Identification of Novel Natural Substrates of Fibroblast Activation Protein-alpha by Differential Degradomics and Proteomics

of and proteolytic activity. sites Quantitative of and shows that has a role in Fibroblast activation protein-alpha (FAP) is a cell-surface transmembrane-anchored dimeric protease. This unique, constitutively active serine protease has both dipeptidyl aminopeptidase and endopeptidase activities and can hydrolyze the post-proline bond. FAP expression is very low in adult organs but is upregulated by activated fibroblasts in sites of tissue remodeling, including fibrosis, atherosclerosis, arthritis and tumors. To identify the endogenous substrates of FAP, we immortalized primary mouse embryonic fibroblasts (MEFs) from FAP gene knockout embryos and then stably transduced them to express either enzymatically active or inactive FAP. The MEF secretomes were then analyzed using degradomic and proteomic techniques. Terminal amine isotopic labeling of substrates (TAILS)-based degradomics identified cleavage sites in collagens, many other extracellular matrix (ECM) and associated proteins, and lysyl oxidase-like-1, CXCL-5, CSF-1, and C1qT6, that were confirmed in vitro . In addition, differential metabolic labeling coupled with quantitative proteomic analysis


In Brief
Fibroblast activation protein-␣ (FAP) can hydrolyze the postproline bond. We identified endogenous substrates of FAP in fibroblasts that were previously naive to both FAP and its proteolytic activity. FAP-dependent cleavage sites were identified in many extracellular matrix (ECM) and ECM-associated proteins including collagens and lysyl oxidase-like-1, and CSF-1, CXCL-5 and C1qT6. Quantitative proteomic analysis implicated FAP in ECM-cell interactions, coagulation and metabolism. This study greatly expands the repertoire of FAP substrates and shows that FAP has a role in coagulation in the mouse.

Highlights
• Natural substrates of FAP were identified using degradomic and proteomic techniques and FAP gene knockout mouse derived embryonic fibroblasts stably transduced with enzymatically active or inactive FAP.
• Terminal amine isotopic labelling of substrates (TAILS) based degradomics identified cleavage sites in collagens, and many other extracellular matrix (ECM) and associated proteins.
• Differential metabolic labelling coupled with quantitative proteomic analysis implicated FAP in regulating proteins that are associated with ECM, ECM-cell interactions, coagulation, metabolism and wound healing.

Identification of Novel Natural Substrates of Fibroblast Activation Protein-alpha by Differential Degradomics and Proteomics* □ S
Fibroblast activation protein-alpha (FAP) 1 is a unique postproline peptidase, which possesses both endopeptidase ac-tivity and dipeptidyl aminopeptidase (DPP) activity, allowing it to cleave a peptide bond two or more residues from the N terminus of a peptide or protein (1,2). FAP is expressed at the cell surface and also shed. Both forms are enzymatically active and have similar substrate preferences (3). Although FAP has attracted much interest because of potential roles in fibrosis, cell migration and apoptosis and as a potential marker for many cancers (4 -6), few of its natural substrates have been identified (1,4,(7)(8)(9)(10)(11).
Only three of the identified endopeptidase substrates of FAP are physiological substrates, namely denatured type I collagen (1,12), ␣2-antiplasmin (3) and fibroblast growth factor-21 (FGF-21) (8 -10). By screening known DPP4 substrates (7), and then measuring cleavage in plasma (13), we have previously identified four physiological DPP-type substrates of FAP, namely neuropeptide Y (NPY), substance P, peptide YY (PYY) and B-type natriuretic peptide (BNP). NPY was found to be the most efficiently cleaved substrate of both human and mouse FAP, whereas all four peptides were found to be efficiently cleaved by endogenous DPP4, indicating that the in vivo degradomes of FAP and DPP4 differ (13).
Identification of additional FAP substrates is essential to fully elucidate its role as a protease. FAP is implicated in fibrinolysis because human ␣2-antiplasmin has 4-fold increased activity following FAP mediated cleavage (14). Inhibition of FAP increases the levels of active FGF-21 in obese mice more than in lean mice, demonstrating a metabolic benefit of FAP inhibition (15), which might be exploited for treating diabetes. Up-regulated expression of FAP is predominantly associated with disease states, including but not lim-ited to tumorigenesis, fibrotic conditions and atherosclerosis (5,6,16,17). Therefore, FAP is emerging as a promising marker of disease and much attention focuses upon exploiting this protease as a therapeutic target (4 -6, 18). Nevertheless, the functions of FAP in both physiological and pathological processes remain unclear. Improved knowledge of the basic biology of FAP is necessary to broaden our perspective on potential outcomes of targeting this protease and predict potential impacts on the proteome when FAP enzyme activity is inhibited or ablated.
Several degradomic techniques have been developed for the discovery of physiological substrates of proteases for the DPP4 family (4, 19 -22). The overall aim of this study was to identify novel substrates of FAP using an unbiased, global, degradomic technique (terminal amine isotopic labeling of substrates, or TAILS), a more targeted data mining approach, and a proteome comparison (stable isotope labeling by amino acids in cell culture, or SILAC). Through our identification of FAP substrates and subsequent investigation of the biological consequence of cleavage of these proteins, knowledge and understanding of the physiological role of FAP was enhanced.

EXPERIMENTAL PROCEDURES
Experimental Design and Statistical Rationale-A total of five replicate analyses were conducted for the differential proteomic investigation of the FAP enzyme-active (eϩ) and FAP enzyme-inactive (eϪ) mouse embryonic fibroblasts (MEFs) (see below for details of cell generation). A total of three replicate analyses were conducted for the differential N-terminomic investigation of the FAP eϩ and FAP eϪ MEFs (see below for details of cell generation). Statistical analysis using linear models for microarray data (limma) (23) allows for the use of linear models to assess differential expression in the context of multifactor designed experiments. In addition, limma enables analyses of complex experiments involving comparisons between many peptides simultaneously in a small sample size, including multiple testing correction. Following the calculation of moderated-t-tests, proteins or cleavage sites with a p value Ͻ 0.05 and a quantitative alteration Ͼ 50% (either increase or decrease), were considered as being significantly affected.
Generation of FAP Enzyme-active (eϩ) and FAP Enzyme-inactive (eϪ) MEFs-MEFs were isolated from FAP gene knockout (GKO) mouse embryos aged 13.5 days. In this FAP GKO mouse strain, FAP has been deleted in both alleles and exons 3 to 26 are absent (24,25). Primary MEFs were immortalized by transduction with the pFU-SV40-LTpuro lentiviral vector encoding the SV40 large T antigen then selected in puromycin (26). These MEFs were then divided into three cultures and transduced with a lentiviral vector pCCLteteGFP to express green fluorescent protein (GFP) as an empty vector control, or linked to one of two human FAP (hFAP) constructs using a picornaviral 2A peptide (27). PCR was used to amplify enzyme active (hFAP eϩ) and inactive (hFAP eϪ; serine724alanine) constructs from GFP-FAP (28) using the primers hFAP-5ЈBmgBI 5Ј-GTGGAA-GAAAACCCCGGGCCCATGAAGACTTGGGTAAAAATCG-3Ј and hFAP-3ЈBstBI 5Ј-AAAAGATTCGAAGTTTAAACTTAGTCTGACAAAGAGA-AACAC-3Ј. PCR fragments (2324bp) were generated using Phusion polymerase (Finnzymes; Thermo Scientific, Waltham, MA) phosphorylated with T4 polynucleotide kinase (NEB, Ipswich, MA) and digested with BstBI. The pCCLteteGFP2Aluc2 lentiviral vector was cut with BmgBI/ClaI to clone in hFAP eϩ and hFAP eϪ genes. Transduced MEFs were selected by GFP-based cell sorting using a BD Biosciences FACS Aria flow cytometer and expanded as pools. An endopeptidase activity assay was performed on the cells and secretomes of FAP eϩ and FAP eϪ MEFs using the fluorogenic substrate Z-Gly-Pro-AMC (1 mM; Bachem, Bubendorf, Switzerland), in the presence or absence of a non-selective DPP inhibitor (Val-boro-Pro at 50 M (kind gift from WW Bachovchin, Boston, MA) and a PEP-selective inhibitor (S17092 at 1 M; Sigma-Aldrich, St Louis, MO).
Collection of Cell Conditioned Medium-Cell conditioned medium (CCM) was generated from three biological replicates of FAP eϩ and eϪ MEF cell lines. Cells were grown in 15 cm dishes to a confluency of ϳ70% in 20 ml culture medium. After 16 h incubation in serum-free supplemented DMEM (Thermo Scientific) without phenol red, CCM was harvested in the presence of the following protease inhibitors: 0.01 mM trans-epoxysuccinyl-l-leucylamido (4-guanidino) butane (E64; Sigma-Aldrich), 5 mM ethylene-diaminetetraacetic acid (EDTA; Sigma-Aldrich) and 1 mM phenylmethylsulfonylfluoride (PMSF; Sigma-Aldrich). CCM was centrifuged at 1500 rpm and filtered using a 0.2 m filter (Amicon; EMD Millipore, Billerica, MA) to remove dead cells and debris. HEPES buffer (Sigma-Aldrich) (pH 7.5) was added to a final concentration of 100 mM and the CCM mildly denatured and reduced with 5 mM dithiothreitol (DTT) (Sigma-Aldrich) at room temperature for 1-2 h. Free sulfide groups were alkylated by incubation with 20 mM iodoacetamide (IAA) (Sigma-Aldrich) for 2 h at room temperature in the dark, then 5 mM DTT was added for 15 min at room temperature to quench any remaining IAA.
Stable Isotope Labeling by Amino Acids (SILAC) In Cell Culture-The SILAC protocol was performed as described (33). Briefly, MEFs were grown in heavy or light labeled DMEM (minus L-Arg and L-Lys; Silantes, Munich, Germany) supplemented with 10% dialyzed FCS and glutamine containing either L-arginine (Arg0) and L-lysine (Lys0) "light" or 13 C6 15 N4 L-arginine (Arg10) and 13 C6 15 N2 L-lysine (Lys8) "heavy" for at least five cell population doublings. After 16 h of seeding cells in heavy or light medium, CCM was collected as described above. Differentially labeled CCM was mixed at a ratio of 1:1 and concentrated to ϳ200 l using Amicon® Ultra 50 ml 5 kDa centrifugal filters (EMD Millipore), 4000 g, 4°C. Samples were run on 12-16% SDS-PAGE (Criterion Tris-HCL gel, Bio-Rad and NuPage® buffers, Thermo Scientific) and proteins visualized using a Colloidal Blue staining kit (Thermo Scientific). After destaining, gels were cut into 1 mm cubes and further destained with three rounds of 0.1 M ammonium bicarbonate, 50% acetonitrile and 50 mM ammonium bicarbonate then 100% acetonitrile, sonicating on ice. Gel pieces were digested with sequencing grade trypsin (Worthington) at pH 7.5, overnight at 37°C. After digestion, peptides were eluted from the gel with 100% ethanol, samples were then concentrated using a Speedvac concentrator (model 5301, Eppendorf, Hamburg, Germany) and desalted using a reversed phase C18 column (GRACE-VYDAC, Deerfield, IL).
Liquid Chromatography-MS/MS and Data Analysis-TAILS samples were then prefractionated by high performance liquid chromatography (HPLC) using a strong cation exchange column (SCX; PolyLC, Columbia, MD). Peptides were eluted in a linear gradient with increasing concentration of 5 mM KH 2 PO 1 M KCl 25% acetonitrile pH 2.7; 10 -14 fractions were collected and desalted using self-packed C18 STAGE tips (Empore; 3 M, St Paul, MN) (34). Differential proteomics samples were pre-fractionated by SDS-PAGE as described previously (31). For nanoflow-LC-MS/MS analysis, samples for quantitative proteomic comparison were analyzed on a Qstar Elite (AB Sciex, Framingham, MA) and TAILS samples on an Orbitrap XL (Thermo Scientific GmbH, Bremen, Germany) mass spectrometer. Both instruments were coupled to an Ultimate3000 micro pump (Thermo Scientific) with a flow rate of 300 nL/min each. Buffer A was 0.5% (v/v) acetic acid, and buffer B was 0.5% (v/v) acetic acid in 80% (v/v) acetonitrile (water and acetonitrile were at least HPLC gradient grade quality). A gradient of increasing organic proportion was used for peptide separation. Column-tips with 75 m inner diameter and a length of 11 cm were self-packed with Reprosil-Pur 120 ODS-3 (Dr. Maisch, Ammerbuch, Germany). The MS operated in data dependent mode, with the SMART being used for the QSTAR Elite. For the Orbitrap XL, each MS scan was followed by a maximum of five MS/MS scans; and for the Qstar Elite three MS/MS scans.
Raw QSTAR LC-MS/MS data in wiff format (quantitative proteomic comparison, SILAC) was converted to the mzXML (35) format, using mzWiff (version 4.3.1, http://sourceforge.net/projects/sashimi/files/ mzWiff%20(Analyst%20converter), with centroiding of MS1 and MS2 data, precursor charge determination, and deisotoping of MS2 data. X!Tandem (version 2013.09.01) (36) was used for spectrum to sequence assignment. The proteome database consisted of (1) mouse UniProt sequences (without isoforms) downloaded from UniProt on April 15th, 2016, (2) the sequence of human FAP, and (3) the contaminant database of MaxQuant (total of 50935 entries). It was appended with an equal number of shuffled decoy entries derived from the original human protein sequences. The decoy sequences were generated with the software DB toolkit (37). X!Tandem parameters for analysis of QSTAR data included: asymmetric precursor mass error of Ϫ50 -150 ppm, fragment ion mass tolerance of 0.2 Da, tryptic cleavage specificity with no missed cleavage sites, residue modifications: cysteine carboxyamidomethylation (ϩ57.02 Da), variable modifications were isotope-labeled (ϩ6.02 Da) arginine and lysine. X!Tandem results were further validated by PeptideProphet (38). Peptides were assembled to proteins using ProteinProphet (39) at a false discovery rate of less than 1%. XPRESS was used for relative peptide and protein quantification (40). Mass tolerance for quantification was 0.1 Da. Ratios of each dataset were calculated as FAP-active/FAP-inactive and log 2 -transformed. Five biological replicates were analyzed. Prior to limma statistical analysis (see Statistical Rationale above), samples were mean normalized. No outlier removal was performed. Isoform quantification was solved by the well-established XPRESS algorithm, which assigns weights to peptides accordingly.
Raw Orbitrap-XL LC-MS/MS data (TAILS, stable isotope labeling with dimethylation) was converted to the mzXML (35) format, using msconvert (41) with centroiding of MS1 and MS2 data and deisotoping of MS2 data. For spectrum to sequence assignment X!Tandem (version 2013.09.01) (36) was used, with the proteome database described above. X!Tandem parameters for analysis of Orbitrap-XL data included: precursor mass error of Ϯ 10 ppm, fragment ion mass tolerance of Ϯ 0.4 Da, semi-ArgC specificity with up to three missed cleavages, static residue modifications: cysteine carboxyamidomethylation (ϩ57.02 Da); lysine and N-terminal dimethylation (light formaldehyde 28.03 Da; heavy formaldehyde 34.06 Da or 36.08 Da; no variable modifications. These X!Tandem results were further validated by PeptideProphet at a confidence level of Ͼ 95%. XPRESS was used for relative peptide quantification (36,38,40). Mass tolerance for quantification was 0.015 Da. Ratios of each data set were calculated as FAP-active/FAP-inactive and log 2 -transformed. Three biological replicates were performed. Prior to limma statistical analysis (see Statistical Rationale above), samples were mean normalized. No outlier removal was performed. An in-house PERL computer script (downloadable at http://www.mol-med.uni-freiburg.de/mom/schilling/ TAILS_v21_xpress-only/at_download/file) then filtered for peptides that had a non-tryptic N terminus generated. Like the original PICS procedure (42), the script then determines bioinformatically, through database lookup, the corresponding prime-or non-prime sequence. The script also used Uniprot and Gene Ontology annotation to distinguish secreted proteins. Web-PICS was used to generate heat-map style representation of protease specificity (43).
Recombinant peptides and proteins at the stated amount were incubated with rhFAP or buffer control in 25 mM Tris and 0.25 M NaCl (pH 8.0) for up to 24 h at 37°C. At each time point, a 1 l sample of enzyme/substrate solution was pipetted onto the surface of a ground steel target plate, allowed to dry and then overlaid with 0.5 l of sinapinic acid matrix solution (10 mg/ml in a solution of 50:50 0.1% trifluroacetic acid and 100% acetonitrile and then sonicated until completely dissolved). Matrix-Assisted-Laser-Desorption-Ionisation (MALDI) Mass Spectrometry analysis was performed as previously described (7,13). Once dry, the target plate was loaded into a Bruker UltafleXtreme MALDI TOF/TOF (Bremen, Germany) and analyzed with the following settings: Linear mode, Mass Range 5-50 kDa, Detector Gain 40ϫ, Sample Rate 0.63 GS/s, Smartbeam Parameter "medium," Frequency 2000 Hz and realtime smoothing set to "high," Laser shots 1000 (or 5ϫ 1000 shots for dilute samples). Once mass spectra were acquired, post processing was performed In Flex analysis (Bruker Daltonics, Germany) with TopHat baseline subtraction followed by spectral smoothing using 1 cycle of the Savitzsky Golay algorithm at a width of 5 m/z. Peak detection was performed manually by labeling the monoisotopic peak.
Immunofluorescence Staining, Confocal Microscopy and Real Time qPCR-The FAP eϩ and eϪ MEFs were stained with the F19 antibody and imaged using a Leica TCS SP5 confocal microscope as described previously (44). Real time qPCR on DPP4, DPP8, and DPP9 were performed as described previously (47) using Taqman gene expression assays (Applied Biosystems; Thermo Scientific), with ␤-actin as a standard.
Leukocyte Migration Assays-Neutrophils and monocytes were isolated from mouse bone marrow as described previously (48,49). Bones harvested from C57BL/6J mice were crushed in sort buffer (2.5% Fetal Bovine Serum [FBS, Hyclone, GE Life Sciences, Pittsburg, PA], 2 mM EDTA in PBS) followed by centrifugation at 500 ϫ g, 5 min, 4°C. For monocyte isolation, cells were then resuspended in 4 ml PBS and underlaid with 3 ml Histopaque® 1083 (Sigma-Aldrich) before centrifuging at 400 ϫ g, 30 min, 22°C without braking. The interface enriched with mononuclear leukocytes was carefully aspirated and washed twice in 10 ml sort buffer at 500 ϫ g, 5 min, 4°C. For neutrophil isolation, cells were immediately subjected to red blood cell lysis by adding 1.5 ml ACK lysis buffer (145 mM NH 4 Cl, 0.1 mM EDTA, 12 mM NaHCO 3 in triple-distilled water) for 2.5 min at room temperature before washing twice in 10 ml sort buffer at 400 ϫ g, 5 min, 4°C. Cells were then stained at 4°C for 50 min with the following antibodies: B220 PE and APC (clone RA3-6B2, BD Biosciences, San Diego, CA), CD115 PE (clone AFS98, eBioscience, San Diego, CA), Ly6C FITC (clone HK1.4, BioLegend, San Diego, CA) and/or Ly6G FITC and A647 (clone 1A8, BD). Cells were washed twice in 10 ml sort buffer at 400 ϫ g, 5 min, 4°C before resuspending in sort buffer containing 0.5 g/ml DAPI (4Ј,6-Diamidino-2-Phenylindole dihydrochloride, #D1306 Invitrogen, Eugene, OR). Monocytes or neutrophils were subsequently sorted on a 7-laser Influx (BD) or 3-laser FACSAria II (BD), whereby DAPI-live cells were initially selected, followed by the exclusion of B220ϩ and Ly6Gϩ cells. Subsequently, for monocyte isolation, CD115ϩ monocytes were selected, and for neutrophil iso-lation, Ly6Gϩ neutrophils were selected. Cells were collected in 2 ml FBS before usage for transwell migration assay. CytoSelect™ 24-well cell migration assay with 5 m pore size (CBA-102, Cell Biolabs, Inc., San Diego, CA) was performed according to the manufacturer's instructions (50). Sorted cells (1.5 ϫ 10 6 cells/well) were added to the insert and chemoattractant was added to the lower well of the migration plate. To examine monocyte migration, mCCL-2 was pre-incubated with buffer or rhFAP at 37°C for 24 h. Where FAP enzyme activity was inhibited, FAP inhibitor ARI-3099 (51) at 20 M was preincubated with rhFAP for 20 min at 37°C. To examine neutrophil migration, mCXCL-5 was preincubated at 37°C for 24 h with buffer or rhFAP in the presence of plasma from FAP enzyme-negative gene knock-in (gki) mice (52). FAPgki mouse plasma contains naturally occurring proteases and cytokines and so is a physiological environment for analyzing substrate cleavage. Cells were allowed to migrate for 1.5 h at 37°C, then cells from cells in the lower well, including cells on the underside of the membrane, were lysed and stained with the CyQuant® GR dye. Fluorescence was read with a fluorescence plate reader at 480 nm excitation/520 nm detection (POLARstar Omega, BMG Labtech). Chemotaxis index (fold change over the media control) was calculated.
Coagulation Assay-Wildtype (WT) and FAP GKO mice were anesthetized using isoflurane then blood was collected via cardiac puncture into tubes at a ratio of 9:1 blood:citrate (3.5%; v/v). Plasma were assayed for prothrombin time (PT) and activated partial thromboplastin time (APTT) by the Institute of Hematology, Royal Prince Alfred Hospital, Australia. Full blood counts on WT, DPP4 GKO and FAP GKO mice were performed by veterinary pathology diagnostic services of the University of Sydney.
Annotated spectra of the TAILS data and of the global quantitative proteome comparison are provided at MS-viewer (54). The experiments and the corresponding Search Keys are listed in supplemental Table S1.

Generation of FAP Enzyme-active (eϩ) and FAP Enzymeinactive (eϪ)
MEFs-To investigate the enzymatic activity of FAP with a focus on identifying novel natural substrates of FAP, cell lines suitable for proteomic and degradomic studies were designed and produced. Uniquely, we used FAP GKO cell lines so that substrates had no exposure to endogenous FAP. We have detected no compensatory changes in DPP8, DPP9 or DPP4 mRNA expression in FAP GKO mice (47). Three biological replicates of the FAP eϩ and eϪ FAP MEF cell lines were generated. Confocal microscopy (Fig. 1A) and flow cytometry (Fig. 1B) indicated cell surface expression of human FAP and intracellular expression of GFP by the FAP eϩ and FAP eϪ mouse cell lines. FAP eϩ and FAP eϪ cells expressed similar amounts of hFAP mRNA, and no hFAP mRNA was detected in the empty vector control cells (Fig.  1C). Only whole live FAP eϩ MEFs (Fig. 1D) and CCM (Fig. 1E) contained FAP enzyme activity. These data confirmed the MEF phenotypes and the serine to alanine catalytic inactivation of FAP.
Expression and activity levels of closely related proteases were then measured in these cell lines, because when one protease is downregulated, compensatory mechanisms might upregulate enzymes that have similar functions. Specifically, DPP4, DPP8, and DPP9 share DPP activity with FAP, and PEP

FAP Substrate Profiling
Molecular & Cellular Proteomics 18. 1 69 exhibits comparable endopeptidase activity; although PEP, DPP8, and DPP9 are exclusively intracellular (2,4,44,55,56). We observed no significant difference in mRNA levels of dpp8 or dpp9, and dpp4 was downregulated to a small extent in the FAP eϪ MEFs compared with FAP eϩ MEFs (supplemental Fig. S2A). PEP activity was not detected in the CCM (supplemental Fig. S2B), so it would be very unlikely that this protease influenced the degradomic analyses of MEF CCM. TAILS N-terminome Analysis on Secretomes of FAP eϩ and FAP eϪ MEF Cell Lines-To identify FAP-dependent proteolytic cleavage sites, we employed the TAILS method for an N-terminomic analysis of CCM produced by FAP eϩ MEFs compared with FAP eϪ MEFs. Three biological replicate analyses were conducted. Identified N termini are summarized in supplemental Tables S2-S4.
In line with earlier TAILS studies (57), we focused on protein N termini that were identified in at least two replicates. Of those 1914 N termini, 1340 mapped to proteins that are annotated in Uniprot as being secreted but not exclusively in exosomes. Compared with previous N-terminomic studies on cultured cells, this was a high proportion of secreted or shed proteins (58). Therefore, our statistical analysis focused on cleavage sites in secreted or shed proteins.
To identify cleavage sites that significantly differ in their abundance depending on FAP activity, we employed limma statistics (59). Limma is superior to a classical Student's t test (e.g. with regard to multiple testing correction and prevention of false-positive discoveries) (60) in the analysis of omics-style data, which is characterized by a large number of analytes in a small number of biological replicates (Fig. 2). We chose the following criteria to distinguish significantly regulated cleavage sites: (1) limma moderated p value Ͻ 0.05 and (2) average increase or decrease in abundance by more than 50% (log2 (FAP-active/FAP-inactive) Ͼ 0.58 for increase; log2 (FAP-active/FAP-inactive) Ͻ Ϫ0.58 for decrease. These strict criteria highlighted 308 up-regulated and 278 downregulated cleavage sites upon increased FAP activity (limma statistics in supplemental Table S5). Log-transformed fold-change distributions before and after normalization appear in supplemental Fig. S3.
Proteases act in networks with strong interdependency (61). Often, altered activity of a protease under investigation has downstream and secondary effects, e.g. stemming from inactivation or activation of other proteases. The sequence specificity of a cleavage site is a useful criterion to distinguish direct substrates of a protease under investigation from downstream effects (62).
FAP predominantly acts as an amino-dipeptidase with a strong preference for P1 proline together with small amino acids (mostly glycine, to a lesser extent alanine) in P2 (63). Endoproteolytic activity has been reported as well (3), but in an in vivo or cell-contextual setting, endoproteolytic activity of FAP remains difficult to distinguish from primary cleavage by a different endoprotease (e.g. a matrix metalloprotease (MMP)), followed by processing of the neo N termini by FAP.
Specificity profiling of the 308 "upregulated" cleavage sites that depend upon FAP activity highlights a noticeable preference for P1 proline and P2 glycine (43) (Fig. 3A and 3B). These data correspond to the reported Gly-Pro cleavage site specificity of FAP (12,63,64). These data also provide some insight into the non-prime amino acids surrounding the cleavage site, including the unique observation of a preference for serine or threonine in P1Ј.
Identification of Candidate FAP Substrates by Parsing TAILS Data-In the set of 308 "upregulated" cleavage sites that depend on FAP activity, we focused on cleavage sites with a P1 proline to identify bona fide FAP substrates.
Candidate Substrates Exhibiting Postproline Cleavage-Following the criteria outlined above, 37 bona fide FAP cleavage sites were identified (Table I, supplemental Table S6). Among the candidate substrates is type 1 collagen (CN-I), which is a well-known substrate of FAP (1,12). Thus, identification of neo-N termini produced from cleavage events after a proline in CN-I from the FAP eϩ secretome acted as an internal positive control and demonstrated the reliability of the TAILS method. Cleavage events were also observed in collagen types III and V.
Further FAP cleavage sites in extracellular matrix components were found in fibrillin-2 and extracellular matrix protein 1 (ECM-1). In fibrillin-2, the cleavage occurred at position 60, in the propeptide domain, after a Gly-Pro sequence. In ECM-1, cleavage occurred at position 142 after a Lys-Pro sequence, which renders a preceding cleavage by a different endoprotease likely. Cleavage of lysyl oxidase homolog 2 (LOX-L2) by FAP was seen near the genome-encoded N terminus at position 36 after a Tyr-Pro sequence that follows the signal peptide.
In addition, there was prominent processing of complement C1q tumor necrosis factor-related protein 6 (C1qT6) by FAP at position 26 after a Val-Pro sequence. After removal of the signal peptide, secreted C1qT6 commences with this Val-Pro sequence. Thus, FAP acts as an aminodipeptidase on C1qT6.
Lastly, TAILS provided evidence for processing by FAP of CXCL-5 at position 42, which is after an Ala-Pro sequence. CXCL-5 has an unusually long signal peptide of 40 residues, so Ala-Pro is the N terminus of secreted CXCL-5. Thus, FAP acts as an aminodipeptidase on CXCL-5.
Cleavage Sites Lacking a P1 Proline That Are Affected By FAP Activity-FAP activity had a strong effect on numerous cleavage sites that do not represent its known active site specificity. There are clear data that FAP can cleave postalanine as well as post-proline (21,22), but cleavage after other residues may be downstream effects of FAP rather than substrates directly hydrolyzed by FAP. Nevertheless, this "downstream degradome" constitutes an important aspect of FAP biology. For example, FAP activity leads to substantially elevated levels of MMP-1 in cancer-associated fibroblasts while simultaneously reducing the levels of tissue plasminogen activator (tPA) (63). MMP-1 and FAP co-operatively digest collagen (12).
Many neo-N termini lacking a P1 proline were affected by FAP activity (supplemental Table S5). These include the ECM proteins fibronectin, CN -I, -III, -V and -VI, laminin, fibulin -1, -4 and -7, fibrillin -1 and -2, filamin -A and -B, plectin, periostin, procollagen C-endopeptidase enhancer 1, the MMP inhibitor TIMP-1, and LOX-L1 and LOX-L2. The LOX enzymes are extracellular copper-dependent amine oxidases that catalyze the first step in the formation of crosslinking in collagens and elastin. LOX enzymes, TIMPs and MMPs alter the ECM, so FAP enzyme activity dependent changes in the relative abundance of any of those enzymes are very likely to cause FAP enzyme activity dependent indirect alterations of abundance or properties of many ECM proteins in CCM.

SILAC Applied to FAP eϩ and FAP eϪ MEF Secretomes and Identification of Candidate FAP Substrates-TAILS
highlighted that FAP activity had a pronounced impact on cellcontextual extracellular proteolysis. This finding prompted us to investigate effects of FAP activity on the composition of the MEF-derived cell conditioned medium (CCM) ("secretome"). For this purpose, we employed a QSTAR device, which was enough for medium proteome coverage. In data interpretation, we were particularly interested in the altered abundance of proteases and protease inhibitors. Differential quantitative proteomics used SILAC in five biological replicates (proteins listed in supplemental Tables S7-S11; peptides listed in supplemental Tables S12-S16).   As outlined for the TAILS approach, we employed limma statistics to delineate proteins that displayed significantly affected abundance in the presence of FAP enzyme activity. We chose the following criteria to distinguish significantly regulated proteins: (1) identification and quantitation in at least three replicate experiments, (2) annotation as a secreted or shed protein, (3) limma moderated p value Ͻ 0.05 and (4) average increase or decrease in abundance by more than 50% (log2 (FAP-active/FAP-inactive) Ͼ 0.58 for increase; log2 (FAP-active/FAP-inactive) Ͻ -0.58 for decrease). These criteria highlighted 31 more abundant and 33 less abundant proteins with increased FAP activity (limma statistics in supplemental Table S17).
Proteins That Were Differentially Abundant In the FAP eϩ MEF secretome-Compared with FAP eϪ, the FAP eϩ secretome contained more MMP-2 ("72 kDa type IV collagenase") and MMP-3 (Stromelysin-1). In addition to our initial interest in proteases and their inhibitors, we observed increased levels of CSF-1 and monocyte chemoattractant protein-1 (MCP-1/ CCL-2) (p limma ϭ 0.055 -but increased in all 5 replicates and further considered because of its biological importance). This impact on CCL-2 is similarly observed in cancer associated fibroblasts (CAFs), where FAP activity leads to substantially elevated levels of CCL-2 (63).
Among proteins that were less abundant in the secretome of FAP eϩ compared with FAP eϪ MEFs of particular interest were annexins A1 and A2 and the collagen -binding protease inhibitor serpin H1 (also seen with limma statistics on TAILS data; supplemental Table S5).
Validation of Candidate Substrates Found By TAILS-Selected candidate substrates identified through TAILS analysis were analyzed by in vitro cleavage assay. Recombinant proteins or synthetic N-terminal peptides corresponding to the secreted form of the protein were incubated with recombinant human FAP (rhFAP), and cleavage was then monitored over time by MALDI-TOF-MS analysis.
Secreted CXCL-5 (41-132) has Ala-Pro at the N terminus that TAILS identified as removed by FAP (Fig. 4A). In vitro cleavage using recombinant mouse CXCL-5 validated that observation. All the CXCL-5 lacked the Ala41-Pro42 after 24 h incubation with rhFAP (Fig. 4B). To examine a biological outcome of FAP-mediated cleavage in ex vivo physiological conditions, neutrophil migration assays were performed using CXCL-5 pre-incubated with FAP gki plasma, into which rhFAP or buffer (control) was added (Fig. 4C). The naturally occurring CXCL-5 in the FAP gki plasma increased neutrophil migration by 2-fold. Following FAP-mediated cleavage, the chemotactic activity of CXCL-5 was not altered (Fig. 4C).
C1qT6 is a 264 -residue protein of four domains and the secreted protein begins with a Val25-Pro26 sequence (Fig.  5A) that TAILS identified as removed by FAP. FAP-cleavage of C1qT6 was confirmed in vitro (Fig. 5B) using a synthetic peptide that represents the N-terminal 21 residues of secreted C1qT6 . Incubation with rhFAP yielded a cleav-age product with a reduced mass of 218 Da within 5 min, with the mass shift corresponding to the removal of Val-Pro. The peptide was almost completely converted to 218 Da by 10 min (Fig. 5B), and within 30 min N-terminal processing of the C1qT6-derived peptide was complete (data not shown). This in vitro cleavage assay substantiates the enzymatic ability of FAP to process the N terminus of C1qT6.
The cleavage of LOX-L1 was observed at position 511 after a Lys-Ala sequence, with a Ͼ 50% average increase in abundance of the cleaved form (p Ͻ 0.05) (supplemental Table S5). Immunoblotting detected more LOX-L1 in the FAP eϪ than in the FAP eϩ MEF CCM (Fig. 6). The whole cell lysates lacked this difference between the FAP eϩ and eϪ MEFs (Fig. 6), consistent with the cleavage of LOX-L1 by FAP occurring extracellularly.
Confirming FAP Alters the MEF Secretome-The secretome analysis highlighted elevated levels of shed CSF-1 in the FAP eϩ MEF secretome (supplemental Table S17). CSF-1 is important in the proliferation, differentiation, and survival of monocytes and macrophages. CSF-1 is a type I cell-surface protein that is released into the extracellular milieu by endoproteolytic shedding. We therefore probed cell-surface levels of CSF-1 using flow cytometry analysis and an antibody specific for the (potentially shed) N terminus of CSF-1 (Lys33-Glu262). That analysis showed 5-fold more cells immunopositive for CSF-1 in the FAP eϪ than in the FAP eϩ MEF (Fig.  7A). Concordantly, immunoblotting on MEF cell lysates for CSF-1 showed that more cellular CSF-1 was in FAP eϪ than eϩ MEFs (Fig. 7B). A possible explanation is that elevated shedding of CSF-1 in the FAP eϩ MEFs leads to its depletion from the cell surface.
However, our data do not show that FAP directly acts as a CSF-1 sheddase. Although one TAILS experiment identified a P1-Pro, FAP-dependent cleavage site in CSF-1 in proximity to its annotated shedding site (Fig. 7C), FAP did not cut a synthetic peptide sequence that encompasses this sequence motif (Fig. 7D). Therefore, the impact of FAP on CSF-1 shedding is possibly indirect.
The present and previous proteomic studies strongly indicate that FAP activity results in increased levels of CCL-2 (63). Both human and mouse CCL-2 proteins have potential postproline cleavage sites after Gln24-Pro25 and Ala30-Pro31. Full-length human and mouse CCL-2 were incubated with rhFAP and specific cleavage after Proline31 was shown by LC-MS/MS (Fig. 8A-8C). Monocyte migration assays assessed biological effects of this cleavage event. CCL-2 was able to induce monocyte migration. However, FAP, with or without the presence of a FAP specific inhibitor, did not alter the chemotactic activity of CCL-2 (Fig. 8D).
Candidate Substrates from Data Mining-Data mining and literature-based approaches were also utilized to search for candidate FAP substrates. FGF-21 was chosen as a candidate FAP endopeptidase substrate. Unlike most other candidate substrates, the cleavage site within FGF-21 is located within a dozen residues of the C-rather than N terminus (Fig.  9A). Recombinant human FGF-21 (hFGF-21) matching residues 29 -209 (lacking the signal peptide) was analyzed in MALDI-TOF-MS and LC-MS/MS analyses to investigate whether this protein undergoes FAP cleavage. By MALDI, the doubly charged ion was identified (peak at ϳ9785 Da) and after 10 min incubation with FAP a smaller peak corresponding to hFGF-21 lacking the last 10 amino acids (ϳ9296 Da) was observed, and after 2 h incubation hFGF-21 was completely digested by FAP (Fig. 9B). No cleavage was observed in the absence of FAP (data no shown). To confirm hFGF-21 as a FAP substrate in vitro, LC-MS/MS analysis was performed using the enzyme AspN rather than trypsin. Only AspN-cleaved peptides were identified in the control. In the FAP-treated sample the peptide C-terminal to the cleavage site was identified but the peptide N-terminal to this cleavage site was not detected. However, 33.7% sequence coverage of FGF-21 was observed in the untreated sample, whereas only 5.5% sequence coverage was observed in the FAP-treated sample. Taken together, the LC-MS/MS data and MALDI-TOF-MS data demonstrated that hFGF-21 undergoes in vitro cleavage by FAP after proline 171 (Fig. 9C).
Influence of FAP on Coagulation-The present work highlights a strong impact of FAP on extracellular proteolysis (both directly and indirectly). Cleavage of human ␣2-antiplasmin by FAP has implicated the activity of FAP in fibrinolysis (14). In the present study, FAP affected the proteolytic processing of several mouse proteins with functions in coagulation, such as prothrombin and thrombospondin-2 (supplemental Table S5), indicating that FAP might influence coagulation. Therefore, plasma was isolated from WT and FAP GKO mice and prothrombin times (PT) and activated partial thromboplastin times (APTT) were measured. PT examines the extrinsic coagulation pathway, whereas APTT test examines the intrinsic coagulation pathway. No significant difference in prothrombin times between WT and FAP GKO mouse plasma was observed (Fig. 10A). However, a significant increase in activated partial prothrombin times was observed in the FAP GKO plasma compared with WT plasma (p ϭ 0.033; Fig. 10B).
Full blood counts on WT and FAP GKO mice showed no difference in platelet numbers or any other major blood cell type (supplemental Table S18), which is consistent with the initial description of this mouse strain (24,65). DISCUSSION A large number of FAP-driven cleavage events were identified using TAILS. A strength of the approach taken was the production of a cellular source of substrates, from GKO MEFs, that had never been exposed to FAP or its enzyme activity. Many of the detected cleavage events are likely to be direct cleavage by FAP, whereas others may be caused indirectly. Global differential proteomics showed differential abundance of several proteases and protease inhibitors that probably contributed indirect cleavage events to our data.
Much of the detected post-proline cleavage by FAP was of collagens and other ECM proteins, including CN-I, CN-III, CN-V, fibrillin-2 and ECM-1 (Table I). Fibroblasts are the main cellular source of ECM proteins, so it is sensible that many ECM proteins were detected in CCM from MEFs. Nevertheless, the preponderance of ECM associated candidate substrates was the dominant finding of this study.
In addition, many proteolytic events were observed in a range of ECM proteins that did not correspond to post-proline processing, suggesting either that FAP is not restricted to cleavage after a proline and/or that FAP drives increased levels of other ECM -degrading proteases. The many ECM associated proteins identified as differentially abundant include collagens, laminins, fibronectin, periostin, cofilin-1, cathepsin-B, fibrillin-1, fibulin-1, ECM-1, serpin-H1, LOX-L1, NCAM-1 and the ECM degrading enzymes MMP-2 and MMP-3 and MMP inhibitors TIMP-1 and TIMP-2, in the FAP eϩ or FAP eϪ secretome (supplemental Table S17). These data may point to a FAP-MMP-TIMP axis of ECM regulation. The lysyl oxidases are of particular interest because they drive collagen crosslinking (66), which is crucial in liver fibrosis (67). In addition to directly causing degradation of LOX-L1 (Fig. 5), we saw that FAP altered the levels of BMP-1 and periostin (supplemental Table S17), which can promote LOX activation (68). Considered together, the data suggests that when FAP eϩ is expressed it results in greater levels of ECM proteolysis, ECM degrading enzymes and their inhibitors and other ECM associated proteins, which points to a significant role for FAP in the regulation of ECM degradation. These insights are consistent with the known importance of FAP in ECM-dependent cancer cell and fibroblast invasion and migration (28, 69 -72) and more generally in cell -ECM interactions (4,12,73,74). Annexin A2 is involved in fibrinolysis (75), and has been shown to interact with FAP (76,77). Several proteins found to be cut or differentially abundant in the presence of FAP enzyme activity are important in liver fibrosis, including CYR-61 (CCN-1) (78,79), and in iron storage, including ferritins and lipocalin (LCN2) (80). These data, together with a recent study (63), highlight that differential FAP activity has a profound effect on the secretome, hence characterizing FAP as a key player in shaping the pericellular environment.
Post-translational modifications (PTM) modify proteins including interaction between a substrate and an enzyme. Although synthetic proteins or peptides may not possess PTM

FAP Substrate Profiling
Molecular & Cellular Proteomics 18. 1 75 that occur in vivo, they are often useful for initial substrate validation. Mouse CXCL-5 contains a DPP substrate cleavage site at Ala-Pro at the N terminus after the signal peptide is removed. The secreted form of mCXCL-5 has 92 amino acids  and can be further cleaved into two isoforms that have 78 residues (41-118) and 70 residues (49 -118) respectively. These truncated isoforms are more potent than the secreted CXCL-5 (41-132) at inducing neutrophil infiltration in vitro (81). Because the FAP cleavage site is quite close to the ELR motif that is important for receptor binding, FAP-truncated CXCL-5 might show enhanced chemotactic activity as well. However, neutrophil chemotaxis was not different between intact and FAP-truncated CXCL-5. Thus, dipeptide removal from the N terminus of CXCL-5 does not affect the biological potency of this chemokine. C1qT6 is a clear candidate substrate, with the mature N terminus present in the FAP eϪ secretome and the proteasegenerated neo-N terminus in the FAP eϩ secretome (Fig. 4). C1qT6 is a member of the C1q/TNF-related protein (CTRP) family, which contains adiponectin (82). C1qT6 also has roles in coagulation and arthritis (83), possibly hepatocellular car-cinoma (84), and metabolic disorders (85). Although removal of just two N-terminal amino acids might not have a significant effect on C1qT6, because the known biological function of this protein is induced via its C-terminal domain (86), DPP cleavage of C1qT6 might confer upon this protein increased susceptibility to further digestion by other proteases, leading to production of the naturally occurring truncated form.
The in vitro endopeptidase cleavage by FAP of both human and mouse CCL-2 ( Fig. 7) is the first report of a natural FAP endopeptidase substrate being cleaved after an alanine-proline sequence. CCL-2 has ubiquitous pro-inflammatory actions (70,87,88) through binding to its cognate receptor CCR2. Removing the first 8 amino acids off the N terminus of CCL-2 converts it into a receptor antagonist that retains the ability to bind to CCR2 but not the ability to elicit a chemotactic response (87,89). Therefore, a truncated CCL-2 analog has been developed as an anti-inflammatory therapeutic (90,91). The cleavage of CCL-2 by FAP, resulting in the removal of the first 8 amino acids, would convert it into a dominant negative receptor antagonist. However, chemotaxis ability of   (Fig. 9). Administering a selective FAP inhibitor acutely increases circulating intact FGF-21 levels in cynomolgus monkeys (8) and prolongs the half-life of circulating human FGF-21 in mice (10), therefore FGF-21 is a physiological substrate for FAP. Proteins that were identified by TAILS and have roles in lipid or energy metabolism include proprotein convertase subtilisin/kexin type 9 (PCSK-9), pyruvate kinase PKM, insulin-2, PYY, clusterin (apolipoprotein J) and low-density lipoprotein receptor 8 (LRP-8) (supplemental Table S5). Thus, FAP may have important influences on energy metabolism. FAP has a well-established role in clotting and fibrinolysis in humans and this role is mediated by cleavage of ␣2-antiplasmin (11,14), but other FAP-interacting molecules might be involved (77). Mouse ␣2-antiplasmin lacks the FAP cleavage site and a role for FAP in clotting in the mouse had not been reported. However, our degradomics and proteomics data pointed to the potential for a role for FAP in coagulation in the mouse. Concordantly, we observed that, as in humans (14), depletion of FAP from plasma slows the clotting time in mice (Fig. 10). The proteins that we identified as potentially mediating this physiological outcome are FSTL-1 (92), C1qT6 (83), SNED-1, filamin-A, serpin-G1, thrombospondin-1, thrombospondin-2, and PAI-1 (supplemental Tables S5 and S17). Our finding that mouse plasma that lacks FAP exhibited slowed coagulation is consistent with the identification of a number of potential FAP substrates that are coagulation -associated. In vitro examination of candidate substrate human CCL-2 (hCCL-2) and mouse CCL-2 (mCCL-2) using LC-MS/MS. A, Schematic of the primary structure of hCCL-2. The Uniprot-annotated N terminus sequence is shown with the proposed FAP cleavage site indicated by a red arrow after Pro31. B, C, Database-derived amino acid sequences of recombinant hCCL-2 (B) and mCCL-2 (C) are shown with red arrows indicating the FAP cleavage sites. The peptide sequence that is colored red corresponds to a peptide produced by FAP-mediated cleavage. Black and gray alternating peptides represent the peptides potentially produced by digestion with trypsin. Recombinant hCCL-2 and mCCL-2 were incubated with rhFAP (ϩ FAP) or a buffer control (no FAP) for 4 h at 37°C. Samples were then treated with trypsin and analyzed by LC-MS/MS. 98% sequence coverage of hCCL-2 was observed in both samples, whereas 14% sequence coverage of mCCL-2 was seen, and the truncated peptides matching to the FAP-cleavage products were identified only in the samples that had been treated with FAP. Some trypsin cleavages were missed (*). Representative data from two technical replicates. D, Monocyte migration assay using mCCL-2 pre-incubated at 37°C for 24 h with buffer (control), rhFAP, or rhFAP with a specific FAP inhibitor (FAPi). Data are representative of two independent experiments. zymes, rather than by direct FAP-mediated cleavage, may provide greater understanding of the pathogenesis of tumors and of fibrotic disorders.  9. In vitro examination of candidate substrate human FGF21 (hFGF21). A, Schematic of the primary structure of hFGF21. The Uniprot-annotated N terminus sequence is shown with proposed FAP cleavage site indicated by a red arrow after Pro199. B, MALDI-TOF-MS spectra showing cleavage after Gly-Pro199. The ϳ9786 Da peak corresponds to the doubly charged full-length protein. After 10 min incubation with rhFAP at 37°C, cleavage resulted in the ϳ9296 Da peak corresponding to the full-length protein lacking the last 10 amino acids. FGF-21 was fully digested at 120 min incubation with rhFAP. C, Database-derived amino acid sequence of rhFGF21, with a red arrow at the FAP cleavage site and the red peptide corresponds to the peptide produced from FAP cleavage. Black and gray alternating peptides represent the peptides produced after digestion with AspN. Recombinant hFGF21 was incubated with rhFAP (ϩ FAP) or a buffer control (no FAP) for 30 min at 37°C, digested with AspN and analyzed by LC-MS/MS. The peptides identified in each sample are listed; 33.7% sequence coverage of rhFGF21 was observed in the untreated sample whereas only 5.5% sequence coverage was observed in the rhFAP-treated sample, and the truncated peptide matching to the FAP-cleavage product was identified only in the sample that was treated with rhFAP. Representative of two technical replicates.