Cathepsin D exacerbates SPARC-driven aggressiveness by limited proteolysis in triple-negative breast cancer

Tumor-specific molecular targets and alternative therapeutic strategies for triple-negative breast cancer (TNBC) are urgently needed. The protease cathepsin D (cath-D) is a marker of poor prognosis in TNBC and a tumor-specific extracellular target for antibody-based therapy. The identification of cath-D substrates is essential for the mechanistic understanding of its role in TNBC and future therapeutic developments. Using degradomic analyses by TAILS, we discovered that the matricellular protein SPARC is a substrate of extracellular cath-D. In vitro, cath-D induced limited proteolysis of SPARC C-terminal extracellular Ca2+ binding domain at acidic pH, leading to the production of SPARC fragments (34-, 27-, 16-, 9-, and 6-kDa). Similarly, cath-D secreted by human TNBC and mouse mammary cancer cells cleaved fibroblast- and cancer-derived SPARC at the tumor pericellular pH. SPARC cleavage also occurred in vivo in TNBC and mouse mammary tumors. Among these fragments, the C-terminal 9-kDa SPARC fragment inhibited MDA-MB-231 TNBC cell adhesion and spreading on fibronectin, and stimulated their migration, endothelial transmigration and invasion more potently than full-length SPARC. These results highlight a novel crosstalk between proteases and matricellular proteins in the TNBC microenvironment through limited proteolysis of SPARC, and reveal that the 9-kDa C-terminal SPARC fragment is an attractive therapeutic target for TNBC. Significance We show that cath-D-mediated limited proteolysis of SPARC promotes its pro-tumor activity in TNBC. Our study will pave the way for the development of strategies for targeting bioactive fragments from matricellular proteins in TNBC.


INTRODUCTION
Breast cancer (BC) is one of the leading causes of death in women in developed countries. Triplenegative breast cancer (TNBC), defined by the absence of oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER-2) overexpression and/or amplification, accounts for 15-20% of all BC cases (1). Chemotherapy is the primary systemic treatment, but resistance to this treatment is common (1). Thus, tumor-specific molecular targets are urgently needed to develop alternative therapeutic strategies for TNBC.
In a previous work using TAILS (18), we isolated the matricellular protein SPARC (Secreted Protein Acidic and Rich in Cysteine), also known as osteonectin or basement membrane 40 (BM40), as a putative cath-D substrate. SPARC is a Ca 2+ -binding glycoprotein that regulates extracellular matrix assembly and deposition, growth factor signalling, and interactions between cells and their surrounding extracellular matrix (19)(20)(21)(22). In cancer, SPARC is mainly secreted by the neighbouring stroma, but also by cancer cells (23)(24)(25). In different cancer types, SPARC plays an oncogenic or a tumor-suppressive role (26,27). For instance, in BC, SPARC has a pro-tumorigenic role and has been associated with worse prognosis (24, [28][29][30][31][32][33]; however, other studies reported anti-tumorigenic functions (34-36). SPARC includes three different structural and functional modules: the N-terminal acidic domain, followed by the follistatin-like domain, and the C-terminal extracellular Ca 2+ binding domain (21). Protein fragments that correspond to these SPARC domains display distinct biological functions in cell de-adhesion and spreading, motility, proliferation, invasion, and in matrix remodelling (21,22,37,38). This suggests that SPARC activity may be modulated by proteolysis, leading to the unmasking of domains with biological functions that are distinct from those described for the full-length (FL) protein.
Here, we found that in the acidic tumor microenvironment of TNBC, cath-D cleaved SPARC exclusively in its C-terminal extracellular Ca 2+ binding domain releasing five main fragments (34-, 27-, 16-, 9-, and 6-kDa). Among these fragments, the 9-kDa C-terminal SPARC fragment (amino acids 235-303) had greater oncogenic activity than FL SPARC, highlighting the importance of limited proteolysis of matricellular proteins in the TNBC microenvironment. This knowledge might pave the way to the development of strategies to target the bioactive fragments of matricellular proteins in cancer.

Identification of SPARC as an extracellular protein affected by cath-D deficiency
Using the TAILS approach, we previously analysed the secretome of immortalized Ctsd -/-MEFs stably transfected with empty vector (Ctsd -/-) or a human cath-D (Ctsd -/cath-D) plasmid to determine cath-D effect on extracellular protein processing (18). We noticed that the SPARC peptide LDSELTEFPLR [156-166] was 5.2-fold less abundant in the secretome of Ctsd -/cath-D MEFs than of Ctsd -/-MEFs (Fig. 1A). To determine whether SPARC is a putative cath-D substrate, we first confirmed that SPARC protein level was reduced in the Ctsd -/cath-D secretome compared with the Ctsd -/secretome (Fig. 1B). Transcriptome analysis of Ctsd -/and Ctsd -/cath-D MEFs, as previously published (18), showed that SPARC reduction in the Ctsd -/cath-D secretome was not due to Sparc gene downregulation in the presence of cath-D (Fig. 1C). These data showed that SPARC protein level in the extracellular environment is reduced in the presence of cath-D.

In vitro, cath-D cleaves SPARC extracellular Ca 2+ binding domain at acidic pH
We investigated whether recombinant cath-D can cleave recombinant SPARC in vitro at acidic pH.
At pH 5.9, SPARC was hydrolysed by cath-D in a time-dependent manner ( Fig. 2A). Moreover, experiments in which pH was gradually reduced from 6.8 to 5.5 showed progressive limited proteolysis of SPARC at lower pH (Fig. 2B). In these two experiments, pepstatin, an aspartic protease inhibitor, inhibited SPARC cleavage by cath-D ( Fig. 2A-B). By amino-terminal oriented mass spectrometry of substrates (ATOMS) analysis, we found that SPARC was hydrolysed by the 51-kDa cath-D form exclusively in its extracellular Ca 2+ binding domain, releasing five main SPARC fragments (34-, 27, 16, 9-, and 6-kDa) at pH 5.9, detected by silver staining (Fig. 2C-E, Table 1). We detected SPARC cleavage fragments of similar size also after incubation with the fully mature 34+14-kDa cath-D form at pH 5.9 ( Fig. 2C-E, Table 1). Thus, in vitro, cath-D triggers the limited proteolysis of SPARC exclusively in its extracellular Ca 2+ binding domain in an acidic environment.

SPARC and cath-D expression in TNBC
To study the pathophysiological relevance of the SPARC/cath-D interplay in TNBC, we first assessed the clinical significance of SPARC and CTSD (the gene encoding cath-D) expression in a cohort of 255 patients with TNBC using an online survival analysis (39). High CTSD mRNA level was significantly associated with shorter recurrence-free survival (HR=1.65 for [1.08-2.53]; p=0.019) ( Supplementary Fig. S1, top panel), as previously observed (16). Similarly, high SPARC mRNA level tended to be associated with shorter recurrence-free survival (HR=1. 6 3B). Cath-D was expressed by TNBC and HMF cells (Fig. 3B, left panel), but was secreted only by TNBC cells (Fig. 3B, right panel). Conversely, SPARC was expressed and secreted by HMF, but only by two out of five TNBC cell lines, namely SUM159 and HS578T (Fig. 3B). Finally, we investigated SPARC and cath-D co-localization in a TNBC patient-derived xenograft (PDX B1995) (40) in which cath-D expression was previously demonstrated (16). Co-labelling with polyclonal anti-SPARC and monoclonal anti-cath-D antibodies (Fig. 3C) showed that SPARC (in red; panel a) partially co-localized with cath-D (in green; panel b) in the PDX B1995 microenvironment (merge; panel c). Together with previously published data on SPARC (24,28-32,41) and cath-D (2)(3)(4)6,7,(9)(10)(11)(13)(14)(15)42) in BC, our results strongly suggest that it is important to investigate the relationship between SPARC and cath-D that are both co-secreted in the TNBC microenvironment.

At acidic pH, cath-D secreted by TNBC and mouse mammary cancer cells cleaves fibroblast-and cancer-derived SPARC in its extracellular Ca 2+ binding domain
As the tumor extracellular environment is acidic (43), we then asked whether cath-D can degrade SPARC in the extracellular medium of TNBC cells at low pH. First, we used conditioned medium from cath-D-secreting TNBC MDA-MB-231 cells co-cultured with SPARC-secreting HMFs for 24h ( Supplementary Fig. S2). SPARC was hydrolysed in a time-dependent manner in the conditioned medium at pH 5.5 (Fig. 4A). By western blot analysis, we detected mainly the 34-kDa and 27-kDa SPARC fragments, and to a lesser extent, the 16-kDa fragment (Fig. 4A). Pepstatin A inhibited SPARC cleavage, confirming the involvement of secreted aspartic protease proteolytic activity (Fig. 4A).
Moreover, TAILS analysis of the secretome in conditioned medium of co-cultured MDA-MB-231/HMF cells at pH 5.5 showed the presence of the five main SPARC fragments (34,  only in the absence of pepstatin A (Table 1). We then assessed SPARC hydrolysis at different pH (6.8 to 5.5), and found that in the MDA-MB-231/HMF conditioned medium, SPARC was significantly degraded up to pH 6.2 (Fig. 4B), similarly to the results obtained with recombinant proteins (Fig. 2B).
In addition, we observed SPARC limited proteolysis also in conditioned medium of TNBC HS578T ( Fig. 4C) and TNBC SUM159 cells (Fig. 4D), which secrete both proteins, at pH 5.5. Finally, we did not observe SPARC cleavage at pH 5.5 in conditioned medium from HMFs co-cultured with MDA-MB-231 cells in which CTSD was silenced by RNA interference, indicating that cath-D was responsible for SPARC proteolysis in acidic conditions (Fig. 4E). We confirmed cath-D direct involvement in SPARC processing also by using a mammary cancer cell line derived from tamoxifen-inducible Cre ERT2, Ctsd fl/fl mice (44) crossed with the transgenic MMTV-PyMT mouse model of metastatic BC (45) (Fig.   4F). In the absence of hydroxytamoxifen (OH-Tam), both cath-D and SPARC were secreted by these cells, whereas cath-D expression and secretion were abrogated by incubation with OH-Tam ( Supplementary Fig. S3). SPARC was hydrolysed in the conditioned medium from this mouse mammary cancer cell line at pH 5.5 only in the absence of OH-Tam when cath-D was secreted (Fig.   4F). These findings demonstrate that cath-D secreted by TNBC and mouse mammary tumor cells cleaves SPARC in its extracellular Ca 2+ binding domain at the acidic pH found in the tumor microenvironment.

SPARC is cleaved in vivo in TNBC and mouse mammary tumors
To validate cath-D-dependent SPARC cleavage in vivo, we first analysed FL SPARC protein level and its cleaved fragments in whole cytosols of mammary tumors from MMTV-PyMT cath-D knock-out mice (Fig. 5A). As expected, cath-D was expressed in the cytosol of mammary tumors from MMTV-PyMT, Ctsd +/+ mice, but not from MMTV-PyMT, Ctsd -/mice (Fig. 5A, left panel). In two of the three tumors from MMTV-PyMT, Ctsd -/mice, SPARC expression level was much higher than in the three tumors from MMTV-PyMT, Ctsd +/+ mice (Fig. 5A, left panel). Unexpectedly, we could not detect any

Cath-D-induced SPARC fragments inhibit TNBC cell adhesion and spreading, and promote their motility, endothelial transmigration and invasion
Previous studies reported that FL SPARC and particularly its C-terminal extracellular Ca 2+ binding domain can modulate adhesion, spreading, motility, endothelial transmigration, and invasion of cancer and stromal cells (22,23,30,(46)(47)(48). Therefore, we compared the effect of the cath-D-induced SPARC fragments (mixture of 34+27+16+9+ 6-kDa fragments) ( Supplementary Fig. S4) Fig. 6D; P<0.001). The effect of cleaved SPARC was 1.5-fold higher than that of FL SPARC ( Fig. 6D; P<0.001). Altogether, these results indicate that FL SPARC inhibits MDA-MB-231 cell adhesion and spreading, and promotes MDA-MB-231 cell motility, endothelial transmigration, and invasion. These effects were increased by incubation with cath-D-induced SPARC fragments, suggesting that in the TNBC microenvironment, cath-D amplifies SPARC pro-tumor activity through proteolysis of its extracellular Ca 2+ binding domain.

The 9-kDa C-terminal SPARC fragment inhibits TNBC cell adhesion and spreading, and promotes their motility, endothelial transmigration, and invasion
To identify the SPARC domain(s) involved in these functions, we produced FL SPARC and its various cleaved fragments in mammalian cells and purified them, as previously described (51,52)  contains the two Ca 2+ binding sequences of the two EF-hand domains ( Supplementary Fig. S8), that are involved in focal adhesion disassembly, and are crucial for SPARC-mediated inhibition of adhesion (46,53). The 16-kDa C-terminal SPARC fragment (amino acids 179-303) reduced cell adhesion by 1.2fold (not significant) ( Fig. 7B; Supplementary Fig. S8), and the 6-kDa SPARC fragment (amino acids 258-303) had no effect ( Fig. 7B; Supplementary Fig. S8). Therefore, among the five cath-D-induced SPARC fragments (Fig. 2E), only the C-terminal 9-kDa fragment could inhibit cell adhesion and more potently than FL SPARC.  Supplementary Fig. S10A). Conversely, we did not observe any significant difference between the 9-kDa C-terminal SPARC and the cath-D-induced SPARC fragments ( Fig. 8 and Supplementary   Fig. S10).

DISCUSSION
This study shows that cath-D secreted by TNBC cells triggers fibroblast-and cancer-derived SPARC cleavage at the acidic pH of the tumor microenvironment, leading to the production of the bioactive 9-kDa C-terminal SPARC fragment that inhibits cancer cell adhesion and spreading, and stimulates their migration, endothelial transmigration and invasion (Fig. 9). The TAILS analysis of the secretomes of conditioned medium from co-cultured TNBC cells and HMFs revealed that five main SPARC fragments (34-, 27-, 16-, 9-, and 6-kDa) are released in the extracellular environment in a cath-D-dependent manner. Our previous TAILS study showed that cystatin C is a substrate of extracellular cath-D and it is completely degraded by multiple cleavage, highlighting the complexity of the proteolytic cascades that operate in the tumor microenvironment (18). Here, we demonstrate that cath-D triggers also the limited proteolysis of the matricellular protein SPARC in an acidic environment to favour TNBC invasion.
Our recent study indicated that extracellular cath-D is a therapeutic target for immunotherapy and a biomarker in TNBC (16). Moreover, Huang et al found that cath-D was overexpressed in 71.5% of the 504 TNBC samples analysed and proposed a prognostic model for TNBC outcome based on node status, cath-D expression and Ki67 index (5). More recently, it was found that co-expression of cath-D and androgen receptor defines a TNBC subgroup with poorer overall survival (6). SPARC protein and mRNA are also overexpressed in TNBC (31,33,54), and this has been associated with poor prognosis in patients with TNBC (30,31,33). Here, we showed that high mRNA expression of CTSD and SPARC tended to be associated with shorter recurrence-free survival in a cohort of 255 patients with TNBC using an on line survival tool (39). Moreover, in a TNBC TMA, we found that cath-D was mainly expressed by cancer cells and some stromal cells, as shown previously (16). Conversely, SPARC was mainly expressed in mesenchymal cells, while its expression level in tumor cells was variable, as previously described (24,25). In cellulo, cath-D was secreted by TNBC cells and SPARC by human breast fibroblasts and some TNBC cell lines, as previously described (6,35,41). Importantly, cath-D and SPARC were co-localized in the microenvironment of TNBC PDX. Overall, these data prompted us to study the interplay between cath-D and SPARC in the TNBC microenvironment.
We then demonstrated that cath-D cleaves SPARC in vitro in an acidic environment exclusively in its extracellular Ca 2+ binding domain, specifically releasing five main SPARC fragments (34-, 27-, 16-, 9-, and 6-kDa). The main peptide bonds cleaved by cath-D at low pH are Phe-Phe, Leu-Tyr, Tyr-Leu, and Phe-Tyr (55) that correspond relatively well to the cleavage sites identified in this study.
Interestingly, other cleavage sites were also detected, such as Leu-Val, Leu-Asp/Glu or Gln-Phe, Gly-Tyr and Ala-Pro, which confirms the preference of cath-D for cleavage sites with at least one hydrophobic residue in P1 or P1'. SPARC biological activity can be modulated by limited proteolysis, leading to the unmasking of distinct or amplified biological functions compared with those of the FL protein (48,56). For instance, matrix metalloproteinases (MMP-1, -2, -3, -9 and -13) cleave SPARC in vitro in its N-terminal acid domain and in its extracellular Ca 2+ binding domain, releasing fragments with higher affinity for collagens that modulate cell-cell and cell-matrix extracellular interactions in the tumor microenvironment (57). In addition, MMP-3-mediated SPARC cleavage in vitro produces fragments that affect angiogenesis (58). More recently, cleavage of SPARC extracellular Ca 2+ binding domain by MMP-8 and MMP-13 has been detected in the serum of patients with lung cancer, indicating their presence also in vivo (59). Similarly, the cysteine cathepsin K (cath-K) also cleaves SPARC in vitro and in vivo in its N-terminal acid domain, and in its extracellular Ca 2+ binding domain in prostate cancer bone metastases, releasing a 10-kDa C-terminal fragment with unknown biological activity (60).
It is striking that the 9-kDa SPARC fragment generated by cath-D in our study is within the 10-kDa SPARC fragment generated by cath-K (60).
We then demonstrated that at acidic pH, cath-D present in conditioned medium from cath-D-  (64), and that promotes cell motility and invasion (50).
Both amino-and carboxyl-terminal (EF-hand) domains of SPARC bind to Ca 2+ that is required for maintenance of its native structure (53). In the EF-hand motif, two helices (E and F) flank a loop of 12 amino acids in which the Ca 2+ ion is coordinated in a pentagonal bipyramidal arrangement (65).
Moreover, FL SPARC binding to the extracellular matrix is Ca 2+ -dependent (53). Interestingly, synthetic small peptides ( 272 TCDLDNDKYIALDEWAGCFG 291 ) with sequences derived from SPARC Cterminal extracellular Ca 2+ binding domain (EF hand-2) inhibit adhesion and spreading of endothelial cells and fibroblasts (46,49,66). In our experimental model, it seems unlikely that only the EF hand-2 (aa 262-294) domain is involved in inhibiting MDA-MB-231 cell adhesion to fibronectin. Indeed, the 6-kDa SPARC fragment (amino acids 258-303) that contains only the EF hand-2 domain did not inhibit MDA-MB-231 cell adhesion, unlike the 9-kDa SPARC fragment (amino acids 235-303) that contains the residues coordinating Ca 2+ in both EF-hands. This suggests that the two Ca 2+ -binding domains are involved in this effect. However, inhibition of cell adhesion by the 16-kDa SPARC fragment (amino acids 179-303) that also contains both EF-hand domains was less important compared with the 9-kDa SPARC fragment. This suggests that the additional N-terminal sequences may alter the EF-hand domain conformation, or may interfere with Ca 2+ binding or with the interaction with fibronectin or a TNBC cell surface receptor. It also suggests that the three-dimensional conformation of the N-terminal sequences of the 16-kDa fragment is different from that of FL SPARC, which significantly inhibited MDA-MB-231 cell adhesion. It remains to be determined whether the 9-kDa C-terminal fragment of SPARC acts directly through a specific receptor, such 51 integrin, as described for FL SPARC (22), or by blocking adhesive interactions.
In addition, we demonstrated that FL SPARC (and more strongly the 9-kDa C-terminal fragment) promoted endothelial transmigration of TNBC cells, an essential step for extravasation and metastasis, as previously shown in melanoma (23). It was previously reported that the C-terminal extracellular Ca 2+ module of SPARC, a domain implicated in binding to endothelial cells (67) and to vascular cell adhesion molecule 1 (VCAM1) (68), is needed to enhance endothelial transmigration of melanoma cells via VCAM1 signalling (23). These findings suggest a role for the 9-kDa C-terminal SPARC fragment in vascular permeability, extravasation and metastasis formation in vivo.
Our current results indicate that cath-D secreted by TNBC cells is part of the proteolytic network in the TNBC acidic microenvironment that generates a bioactive 9-kDa C-terminal fragment of the matricellular protein SPARC with enhanced oncogenic activity. We dissected the molecular mechanisms that link SPARC limited cleavage by cath-D in TNBC microenvironment to the amplified oncogenic activity of a 9-kDa C-terminal fragment of SPARC, highlighting a novel paradigm of alteration of the extracellular milieu of TNBC by proteolysis. Overall, these results indicate that the 9-kDa C-terminal SPARC fragment is an attractive target for cancer therapies in TNBC, and open the way for developing novel targeted therapies against bioactive fragments from matricellular proteins, for both restructuring the surrounding microenvironment and reducing tumorigenesis (69).

MATERIALS AND METHODS
Additional Materials and Methods are in Supplementary files.

Secretome preparation
Secretomes from Ctsd -/and Ctsd -/cath-D cells were prepared as previously described (18). Briefly, cells were washed extensively with phenol red-free, serum-free medium to remove serum proteins and grown overnight in phenol red-free, serum-free medium. Conditioned medium was immediately incubated with protease inhibitors (1 mM EDTA, protease inhibitor cocktail (Complete; Roche Applied Science)), clarified by centrifugation (500 g for 5 min; 8,000 g for 30 min) and filtered (0.45 µM). Proteins present in conditioned medium in 50 mM HEPES (pH 7.5) were then concentrated to 2 mg/ml through Amicon filters (3 kDa cut-off, Millipore). To prepare secretomes from MDA-MB-231/HMF co-cultures, cells (ratio 1:5, respectively) were plated in 150 mm Petri dishes in DMEM with 10% FCS. At a 90% confluence, MDA-MB-231/HMF cells were washed extensively as described above. The 24hconditioned medium in 50 mM HEPES (pH 7.5) was then concentrated to 0.2 mg/ml through Amicon filters (3 kDa cut-off, Millipore), and incubated in cleavage buffer with or without pepstatin A (12.5 µM) at pH 5.5 and at 37° for 60 min. Samples were concentrated by TCA/acetone precipitation (73).

Mass spectrometry analysis of protein N-termini in cell culture samples (TAILS)
Enrichment of N-terminal peptides by TAILS (42) by isotopic labelling of proteins with iTRAQ reagents N-terminal peptides with significant changes between conditions were identified by calculating the log2 of the intensity ratios, correcting the mean of all ratios, and applying a 3-fold change cut-off (meancorrected log2>1.58 or <-1.58). The abundance of N-terminal peptides was visualized using the raincloud plot R tool (76).  Table 1.

Study approval
For TMA, TNBC samples were provided by the biological resource centre (Biobank number BB-0033- The study approval for PDXs was previously published (40).

Construction of tissue microarrays
Tumor tissue blocks with enough material at gross inspection were selected from the Biological Resource Centre. After haematoxylin-eosin-safranin (HES) staining, the presence of tumor tissue in sections was evaluated by a pathologist. Two representative tumor areas, to be used for the construction of the TMAs, were identified on each slide. A manual arraying instrument (Manual Tissue Arrayer 1, Beecher Instruments, Sun Prairie, WI, USA) was used to extract two malignant cores (1 mm in diameter) from the two selected areas. When possible, normal breast epithelium was also sampled as internal control. After arraying completion, 4 μm sections were cut from the TMA blocks. One section was stained with HES and the others were used for IHC.

TMA immunohistochemistry
For SPARC and cath-D immunostaining, serial tumor sections from a TNBC TMA were incubated with 0.2 µg/ml anti-human SPARC mouse monoclonal antibody (clone AON-5031) for 30 min or with 0.4 µg/ml anti-human cath-D mouse monoclonal antibody (clone C-5) for 20 min after heat-induced antigen retrieval with the PTLink pre-treatment (Dako) and the High pH Buffer (Dako) and endogenous peroxidase quenching with Flex Peroxidase Block (Dako). After two rinses in EnVision TM Flex Wash buffer (Dako), sections were incubated with a HRP-labelled polymer coupled to a secondary anti-mouse antibody (Flex® system, Dako) for 20 min, followed by incubation with 3,3'-diaminobenzidine as chromogen. Sections were counterstained with Flex Hematoxylin (Dako) and mounted after dehydration. Sections were analysed independently by two experienced pathologists, both blinded to the tumor characteristics and patient outcomes at the time of scoring. SPARC signal was scored as low (<50%), or high (>50%), and cath-D signal was scored as absent, low, medium (<50%), high, or very high (>50%) in cancer and stromal cells.

RT-qPCR
For gene expression analysis, fresh tumor tissues were homogenized in an Ultra-Turrax. RNA was isolated using the RNeasy Mini Kit (Qiagen, Hilden, Germany) and 1 µg of total RNA was reverse

Expression and purification of recombinant proteins
The cDNA encoding human SPARC (303 amino acids according to the GenBank reference NP_003109) and its truncated fragments were PCR-amplified using the pcDNA3.1-SPARC plasmid as template (78), cloned into pGEM®-T Easy Vector (Promega), and then into the pSec-Tag2/hygroA vector (Thermo Fisher Scientific) by Not I digestion. Orientation and sequence were verified (Supplementary Table S1).
Human embryonic kidney 293 (HEK-293T) cells were stably transfected with the vectors using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions, and were selected with 400 µg/ml hygromycin B Gold TM (Invivogen). The recombinant His-tagged proteins were purified from cell lysates on a nickel-chelating column (Ni-nitrilotriacetic acid agarose; His-select high flow nickel affinity gel; Sigma-Aldrich), as described previously (52). The isolated recombinant proteins were analysed by western blotting using anti-mouse Myc (clone 9B11) and anti-SPARC (clone AON-5031) antibodies and quantified using the Image J densitometric software (National Institutes of Health). To immunodeplete purified SPARC or its fragments, protein supernatants were incubated with an anti-Myc antibody (clone 9B11) overnight and protein G-Sepharose at 4°C for 4h, and supernatants were analysed by immunoblotting to validate SPARC depletion. SPARC-immunodepleted supernatants were used as internal controls in the biological assays.

Table 1. Sequences of the SPARC fragments identified by ATOMS and TAILS
High-confidence peptides with N-terminal iTRAQ labelling from the FL SPARC protein (Uniprot accession number P09486) identified by iTRAQ-ATOMS after in vitro cleavage of recombinant SPARC by recombinant cath-D, or by TAILS in the conditioned medium of co-cultured MDA-MB-231/HMFs. Peptides defining cleavage sites with iTRAQ ratios >2 or <0.5 for ATOMS or TAILS are shown. CM, conditioned medium; pepst., pepstatin A; *, according to the silver staining.