re-attach, and it was therefore not possible to determine the effects of the mutation. All other HeLa cells expressing constructs succeeded in cell attachment. In cells expressing mutant forms for EZR and SCRIB, cell adhesion was significantly reduced (Figure 6A). Changes in EPS8 and CDH3, the same proteins where mutation to alanine did not change the subcellular localization, also failed in reducing cell adhesion.

Figure 6.

(A) Selected proteins Ezr, Scrib, Eps8, and Cdh3 were tested on cell adhesion on fibronectin-1. Cells attached were quantified using a colorimetric assay and measured at 567 nm. (B) Cell adhesion assay only quantifying transfected cells. Fluorescent protein tags were added as CFP-ezr, GFP-Scrib and dsRed-Eps8. Attached cells were imaged using a confocal microscope and quantified using Open-CV software. Representative images and (C) quantification. (D) Tracks in x/y axis of single-cell migration of cells expressing ezr wt and mutant (E) and box-plot quantification of the total migrated distance. (F) Tracks in x/y axis of single cell migration of cells expressing Eps8 wt and mutant (G) Statistics T-test, p-value * 0.05.

Since EPS8 was not statistically significant, but showed some reduction in adhesion, the assay was also repeated only counting transfected cells, removing the effect of untransfected cells in the colorimetric assay. Fluorescent tagged proteins for EZR, SCRIB and EPS8 were re-analyzed using a fluorescent microscope, and fluorescent cells were counted. Attached cells were automatically counted in 3 independent experiments (See Methods section). Similar results were obtained, and only mutants of EZR and SCRIB showed statistically significant reduction of cell adhesion, (Figure 6B, quantified in 6C).

Previously, we found that loss of cell adhesion caused by PM-ceramide resulted in a gain in cell migration. Moreover, the results in Figures 2 and 3 show that the effects on adhesion and migration were mediated by PP1 alpha phosphatase. To evaluate if the identified phosphosites mediated these effects, we evaluated three of the selected proteins. Cell migration was evaluated for EZR WT and mutant (Figure 6D, quantified in 6E) and SCRIB WT and mutant (Figures 6F and G) as representative for proteins with effect on cell adhesion, and EPS8 (Figure 6H, quantified in 6I) for proteins without effects on cell adhesion. HeLa cells were transfected with the corresponding DNA constructs for 24h, and then they were replated on fibronectin-1-coated dishes and evaluated for single cell migration for 16h. Similar transfection between WT and mutant forms were evaluated by western blot (supplemental Figure S4). Both EZR and SCRIB but not EPS8 mutant increased cell migration. The meandering index³⁸ was also calculated from the single tracks in figure 6D, F, and H. WT and mutant cells from EPS8 and SCRIB did not show any difference in persistence to directional migration, and differences in EZR were not statistically significant (p > 0.05), suggesting the effects were more on random migration rather than on persistence (supplemental Figure S5). Thus, PP1 alpha targets appear to play important roles in regulating adhesion and migration.

DISCUSSION

Altogether, the results from this work suggest that the effects observed on induction of PM-ceramide on cell adhesion and cell migration are caused by activation of PP1 alpha acting on a set of proteins identified using phospho-proteomic mass spectrometry. The bioinformatics approach allowed us to reduce the number of candidates to a selected few which functioned as key regulators of cell adhesion and cell migration.

The results of this work shed light on a mechanism triggered by doxorubicin on cancer cells leading to cell migration and loss of adhesion^{25, 26}. This may be relevant to cancer progression once the induction of cell death has failed. There are many reasons why a cancer cell can escape from the lethal effects of doxorubicin, for example, reduced doses of drug administration to avoid tissue damage, resistance mechanism to chemotherapy, and tissue accessibility²⁷. In our very recent work, we described a new pathway that mediated doxorubicin chemotherapy’s effects in improving pro-metastatic features, such as loss of cell adhesion and gain of cell migration. We identified that nSMase2 and ceramide in the plasma membrane were necessary and sufficient to execute the doxorubicin phenotype. Here, we report that PP1 alpha is a crucial mediator of the nSMase2-ceramide axis for the doxorubicin effects. We identified 14 proteins as downstream substrates of PP1 alpha. From these hits, we evaluated EZR, scrib, pak2, cadherin-3 and EPS8 and validated the role of scrib and ezrin, and their phosphosites, as part of the biological network activated by doxorubicin leading to loss of cell adhesion and gain of cell migration. Under plasma membrane ceramide, PP1 alpha seems to coordinate the function of several targets to achieve the effects on adhesion and migration. Therefore, the loss of some of these targets might not be totally critical to trigger the biology, whereas other phospho-sites seem to play a crucial role in this biology.

The effects of doxorubicin on the tumor tissue appear to be a complex response from the cancer cells and the cancer tissue microenvironment. Defining the molecular mechanism on how cancer cells activate pro-metastatic programs is a must to design strategies to block the metastatic effect of doxorubicin, allowing the lethal effects. We have previously pointed to nSMase2 as a potential target to block the metastatic effect of doxorubicin. Although this target is still valid, and currently being evaluated in vivo, there are other cellular mechanisms to generate ceramide at the plasma membrane, such as secretion of acid sphingomyelinase²⁸, or inverse activity of sphingomyelin synthase 2 (SMS2)²⁹, which could also trigger the same phenotype. These enzymes have also been implicated in cancer progression^{30, 31}, and the mechanisms under their effects are unknown. Blocking the pool of PP1 alpha that responds to ceramide, or targeting key players in the downstream network, might be a more effective or complementary strategy. It is also worth mentioning that even though HeLa cells are an exceptional model for mechanistic studies, they might not be the most physiologically relevant model to expand the studies on the new phenotypes in doxorubicin-treated cancer cells. In that sense, we are currently preparing another work where we apply the knowledge on sublethal doxorubicin, enhancing pro-apoptotic features in breast cancer.

From three PP1 isoforms³², specifically alpha and gamma isoforms have been shown to be activated by ceramide in vitro and in cellulo. Curiously, PP1 beta has not been tested for activation and has not been identified downstream of ceramide in cells. PP1 gamma is the most studied PP1 downstream of ceramide and has been reported to mediate ceramide effects on cell growth arrest³³, insulin signaling³⁴, and cell death by apoptosis³⁵. We previously showed that PP1 alpha mediated effects of artificially generated ceramide on ezrin protein, although no biological context was reported^{10, 11}. In this work, genetic depletion of PP1 beta and gamma found that these isoforms did not mediate the effects of plasma membrane ceramide on the regulation of cell adhesion. However, PP1 alpha was found to be necessary to mediate the effects of plasma membrane ceramide on cell adhesion. These results prove that the relation between PP1 alpha and ceramide is not just circumstantial, but it gives a real physiological context, specifically is a regulatory mechanism on how the cell regulates relevant biologies such as adhesion and migration.

In our previous and present manuscripts, we have used a novel approach to measure precisely ceramide in the plasma membrane. Current techniques used to measure ceramide changes in the response of cell treatment often include measuring ceramide levels in the whole cell lysate or measure ceramide from specific subcellular membranes using cell fractionation in gradient centrifugation and also the extraction of outer-leaflet lipids using dextrans. Total cellular levels can miss ceramide changes in membranes where ceramide is low in abundance. Protocols focusing on specific membranes require many hours to separate membranes or extract lipids, disrupt the membranes, and release enzymes in the lysate. Treatment with pCDase to hydrolyze PM-Ceramide followed by lipid extraction with organic solvents requires 5 min treatment in intact cells. Therefore, this protocol minimizes potential artifacts and concentrates on PM-ceramide measurement, ignoring the other subcellular membranes, which are much more rich in ceramide.

Importantly, we have established a protocol to start dissecting ceramide functions dependent on their subcellular localization. Far from being ceramide a unique entity regulating many biological processes, ceramide might exists as independently regulated pools in different membranes regulating different functions³⁶. In previous work, we already evidenced the existence of independently regulated organelle pools of ceramides and how these can be artificially modulated³⁷. However, the unique relationship between defined pools of ceramide and specific biologies are still not determined. Focusing on the plasma membrane, we have demonstrated that not all ceramides are equal. Interrogation of the plasma membrane led us to discover a small pool of ceramide which functions are not typically attributed to ceramide, and the commonly related ceramide functions were unaffected by this pool.

In summary, we have identified PP1 alpha as a direct mediator of the ceramide signaling in vivo and the specific downstream signaling of this ceramide-PP1 alpha axis leading to an increase in cell migration. This mechanism is key to transmit the doxorubicin pro-metastatic effects in cultured cells.

Abbreviations:

BP

biological process

bSMase

bacterial sphingomyelinase

CBP

Ceramide binding proteins

CC

cellular compartment

GO

gene ontology

MF

molecular function

nSMase

neutral sphingomyelinase

nSMase2

neutral sphingomyelinase2

pCDase

Pseudomonas bacterial ceramidase

PM

Plasma membrane

PP1

protein phosphatase 1

SMase(s)

sphingomyelinase(s)