Development of a Proteomic Workflow for the Identification of Heparan Sulphate Proteoglycan‐Binding Substrates of ADAM17

ABSTRACT

Ectodomain shedding, which is the proteolytic release of transmembrane proteins from the cell surface, is crucial for cell‐to‐cell communication and other biological processes. The metalloproteinase ADAM17 mediates ectodomain shedding of over 50 transmembrane proteins ranging from cytokines and growth factors, such as TNF and EGFR ligands, to signalling receptors and adhesion molecules. Yet, the ADAM17 sheddome is only partly defined and biological functions of the protease have not been fully characterized. Some ADAM17 substrates (e.g., HB‐EGF) are known to bind to heparan sulphate proteoglycans (HSPG), and we hypothesised that such substrates would be under‐represented in traditional secretome analyses, due to their binding to cell surface or pericellular HSPGs. Thus, to identify novel HSPG‐binding ADAM17 substrates, we developed a proteomic workflow that involves addition of heparin to solubilize HSPG‐binding proteins from the cell layer, thereby allowing their mass spectrometry detection by heparin‐treated secretome (HEP‐SEC) analysis. Applying this methodology to murine embryonic fibroblasts stimulated with an ADAM17 activator enabled us to identify 47 transmembrane proteins that were shed in response to ADAM17 activation. This included known HSPG‐binding ADAM17 substrates (i.e., HB‐EGF, CX3CL1) and 14 novel HSPG‐binding putative ADAM17 substrates. Two of these, MHC‐I and IL1RL1, were validated as ADAM17 substrates by immunoblotting.

Abbreviations

ADAM17

a disintegrin and metalloprotease 17

EGFR

epidermal growth factor receptor

FASP

filter aided sample preparation

HSPG

heparan sulphate proteoglycan

LC‐MS/MS

liquid chromatography‐tandem mass spectrometry

STAGE tip

STOP and GO extraction tip

TNF

tumour necrosis factor alpha

1. Introduction

Summary

• ADAM17 is a “sheddase” with crucial roles in health and disease.

• The systematic identification of ADAM17 substrates (sheddome) can contribute to uncover new functions of the sheddase.

• We developed a proteomic workflow involving addition of heparin to solubilize HSPG‐binding proteins (HEP‐SEC) that allowed the identification of both known and putative novel ADAM17 substrates that were underrepresented in traditional secretome analyses.

• This study thus uncovered new biological functions of ADAM17 by identifying novel proteins that are released by the protease.

• HEP‐SEC is broadly applicable and has the power to deepen our understanding of mechanisms, including proteolysis, that regulate cell surface abundance of HSPG‐binding proteins.

The proteolytic release of transmembrane proteins, also known as ectodomain shedding, is a post‐translational modification that plays a crucial role in cell‐cell communication and other biological processes [1]. ADAM17, a member of the “disintegrin and metalloproteinase” family that was first identified as the enzyme responsible for the proteolytic cleavage of TNF [2, 3], mediates ectodomain shedding of over 50 proteins, ranging from signalling molecules, such as cytokines and growth factors, to cell receptors, adhesion molecules and endocytic proteins (reviewed in refs. [4, 5]). In agreement with its high number of substrates, ADAM17 plays a crucial role in many biological processes, including development, inflammation, cell proliferation and survival [6]. Functions of a protease are determined by the collection of its substrates, therefore the identification of novel ADAM17 substrates may uncover additional biological functions of this protease in health and disease. Mass spectrometry (MS)‐based analysis of the cellular secretome has become a powerful method to identify candidate substrates of sheddases like ADAM17, both in vitro and in vivo [6]. Activation, overexpression or ablation of sheddases leads to changes in soluble levels of their substrate ectodomains in the secretome, which can be quantified by MS to identify novel candidate substrates of the sheddase. However, despite technological improvement in mass spectrometers and innovative acquisition methods that have improved their resolution, the detection of low‐abundance proteins is still difficult to achieve given the large dynamic concentration range of proteins in the secretome [7]. For example, if secreted proteins are sequestered by cell surface or pericellular heparan sulphate proteoglycans (HSPGs), then their soluble levels could be too low for detection by MS.

HSPGs are localised in the pericellular and extracellular matrix (e.g., perlecan, agrin) or on cell membranes (e.g. syndecans, glypicans), where they function as coreceptors, signalling receptors, or endocytic receptors [8, 9]. More than 400 human proteins are known to bind HSPGs [10, 11], including ADAM17 substrates such as HB‐EGF [12]. We hypothesised that the small number of known HSPG‐binding ADAM17 substrates may be an underestimation, arising from the technical constraints of secretome‐based MS approaches. Here, we tested this hypothesis by treating murine embryo fibroblasts (MEFs) with heparin, a highly sulphated form of HS, to solubilize HSPG‐binding proteins [13] while activating ADAM17‐mediated shedding by addition of phorbol‐12‐myristat‐13‐acetate (PMA) [14]. This “heparin‐treated secretome” (HEP‐SEC) analysis identified a number of known ADAM17 substrates, including H‐2 class I histocompatibility antigen, in addition to 14 new putative ADAM17 substrates that standard LC‐MS/MS analysis and label‐free quantification (LFQ) were not able to detect. One of these proteins, interleukin 1 receptor‐like 1 (IL1RL1), was validated as a novel ADAM17 substrate by Western blotting.

2. Materials and Methods

2.1. Cell Lines and Reagents

Wild‐type (WT) and ADAM17 knockout (KO) MEFs were cultured in Dulbecco's Modified Eagle's Medium (DMEM), containing 1% L‐glutamine, 1% penicillin and streptomycin, 1% sodium pyruvate and 10% foetal bovine serum (FBS) at 37°C, 5% CO₂ (all reagents were purchased from Euroclone, Pero, Italy). HTB94 were kindly provided by Prof. Hideaki Nagase and cultured in DMEM with the same manner as above. Anti‐APP clone 22C11 was purchased from Millipore (part of Merck/Sigma Aldrich, Darmstadt, Germany); anti‐FLAG M2 (ab205606), anti‐HLA‐ABC (EMR8‐5) and anti‐ADAM17 (ab39162) antibodies were purchased from Abcam (Cambridge, UK); anti‐actin antibody (A2066) was purchased from Sigma Aldridge, anti‐calnexin antibody (ADI‐SPA‐860‐F) was purchased from Enzo Life Science (Farmingdale, New York, USA), HRP‐coupled anti‐mouse and anti‐rabbit secondary antibodies were from Promega (Madison, Wisconsin, USA).

2.2. Evaluation of Heparin Effects on Sample Preparation

WT MEFs were seeded in T175 flasks and grown in complete DMEM until confluence. Cells were washed twice with PBS, incubated with 20 mL serum free DMEM for 24 h. Then, conditioned medium (CM) was harvested and aliquoted. Next, 1 mL of conditioned media was treated with 200 µg/mL heparin (heparin sodium salt–H3393 – from Sigma‐Aldrich) or equal volumes of PBS (untreated controls – CT) for 20 min at RT under agitation. After centrifugation to remove cell debris (7000 x g, 10 min), heparin‐treated or untreated conditioned media were subjected to filter aided sample preparation (FASP) with 10 kDa Vivacon 500 spin filters (Sartorius, Göttingen, Germany), as previously described [15]. Briefly, samples were reduced and denatured by incubation with 50 mM DTT in UA buffer (8 M urea, 100 mM Tris/HCl, pH 8.5) for 30 min. Subsequently, samples were centrifuged to remove excess DTT (14,000 × g, 20 min) and alkylated with 50 mM iodoacetamide (IAA) in UA buffer for 5 min, at 37°C in the dark. After reduction and alkylation, samples were washed twice with UB buffer (8 M urea, 100 mM Tris/HCl, pH 8) and digested with 0.2 µg LysC (Promega, Madison, Wisconsin, US) in UC buffer (2 M urea, 25 mM Tris/HCl, pH 8) overnight at 37°C. Then, samples were subjected to another digestion step with 0.1 µg trypsin (Promega) for 4 h at 37°C in 50 mM ammonium bicarbonate. Peptides were eluted in 120 µL of 0.5 M NaCl by subjecting the filter columns to centrifugation (14000 x g for 1 h) and acidified with 20 µL of 8% formic acid (FA). The buffer of the eluted peptides was exchanged by stop‐and‐go extraction (STAGE) on reverse phase tips packed with C18 disks in‐house (Empore SPE Disks, Sigma‐Aldrich), as previously described [16]. Briefly, after activation with 100 µL methanol, the C18 resin was washed four times with 0.1% FA. Peptides were loaded onto the C18‐packed STAGE‐tips, washed with 0.1% FA and finally eluted with 40 µL of 60% acetonitrile and 0.1% FA in MS grade water (all reagents from Sigma‐Aldrich). Peptides were then dried by vacuum centrifugation and resuspended in 20 µL of 0.1% FA. Alternatively, samples were processed with the iST kit from PreOmics (Planegg/Martinsried, Germany), according to the manufacturer's instructions. Peptide concentration was measured by Nanodrop 2000 (Thermo Scientific, Waltham, Massachusetts, US). Subsequently, 1 µg of peptide mixture per sample was loaded onto a Dionex Ultimate 3000 RSLCnano LC system, which was coupled online via a Nanospray Flex Ion Source to a Q‐Exactive + mass spectrometer (all instruments from Thermo Scientific). Peptides were separated on Acclaim PEPMap C18 column (50 cm × 75 µm ID, Thermo Scientific) with 250 nL/min flow using a binary gradient of water (A) and acetonitrile (B) supplemented with 0.1% FA for 350 min. Data‐dependent acquisition (DDA) was used for label free quantification (LFQ). Full MS scans were acquired at a resolution of 70000 (m/z range: 300−1400; automatic gain control [AGC] target: 1 × 10⁶; max injection time 50 ms). The DDA was used on 10 most intense peptide ions per full MS scan for peptide fragmentation (resolution: 17,500; isolation width: 2 m/z; AGC target: 1 × 10⁵; normalized collision energy (NCE): 25%, max injection time: 120 ms). A dynamic exclusion of 120 s was used for peptide fragmentation. Raw data were analysed with Maxquant software (version 2.0.1.0 downloaded from maxquant.org, Max‐Planck Institute Munich) and searched against a reviewed canonical FASTA database of Mus musculus from UniProt (download: November 5th, 2020). Trypsin/P was defined as a protease. For the main search peptide and peptide fragment mass tolerances were set to 4.5 and 20 ppm, respectively. Carbamidomethylation of cysteine was defined as static modification. Acetylation of protein N‐termini as well as oxidation of methionine were set as variable modifications. The “match between runs” option was enabled according to the experimental setup with a match time window of 1.5 min. LFQ of proteins required at least one ratio count of unique peptides. Unique and razor peptides were used for quantification. Data normalization was enabled. Perseus was used for the evaluation of MS data [17].

2.3. HEP‐SEC to Identify Proteoglycan‐Binding ADAM17 Substrates

WT and ADAM17 KO MEFs were seeded onto a 6‐well plate and grown in complete DMEM until confluency. After two PBS washes, cells were incubated for 3 h with 2 mL serum‐free DMEM with or without 200 µg/mL heparin, in the presence or absence of 25 ng/mL PMA. Although serum deprivation may affect ectodomain shedding (i.e., many sheddases are less active in the absence of serum), its removal is necessary as serum masks the signal of low abundance proteins in the secretome, thus preventing their detection by MS [18, 19]. Conditioned media were freshly collected and processed for MS analysis as described above. Peptides were quantified by Nanodrop 2000 and stored at −80°C until the LC‐MS/MS analysis. MS results were analysed by Perseus software [17]. Protein LFQ intensities were log2 transformed, and volcano plots obtained by plotting log2 transformed protein ratios of PMA‐treated cells versus CT, against the −log10‐transformed p value of protein changes, calculated by a Student t‐test. Six biological replicates of both PMA‐treated cells and controls were analysed, false discovery rate correction was applied to the analysis, and a corrected p value of 0.05 was set as a threshold for statistical significance. QARIP (quantitative analysis of regulated intramembrane proteolysis) analysis was used to match peptides detected by MS to the ectodomain of transmembrane proteins identified as potential ADAM17 substrates [20]. This analysis was performed through the QARIP web server (http://webclu.bio.wzw.tum.de/qarip/).

2.4. Generation of HTB94 Cells ADAM17KO by CRISPR/Cas9 Technology

CRISPR guides for ADAM17 (5′‐GGTCGCGGCGCCAGCACGAA‐3′) and non‐targeting control sequences (5′‐TCCGGAGCTTCTTTCAGTCAA‐3′) were cloned into a lentiCRISPRv2 vector (Addgene, cat. No. 52961), as previously described [21]. In brief, HEK293T packaging cells were used for CRISPR/Cas9 lentiviral particle production. After growing the cells until 70% confluence, they were transfected by using Lipofectamine 3000 with the packaging plasmid pxPAX2 and pcDNA3.1‐VSVG, and the lentiCRISPRv2 construct (containing either the ADAM17 CRISPR guide or non‐targeting control guide—NTC). After 24‐h incubation, conditioned media containing the viral particles were harvested, centrifuged for 10 min at 1000 × g to remove cell debris and aliquoted (1 mL) for storage at −80°C. A total of 1 × 10⁶ HTB94 cells were used for lentiviral transduction of each CRISPR construct (ADAM17 KO or NTC) and 500 µL of conditioned media containing viral particles were added to 500 µL of cell suspension, in the presence of 5 µg/mL polybrene. Then, cells were transferred into a six‐well plate, spin‐infected by centrifugation (2 h, 1800 rpm, 33°C) and selected with 1 µg/mL puromycin. Generation of an ADAM17 KO cell line was evaluated by Western blot. ADAM17 KO and WT control cells were lysed in 150 mL of STET lysis buffer (10 mM Tris‐HCl, 1 mM EDTA, 100 mM NaCl, 1% Triton X‐100, 1× Proteinase Inhibitor mixture [Sigma Aldrich], 10 mM 1–10 phenanthroline). Finally, 20 µg of proteins were separated by SDS‐PAGE electrophoresis, transferred to poly‐(vinylidene difluoride) membranes (BioRad, Hercules, California, USA) and incubated with anti‐ADAM17.

2.5. Ectodomain Shedding of H2‐D1 and IL1RL1

Sequences to express recombinant murine H2‐D1 or human IL1RL1 were cloned into a pcDNA3.1/Zeo (+) plasmid using HindIII and NotI. Both H2‐D1 and IL1RL1 are type I transmembrane proteins, and they were cloned with a CD5 signal peptide (MPMGSLQPLATLYLLGMLVASVLG) followed by three FLAG‐tag sequences (DYKXXD) at the protein N‐terminus. Three HA‐tag sequences (YPYDVPDYA) were added to the protein C‐terminus. Recombinant IL1RL1 was expressed in ADAM17 KO and WT control fibroblast‐like HTB94 cells generated by CRISPR‐Cas9, and H2‐D1 was expressed in MEFs isolated from WT or ADAM17KO mice [22] using Lipofectamine 3000 according to manufacturer's instructions. After 48 h from transfection, cells were washed twice with PBS and stimulated with 25 ng/mL PMA for 3 h, in the presence or absence of 10 mM TAPI or 200 µg/mL heparin. Conditioned media were collected and proteins precipitated overnight with trichloroacetic acid (TCA) and resuspended in 50 µL of Leammli Sample Buffer (BioRad). Cells were collected in 150 µL of STET lysis buffer and analysed by SDS‐PAGE electrophoresis and Western blotting as described above. An anti‐FLAG M2 antibody (Sigma) was used to detect recombinant H2‐D1 or IL1RL1 in conditioned media and cell lysates. Protein bands were detected by a Chemidoc image analyser (Bio‐Rad) and their intensity quantified by ImageLab (BioRad). Statistical analysis was performed using the Student t‐Test. Nine biological replicates were used for the analysis of H2‐D1 shedding, and 3 were used for the analysis of IL1RL1.

2.6. Molecular Docking of Heparin‐Binding Sites

H2‐D1, IL1RL1, HB‐EGF and PTK7 structures were downloaded from Alphafold protein structure database and prepared for docking by removing signal peptide, transmembrane and intracellular sequences. HLA‐A was downloaded from Protein Data Bank (accession number 8ESB). In this structure HLA‐A is in complex with B‐2‐microglobulin and bound to a 9 amino acid peptide antigen from MAGEA8 (232‐241), which was removed by the ChimeraX tool for further analysis. A 3D model of a heptasaccharide heparin fragment with sequence GlcNS6S‐IdoUA2S(2SO)‐GlcNS6S‐IdoA2S(2SO)‐GlcNS6S‐IdoA2S(2SO)‐GlcNS6S‐OMe was built using the GAG builder as previously described [23]. Molecular docking and prediction of heparin binding affinity were performed with LeDock software (LePhar). Predictive models of heparin binding to H2‐D1 and IL1RL1, and the basic amino acid residues responsible for such an interaction, were visualized by PyMol (version 2.5.7, Schrödinger).

2.7. Ectodomain Shedding of HLA and IL1RL1 in Response to Physiological Stimuli

WT and ADAM17KO HTB94 cells were grown in 6‐well plates until confluent, washed and incubated with serum‐free DMEM for 3 h, in the presence or absence of 25 ng/mL PMA (Sigma), 10 µM lysophosphatidic acid (LPA, from Sigma), 10 ng/mL TNF or 50 ng/mL platelet‐derived growth factor subunit A (PDGFA) (both from Peprotech, part of Thermo Fischer Scientific). Conditioned media were collected, proteins precipitated with TCA and loaded onto a gel for SDS‐PAGE electrophoresis and Western blotting. Lysates were collected in STET buffer, 10 µg of which was loaded onto a gel. After blocking with 5% non‐fat milk (Sigma) in PBS with 0.1% Tween, the membranes were incubated with anti‐HLA Class 1 ABC antibody (clone EMR8‐5, from Abcam, Cambridge, UK).

WT MEF cells were grown in six‐well plates and stimulated for 3 h with 25 ng/mL PMA, 10 µM LPA, 10 ng/mL TNF or 50 ng/mL PDGFA as described above. Conditioned media were collected and applied to FASP with 10 kDa Vivacon 500 spin filter columns for tryptic digestion. Generated peptides were analysed by LC‐MS/MS as described above. Intensity Based Absolute Quantitation (iBAQ) was used to compare the abundance of IL1RL1 in the conditioned media of differentially stimulated cells. Two‐way ANOVA, applied to three biological replicates, was used for statistical analysis of the results.

3. Results

3.1. Secretome Analysis Identified Novel ADAM17 Substrates

ADAM17 is a regulated protease that needs to be activated by precise stimuli in order to cleave its substrates, including G‐protein coupled receptor ligands [5]. In addition to physiological stimuli, ADAM17‐mediated shedding can be stimulated by PMA [14]. Thus, in order to perform a systematic identification of ADAM17 substrates, we analysed the conditioned, serum‐free media of MEFs stimulated with PMA for 3 h. Conditioned media were applied to an advanced workflow for secretome analysis [24], comprising filter‐aided sample preparation (FASP), followed by “stop and go extraction tips” (STAGE tips) for peptide recovery and LFQ [15, 16]. A total of 1390 proteins were detected in the conditioned media of PMA‐stimulated and control MEFs (Figure 1A, Table S1). ADAM17 mostly cleaves single pass transmembrane and GPI‐anchored proteins, thus we focused our analysis on proteins with such a topology [4, 5]. LFQ analysis indicated that levels of 40 single pass transmembrane/GPI‐anchored proteins (of which 33 were type 1, 3 were type 2 and 4 were GPI‐anchored proteins) increased upon PMA stimulation, and were therefore potential ADAM17 substrates (Figure 1A, Table S1). PMA is a pleiotropic stimulator of different signalling pathways, leading to transcriptional activation, apoptosis and release of extracellular vesicles, as well as ADAM17 activation. However, PMA stimulation of ADAM17 KO MEFs had only minor effects on their secretome, further confirming that protein shedding in response to PMA primarily reflects the activity of ADAM17 (Figure 1B, Table S1). Nevertheless, levels of six transmembrane proteins (CPD, APLP2, LY75, LMAN1, COLEC12 and CDH2) increased even in the absence of ADAM17, potentially due to compensatory effects of other proteases [1]. This reduced the list of potential ADAM17 substrates identified by the analysis to 34, including, as expected, 19 proteins already validated as ADAM17 substrates (e.g., APP, LDLR, AXL and others, reviewed in refs. [4, 5]—Figure 1A, Tables 1 and 2). A further 15 proteins were also identified and are potentially novel ADAM17 substrates (e.g., MXRA8, SGCE and FLRT2).

FIGURE 1.

High resolution proteomics identified ADAM17 substrates in MEFs. (A) Volcano plot showing the log2 of protein ratio between PMA‐stimulated and control MEFs versus the −log10 of p value of 1391 proteins (n = 6). Permutation‐based false discovery rate (FDR, p = 0.05, s0 = 0.1) estimation is visualized with black hyperbolic curves. Proteins with an FDR corrected p value below 0.05 are displayed with a grey open circle, and proteins above the p value with a grey filled circle. Single‐pass transmembrane/GPI anchored proteins with a higher abundance in PMA‐stimulated cells that have already been validated as ADAM17 substrates are displayed as pink circles (known ADAM17 substrates), and those that have not already been validated as red circles (putative novel ADAM17 substrates). (B) Volcano plot showing the log2 of protein ratio between PMA‐stimulated and ADAM17KO MEFs versus the −log10 of p value of 1793 proteins (n = 6). Proteins with an FDR corrected p value below 0.05 are displayed with a grey open circle, and proteins above the p value with a grey filled circle. Single‐pass transmembrane proteins with a higher abundance in PMA‐stimulated cells are displayed as black filled circles. To note, six of these proteins were found increased in the WT cells, and therefore are not highlighted as ADAM17 substrates in A, as their abundance change cannot be dependent on ADAM17. (C) QARIP analysis, which matches peptides identified by LC‐MS/MS to the protein topology, was applied to single pass/GPI anchored transmembrane proteins that increased in the conditioned media of MEFs upon PMA stimulation. (D) Venn diagram showing the collection of known ADAM17 substrates and the collection of putative novel substrates emerging from the secretome analysis of PMA‐stimulated MEFs.

TABLE 1.

List of validated ADAM17 substrates identified by secretome analysis of PMA‐stimulated MEFs.

Protein	Gene	ID	Ratio	p value	Peptides
Neogenin	Neo1	P97798	7.17	1.27E−09	32
Inactive tyrosine‐protein kinase 7	Ptk7	Q8BKG3	9.93	2.08E−09	39
C‐type mannose receptor 2	Mrc2	Q64449	17.36	4.46E−09	35
Vascular cell adhesion protein 1	Vcam1	P29533	4.00	1.14E−08	25
Very low‐density lipoprotein receptor	Vldlr	P98156	18.72	1.80E−08	30
Tyrosine‐protein kinase receptor UFO	Axl	Q00993	5.60	2.91E−08	12
Endothelial protein C receptor	Procr	Q64695	11.12	3.41E−08	4
Syndecan‐4	Sdc4	O35988	7.09	8.48E−08	6
Activated leukocyte cell adhesion molecule	Alcam	Q61490	2.70	2.44E−07	18
Vasorin	Vasn	Q9CZT5	5.48	1.39E−06	13
Low‐density lipoprotein receptor	Ldlr	P35951	2.76	1.55E−06	36
Semaphorin‐4B	Sema4b	Q62179	5.51	2.32E−06	10
Amyloid precursor protein	App	P12023	1.71	6.64E−06	29
Integrin beta‐1	Itgb1	P09055	1.88	1.42E−05	19
Neural cell adhesion molecule 1	Ncam1	P13595	2.27	8.70E−05	23
Glypican‐1	Gpc1	Q9QZF2	1.77	1.50E−04	21
Receptor‐type tyrosine‐protein phosphatase F	Ptprf	A2A8L5	4.29	5.72E−04	25
Low‐density lipoprotein receptor‐related protein 1	Lrp1	Q91ZX7	1.60	1.61E−03	122
Mesothelin	Msln	Q61468	1.88	1.85E−03	23

Note: The table contains a list of 19 proteins that were significantly increased in the secretome of WT MEFs upon PMA stimulation and have already been validated as ADAM17 substrates (reviewed in refs. [4, 5]). Indicated are the names of the proteins, the gene name, the protein ID, the mean of the ratio between PMA‐stimulated MEFs and untreated controls of six biological replicates, the p value calculated with a two‐sided, heteroscedastic t‐test based on the intensity ratios for PMA‐stimulated and control MEFs and the number of unique peptides.

TABLE 2.

List of putative novel ADAM17 substrates identified by secretome analysis of PMA‐stimulated MEFs.

Protein	Gene	ID	Ratio	p value	Peptides
Chondroitin sulphate proteoglycan 4	Cspg4	Q8VHY0	2.56	5.53E−07	42
Fibronectin leucine rich transmembrane protein 2	Flrt2	Q8BLU0	4.85	5.90E−07	14
Transferrin receptor protein 1	Tfrc	Q62351	1.49	2.15E−05	28
Dystroglycan 1	Dag1	Q62165	2.59	3.96E−05	29
Neuropilin‐2	Nrp2	O35375	1.76	6.78E−04	12
Amino acid transporter heavy chain SLC3A2	Slc3a2	P10852	1.37	6.96E−04	19
Basigin	Bsg	P18572	1.55	1.29E−03	7
Matrix remodeling‐associated protein 8	Mxra8	Q9DBV4	10.88	1.46E−03	17
Kin of IRRE‐like protein 1	Kirrel	Q80W68	1.81	2.20E−03	8
Dolichyl‐diphosphooligosaccharide‐protein glycosyltransferase subunit 1	Rpn1	Q91YQ5	1.72	2.55E−03	10
Epsilon‐sarcoglycan	Sgce	O70258	7.26	4.89E−03	9
Receptor‐type tyrosine‐protein phosphatase S	Ptprs	B0V2N1	2.85	9.91E−03	21
Discoidin, CUB and LCCL domain‐containing protein 2	Dcbld2	Q91ZV3	1.58	8.29E−03	10
Vesicular integral‐membrane protein VIP36	Lman2	Q9DBH5	1.89	1.34E−02	13
Protocadherin‐19	Pcdh19	Q80TF3	2.82	1.35E−02	12

Note: The table contains a list of 15 proteins that were significantly increased in the secretome of WT MEFs upon PMA stimulation and have NOT already been validated as ADAM17 substrates (reviewed in refs. [4, 5]). Indicated are the names of the proteins, the gene name, the protein ID, the mean of the ratio between PMA‐stimulated MEFs and untreated controls of six biological replicates, the p value calculated with a two‐sided, heteroscedastic t‐test based on the intensity ratios for PMA‐stimulated and control MEFs and the number of unique peptides.

PMA is known to affect the release of extracellular vesicles [25]. To provide further evidence that only the ectodomain of these proteins was found in the conditioned media and prove that their increase upon PMA stimulation was a consequence of ectodomain shedding rather than unconventional secretion or other biological mechanisms, we took advantage of the webserver ‘Quantitative Analysis of Regulated Intramembrane Proteolysis’ (QARIP), which allows mapping of identified peptides to the protein sequence and topology [20]. The QARIP analysis confirmed that peptides identified for the large majority of these transmembrane proteins arose from their ectodomains, and not their transmembrane or intracellular domains (Figure 1C, Table S2). However, QARIP also identified peptides matching to the intracellular domain of two known ADAM17 substrates (PTK7 and ITGB1) and five potentially novel ADAM17 substrates (TFRC, SLC3A2, RPN1, KIRREL and PTPRS). Intensity of peptides matching to the intracellular domains of PTK7, ITGB1, SLC3A2 and PTPRS were not altered upon PMA stimulation, unlike from those from their ectodomains, whose intensity increased in response to PMA (Figure S1A, Table S2). These results suggest that these four proteins can be found in the secretome as a full‐length form that is not altered by PMA stimulation (potentially tethered to extracellular vesicles), and also in a shed form that is augmented by PMA stimulation, as we demonstrated for PTK7 (Figure S1B). Intracellular peptides of RPN1, KIRREL and TFRC were only found in a few samples (2 out of 12, Table S2), indicating that mostly the ectodomain of these proteins was present in the conditioned media. In conclusion, secretome analysis identified 34 proteins whose shedding increases in response to ADAM17 activation with PMA, 19 of which were already validated ADAM17 substrates and 15 of which are putative novel substrates (Figure 1D, Tables 1 and 2).

3.2. Set‐Up of a Proteomic Workflow for Analysis of Heparin‐Treated Secretomes (HEP‐SEC Analysis)

While our analysis identified a number of known and putative ADAM17 substrates, other well‐characterized substrates, including heparin‐binding epidermal growth factor (HB‐EGF) [6], were not detected. We reasoned that, in addition to ADAM17 substrates that are minimally or not expressed in MEFs, soluble levels of specific substrates could be lowered by their affinity for ECM or cell surface components such as HSPGs. A number of HSPG‐binding proteins, including TIMP‐3 [26], CTGF [27] and TGFβ [28], are known to be solubilized from the ECM and/or cell surface to the conditioned media by addition of heparin [13]. Thus, we considered enriching the conditioned media with HSPG‐binding proteins, including ADAM17 substrates, by treating cells with heparin before secretome analysis by MS. Before doing this, we assessed the effect of heparin on the different steps of the proteomic workflow. First, we tested effect of heparin on the preparation of samples for LC‐MS/MS analysis through the FASP protocol [15]. To assess the efficiency of protein digestion in the presence of heparin, we collected CM from MEFs that were incubated for 24 h with serum‐free DMEM. Then, we treated 1 mL of CM with 200 µg/mL heparin or equal volume of PBS, before conducting FASP. Heparin reduced the concentration of recovered tryptic peptides by about 30% (Figure 2A). After buffer exchange through STAGE tips, 1 µg of peptides arising from heparin‐treated or untreated CM were analysed by LC‐MS/MS, and peptides and proteins were quantified by LFQ. We detected an average of 8951 peptides in heparin‐treated CM and 9699 in CT (Figure 2B). Trypsin cleaves at the carboxyl side of positively‐charged lysine and arginine residues. Consequently, tryptic digestion of complex protein samples generates most peptides containing only one lysine/arginine residue, and additional peptides, arisen from inefficient digestion, containing more than one lysine/arginine residues. Therefore, fully tryptic peptides usually have a +2 charge, given by the amino group at the N‐terminus and within the side chain, while incompletely‐digested tryptic peptides have higher number of positive charges. Peptides generated from heparin‐treated media were mostly double‐charged and had a lower frequency of 3 or 4 positive charges compared to peptides from control media (Figure 2C). In line with charge distribution, almost 80% of the peptides arising from heparin‐treated media had no missed cleavages, with this percentage being significantly reduced for peptides from untreated control media (Figure 2D). On the other hand, heparin reduced the number of peptides with one missed cleavage from 24.2% to 21.3%, and with two missed cleavages from 4.0% to 2.4% (Figure 2D). In line, peptides arisen from heparin‐treated samples had lower length and mass than controls (Figure S2A,B). A similar reduction in peptide concentration after FASP and in the number of MS‐detected peptides, as well as an increase in the percentage of fully tryptic peptides in heparin‐treated samples was observed when 30 kDa cut‐off was used for FASP (not shown). These results suggest that heparin binds to peptides with a higher number of positive charges, thereby reducing their recovery after FASP, but it has no major effects on fully tryptic double‐charged peptides. Furthermore, while heparin reduced the number of total peptides identified by the LC‐MS/MS analysis, such a reduction did not result in the identification of a lower number of proteins after database searching, as the total number of proteins identified was slightly higher in heparin‐treated samples compared to the number of proteins identified in the untreated control samples (Figure 2E). Out of 1074 proteins, 825 were found in both heparin‐treated CM and CT, accounting for about 77% of total proteins (Figure 2F). A similar percentage overlap in protein identity is found when multiple biological replicates are analysed by LC‐MS/MS (Table S1). Overall, these data indicate that heparin slightly yet significantly inhibited FASP elution of peptides that were not fully digested, but that it had minimal effects on recovery of fully tryptically‐digested peptides and on the number and identity of proteins identified by LC‐MS/MS analysis.

FIGURE 2.

Effects of heparin on sample preparation through FASP and LC‐MS/MS analysis. (A) Heparin‐supplemented (HEP) and control media (CT) were processed by FASP and STAGE‐tips and assessment by Nanodrop showed that heparin reduced the concentration of generated tryptic peptides. (B) Bar chart showing the average number of peptides detected and quantified in heparin‐supplemented or control samples. (C and D) Bar chart showing that peptides arising from HEP had a different charge distribution to controls, with an increase in double charged peptides (C), and a higher percentage of fully tryptically‐digested peptides (0 missed cleavages – D) (n = 3). (E) Bar chart showing the total number of proteins identified and quantified after LC‐MS/MS analysis and LFQ of HEP and CT. (F) Venn diagram displaying percentage and number of proteins identified in both HEP and CT. (G and H) LFQ intensity distribution of identified peptides (G) and proteins (H) in HEP and CT. (I) Pearson correlation of protein intensities across HEP or CT, and between heparin‐supplemented and controls. Pearson correlation coefficients (r) were larger than 0.85 for each pair of datasets compared, which indicates a strong correlation across different samples.

A central task in proteomics is the accurate quantification of protein abundance across multiple biological samples. The distribution of LFQ intensities of peptides and proteins was comparable between heparin‐treated and control samples (Figure 2G,H). To evaluate whether heparin could affect run‐to‐run reproducibility we analysed the Pearson correlation of LFQ protein intensities between heparin‐treated samples and CT. We found an excellent linearity over 200 orders of magnitude in protein abundance, with an average Pearson correlation coefficient R² of 0.93 across controls and 0.96 across heparin‐treated samples (Figure 2I). Moreover, the correlation between heparin‐treated and untreated samples was also strong (R²  = 0.88). This indicates that addition of heparin to conditioned media did not alter the reproducibility of MS/MS quantifications.

Finally, we wanted to prove that processing methods other than FASP were not affected by the presence of heparin in the conditioned media. Thus, we processed heparin‐treated and control conditioned media with a commercial kit based on the in‐StageTip (iST) method. Compared to FASP, iST improved peptide collection (about 1 µg/µL), but there was no difference in peptide concentration between heparin‐treated and control samples (Figure S2C). Similarly, number of peptides and their charge distribution was alike in heparin‐treated and control samples after LC‐MS/MS analysis, and both showed a percentage of missed cleavages slightly above 10 % (Figure S2D–S2F). Consistent with this, the number of identified proteins and Pearson correlation were also similar between heparin‐treated and control samples (Figure S2G and S2H), indicating that the effect of heparin on sample processing by iST technology were negligible.

3.3. HEP‐SEC Analysis Identified HSPG‐Binding ADAM17 Substrates

Having validated that heparin did not have major effects on sample preparation and LC‐MS/MS analysis, we investigated how it altered the secretome when added to cells. To this end, MEFs were grown in six‐well plates until confluent, washed and incubated with serum‐free DMEM for 3 h in the presence or absence of 200 µg/mL heparin. Then, conditioned media were subjected to the previously described proteomic workflow, which includes FASP and LFQ. A total of 2112 proteins were identified in the secretome of heparin‐treated MEFs and 2057 in that of CT (Figure 3A, Table S1). The number of secreted proteins and single‐pass transmembrane proteins was similar in both conditions. A total of 199 secreted proteins were identified in the HEP‐SEC and 201 in CT. Eighty‐seven type 1, 31 type 2 and 17 GPI‐anchored proteins were identified in heparin‐treated samples, and 92 type 1, 28 type 2 and 17 GPI‐anchored proteins in CT (Figure 3A, Table S1). Pearson correlation revealed high reproducibility of quantifications between heparin‐treated and control samples, with an average coefficient of 0.906 (Figure 3B). Quantitative analysis found that 24 proteins annotated as secreted were more abundant in MEFs treated with heparin (Figure 3C, Table S1). Within this group, we found 18 are known heparin‐binding proteins such as CTGF, THBS1, CYR61, SLIT2 and PCOLCE (according to the glycosaminoglycan interactome 2.0 [11]). Moreover, 9 transmembrane type 1, 4 transmembrane type 2 and 2 GPI‐anchored proteins increased upon addition of heparin. Importantly, heparin did not affect shedding of ADAM17 substrates per se, since their abundance in conditioned media was not significantly altered by heparin treatment (Figure 3C).

FIGURE 3.

Heparin‐treated cells (HEP‐SEC) increased the number of proteins identified and quantified. (A) Bar chart showing the total number of proteins, and the number of proteins annotated as secreted or transmembrane (TM, including type 1, type 2 and GPI‐anchored) identified and quantified in unstimulated MEFs by HEP‐SEC or standard secretome analysis. (B) Representative Pearson correlation between conditioned media from HEP‐SEC and untreated controls (CT). (C) Volcano plot showing the log2 of protein ratio between heparin‐treated and CT MEFs versus the −log10 of p value of 1515 proteins (n = 6). Proteins with an FDR corrected p value below 0.05 are displayed with a grey open circle, and proteins above the p value with a grey filled circle. Known and putative ADAM17 substrates are displayed as filled pink and red circles, respectively. Heparin‐binding proteins that increased upon heparin treatment are displayed as filled green circles.

Then, we applied HEP‐SEC to PMA‐stimulated MEF cells for the identification of ADAM17 substrates. Thus, MEFs were treated with or without PMA for 3 h in the presence of heparin, and conditioned media were subjected to LC‐MS/MS analysis and label‐free quantification. A total of 1475 proteins were identified and quantified by the analysis. Fifty‐five single pass transmembrane/GPI‐anchored proteins were significantly increased upon PMA stimulation (Figure 4A, Table S1). Among them, 48 were type‐1‐transmembrane, 2 type‐2 and 5 GPI‐anchored. Eight of these proteins were also augmented by PMA stimulation in the conditioned media of ADAM17 KO cells (LMAN1, APP and LY75 detected by HEP‐SEC—Figure 4B, and CPD, APLP2, COLEC12, IGF2R and CDH2 detected by standard proteomics) and were therefore eliminated from the list of putative ADAM17 substrates. QARIP analysis matched identified peptides with the intracellular domain of five proteins: PTPRS, EFNB2, SCARF2, CANX, EPHA2 and ADAM10 (Figure 4C, Table S2). The intensity of peptides matching to the intracellular domain of EPHA2 and ADAM10 was not altered upon PMA stimulation, while CANX intracellular peptides decreased upon stimulation (Figure S3A, Table S2). The intracellular peptides of PTPRS, EFNB2 and SCARF2 were only identified in two biological replicates and therefore a similar analysis of their intensities could not be performed. Thus, we did not eliminate these proteins from the list of ADAM17 substrate candidates. Of these 47 single pass transmembrane/GPI‐anchored proteins that increased in the heparin‐treated conditioned media of PMA‐stimulated cells, 26 proteins, including several known ADAM17 substrates, were also found to be increased by standard secretome analysis. Conversely, 21 proteins were exclusively identified by HEP‐SEC, including H2‐D1, a protein that we have recently discovered as an ADAM17 substrate in macrophages [29] (Figure 4A,D, Tables 3 and 4, Table S1). In line with our hypothesis, HB‐EGF and CX3CL1, known ADAM17 substrates that bind to HS proteoglycans [10], were only detected by HEP‐SEC, not standard secretome analysis (Figure 4A,D, Table 3) [30, 31, 32]. In addition, HEP‐SEC identified 14 putatively novel ADAM17 substrates including interleukin‐1 receptor‐like 1 (IL1RL1) (Figure 4A,D, Table 4).

FIGURE 4.

Heparin‐treated cells (HEP‐SEC) identified additional novel ADAM17 putative substrates. (A) Volcano plot showing the log2 of protein ratio between PMA‐stimulated and control MEFs versus the −log10 of p value of 1475 proteins (n = 6) analysed by HEP‐SEC. Proteins with an FDR corrected p value below 0.05 are displayed with a grey open circle, and proteins above the p value with a grey filled circle. Single‐pass transmembrane/GPI‐anchored proteins with a higher abundance in PMA‐stimulated cells and therefore known or putative ADAM17 substrates, which were also identified by standard proteomics are displayed as filled pink and red circles, respectively. Known or potentially novel substrates that were exclusively detected by HEP‐SEC are displayed by light and dark blue, respectively. (B) Volcano plot showing the log2 of protein ratio between PMA‐stimulated and control ADAM17 KO MEFs versus the −log10 of p value of 1920 proteins (n = 6) analysed by HEP‐SEC. Proteins with an FDR corrected p value below 0.05 are displayed with a grey open circle, and proteins above the p value with a grey filled circle. Single‐pass transmembrane proteins with a higher abundance in PMA‐stimulated cells are displayed as black filled circles. (C) QARIP analysis of single pass transmembrane /GPI anchored proteins exclusively identified by HEP‐SEC as augmented in the conditioned media of MEFs upon PMA stimulation. (D) Venn diagram showing validated/putative substrates identified by standard secretome and HEP‐SEC analysis. Proteins in bold are putative novel substrates, while proteins in regular are validated ADAM17 substrates.

TABLE 3.

List of validated ADAM17 substrates exclusively identified by HEP‐SEC

Protein	Gene	ID	Ratio	p‐value	peptides
Cation‐independent mannose‐6‐phosphate receptor	Igf2r	Q07113	1.73	1.09E–08	59
Heparin‐binding EGF‐like growth factor	Hbegf	Q06186	8.89	1.52E–05	6
Amyloid beta precursor like protein 2	Aplp2	Q06335	2.02	5.72E–05	17
H‐2 class I histocompatibility antigen	H2‐D1	P01899	3.98	2.50E–04	15
CD44 antigen	Cd44	P15379	250	1.17E–04	5
Macrophage colony‐stimulating factor 1	Csf1	P07141	1.64	5.03E–03	15
Semaphorin 7A	Sema7a	Q9QUR8	1.55	7.78E–03	14
Monocyte differentiation antigen CD14	Cd14	P10810	1.59	1.92E–02	5
Fractalkine	Cx3cl1	O35188	1.61	2.71E–02	5

Note: The table contains a list of 9 proteins that were significantly increased in the secretome of heparin‐treated WT MEFs upon PMA stimulation and have already been validated as ADAM17 substrates (reviewed in (4, 5)). Indicated are the names of the proteins, the gene name, the protein ID, the mean of the ratio between PMA‐stimulated MEFs and untreated controls of 6 biological replicates, the p‐value calculated with a two‐sided, heteroscedastic t‐test based on the intensity ratios for PMA‐stimulated and control MEFs and the number of unique peptides.

TABLE 4.

List of putative novel ADAM17 substrates exclusively identified by HEP‐SEC.

Protein	Gene	ID	Ratio	p value	Peptides
Interleukin‐1 receptor‐like 1	Il1rl1	P14719	2.34	4.03E−06	7
V‐type proton ATPase subunit S1	Atp6ap1	Q9R1Q9	1.44	6.39E−05	7
Ephrin‐B2	Efnb2	P52800	2.38	0.000158	9
Scavenger receptor class F member 2	Scarf2	P59222	1.83	0.000258	12
Receptor‐type tyrosine‐protein phosphatase gamma	Ptprg	Q05909	2.26	0.001035	8
Cysteine‐rich motor neuron 1 protein	Crim1	Q9JLL0	1.30	0.001078	10
Ephrin type‐A receptor 2	Epha2	Q03145	1.54	0.00186	29
CD109 antigen	Cd109	Q8R422	1.79	0.002758	31
Plexin‐B2	Plxnb2	B2RXS4	1.33	0.006602	24
Lysosome‐associated membrane glycoprotein 1	Lamp1	P11438	1.71	0.006747	3
A disintegrin and metalloprotease 10	Adam10	O35598	1.33	0.007441	21
Exostosin‐1	Ext1	P97464	1.38	0.010327	3
Lysosome‐associated membrane glycoprotein 2	Lamp2	P17047	1.58	0.020775	7
Calnexin	Canx	P35564	1.63	0.055445	13

Note: The table contains a list of 14 proteins that were significantly increased in the secretome of heparin‐treated WT MEFs upon PMA stimulation and have NOT already been validated as ADAM17 substrates (reviewed in [4, 5]). Indicated are the names of the proteins, the gene name, the protein ID, the mean of the ratio between PMA‐stimulated MEFs and untreated controls of six biological replicates, the p value calculated with a two‐sided, heteroscedastic t‐test based on the intensity ratios for PMA‐stimulated and control MEFs and the number of unique peptides.

3.4. Validation of Novel ADAM17 Substrates Identified by HEP‐SEC Analysis

Among proteins identified by HEP‐SEC as ADAM17 substrates, H2‐D1 was one of the most regulated (Figure 4A, Table 3), in line with our previous results [29]. In contrast with HB‐EGF and CX3CL1, H2‐D1 has not been validated as an HSPG‐binding protein. Thus, we sought to confirm H2‐D1 as an ADAM17 substrate and to show that heparin increased its extracellular levels by orthogonal means. Recombinant H2‐D1 was ectopically expressed in WT MEFs, and cells stimulated with PMA to activate ADAM17‐mediated shedding, in the presence or absence of heparin. Levels of shed H2‐D1 were negligible in the conditioned media in the absence of heparin, while its levels were markedly augmented in the presence of the sugar (Figure 5A). Two HSPG‐binding proteins, TIMP‐3 and CTGF, showed a similar accumulation in the conditioned media of heparin‐treated cells, while shed vasorin, an ADAM17 substrate identified by standard secretome analysis, was evident in the conditioned media also in the absence of heparin and its levels were not markedly altered by the sugar. Furthermore, LeDock software (https://doi.org/10.1039/C6CP01555G) predicted a binding affinity of heparin to H2‐D1 of −1.58 Kcal/mol. H2‐D1 is a murine haplotype of major histocompatibility complex class I (MHC‐I) molecules, that in human is represented by the human leukocyte antigen (HLA) system, comprising three polymorphic genes (HLA‐A, ‐B and ‐C). LeDock predicted an affinity of HLA‐A to heparin of −3.70 Kcal/mol. We used PyMol to dock heparin onto H2‐D1 and visualize the basic amino acid residues Arg30, Arg135 and Lys267 that were predicted to be responsible for their interaction (Figure 5B). Then, we analysed the ADAM17‐dependent shedding of H2‐D1. H2‐D1 was expressed in WT MEFs, and levels of shed H2‐D1 increased upon PMA stimulation, while they decreased when stimulated cells were treated with the ADAM inhibitor TAPI‐1 (Figure 5C–E). These results were compatible with ADAM17 shedding and in line with the MS results. However, we could not further prove that H2‐D1 shedding was due to ADAM17 in this model as H2‐D1 was not found in the lysate of transfected ADAM17 KO MEFs (Figure 5C). Thus, to further confirm that MHC class I molecules, including H2‐D1, were shed by ADAM17, we analysed shedding of its human haplotype HLA (against which antibodies for Western blotting are available) in human fibroblast‐like HTB94 cells. ADAM17 was knocked out in HTB94 cells by CRISPR/Cas9 (Figure S3B). PMA induced an increase in HLA in the conditioned media of WT cells and did not alter HLA abundance in the lysate, thus indicating that the protein is subject to stimulated shedding, in line with results of MS and Western blotting analysis in murine fibroblasts. In contrast, PMA stimulation of ADAM17 KO HTB94 cells did not cause a similar increase in extracellular HLA. APP, a known ADAM17 substrate that was taken as a positive control, showed a similar behaviour, with its levels increased in the conditioned media of PMA‐stimulated WT cells but not PMA‐stimulated ADAM17 KO cells. These results confirmed that in human cells, MHC‐I shedding depends on ADAM17 (Figure 5F,G). Although PMA is commonly used to stimulate ADAM17 [14, 33], it is a pleiotropic activator of signalling pathways and an artificial way to stimulate the protease. Thus, we stimulated WT and ADAM17KO HTB94 cells with tumour necrosis factor (TNF), LPA and PDGFA, three known physiological stimuli of ADAM17 [14, 33], and analysed shedding of HLA by Western blotting. All stimuli had no effect on HLA shedding, while PMA clearly induced its shedding in these cells (Figure 5H), suggesting that the HTB94 cells are not very responsive to these stimuli and that the ability of physiological stimuli to induce ADAM17‐dependent shedding of HLA can be cell‐dependent.

FIGURE 5.

MHC class I molecules are shed by ADAM17. (A) Western blotting showing levels of shed H2‐D1 (sH2‐D1) in the conditioned media and full length H2‐D1 (flH2‐D1) in cell lysates of PMA‐stimulated MEFs, transfected with N‐terminally FLAG‐tagged H2‐D1, in the presence or absence of heparin. Heparin treatment increased levels of sH2‐D1 in the conditioned media (detected by an M2 α‐FLAG antibody), while its levels in the lysate did not change. Actin was used as a loading control. Similar to sH2‐D1, extracellular levels of TIMP‐3 and CTGF, two known heparin‐binding proteins, increased in the conditioned media when cells were treated with heparin. Conversely, levels of shed vasorin (sVASN), an ADAM17 substrate identified by standard proteomics, were not altered by heparin. (B) Computational prediction of the heparin‐binding site of H2‐D1. Amino acids that participated in strong interactions with the heparin fragment are highlighted in white. (C) Representative Western blot and band quantification showing (D) the increase of sH2‐D1 in the conditioned media of H2‐D1‐FLAG transfected MEFs upon stimulation with 25 ng/mL PMA (n = 9), and (E) the reduction of sH2‐D1 in the conditioned media of PMA‐stimulated H2‐D1‐FLAG transfected MEFs in the presence of the ADAM inhibitor TAPI‐1. Levels of full length H2‐D1 (flH2‐D1) in the lysate were not altered by TAPI‐1 treatment. Calnexin (CANX) was used as a loading control. Uncropped blots showing full‐length H2‐D1 in the lysate and shed H2‐D1 in the conditioned media of transfected cells are displayed in Figure S4. (F and G) Blots and relative band quantifications showing levels of HLA and the known ADAM17 substrate APP, in the conditioned media and cell lysates of ADAM17 KO or wild‐type control HTB94 cells (WT), stimulated with or without 25 ng/mL PMA. Levels of shed HLA and APP increased in the conditioned media of CT cells upon PMA‐stimulation, while their levels did not change in the conditioned media of ADAM17 KO cells. Levels of full‐length APP (flAPP) in the lysate of WT cells slightly decreased upon PMA stimulation, compatible with APP being an ADAM17 substrate, and levels of full length HLA molecules (flHLA‐ABC) did not change. Levels of both flAPP and flHLA did not change in the lysate of ADAM17 KO cells. Calnexin (CANX) was used as a loading control. (H) WT and ADAM17KO HTB94 cells were stimulated with 25 ng/mL PMA, 10 ng/mL TNF, 10 µM LPA or 50 ng/mL PDGFA. After 3 h of stimulation, conditioned media were collected, proteins precipitated with TCA and loaded onto a gel for SDS‐PAGE electrophoresis and Western blotting. Lysates were collected in STET buffer, 10 µg of which was loaded onto a gel. A representative Western blot of shed HLA in the conditioned media and full‐length HLA in the lysate of WT and ADAM17KO HTB94 cells is displayed. Calnexin (CANX) was used as a loading control.

Next, we analysed heparin‐binding and shedding of IL1RL1, a putative ADAM17 substrate identified by HEP‐SEC. IL1RL1 was expressed in HTB94 cells and, similar to H2‐D1, levels of its ectodomain accumulated in the conditioned media of PMA‐stimulated cells when treated with heparin (Figure 6A). Heparin treatment did not lead to a similar increase of shed PTK7, a known ADAM17 substrate that was detected by standard proteomics. LeDock predicted an affinity of IL1RL1 to heparin of −4.41 kcal/mol (as a reference, a similar analysis predicted an affinity of −2.23 kcal/mol for HB‐EGF). Furthermore, computational docking indicated that the basic amino acid residues Arg35, Arg 198 and Lys203 may interact with the sulphate groups of heparin (Figure 6B).

FIGURE 6.

Validation of IL1RL1 as a novel substrate of ADAM17. (A) Western blotting showing levels of shed IL1RL1 (sIL1RL1) in the conditioned media and full length IL1RL1 (flIL1RL1) in cell lysates of PMA‐stimulated HTB94 cells, transfected with N‐terminally FLAG‐tagged IL1RL1, in the presence or absence of heparin. Heparin treatment increased levels of sIL1RL1 in the conditioned media (detected by an M2 α‐FLAG antibody), while its levels in the lysate did not change. Actin was used as a loading control. Heparin had negligible effect on levels of shed PTK7 (sPTK7), an ADAM17 substrate identified by standard proteomics. Uncropped blots showing full‐length IL1‐RL1 in the lysate and shed IL1RL1 in the conditioned media of transfected cells are displayed in Figure S4. (B) Computational prediction of the heparin‐binding site of IL1RL1. Amino acids that participated in strong interactions with the heparin fragment are highlighted in white. (C‐D) Western blots and relative band quantification showing levels of IL1RL1‐FLAG in conditioned media and lysates of transfected ADAM17KO and control HTB94 cells, in the presence or absence of PMA. When stimulated with 25 ng/mL PMA, levels of shed Il1RL1 increased in the conditioned media of control cells, while such an increase was not evident in ADAM17KO cells. IL1RL1 levels in the lysate of both cell lines did not change upon PMA stimulation. Calnexin (CANX) was used as a loading control. (E) WT MEFs were stimulated with 25 ng/mL PMA, 10 ng/mL TNF, 10 µM LPA or 50 ng/mL PDGFA. After 3 h of stimulation, conditioned media were collected, applied to FASP and generated peptides analysed by LC‐MS/MS. iBAQ intensity was used to compare the abundance of IL1RL1 in the conditioned media of differentially stimulated cells. Two‐way ANOVA was used for statistical analysis of the results (**p < 0.01; ***p < 0.005, ns = non‐significant).

Extracellular levels of shed IL1RL1 increased in PMA‐stimulated WT cells while its levels in the lysate did not change. Conversely, when IL1RL1‐expressing ADAM17 KO cells were stimulated with PMA, levels of IL1RL1 in the conditioned media and lysates were unaltered, indicating its shedding did not occur in ADAM17 KO cells and thus validating IL1RL1 as a novel substrate of ADAM17 (Figure 6C,D). Finally, we evaluated whether the physiological stimuli by TNF, LPA and PDGFA induced shedding of IL1RL1 by MS. As shown in Figure 6D, similar to PMA, TNF increased levels of shed IL1RL1 in the conditioned media of MEFs, while LPA and PDGFA were ineffective.

4. Discussion

Since its discovery as the TNF convertase, ADAM17 has been considered of major pharmaceutical interest. Aberrant release of TNF is associated with several autoimmune diseases, and, subsequently, ADAM17 ablation is beneficial in murine models of rheumatoid arthritis, psoriasis, bowel disease and many others [4]. Furthermore, ADAM17 plays a crucial role in the progression of various cancers as it induces EGFR activation by releasing its ligands, and consequently, its ablation has been proven beneficial in several disease models. In humans, genetic deficiency of ADAM17 causes severe skin and intestinal barrier defects, most likely due to lack of EGFR signalling [34, 35]. Similarly, because of reduced EGFR activation, ADAM17 KO mice are born with open eyes and die perinatally [36]. Although shedding of EGFR‐ligands is considered functionally dominant during development, ADAM17 cleaves over 80 different transmembrane proteins, ranging from cytokine and growth factors to cell receptors and adhesion molecules. Full characterization of the ADAM17 sheddome will have the dual effect of uncovering new biological functions of the protease and, potentially, better understanding the molecular basis of pathological conditions that are associated with dysregulated ADAM17 activity. High‐resolution proteomics has proved a powerful tool for the systematic identification of protease substrates [6]. For sheddases such as ADAM17, the ectodomain of its cleaved substrates are released into the conditioned media, and so levels of the ectodomains increase in the conditioned media when the protease is activated. High‐resolution quantitative proteomics can be used to quantify this increase, and consequently, to identify proteins cleaved by the protease. In this study, we applied an advanced approach for the quantitative analysis of secretome proteins to murine fibroblasts in which ADAM17 is activated by PMA. This allowed us to comprehensively characterize the sheddome of ADAM17 in these cells. Besides 19 already validated substrates that supported the robustness of the analysis, we uncovered 15 novel putative substrates that would further broaden the list of biological functions of ADAM17. Among them, the most regulated were matrix remodelling associated 8 (MXRA8), which has been recently uncovered as the entry receptor for arthritogenic viruses [37], and ε‐sarcoglycan, whose function is still largely unknown although its mutation leads to a rare movement disorder called myoclonus dystonia [38]. Both proteins have not been previously reported to be shed by proteases. Chondroitin sulphate proteoglycan 4 (CSPG4), fibronectin leucine‐rich repeat transmembrane protein (FLRT2), dystroglycan (DAG1) and transferrin receptor protein 1 (TFRC) were found well above the FDR curve in our analysis, suggesting that shedding of these proteins can also be mediated by ADAM17. CSPG4, a protein that plays an important role in regulating neuronal networks, was previously shown to be processed by ADAM10 [39]. Several transmembrane proteins that are constitutively shed by ADAM10 can also undergo stimulated shedding by ADAM17 [1], and we speculate that this is also the case for CSPG4. FLRT2, DAG1 and TFRC are also known to undergo shedding, although the sheddase(s) responsible have not been identified to date. FLRT2 modulates migration of cortical neurons during development, and its ectodomain shedding by an undefined metalloproteinase is pivotal in regulating this process [40]. DAG1 shedding from keratinocytes is promoted by IL‐1β and phorbol esters and inhibited by metalloproteinase inhibitors [41]. TFRC can also be released from the cell membrane of leukocytic cell lines by an undefined metalloproteinase [42]. Our results strongly indicate that ADAM17 is a main sheddase for FLRT2, DAG1 and TFRC. The other nine candidate substrates of ADAM17, NRP2, PCDH19, DCBLD2, BSG, LMAN2, SLC3A2, RPN1, KIRREL and PTPRS fall close to the FDR curve in our dataset. This can occur when a protein is released by multiple proteases, and the activation or ablation of one of them does not lead to a marked alteration in their abundance [6]. This is likely to be the case for BSG, which is known to be an ADAM12 substrate [43]; DCBLD2 and KIRREL, which are released by the rhomboid protease RHBDL2 [44]; and RPN1, which is released by RHBDL4 [45]. Previous studies have shown that shedding of PTPRS, NRP2, PCDH19, LMAN2 and SLC3A2 is promoted by stimuli known to activate ADAM17 (e.g., phorbol esters, LPS, neural activity – reviewed in refs. [4, 5]), or in pathological conditions characterized by aberrant ADAM17 activity, supporting our identification of these as putative ADAM17 substrates. For example, NRP2 shedding in macrophages was induced by LPS [46], and the epilepsy‐associated protein PCDH19 was released in response to neural activity [47]. Similarly, PTPRS was reported to be released in response to phorbol esters [48]. Moreover, serum levels of shed LMAN2 increase in sepsis patients [49], and shedding of SLC3A2 is increased in both lung adenocarcinoma and lung squamous cell carcinoma patients compared to healthy volunteers [50].

Despite the overall soundness of the data gathered using standard proteomics, we observed that several known ADAM17 substrates, which are expressed in fibroblasts, were not detected in the secretome. Most recent mass spectrometers have reached a very high resolution, especially when coupled with advanced acquisition methods, enabling identification of an unprecedented high number of proteins from complex protein samples, such as secretomes. Yet, detection of low‐abundance proteins can be difficult, given the large dynamic concentration range of proteins in typical secretome samples. Sheddase substrates can fall into this low abundance category, due to localisation on the cell surface or inefficient cleavage by sheddases. In such cases, methods to reduce sample complexity by protein enrichment are required to identify candidate substrates by secretome analysis. For instance, most of the conventionally shed/secreted proteins are highly glycosylated, and so glycocapturing or other glycan‐labelling methods have proved effective in enriching shed/secreted proteins, thereby allowing systematic identification by secretome analysis of substrates of different sheddases, such as BACE1 and ADAM10 [6, 51, 52]. In addition to proteins that are lowly expressed or cleaved, soluble levels of several extracellular proteins are low as they can bind to components of the ECM or cell membrane‐associated proteins and/or proteoglycans. This has already been shown for several ECM‐binding proteins, including TIMP‐3 and CTGF, and we reasoned that it could also be the case for soluble ectodomains of proteins that are cleaved at the cell surface. Thus, in this study, we developed a novel workflow for secretome analysis that is particularly suitable for the detection of proteins that, due to binding to cell‐associated HSPGs, can hardly be found in conditioned media. We exploited the ability of heparin to solubilize HSPG‐bound proteins into the secretome, thus enhancing their detection by MS. Heparin can also increase abundance of proteins in the medium by inhibiting their endocytosis and lysosomal degradation via the endocytic receptor low‐density lipoprotein receptor‐related protein 1 (LRP1). LRP‐1 is known to internalize about 50 extracellular proteins, and their binding to the receptor is often mediated by the same amino acid residues that are responsible for binding to heparin (among others [53, 54]). Thus, in addition to displacing proteins from pericellular HSPGs, heparin can potentially inhibit endocytosis of proteins by LRP‐1, further increasing their levels in the secretome and facilitating their MS detection.

First, we confirmed that heparin had only minor effects on sample preparation and LC‐MS/MS analysis. We found that it slightly reduced detection of peptides with missed cleavages, but not that of fully tryptic peptides. We speculate that this effect could be due to the ability of heparin to bind to peptides with a higher number of positive charges, thereby reducing their recovery after FASP, but we cannot exclude that heparin binding allows the target proteins to be more accessible for tryptic digestion. The identity and number of proteins detected in HEP‐SECs were not significantly altered compared to CT. We then went on to perform LC‐MS/MS analysis of HEP‐SECs to detect ADAM17 substrates that may bind to HSPGs. As hypothesised, HEP‐SEC identified a number of heparin‐binding proteins that are known (e.g., HB‐EGF, CX3CL1, CD44 – reviewed in [4, 5]) or potentially novel (e.g. PTPRG, CD109, LAMP1 and PLXNB2) ADAM17 substrates not detected by standard analysis [11]. In addition, among the group of ADAM17 substrates that were exclusively detected by HEP‐SEC, there were three proteins known to bind other sulphated glycosaminoglycans (GAGs), namely, EPHA2 and EFNB2 that bind to keratan sulphate and LAMP2 that binds to dermatan sulphate [11]. Although their direct binding to heparin has not been proved, it is likely that these proteins can also bind to HSPGs or be displaced from their interaction with KS or DS proteoglycans by more highly sulphated heparin. Another group of four ADAM17 substrates identified by HEP‐SEC has not been reported to bind heparin, but shows high homology with family members that bind the sugar. H2‐D1 falls into this group, as the human haplotypes of MHC class I molecules HLA‐A and HLA‐B are reported to bind heparin [11]. Similarly, CSF1 exhibits almost 60% sequence homology with the heparin/HSPG‐binding CSF2 [11, 55]; and family members of SEMA7A, including SEMA5A and SEMA5B, bind heparin and other HS, CS and KS proteoglycans [11, 56]. Moreover, a number of components of multisubunit ATPases (ATP6V0A1, ATP6V0A2, ATP60D1, etc.) bind heparin, suggesting that ATP6AP1 may also be a heparin‐binding protein [11]. The remaining putative ADAM17 substrates identified by HEP‐SEC, including IL1RL1, CRIM1, ADAM10 and SCARF2, are not listed in publicly available databases of heparin‐binding proteins (e.g., MatrixDB, Uniprot and those reviewed in refs. [10, 11]). These databases are, however, incomplete, as new experimental evidence continues to increase the list of known heparin‐binding proteins. Our results support this possibility, because H2‐D1 and IL1RL1 accumulated in conditioned media when cells were treated with heparin and in silico docking analysis inferred heparin binding sites for both proteins. Alternatively, it is also possible that these putative substrates may not interact with heparin directly, but that heparin may improve their identification via a different mechanism. For example, we found that addition of heparin reduced identification of peptides that were not fully digested by trypsin (i.e., peptides with one or more missed cleavages). About 30% of all peptides identified in high‐resolution proteomic experiments contain one or more tryptic sites, which negatively influences protein identification and for which improved digestion protocols and/or bioinformatic tools are continuously sought [57, 58]. Heparin may reduce identification of such incompletely cleaved peptides by promoting trypsin activity, or by increasing retention of such peptides on the filter column used for sample processing, potentially through formation of higher‐order complexes between negatively‐charged heparin and the positively‐charged peptides.

In conclusion, we developed a proteomic method, HEP‐SEC, which allowed better identification of HSPG‐binding ADAM17 substrates due to the ability of heparin to increase their levels in the medium, as we showed for H2‐D1 and IL1RL1. HEP‐SEC could also be applied to other sheddases, whose HSPG‐binding substrates could be difficult to identify by standard proteomic methods, and, more generally, represents a useful method for characterizing the abundance of cell‐associated HSPG‐binding proteins, including growth factors, chemokines, and other bioactive molecules. HEP‐SEC MS could thus be a powerful tool to investigate HSPG interactomes and functions in response to physiological and pathological stimuli. Combined with transgenic approaches like Crispr/Cas9 to modulate expression of specific HS polymerases and sulfotransferases, HEP‐SEC could also deepen our understanding of how specific HS modifications and structures drive the HSPG‐binding repertoire in a complex biological context.

Author Contributions

Matteo Calligaris, Donatella Pia Spanò and Maria Chiara Puccio designed, acquired, analysed and interpreted the data and edited the manuscript. Stephan A. Müller aided with the generation and interpretation of proteomic data. Simone Bonelli generated the ADAM17 KO cells. Margot Lo Pinto processed samples for mass spectrometry. Giovanni Zito, Carl P. Blobel and Stefan F. Lichtenthaler interpreted data and edited the manuscript. Linda Troeberg and Simone Dario Scilabra designed the study, provided supervision, interpreted data and prepared the manuscript. All authors have read and agreed to the published version of the manuscript.

Conflicts of Interest

C.P.B. holds a patent on a method of identifying agents for combination with inhibitors of iRhoms. C.P.B. and the Hospital for Special Surgery have identified iRhom2 inhibitors and have co‐founded the start‐up company SciRhom in Munich to commercialize these inhibitors.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Data Availability Statement

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD055223 and PXD055002.