Proteolytic control of TGF-β co-receptor activity by BMP-1/tolloid-like proteases revealed by quantitative iTRAQ proteomics

Abstract

The metalloproteinase BMP-1 (bone morphogenetic protein-1) plays a major role in the control of extracellular matrix (ECM) assembly and growth factor activation. Most of the growth factors activated by BMP-1 are members of the TGF-β superfamily known to regulate multiple biological processes including embryonic development, wound healing, inflammation and tumor progression. In this study, we used an iTRAQ (isobaric tags for relative and absolute quantification)-based quantitative proteomic approach to reveal the release of proteolytic fragments from the cell surface or the ECM by BMP-1. Thirty-eight extracellular proteins were found in significantly higher or lower amounts in the conditioned medium of HT1080 cells overexpressing BMP-1 and thus, could be considered as candidate substrates. Strikingly, three of these new candidates (betaglycan, CD109 and neuropilin-1) were TGF-β co-receptors, also acting as antagonists when released from the cell surface, and were chosen for further substrate validation. Betaglycan and CD109 proved to be directly cleaved by BMP-1 and the corresponding cleavage sites were extensively characterized using a new mass spectrometry approach. Furthermore, we could show that the ability of betaglycan and CD109 to interact with TGF-β was altered after cleavage by BMP-1, leading to increased and prolonged SMAD2 phosphorylation in BMP-1-overexpressing cells. Betaglycan processing was also observed in primary corneal keratocytes, indicating a general and novel mechanism by which BMP-1 directly affects signaling by controlling TGF-β co-receptor activity. The proteomic data have been submitted to ProteomeXchange with the identifier PXD000786 and doi:10.6019/PXD000786.

Electronic supplementary material

The online version of this article (doi:10.1007/s00018-014-1733-x) contains supplementary material, which is available to authorized users.

Introduction

Bone morphogenetic protein-1 (BMP-1) was originally isolated from a demineralized bone extract for its ability to induce ectopic bone formation [1]. However, it strongly differs from other BMPs in that it is a metalloproteinase and not a transforming growth factor-β (TGF-β)-related protein. BMP-1 actually belongs to a small family of metalloproteinases known as the BMP-1/tolloid-like proteinases (BTPs) which also include, in humans, a longer splice variant named mammalian tolloid (mTLD) and two other isoforms named mammalian tolloid-like 1 and 2 (mTLL-1 and mTLL-2). Since its initial discovery, BMP-1 was shown to be crucial for development, despite redundancies with other BTP isoforms, and BMP-1-null mice die soon after birth from failure of the ventral body wall to close [2]. Conditional ablation of the Bmp1 and Tll1 genes in mice also leads to severe bone defects [3] and mutations in the Bmp1 gene in humans cause osteogenesis imperfecta [4, 5]. Finally, BMP-1 and other BTPs are thought to be important for tissue repair [6] and have been shown to be up-regulated in several types of fibrotic disorders [7–10].

Around 25 substrates have been described for BTPs, most of them being extracellular matrix molecules [11, 12]. They include several fibrillar procollagens (I, II, III, V and XI), small leucine-rich proteoglycans (decorin, biglycan and osteoglycin), basement membrane components (laminin 332, procollagen VII and perlecan), lysyl oxidases (LOX and LOXL) and mineralization factors (dentin matrix protein-1 and dentin sialophosphoprotein). More recently, BTPs have also been widely implicated in the activation of several TGF-β superfamily growth factors (TGF-β1, BMP-2/4, GDF-8 and GDF-11). For TGF-β1 itself, BTPs release the small latent complex (SLC, consisting of TGF-β1 bound non-covalently to its propeptide) by cleavage of LTBP-1 (latent TGF-β binding protein 1) through which SLC is immobilized in the ECM [13]. This proteolytic event triggers further activation of TGF-β1, especially through the action of additional proteases such as MMPs and plasmin, and contributes to increasing the pool of active TGF-β1 in tissues. Interestingly, expression of BTPs is up-regulated in the presence of TGF-β1 [14], thereby creating a positive feedback loop.

TGF-β1, like the two other isoforms TGF-β2 and TGF-β3, is a multifunctional cytokine known to play critical roles in the developing embryo and in adult tissues [15]. Though TGF-βs control multiple aspects of cell phenotype (proliferation, survival, migration and differentiation) with major consequences for tissue homeostasis and remodeling, the pathophysiological processes that seem to be most affected by TGF-βs include wound healing and fibrosis, tumor progression and auto-immune diseases [16]. While TGF-βs signal through type I and type II receptor heterotetramers via SMAD-dependent or -independent pathways [17], several co-receptors also play important roles, leading to potentiation or inhibition of signaling. Among the co-receptors, endoglin and betaglycan (type III class of TGF-β receptors) were discovered first [18–21] but recent work has also pointed to CD109 [22, 23] and neuropilins 1 and 2 [24, 25] as important regulators of TGF-β receptor activity. These co-receptors can be secreted or shed from the cell surface, without losing their ability to bind TGF-β, thereby becoming antagonists [26–30].

In this study, we used iTRAQ™ (isobaric tags for relative and absolute quantification) labeling of tryptic peptides from the conditioned medium of HT1080 cells to reveal the BMP-1-dependent proteolysis of cell-membrane or extracellular matrix proteins, a degradomics approach shown to reliably identify protease-shed proteins and substrates [31]. Among the candidate substrates identified, the TGF-β co-receptors betaglycan, CD109 and neuropilin-1 were selected for further biochemical and functional validation. Betaglycan and CD109 were found to be substrates of BMP-1 and the corresponding cleavage sites were localized at sites previously suggested to be critical for their activity. We further show that BMP-1 cleavage has diverse consequences on the binding of mature TGF-β to its co-receptors and that important changes in SMAD-dependent signaling responses are observed when BMP-1 expression is modulated. This clearly demonstrates a new role for BTPs in the control of TGF-β signaling and stability, which is both independent of and complementary to their previously known function in TGF-β activation.

Materials and methods

Proteins and antibodies

Soluble human CD109 (residues 22-1268; with a C-terminal His-tag) and mouse neuropilin-1 (residues 22-856; with a C-terminal His-tag) were from R & D Systems. Recombinant human TGF-β1 and TGF-β2 produced in mammalian cells were purchased from Peprotech. Soluble forms of rat and human betaglycan with C-terminal His-tags were produced in insect cells, as previously described [32, 33], and despite being wild type at the GAG-attachment sites, were not endowed with GAG chains [33]. Recombinant BMP-1 and mTLD (non-tagged) and mTLL-1 (His-tagged at the C-terminus) were produced in 293-EBNA cells as described [34, 35]. Recombinant human procollagen I heterotrimer was purified as described [36] from Pichia pastoris cells kindly supplied by Fibrogen, Inc (San Francisco, USA). Rat monoclonal anti-BMP-1 (MAB1927) and goat polyclonal anti-betaglycan (AF-242-PB) antibodies were from R & D Systems, mouse monoclonal anti-CD109 antibody (TEA 2/16) from BD Pharmingen, mouse monoclonal anti-SPARC antibody (sc-33645) from Santa-Cruz Biotechnology, rabbit polyclonal anti-Smad2/3 (3102) and anti-Phospho-Smad2 (Ser465/467; 3108) antibodies from Cell Signaling Technology and mouse monoclonal anti-actin antibody (A3854) from Sigma-Aldrich. The rabbit polyclonal antibody LF-41 which detects the C-terminal domain of collagen I (C-propeptide) was a kind gift from L. Fisher (NIH Bethesda, USA). HRP-coupled secondary antibodies were purchased from Cell Signaling Technology or Dako.

Stably transfected HT1080 cells

Human BMP-1 cDNA fused with a 6His-tag at the C-terminus and inserted into the pIRESneo2 vector was a kind gift from W. Stöcker (University of Mainz, Germany). The E94A mutant of BMP-1 in pIRESneo2 was obtained by PCR with the QuikChange site-directed mutagenesis kit (Stratagene). These constructs were then used to transfect HT1080 cells (ATCC CCL-121) in the presence of lipofectamine (Life Technologies). Stable transfectants were selected and maintained in high glucose DMEM (containing l-glutamine and sodium pyruvate) with 10 % fetal bovine serum and 0.5 mg/ml G418 sulfate, unless otherwise stated. All culture products were from PAA laboratories.

Sample preparation for iTRAQ proteomics

Transfected HT1080 cells were seeded in P100 or square 500 cm² culture plates (Corning) at a cell density of 2,000 cells/cm² and grown in high glucose DMEM (containing l-glutamine and sodium pyruvate) with 10 % fetal bovine serum and 0.5 mg/ml G418 sulfate for 4 days. On day 5, cells were washed extensively with PBS and left in serum-free, phenol red-free DMEM containing 50 µg/ml gentamicin. After 2–3 h, the medium was replaced and cells were maintained in this new medium for 9 or 36 h. Then, the conditioned medium was collected, clarified by centrifugation (5 min at 200 g and 15 min at 10,000 g) after addition of 1 mM EDTA and 2 mM PMSF and processed either directly using solid-phase extraction (SPE) columns (see below) or after separation into two equal halves. In the latter case, the first half was concentrated by ultrafiltration using Centricon-Plus 70 units (Millipore, cut-off 10 kDa) to a concentration higher than 2 mg/ml (as determined with the Coomassie plus protein assay reagent, Thermo Scientific). The second half was purified on two C4 and C18 SPE columns (Vydac) connected in series, as previously described [31]. Each column was then separately eluted with 1 mL of 95 % acetonitrile, 0.1 % trifluoroacetic acid (TFA) and eluates were pooled and evaporated in a Speed-Vac to obtain a final concentration of more than 2 mg/ml. Aliquots of 100 µg of each sample were then adjusted to pH 8 with iTRAQ™ dissolution buffer (iTRAQ™ Reagent Multi-plex kit, AB Sciex), reduced with 3.5 mM Tris(2-carboxyethyl)phosphine-hydrochloride (TCEP) for 1 h at 60 °C, alkylated with 6.7 mM methyl methanethiosulphonate for 10 min at room temperature and digested overnight at 37 °C with sequencing grade trypsin (Promega) at a 1:10 trypsin:protein w/w ratio. One tube of iTRAQ label (114–117) was then dissolved in 70 µl of ethanol and added to each sample. After 60 min at room temperature, samples processed with the same concentration protocol (SPE cartridges or ultrafiltration) were pooled to be analyzed as duplex experiments. The mixture was then diluted to a final volume of 1.9 ml with rehydration buffer (Offgel Low Resolution kit, Agilent) before fractionation with the 3,100 Offgel fractionator using a pH 3–10 strip (Agilent). After 36–48 h, 12 fractions/duplex were collected, evaporated using a Speed-Vac and resuspended in 0.1 % TFA before mass spectrometry analysis.

In vitro cleavage assays

All cleavage assays were performed at 37 °C in 40–100 µl of 50 mM HEPES pH 7.4, 0.15 M NaCl, 5 mM CaCl₂ and 0.02 % n-octyl-β-D-glucopyranoside.

Sample preparation for N-terminal sequencing

10 µg of each substrate were incubated with and without BMP-1 in the conditions described above for 8 h for rat soluble betaglycan, 4 h 30 for human soluble betaglycan and overnight for human soluble CD109. The two samples were then either processed for Edman sequencing on PVDF membranes or for iTRAQ-ATOMS, according to a protocol adapted from Doucet et al. [37]. Briefly, for the latter, samples were denatured in 2.5 M guanidine hydrochloride and 0.25 M HEPES pH 8.0 at 65 °C for 10 min, reduced with 1 mM TCEP for 45 min at 65 °C and alkylated with iodoacetamide for 30 min at 65 °C in the dark. After iTRAQ labeling in DMSO (half a tube of label/sample), the two samples (with and without BMP-1) were mixed and precipitated with eight volumes of cold acetone and one volume of cold methanol. The pellet was washed extensively with cold methanol, dried and resuspended in 5 µl of 50 mM NaOH. The pH was decreased with 1.8 M HEPES pH 8.0 and the sample was digested for 2 h at 37 °C with sequencing grade trypsin (Promega; 1:10 trypsin:protein w/w ratio). In the case of human betaglycan, tryptic peptides were subsequently submitted to pepsin (Sigma-Aldrich) digestion at pH < 2 (1:10 w/w ratio).

Edman sequencing

Edman degradation of samples was performed on a Procise-492A sequencer (Applied Biosystems) using standard PVDF cycle conditions. Analysis was done through in-line microbore reversed-phase chromatography (140 C Microgradient System, Applied Biosystems) with UV detection at 269 nm and the 610A software (Applied Biosystems) was used for calculations.

Mass spectrometry

LC–MS/MS experiments were performed on a QSTAR XL mass spectrometer (AB Sciex), equipped with Analyst software version 1.1 and with an Ultimate nanoLC pump (Dionex). A C₁₈ column (Dionex, 75 µm id × 15 cm; 300 Ǻ pore and 5 µm particle size) was coupled in line to the mass spectrometer through a nanospray needle (New Objective). A 120 min linear gradient was applied starting with 100 % mobile phase A (0.1 % formic acid and 5 % acetonitrile in water) and ending with 50 % mobile phase A and 50 % mobile phase B (0.1 % formic acid and 80 % acetonitrile in water) at 300 nl/min. The mass spectrometer was operated in data dependant acquisition mode. For each sample, the mass analyses were conducted on two different mass ranges (400–650 and 650–1,600 Da).

MS data analysis and secondary validation

For iTRAQ proteomics, peak lists (.mgf files) were generated for each MS/MS file with Analyst 1.1 software (Mascot.dll script, default parameters) and used for database searches with MASCOT 2.2.2. Mass spectrometry data were searched against the human Swiss-Prot database (release 2012_01, 40 400 entries including reversed decoy sequences) with a mass tolerance of 0.4 Da for precursor and fragment ions. Cysteine carboxymethylation was set as a fixed modification; methionine oxidation, N-terminal acetylation, glutamine and asparagine deamidation and iTRAQ labeling of the N-terminus and lysines were fixed as variable modifications. One miscleavage was allowed and trypsin or semi-trypsin cleavage specificity was selected for iTRAQ proteomics or iTRAQ-ATOMS, respectively. A secondary peptide and protein validation was achieved with the Trans-Proteomic Pipeline (TPP) version 4.4 [38, 39]. Briefly, MS/MS data (.wiff) were converted to mzXML files with mzWiff software (default parameters). Mascot results files (.dat) were converted to pepXML files and combined using the XInteract, PeptideProphet and ProteinProphet tools. Quantitation was achieved using the LIBRA tool of the TPP (default parameters). The final protein lists were compiled using ProteinProphet with a probability of 0.8 resulting in low error rates of 2 %. For iTRAQ proteomics, log₂(BMP-1:E94A) was calculated for each protein, normalized and the standard deviation of the corresponding normal distributions of ratios was used for the determination of the cut-off of the experiment (=2 × standard deviation, corresponding to a p value <0.05). Mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository [40] with the data set identifier PXD000786 and doi: 10.6019/PXD000786.

Mass spectra corresponding to iTRAQ-ATOMS experiments were analyzed with ProteinPilot 3.0 and peptides with iTRAQ labeling on their N-terminus were manually validated.

Cell extracts and Western blots

HT1080 cell supernatants were prepared as described above for proteomic analysis (except that they were collected after 15–16 h and that complete protease inhibitor cocktail (Roche) was added instead of EDTA and PMSF) then concentrated by ultrafiltration. For extraction of the cell layer, cells were washed with PBS, scraped with cold RIPA buffer (50 mM Tris pH 8, 0.15 M NaCl, 1 % v/v NP-40, 0.5 % w/v sodium deoxycholate, 0.1 % w/v sodium dodecyl sulfate) containing protease inhibitors (Complete, Roche) and centrifuged at 15,000 g for 30 min to remove cell debris. To harvest the ECM, the cells were detached by a 5-min treatment with 20 mM NH₄OH and the underlying ECM was washed extensively with water before being scraped with 0.325 M Tris–HCl pH 6.9, 25 % glycerol, 10 % sodium dodecyl sulfate at 100 °C. Protein concentrations in cell extracts (except ECM) were determined using the Bradford assay (Coomassie plus protein assay reagent, Thermo scientific). For Western blots, equal amounts of total protein (medium and cell layer extracts) or equal volumes (ECM) were loaded onto the gels.

Isolation and cultures of human primary keratocytes

Keratocytes were isolated from human corneas, harvested in accordance with ethical regulations, by the Cornea Bank at the Hospices Civils de Lyon, as previously described [41]. The cells used in this study were at passage 6. They were cultured in a medium previously shown to maintain the keratocyte phenotype [41] and composed of DMEM/Ham F12 (1:1; PAA laboratories), 10 % iron-supplemented calf serum (Hyclone), 5 ng/mL bFGF (Sigma-Aldrich) and 1 % antibiotic/antimycotic solution (PAA laboratories). At confluency, cells were washed with PBS and placed for 48 h in phenol-red free DMEM containing bFGF and antibiotic/antimycotic solution only. Proteolysis in these cells was studied in the absence or presence of 10 µM of BTP inhibitor described below. Medium was collected and analyzed by Western blot as described for HT1080 cells.

BTP inhibitor

The BTP-specific inhibitor used in cell cultures of HT1080 cells and keratocytes was the N-substituted aryl sulfonamide hydroxamate 90, described in [42]. It was synthesized as reported therein and dissolved in DMSO before addition to the culture medium. This hydroxamate compound was previously shown to efficiently inhibit BMP-1 with an IC₅₀ of 30 nM [42].

Surface plasmon resonance

SPR experiments were performed using a Biacore T100 (GE Healthcare) at the Protein Production and Analysis facility of the UMS3444 (Lyon). Prior to immobilization, recombinant CD109 was incubated alone or in the presence of BMP-1 (1:1 molar ratio to obtain 50–60 % cleavage) for 7 h at 37 °C. In the latter case, BMP-1 was removed by a 30 min incubation with Protein A-Sepharose 4B (Invitrogen) loaded with an antibody directed against the N-terminus of BMP-1 [43] and the flow-through was used for immobilization. Intact and cleaved CD109 were immobilized on channels two and four of a CM5 sensorchip using amine coupling chemistry (2,686 and 2,559 RU, respectively) while channels 1 and 3 were submitted to the same activation/deactivation procedure (omitting protein) and served as control channels. Sensorgrams were recorded at 25 °C using 10 mM HEPES pH 7.4, 0.15 M NaCl, 0.05 % P-20 as running buffer. TGF-β1 was injected at 30 µl/min for 90 s and regeneration was achieved with 2 M guanidinium chloride. Model fitting was carried out using BiaEvaluation software version 4.1.

Detection of TGF-β activity with luciferase reporter assay

HT1080-E94A or -BMP-1 cells at 70 % confluency were co-transfected with the pGL3-basic reporter plasmid (Promega) encoding firefly luciferase under the control of 9 repeats of a SMAD-binding element ((CAGA)₉-Luc; [44] ) and the phRLCMV vector (Promega) encoding Renilla luciferase under the control of the cytomegalovirus ubiquitous promoter to determine the transfection efficiency and to normalize luciferase values. The following day, transfected cells were stimulated with recombinant TGF-β1 or TGF-β2 at the indicated concentration for 24 h. Proteins were extracted from HT-1080 cells in Passive Lysis Buffer (Promega) and transcriptional responses were determined as described previously [45] with the Dual-Luciferase Reporter Assay System (Promega).

To compare the ability of intact and cleaved betaglycan to block the activity of TGF-β2, 150 µg of recombinant rat soluble betaglycan was first incubated with 26 µg of BMP-1 at 37 °C in 700 µL total volume. Control incubations were also run with soluble betaglycan alone (150 µg) or BMP-1 alone (26 µg). After 4 h, the reaction was stopped with 25 mM EDTA and the buffer was exchanged for PBS with Zeba™ Spin desalting columns (7 K MWCO, 2 mL; Thermo Scientific). Measurement of TGF-β2 neutralizing activity was performed as previously reported [33]. Briefly, Mv1Lu cells were transfected with TGF-β reporter plasmid pCAGA₁₂-Luc [44] and seeded at 50,000 cells/well in 24-well plates 24 h after transfection. The following day, BMP-1/betaglycan samples prepared as above were added at various dilutions, together with 20 pM TGF-β2, and after another 24 h, cell lysates were prepared and assayed for luciferase activity.

¹²⁵I- TGF-β affinity labeling of membrane-bound betaglycan

TGF-β iodination and affinity labeling were performed as described [46]. Briefly, HT1080 cell monolayers were grown until 80–90 % confluency and washed with cold PBS. Iodinated TGF-β was added at 100 pM in Krebs–Ringer HEPES solution containing 0.5 % bovine serum albumin (BSA) and gently rocked for 3–4 h at 4 °C. After washing the unbound ligand, the dissuccimidyl suberate cross-linker (Pierce Chemicals) was added and reacted for 15 min. The excess of cross-linker was quenched with 10 mM Tris–HCl pH 7.4, 1 mM EDTA and 8.6 % sucrose and cells were scraped from the dish and pelleted in the same sucrose Tris–EDTA solution. Labeled cells were lysed in 200–400 µl of 1 % Triton X-100, 10 mM Tris–HCl pH 7.4 containing a cocktail of protease inhibitors [46]. After clearing cell debris by centrifugation, total radioactivity was counted in a Cobra II D5002 gamma counter (Packard Instruments). Then, the samples were used for immunoprecipitation with anti-rat betaglycan polyclonal antibody # 822 [47] and counted again.

Detection of TGF-β activity by monitoring SMAD phosphorylation

HT1080-E94A and HT1080-BMP-1 cells were seeded in a 6-well plate (400,000 cells/well) and treated with recombinant human TGF-β1 or TGF-β2 (50 pM) or vehicle (0.1 % w/v BSA, 4 mM HCl) for 1–48 h. Cells were washed twice with PBS and scraped in the presence of lysis buffer (150 mM NaCl, 1 % Triton X-100, 20 mM Tris–HCl pH 7.4) containing protease inhibitors (1 mM PMSF, 1 μg/mL aprotinin) and phosphatase inhibitors (1 mM Na₃VO₄). Lysates were incubated on ice for 10 min, then centrifuged at 13,000 g for 10 min. Proteins contained in supernatants were quantified using the Bradford Reagent (Sigma-Aldrich) and analyzed by Western blotting (20 µg/lane), using anti-SMAD2/3, anti-PhosphoSMAD2 and anti-actin antibodies. Relative protein levels were quantified using the densitometric software of Image J.

Results

iTRAQ quantitative proteomics

Use of iTRAQ labeling permits the comparison of multiple proteomes, each labeled with specific isobaric mass tags, and the ratio of the intensities between the different reporter ions can be used to measure increases or decreases in the amount of the corresponding peptides [48, 49]. Here, we chose to analyze the conditioned medium of HT1080 cells as a model system to identify fragments released from both cell layer and ECM upon BMP-1 cleavage of parent molecules (Fig. 1a). The HT1080 fibrosarcoma cell line was selected because it does not produce any of the fibrillar procollagens cleaved by BMP-1 [2, 11, 12], which are usually abundant in the medium of fibroblast-like cells, and seemed appropriate to reveal novel substrates present in smaller quantities. To achieve this, HT1080 cells were stably transfected with constructs encoding either wild-type BMP-1 (HT1080-BMP-1) or an inactive form of BMP-1 where the glutamate E94 required for catalysis had been mutated to alanine ([43]; HT1080-E94A), each fused with a C-terminal His-tag. As shown in Fig. 1b, equivalent amounts of protein were detected by Western blot in the supernatant of both cell types, with relatively large amounts of BMP-1 or mutant also present in the ECM. Interestingly, in the cell medium but not the ECM, BMP-1 seems to partially process its own C-terminus, resulting in a product migrating slightly faster by SDS-PAGE that is not detected with an anti-His antibody, while the inactive E94A mutant appeared as a single band (Fig. 1b).

Fig. 1.

Proteomic analysis of HT1080-conditioned medium. a Quantitative proteomics is used here to reveal the release of protein fragments from the extracellular matrix or the cell surface but indirect effects leading to up- or down-regulation of some proteins in cell medium can also be observed. b BMP-1 immuno-detection in the extracellular matrix (ECM) and conditioned medium of non-transfected HT1080 cells (−), HT1080-E94A cells (*) and HT1080-BMP-1 cells (+). c Main steps of the protocol. HT1080-E94A and HT1080-BMP-1 cells are grown for 4 days and placed in serum-free medium for 9 or 36 h. After clarification by centrifugation, the collected conditioned medium is concentrated using two C4/C18 SPE cartridges connected in tandem or ultrafiltration. Proteins in both concentrated samples are then reduced, alkylated and digested with trypsin. After isotopic labeling with iTRAQ, samples are mixed, fractionated on the basis of their isoelectric points and analyzed by LC–MS/MS. Peptide identification is performed using the MASCOT search engine and secondary validation and relative quantification at the peptide and protein levels using the Trans-Proteomic Pipeline

Both types of HT1080 cells were grown for 4 days then incubated in serum-free medium for 9 or 36 h. Sample preparation for iTRAQ labeling and LC–MS/MS analysis was then optimized and resulted in the protocol described in Fig. 1c and materials and methods. A critical step was the concentration of conditioned medium, for which two different protocols were tested. In the first experiment (Exp 1), cell supernatants were concentrated on two solid-phase extraction (SPE) columns (C₄ and C₁₈) connected in series [31] while in the second and third experiments (Exp 2 and Exp 3), cell supernatants were split and concentrated separately, either using SPE columns or ultrafiltration. As shown in Supplemental Fig. S1a, parallel use of different concentration procedures led to a significant increase in the number of proteins identified.

Candidate substrates

Distributions of iTRAQ ratios (BMP-1:E94A) were fitted with normal Gaussian distributions to establish the statistical cut-offs used to identify candidate substrates or differentially expressed proteins (±2 × standard deviation; Supplemental Fig. S2). For the 757 unique proteins identified in the conditioned medium of the HT1080 cells (Supplemental Table S1; Supplemental Fig. S1b), we considered that they were potential BMP-1 targets if they had one iTRAQ ratio above or below the cut-offs, in at least one experiment. Note that potential substrates may have a high iTRAQ ratio if a fragment is released from the cell layer into the medium or a low iTRAQ ratio if the released fragment is more susceptible to degradation than the intact protein, captured by a receptor or rapidly internalized (Fig. 1a). Table 1 shows the 38 extracellular and cell surface proteins that fitted the criteria defined above. This list includes BMP-1 itself, confirming that auto-proteolysis can occur, and the laminin γ2 chain, which is already known to be cleaved by BTPs [50]. We also note that three other previously described BTP substrates (LTBP-1, perlecan and apolipoprotein A1) were found in the conditioned medium of HT1080 cells (Supplemental Table S1) but that their iTRAQ ratios were not significantly different from 1. This suggests that they were not cleaved in the conditions of this specific cell context or that the cleavage of these substrates did not lead to significant changes in protein amounts in the conditioned medium of these cells.

Table 1.

Extracellular proteins with high or low iTRAQ ratios

Proteins (UniProtKB accession numbers)	BMP-1:E94A iTRAQ ratios
	Exp 1 (36 h)	Exp 2 (36 h)		Exp 3 (9 h)
Membrane-bound proteins	SPE	SPE	Ultra	SPE	Ultra
ADAM 9 (Q13443)	0.23a (1)	–	–	0.90 (3)	–
Cadherin 2 (P19022)	0.21a (3)	0.83 (3)	0.88 (1)	0.85 (14)	0.91 (4)
Cadherin 13 (P55290)	0.24a (3)	–	–	0.94 (1)	–
Cadherin 11 (P55287)	1.18 (4)	–	–	0.90 (2)	0.57a (2)
CD44 (P16070)	0.27a (7)	0.58a (11)	0.69 (12)	0.94 (10)	0.98 (22)
CD109 (Q6YHK3)	–	–	–	2.09a (4)	1.31 (2)
Dystroglycan (Q14118)	0.27a (2)	0.89 (5)	0.78 (1)	0.76 (15)	0.89 (9)
Glypican (P35052)	–	–	0.84 (4)	3.52a (1)	1.31 (4)
Major prion protein (P04156)	0.17a (2)	–	–	–	–
MHC class I antigen B35 (P30685)	–	–	–	0.37a (1)	–
Neuropilin 1 (O14786)	0.47 (2)	–	1.03 (1)	0.59a (2)	1.05 (11)
Semaphorin 7A (O75326)	0.45 (2)	0.85 (4)	0.78 (6)	0.94 (6)	0.54a (1)
TNF receptor superfamily member 12A (Q9NP84)	0.11a (1)	0.71 (2)	–	0.76 (1)	–
Betaglycan (Q03167)	2.28 (2)	–	–	4.84a (5)	2.92a (2)
Secreted proteins	SPE	SPE	Ultra	SPE	Ultra
Bone morphogenetic protein-1 (P13497)	0.30a (10)	0.80 (17)	0.88 (11)	0.64 (22)	0.74 (31)
Calreticulin (P27797)	1.32 (3)	0.58a (5)	0.65(6)	1.02 (6)	0.74 (6)
Calumenin (O43852)	0.22a (7)	0.65 (10)	0.65 (14)	0.73 (16)	0.49a (11)
Carboxypeptidase A4 (Q9UI42)	–	–	0.57a (2)	1.25 (23)	0.98 (21)
C-type lectin domain family 11 member A (Q9Y240)	0.30a (10)	0.76 (11)	0.92 (16)	0.74 (15)	0.86 (27)
CYR61 (O00622)	–	2.19a (1)	2.95a (1)	–	–
Dermcidin (P81605)	–	–	–	0.43a (1)	–
Endothelial protein C receptor (Q9UNN8)	–	–	–	0.43a (1)	1.33 (2)
Exostosin-like 2 (Q9UBQ6)	–	–	–	0.73 (3)	0.58a(2)
Fetuin A (P02765)	0.58 (2)	0.63 (6)	1.11 (4)	1.87a (10)	1.73a(10)
Fibronectin (P02751)	3.72a (18)	0.90 (6)	0.81 (6)	2.18a (30)	1.37 (36)
Fibulin-3 (Q12805)	0.87 (1)	–	–	0.76 (11)	0.55a (9)
γ-glutamyl hydrolase (Q92820)	0.29a (1)	0.87 (2)	0.61 (1)	0.92 (7)	0.46a (1)
Insulin-like growth factor-binding protein 7 (Q16270)	0.50 (5)	0.58a (16)	0.43a (11)	0.71 (37)	0.68 (33)
Laminin γ2 (Q13753)	–	0.74 (1)	0.57a (1)	–	0.83 (5)
β2-microglobulin (P61769)	0.29a (18)	0.90 (12)	0.85 (13)	0.74 (20)	0.91 (33)
Serine protease 23 (O95084)	0.23a (5)	0.71 (6)	0.94 (3)	0.78 (8)	1.10 (4)
Serine protease HTRA1 (Q92743)	3.35a (1)	0.58a (1)	0.75 (1)	1.41 (1)	–
Suprabasin (Q6UWP8)	0.17a (1)	0.53a (2)	0.92 (2)	0.76 (16)	–
TIMP-2 (P16035)	0.53 (6)	0.54a (8)	1.08 (4)	0.63 (15)	1.07 (3)
Tissue-type plasminogen activator (P00750)	–	–	0.50a (1)	–	–
Transcobalamin-2 (P20062)	–	–	1.00 (1)	1.06 (2)	0.42a (1)
Trypsin-2 (P07478)	0.24a (2)	–	–	–	–
Tryptase 1 (Q15661)	–	–	–	0.56a (1)	–

Proteins were selected when at least one of their iTRAQ ratios, after normalization, was below or above the cut-offs of the experiment (^a; Supplemental Fig. S2). Numbers in brackets correspond to the number of iTRAQ-labeled peptides identified and used for quantification. TGF-β co-receptors are highlighted in grey

– not detected

New candidate substrates listed in Table 1 included adhesion/matrix molecules and growth factors, in agreement with known BMP-1 functions, together with novel types of substrates such as proteases and cell receptors. Most of these new candidate substrates had decreased ratios in the presence of catalytically active BMP-1 and only 18 % of the proteins (7/38) had at least one iTRAQ ratio significantly higher than 1. In addition, most of these proteins with high ratios can be linked to TGF-β activity as co-receptors (betaglycan and CD109), a negative regulator (serine protease HTRA1) or target genes (fibronectin, CYR61). Since BMP-1 already has a demonstrated activity in the control of TGF-β1 activity through the cleavage of LTBP-1 [13], we decided to start substrate validation with the study of the TGF-β co-receptors: CD109, betaglycan and neuropilin-1 (which also appears in Table 1 albeit with iTRAQ ratios below 1). Noteworthy, almost all the peptides identified for each of these proteins were located in the N-terminal/distal region of the ectodomains for betaglycan (endoglin-like domain) and neuropilin-1 (CUB and coagulation factor 5/8 domains) and in the 180 kDa N-terminal fragment generated after furin maturation for CD109 (Supplemental Table S2).

Analysis of CD109 and neuropilin-1 cleavage by BMP-1

As a first approach to study the cleavage of CD109 by BMP-1, we analyzed the cell layer and conditioned medium of HT1080-E94A and -BMP-1 cells and searched for fragments from this protein. CD109 is known to be synthesized as a 205 kDa precursor form that is processed in the Golgi apparatus by furin leading either to a GPI-anchored heterodimer or to a secreted 180 kDa protein [27]. The expected mature 180 kDa form of CD109 was observed with cells expressing both inactive and active BMP-1 but, in agreement with iTRAQ degradomics analyses, an additional product around 120 kDa was also present in the cell medium, in relatively higher amounts, with HT1080-BMP-1 cells (Fig. 2a). To further demonstrate that BMP-1 could be directly responsible for the observed cleavage product, we incubated the HT1080-BMP-1 cells with a specific inhibitor of BTPs [42] and observed that the level of CD109 processing in the culture medium was significantly reduced (Fig. 2b). In contrast, the amount of CD109 cleavage product did not seemed to be modified, in the presence of the inhibitor, in non-transfected HT1080 cells and in HT1080-E94A cells. This result suggests that BMP-1 is directly involved in CD109 cleavage and that the residual amount of cleavage product observed in cells that do not overexpress active BMP-1 is due to background proteolysis by a protease unrelated to BTPs. Interestingly also, the total amount of CD109 in the medium of HT1080-BMP-1 cells seems to be reduced upon inhibitor treatment to a level similar to that observed in control cells, suggesting an indirect effect of BMP-1 on CD109 expression.

Fig. 2.

Analysis of CD109 cleavage by BMP-1. a CD109 immuno-detection in the conditioned medium and cell layer of HT1080-E94A cells (*) and HT1080-BMP-1 cells (+) and after incubation of human soluble CD109 (rec CD109; 236 nM) with (+) or without (−) BMP-1 (236 nM) for 8 h. Detection was by Western blot with anti-CD109 antibody after SDS-PAGE (8 % (left) or 4–20 % (right) acrylamide, reducing conditions). Arrowheads indicate full-length mature proteins and arrows BMP-1-dependent cleavage products. An additional cleavage product, only observed when rec CD109 is incubated with BMP-1, is indicated by #. b CD109 immuno-detection in the conditioned medium of non-transfected HT1080 cells (NT), HT1080-E94A cells (E94A) and HT1080-BMP-1 cells (BMP-1) treated for 16 h with 2.5 µM of BTP inhibitor (+) or with vehicle (DMSO; −). Same conditions as in a for detection. SPARC was used as a loading control. c Human CD109 (rec CD109; 536 nM) was incubated at 37 °C for 8 h with or without BMP-1 in a 10:1 (+) or 1:1 (++) substrate:protease molar ratio (8 % acrylamide, reducing conditions). Detection was with Coomassie blue (upper panel) or Western blot with anti-His antibody (lower panel). Closed arrows indicate the cleavage products. d Schematic view of the truncated His-tagged soluble form of human CD109 (rec CD109) used in in vitro studies. BMP-1 cleavage site is indicated with dotted lines and corresponding fragments numbered as in a and b. TED thioester domain. e SPR analysis of the binding of TGF-β1 to intact and cleaved recombinant CD109. A single concentration (15.6 nM; left panel) or increasing concentrations of TGF-β1 (4–250 nM; twofold serial dilutions; middle and right panels) was (were) injected on similar levels of immobilized intact and half-cleaved CD109 (2,686 and 2,559 RU respectively). Recorded sensorgrams are shown as solid lines and best fits obtained with the “heterogenous ligand” model as dotted lines (χ ² = 1.2 for intact CD109 and 0.95 for cleaved CD109)

Then, human soluble recombinant CD109 was incubated in vitro with recombinant human BMP-1. By Western blot, the fragment previously observed in the BMP-1 cell medium was also detected, confirming the direct cleavage of CD109 by BMP-1 (Fig. 2a). This cleavage, leading to a 120 kDa product, could also be observed by SDS-PAGE with Coomassie blue staining (fragment 1, Fig. 2c, d). A second cleavage product (fragment 2), which was not detected with the anti-CD109 monoclonal antibody, was found to migrate very close to BMP-1 and was more easily detected with an anti-His antibody (Fig. 2c). A further fragment, at ~150 kDa, was also detected in vitro (indicated by# in Fig. 2; not seen in cell extracts), showing that partial cleavage of recombinant soluble CD109 towards the N-terminus of the protein can also occur. This additional cleavage site might be due to small differences in protein folding and/or post-translational modifications between the native CD109 produced by HT1080 cells and the commercial recombinant form produced in the murine myeloma cell line NSO.

To identify the main cleavage site in CD109, rather than Edman sequencing, we adapted the ATOMS method recently described [37]. ATOMS exploits mass spectrometry to allow the identification of one or multiple cleavage site(s) in one single experiment, even when starting with low amounts of complex proteins. Here, N-termini of proteins incubated with BMP-1 or buffer were labeled with iTRAQ reagents rather than by isotopic dimethylation, as used previously, and analyzed by LC–MS/MS after trypsin digestion. Using the ATOMS technique, we found that the main cleavage in CD109 occurred between glutamate 666 and glutamate 667 (Table 2; Supplemental Fig. S4a and Table S3) in the putative “bait region” of CD109 (Fig. 2d), defined by analogy with the homologous broad-spectrum protease inhibitor α₂-macroglobulin [51].

Table 2.

Cleavage sites deduced from peptides identified by classical Edman sequencing and by iTRAQ-ATOMS

	Edman sequencing	iTRAQ-ATOMS sequencing
		Peptide sequence	BMP-1:buffer iTRAQ ratio (# of peptide identifications)
CD109	nd	22VAPGPRa	–
		667 ENEGHIVDIHDFSLGSSPHVR	6.3 (2)
Rat betaglycan	24GPEPSTa	24GPEPSTRa	1.3 (6)
	366 DEEVHXI	366 DEEVHTIPPELR	10.5 (2)
		554 DGDEGETAPLSR	44.9 (6)
		556 DEGETAPLSR	34.1 (6)
Human betaglycan	19TAGPXP	19TAGPEPGALCELSPVSASHPVQALMESFTVLSGCASR	1.3 (4)
	21GPEPGAa	21GPEPGALCELSPVSASHPVQALMESFTVLSGCASRa	1.1 (13)
		364 DEEVHT b IPPELR	14.6 (6)
		379 DPGALPALQNPPIR	16.8 (16)
		553 DMDEGDASL	26.0 (1)

Note that only peptides with labeled N-termini are reported for iTRAQ-ATOMS (see Supplemental Table S3 for complete iTRAQ-ATOMS datasets and Supplemental Fig. S4–S8 for mass spectra)

Peptides defining cleavage sites with iTRAQ ratios >5 are underlined

^aIndicates the expected mature N-terminal peptide of the protein (based on UniProtKB annotations); note that alternative signal peptide processing is observed for human betaglycan (the GPEPGA sequence being 10-fold more abundant by Edman sequencing than the TAGPEP sequence)

^bIndicates O-glycosylation of the threonine

nd not determined; – not found

In contrast, when the recombinant soluble form of neuropilin-1 was incubated with BMP-1, even at high enzyme:substrate ratios, there was no evidence of proteolysis (Supplemental Fig. S3). This result suggests that the low iTRAQ ratio observed for neuropilin-1 in proteomic studies was due to indirect effects of BMP-1, leading to changes in protein expression. Because of this, neuropilin-1 was not analyzed further.

Effect of CD109 cleavage on its ability to bind TGF-β1

We then used surface plasmon resonance to determine how cleavage of CD109 by BMP-1 affects its interaction with TGF-β1. To this end, CD109 was first incubated alone or with BMP-1 (to obtain 50-60 % conversion into cleavage products) and then, after removal of BMP-1, covalently immobilized at similar levels on a sensorchip. When the same TGF-β1 preparation was injected on the two surfaces, sensorgrams had the same overall shape but the recorded signal was 30 % higher on cleaved CD109 than on intact CD109 (Fig. 2e). A more quantitative analysis with serial dilutions of 4–250 nM TGF-β1 further revealed that binding curves could be fitted with the “heterogenous ligand” model yielding two K_Ds of 25 and 250 nM with intact CD109. In agreement with the preliminary injections, the dissociation constants were found to be decreased to 13 and 143 nM with cleaved CD109, indicating that TGF-β1 binds CD109 with a slightly higher affinity when the latter is cleaved.

Analysis of betaglycan cleavage by BMP-1

Of all the potential BMP-1 substrates identified (Table 1), betaglycan consistently gave the highest iTRAQ ratios. When examined biochemically, Western blotting showed that the conditioned medium of control (HT1080-E94A) cells contained the shed form of betaglycan [30, 52], which appeared as a band above 250 kDa with a broad leading smear (Fig. 3a, left panel), probably due to glycosaminoglycan (GAG) chains [29], and as another band around 55–60 kDa. In contrast, the pattern was clearly different in the conditioned medium of HT1080-BMP-1 cells where the smear moved to lower molecular weights and two new products appeared around 42 and 21 kDa. In the cell layer of HT1080-BMP-1 cells, the 21 kDa product was also present (Fig. 3a).

Fig. 3.

Analysis of betaglycan cleavage by BTPs. a Betaglycan immuno-detection in the conditioned medium and in the cell layer of HT1080-E94A cells (*) and HT1080-BMP-1 cells (+) and after incubation of human soluble betaglycan (rec hBG; 833 nM) with (+) or without (−) BMP-1 (140 nM) for 8 h. Detection was by Western blot with anti-betaglycan antibody after SDS-PAGE (4–20 % acrylamide, reducing conditions). Arrows indicate fragments resulting from BMP-1 cleavage. b Effect of reduction with dithiothreitol (DTT) on the detection of fragments 1–4 from betaglycan. Recombinant human soluble betaglycan (1.25 µM) was incubated without or with BMP-1 (250 nM) for 4 h at 37 °C. Detection was by SDS-PAGE (12 % acrylamide) with Coomassie blue staining or by Western blot with anti-His antibody. In the absence of DTT, fragments 1 and 2 are shifted towards lower molecular weights (named 1′ and 2′) while fragments 3 and 4 remain linked by a disulfide bridge and migrate at the same molecular weight as fragment 1. c Rat soluble betaglycan (780 nM) was incubated at 37 °C for 4 h with or without 78 nM BMP-1, mTLL-1 or mTLD (SDS-PAGE, 4–20 % acrylamide, non-reducing conditions). Note that mTLL-1 and mTLD are similar in size to intact betaglycan and that products 3 and 4 are not observed in the conditions of this experiment. d Schematic view of the truncated His-tagged soluble forms of human (rec hBG) and rat (rec rBG) betaglycan used in this study. BMP-1 cleavage sites are indicated with dotted lines and corresponding fragments numbered like in a–c. The disulfide bridge linking fragments 3 and 4 is also indicated. Open triangles indicate the position of GAG attachment sites (not used in the recombinant forms used here)

In subsequent experiments, direct cleavage of betaglycan by BMP-1 was shown using recombinant human soluble GAG-free betaglycan. When analyzed by Western blotting (Fig. 3a, right panel), the cleavage pattern of human soluble betaglycan revealed three bands at 60, 42 and 21 kDa, similar to those previously observed in the cell-based experiments, and an additional faint band at 33 kDa. These bands are numbered fragments 1–4 in Fig. 3a and are also seen by Coomassie blue staining (Fig. 3b, reducing conditions). When analyzed in non-reducing conditions, however (Fig. 3b), bands 3 and 4 disappeared while the intensity of the band corresponding to fragment 1 increased. This shows that fragments 3 and 4 are products of fragment 1, which remain linked by a disulfide bridge in non-reducing conditions. Furthermore, Western blotting using an anti-His antibody shows that fragment 3 is at the C-terminus of the protein (Fig. 3b). Finally, betaglycan processing into fragments 1 and 2 was also observed when recombinant rat soluble betaglycan was incubated with either BMP-1 or the other BTP isoforms mTLL-1 and mTLD (Fig. 3c).

By Edman sequencing, fragment 2 (42 kDa) derived from soluble rat betaglycan (100 kDa) was shown to correspond to the N-terminal domain, starting with ²⁴GPEPST while fragment 1 (60 kDa) started with ³⁶⁶DEEVHXI (Table 2), clearly localizing the main cleavage site in the linker region between the endoglin (E; membrane-distal) and zona pellucida (ZP; membrane-proximal) domains of betaglycan (Fig. 3d). To identify the second cleavage site, the more sensitive iTRAQ-ATOMS technique was also applied to rat betaglycan. As expected, the identity of the cleavage site previously determined by Edman degradation was confirmed (Table 2; Supplemental Fig. S5a; Table S3). In addition, two peptides starting at aspartate 554 or 556 near the GAG attachment sites in the ZP domain (Table 2; Supplemental Figs. S5b, S6a; Fig. 3d) were also identified. These peptides define the cleavage sites between fragments 3 and 4 previously observed by Western blot and SDS-PAGE (Fig. 3a, b). These sites are in agreement with the observed behavior of the cleavage products upon reduction (Fig. 3b) since, based on the known structure of the N-terminal moiety of the ZP domain [53, 54], fragments 3 and 4 are expected to remain linked by a disulfide bridge between the first and fourth cysteines of this domain (Fig. 3d).

In human betaglycan, a similar cleavage pattern was identified using iTRAQ-ATOMS sequencing (Tables 2, S3; Supplemental Fig. S7b, S8; Fig. 3d) with two cleavage sites in the linker between the E and ZP domains and one cleavage site inside the ZP domain. Cleavages at ³⁶⁴D and ⁵⁵³D exactly correspond to cleavages at ³⁶⁶D and ⁵⁵⁴D in rat betaglycan, showing that the main cleavage sites are conserved in both species.

Betaglycan cleavage in human primary keratocytes

Betaglycan is known to be expressed in a large number of mesenchymal cells which, most of the time, also express BTPs. Here, we analyzed the conditioned medium of corneal stromal cells (keratocytes), isolated from human donors [41], to see if we could detect the betaglycan fragments identified above and if these fragments were affected in the presence of the synthetic BTP inhibitor already used for CD109. We first analyzed the cleavage of procollagen I, a well-known substrate of BTPs, to check the efficiency of the BTP inhibitor in keratocyte culture conditions. In the presence of 10 µM of the synthetic inhibitor, the amount of the C-propeptide product, released upon BTP-dependent procollagen I maturation, was significantly decreased in the keratocyte medium when compared to the control condition (Fig. 4). Similarly, a fragment with a size identical to betaglycan fragment 1 was detected in the keratocyte medium of cells treated with DMSO (Fig. 4), which was clearly decreased when the cells were cultured for 48 h in the presence of the BTP inhibitor. This strongly suggests that the cleavage of betaglycan observed in keratocytes is mainly due to one of the BTP members.

Fig. 4.

Analysis of procollagen I and betaglycan proteolysis in human primary keratocytes. Immuno-detection of the C-propeptide (CPI) of procollagen I (Procoll I; left panel) or betaglycan (right panel) in the conditioned medium of human keratocytes, treated for 48 h with 10 µM of hydroxamate 90 (inhibitor) or with vehicle (DMSO). Control lanes (in vitro) show the fragments observed when recombinant procollagen I or soluble human betaglycan was incubated with recombinant BMP-1. 8–10 % acrylamide gels, reducing conditions

Effect of cleavage on betaglycan function

We then compared the cleaved and non-cleaved forms of soluble betaglycan for their ability to antagonize TGF-β activity. Thus, recombinant soluble rat betaglycan was incubated with buffer or with BMP-1 to obtain complete cleavage and added to Mv1Lu cells transfected with pCAGA₁₂-Luc, to assess its ability to antagonize TGF-β2 activity through the well-characterized TGF-β luciferase reporter. As shown in Fig. 5a, cleaved betaglycan was much less efficient in antagonizing TGF-β2 activity than the parent molecule. Its IC₅₀ was increased by a factor of more than 20 (from 2.5 nM to at least 50 nM), demonstrating that proteolytic fragments derived from soluble betaglycan bound TGF-β2 with much less affinity or in a non-productive manner.

Fig. 5.

Effect of BMP-1-dependent cleavages on betaglycan ability to bind TGF-β. a Effect of cleavage on soluble betaglycan antagonist activity. This activity was measured using a luciferase reporter assay in the presence of increasing concentrations of recombinant soluble betaglycan, incubated (cleaved rBG) or not (intact rBG) with BMP-1 at 37 °C for 4 h prior to the assay, and expressed in percent of control activity obtained when 20 pM TGF-β2 alone is added to Mv1Lu cells transfected with pCAGA₁₂-Luc. Squares represent the luciferase activity obtained without betaglycan but with the same BMP-1 concentration that was present in the corresponding “cleaved rBG” condition. One representative experiment out of 3 is shown (mean ± SD of quadruplicate measurements). b Quantitative analysis of ¹²⁵I-TGF-β1 or -β2 affinity labeling of membrane-bound betaglycan. The radioactivity immuno-precipitated with anti-betaglycan antibody is expressed as the percent of the total ¹²⁵I-TGF-β bound in the cell lysates before immunoprecipitation (IP). Mean ± SD of 5 experiments

To determine if BMP-1 activity could also affect the ability of membrane-bound betaglycan to bind TGF-β, we labeled HT1080-E94A and HT1080-BMP-1 cells with ¹²⁵I-TGF-β1 and -β2, immuno-precipitated betaglycan located at the cell surface and counted the associated radioactivity. Figure 5b shows that there is no significant difference between the amounts of labeled betaglycan quantified in the two types of HT1080 cells. This indicates that BMP-1-dependent modification of betaglycan at the cell surface only affects a relatively minor proportion of the membrane-bound protein.

Effect of BMP-1 activity on TGF-β signaling

To analyze the direct consequences of co-receptor/antagonist proteolysis on TGF-β signaling, we monitored the responses induced by the addition of TGF-β1 or TGF-β2 in HT1080 cells. First, luciferase reporter activity was measured in the presence of increasing amounts of both TGF-β isoforms and we found that HT1080-BMP-1 cells were significantly more responsive to TGF-β than control cells in a concentration range between 40 and 100 pM (Fig. 6a).

Fig. 6.

Effect of BMP-1 on TGF-β signaling. a Luciferase activity detected in cell lysates of HT1080-E94A and HT1080-BMP-1 cells transfected with the TGF-β reporter plasmid pGL3-basic and treated for 24 h with increasing amounts of TGF-β1 or TGF-β2. Means ± ranges of two independent experiments run in duplicate. b SMAD2 phosphorylation, detected by Western blot (left), in cell lysates of HT1080-E94A and HT1080-BMP-1 cells, cultured in the absence of TGF-β for 48 h or in the presence of 50 pM TGF-β1 or TGF-β2 for 1–48 h. Ratios of phospho-SMAD2 (P-SMAD2) signal intensities over signal intensities for actin or total SMAD2/3 are also shown (right)

We then monitored the phosphorylation of SMAD2 over time after the addition of a fixed TGF-β concentration (50 pM). Again, higher levels of phosphorylated SMAD2 were detected at all time points in HT1080-BMP-1 cells (Fig. 6b). In particular, at 48 h when SMAD2 phosphorylation had disappeared or was strongly diminished in HT1080-E94 cells, the signal remained relatively high in HT1080-BMP-1 cells suggesting prolonged SMAD-dependent signaling in these cells. These results indicate that both TGF-β stability and signaling were strongly affected by the presence of catalytically active BMP-1.

Discussion

In this study, we show iTRAQ-based relative quantitation to be a powerful degradomics tool for the discovery of novel cellular substrates and pathways controlled by BMP-1. Thirty-eight secreted or membrane-bound proteins were found to be modified in the presence of BMP-1, either through proteolytic processing or through indirect mechanisms. Their nature further strengthen the importance of BTPs in the regulation of ECM, angiogenesis and growth factor activities (TGF-β co-receptors, CYR61, fibronectin, fibulin-3, IGFBP7, C-type lectin domain family 11 member A) and suggest new consequences of BMP-1 activity on cell adhesion and migration (cadherins, CD44, dystroglycan, glypican, semaphorin 7A), on the control of other proteolytic activities (ADAM9, serine protease 23, HTRA1, tissue-type plasminogen activator, trypsin-2, tryptase 1, TIMP-2) or on immunity (MHC class I antigen B35, β2-microglobulin, calreticulin, dermcidin). Here, we analyzed the effects of BMP-1 on three TGF-β co-receptors to shed a new light on the relationships between BMP-1 and the TGF-β pathway. Among these co-receptors, CD109 and betaglycan were confirmed to be directly cleaved by BMP-1 while neuropilin-1 seemed to be the target of downstream effects of BMP-1 cleavages, as can occur in cell-based assays by interactions and connectivity of proteases and inhibitors in the protease web [55].

The role of CD109 has been well described in keratinocytes where it is known to negatively regulate TGF-β signaling by forming a heteromeric complex with TGF-β signaling receptors and betaglycan [22] and promoting their internalization and degradation [23]. Furin maturation of CD109 [27] leads either to the formation of a heterodimer at the cell surface or to a secreted form which can also sequester TGF-β from its receptors, thereby also contributing to decrease TGF-β-dependent signaling. Here, we show that BMP-1 further cleaves CD109 in its so-called “bait region” and that it is not the sole protease which can do so since a BTP inhibitor fails to completely inhibit this cleavage. The identity of this alternative protease remains to be determined. We also provide the first quantitative analysis of CD109 binding to TGF-β1 and found that the dissociation constants which were determined (25 and 245 nM) fall in the same range as the dissociation constants that were previously determined for the interactions between soluble betaglycan and TGF-β1-3 using the same technique (36–143 nM; [56]). Interestingly also, we observed that cleavage of soluble CD109 by BMP-1 does not disrupt the TGF-β binding site on the co-receptor and even seems to lead to a moderate but significant increase in the stability of the CD109/TGF-β1 complex, suggesting that the binding site is more accessible when CD109 is cleaved.

The last candidate to be analyzed was betaglycan. We found that it is efficiently cleaved by BMP-1 and other BTPs, both in cellular assays and in vitro, at two main locations. The first cleavage site was mapped to the long linker between the E and ZP domains while the second cleavage site was located inside the ZP domain. We could further show that these cleavages severely impair betaglycan antagonist activity towards TGF-β, clearly supporting the idea that efficient TGF-β binding was prevented upon cleavage. Noteworthy, in vitro processing of the soluble form of betaglycan by other proteases has previously been reported [56–58]. Plasmin and trypsin were shown to cleave betaglycan in the linker region, approximately 50 residues downstream of the main BMP-1 cleavage, leading to a loss of quaternary structure from a dimeric to a monomeric form [56]. Furthermore, after cleavage, individual E and ZP domains bind the three TGF-β isoforms much less efficiently, suggesting that cooperativity between the two tethered domains is crucial for TGF-β binding. Besides, granzyme B cleaves human betaglycan at several sites, one of them being located in the ZP domain between ⁵⁵⁸D and ⁵⁵⁹A, also in close proximity with one of the cleavage sites described here for BMP-1. The ability of betaglycan to bind TGF-β also seems to be impaired after cleavage by granzyme B since active TGF-β1 can be released from immobilized betaglycan in the presence of the protease [57]. While there is no known structure for a betaglycan ZP-N domain, structural alignment with the homologous domain of ZP3 (Supplemental Fig. S9) suggests that residues defining BMP-1 and granzyme B cleavage sites reside in a loop between the strands F and G defined for ZP3 [54] and are thus readily accessible to proteases. Collectively, these results show that the regions of betaglycan corresponding to the long linker and to the residues located just after the GAG attachment sites (Fig. 3d) can be used by proteases as proteolytic switches to turn off the activity of soluble or matrix-bound betaglycan. Surprisingly, GAGs themselves do not seem to play a crucial role in this mechanism since our results show that BMP-1 cleaves both the cellular (GAGged) and recombinant (unGAGged) forms of betaglycan.

We focused on TGF-βs as the main ligands of betaglycan but other TGF-β superfamily members such as BMP-2/4/7, GDF-5 or inhibin A can also bind betaglycan and could be affected by proteolysis of the receptor. While BMP-2/4/7 and GDF5 binding actually engages both betaglycan domains like TGF-βs [59], this is not the case for inhibin A which only requires the C-terminal membrane-proximal part of the ZP domain for efficient binding [53, 60]. Thus, BTP cleavage of betaglycan should be without consequences on the inhibin–betaglycan interaction and could allow the differential control of betaglycan activity regarding its various ligands.

A striking result from the present study is that BMP-1 proteolytic activity has a strong effect on TGF-β signaling, both in terms of intensity and duration. While no difference was observed between HT1080 overexpressing BMP-1 and control cells in the absence of exogenous TGF-β, there was a strong increase in phospho-SMAD2 and luciferase signals when recombinant active forms of TGF-β1 or -β2 were added to the cells. The latter results indicate that signaling enhancement cannot be the sole consequence of the previously described role of BMP-1 in TGF-β activation which is dependent on LTBP1 cleavage [13] and clearly point to other regulatory mechanisms. In this context, betaglycan and CD109 co-receptors are very good candidates to explain BMP-1 effects as they directly control the bioavailability of TGF-βs and their interactions with signaling receptors. However, both co-receptors can be present both at the cell surface and in the extracellular space with potentially divergent outcomes and it is important to determine in which cell compartment these proteins are preferentially cleaved.

A previous report showed that in HT1080-BMP-1 cells, BMP-1 can be found at the plasma membrane [61]. Here, we further demonstrate for the first time that BMP-1 can work directly at the cell surface and cleave membrane-bound betaglycan to leave a shorter form of the receptor at the membrane. However, the amount of betaglycan cleaved at the cell surface seems to represent a minor proportion of the total amount of cellular betaglycan (less than 20 % based on Western blots) and HT1080-BMP-1 cells were found to be as efficiently labeled by radioactive TGF-β as control cells. The latter results suggest that the pool of functional betaglycan at the cell surface is not severely modified in the presence of BMP-1.

In contrast, BMP-1 has a major impact on the soluble (shed) form of betaglycan, present in the conditioned medium, which seems to be totally processed in HT1080-BMP-1 cells. This is in agreement with the observation that BMP-1 is efficiently secreted in the conditioned medium of these cells and suggests a mechanism by which BMP-1 could inactivate the soluble form of betaglycan, which act as an antagonist, to regulate TGF-β bioavailability and increase the pool of free and active cytokine that can bind to the signaling receptors, ultimately leading to the increased TGF-β signaling that is observed. This new role fits well with the previous demonstration that BMP-1 participates in the activation of latent forms of TGF-β immobilized in the ECM through the cleavage of LTBP-1 [13]. Proteolytic inactivation of soluble betaglycan could actually prevent the immediate uptake of active TGF-β by this antagonist to increase the local concentration of TGF-β and potentiate its signaling activity. The situation for CD109 is less clear and requires further investigation to conclude on the potential effects of proteolysis. Even if TGF-β still binds cleaved CD109 with high affinity, it cannot be excluded that interactions with other important partners, especially signaling receptors and betaglycan, are affected and could also play a role in the regulation of TGF-β signaling. Interestingly also, we observed that CD109 expression was significantly and reproducibly higher in the supernatant of HT1080-BMP-1 cells than in the supernatant of control cells, suggesting that BMP-1 could also have an indirect effect on CD109, through the regulation of its synthesis and/or secretion. Whether increased TGF-β signaling is involved in this regulation remains to be determined but increased expression of CD109 in the presence of TGF-β1 was previously observed in HEK293 cells [27].

BTP-dependent inactivation of soluble betaglycan could have significant consequences in cancer, as BTPs were recently shown to be present in the ECM surrounding tumors and the TGF-β neutralizing activity of soluble betaglycan has been found to be a major player in the regulation of local immune response during tumorigenesis [62, 63]. Intriguingly also, a common feature of the targeted disruption of mTLL-1 and betaglycan genes in mice is a ventricular septal defect [64, 65] which significantly contributes to lethality. In both cases, it is thought that important changes in the levels of activity of TGF-βs and BMPs are playing a major role but the possible interplay between BTPs and betaglycan in the control of this particular morphogenetic event still requires further investigation. Finally, our work shows that betaglycan is detected in the conditioned medium of human primary keratocytes and that it is naturally processed to produce a fragment which is significantly reduced in the presence of a BTP inhibitor. This demonstrates that BTP-dependent betaglycan processing can occur in physiological conditions despite the low levels of both proteinase and substrate. Control of betaglycan activity by BTPs could thus become even more relevant in conditions where increased levels of BTPs are observed, such as fibrosis [7–10] or corneal healing and scarring [66]. Interestingly, after corneal injury, TGF-β2, which is highly dependent on betaglycan to trigger signaling, is the major TGF-β isoform produced by injured epithelial cells to activate stromal cells [67]. Cleavage of the soluble form of betaglycan by BTPs could then increase the bioavailability of TGF-β2 and potentiate its pro-fibrotic activity. In support of this mechanism, betaglycan inhibition was previously shown to aggravate fibrosis [68, 69].

In summary, our results reveal a completely new aspect of the complex relationships between TGF-β and BMP-1/tolloid-like proteases. It was already known that BTPs were involved in the release of latent TGF-β from the ECM through the cleavage of LTBP-1, thereby promoting degradation of the propeptide and dissociation of active TGF-β from the latent complex. In addition, TGF-β1 and -β2 were shown to induce the expression of BMP-1 in various cell types [9, 14], possibly contributing to the establishment of a positive feedback loop which can have important consequences in pathological conditions such as fibrosis. Our present results provide strong evidence that the control of TGF-β activity by BTPs extends far beyond the release of the active form of the growth factor and also affects its stability after activation and its interaction with soluble co-receptors. In this context, BMP-1 can be seen as a signaling molecule playing a direct and irreversible role in the TGF-β pathway.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

BG

Betaglycan

BMP

Bone Morphogenetic Protein

BSA

Bovine Serum Albumin

BTP

BMP-1/Tolloid-like Proteinase

DTT

Dithiothreitol

ECM

Extracellular Matrix

GAG

Glycosaminoglycan

mTLD

Mammalian Tolloid

mTLL-1

Mammalian Tolloid-like 1

SDS-PAGE

Sodium Dodecyl Sulfate Poly-Acrylamide Gel Electrophoresis

TGF-β

Transforming Growth Factor-β

ZP

Zona Pellucida