Matricellular Protein WISP2 Is an Endogenous Inhibitor of Collagen Linearization and Cancer Metastasis

Metastasis accounts for the vast majority of cancer-related deaths but its underlying mechanisms are not completely understood. During tumor progression, extensive remodeling of the tumor microenvironment occurs and facilitates metastasis (1, 2). In particular, linearization of collagen, one of the most abundant extracellular matrix proteins in solid tumors (3, 4), is recognized as a hallmark of aggressive cancers and is associated with an increased risk of metastasis and poor survival (5).

At the cellular level, linearized collagen facilitates tumor cell invasion and metastasis by providing tracks on which tumor cells can easily migrate (6–12). Therefore, preventing collagen linearization may be a valuable therapeutic approach to limit metastasis (1, 2). However, limited knowledge of the cellular and molecular mechanisms that govern collagen linearization has impeded the development of such therapies. Historically, cell-generated mechanical tension has been proposed to play a role in collagen linearization (13). However, a direct evaluation of its contribution to the organization of type I collagen (Col I) in solid tumors is still lacking due to the difficulty of precisely targeting cell-generated mechanical forces without affecting other cellular functions. Therefore, there is a pressing need to identify druggable processes to prevent Col I linearization and metastasis.

We have recently identified a novel mechanism of Col I linearization that is independent of cell-generated mechanical forces and that is based on the secreted matricellular protein WISP1 (14). WISP1 (CCN4) is a TGFβ1-induced cell-secreted factor that binds to Col I and promotes Col I linearization independently of cell-generated mechanical forces (14). Knockdown of WISP1 in cancer cells reduces collagen linearization and inhibits tumor cell invasion and metastasis (14). Thus, collagen linearization is modulated by WISP1 in tumors, either independently or in concert with cell-generated mechanical forces (13, 14). However, a key outstanding question is whether there are molecular antagonists of WISP1 that would actively prevent collagen linearization in normal tissues and that are dysregulated in tumors.

WISP1 is part of a family of structurally related proteins that includes WISP2 (CCN5). These two proteins share three domains, which are 73% homologous between WISP1 and WISP2: an IGF binding protein-like module (IGFBP), a von Willebrand factor type C repeat module (vWC), and a thrombospondin type-1 repeat module (TSP-1). In addition, WISP1 contains a cysteine-rich knot-like C-terminal domain (CT domain) but this domain is absent in WISP2 (15). Whereas high WISP1 expression in tumors is indicative of poor prognosis (14, 16, 17), high WISP2 levels correlate with good prognosis in patients with breast cancer (18). Altogether, the lack of a CT domain and a prognosis significance that inversely correlates with that of WISP1 suggest that WISP2 could work opposite to WISP1 in regulating collagen linearization and cancer cell metastasis. Here, we have tested this hypothesis and found that WISP2 is an endogenous inhibitor of collagen linearization that has therapeutic potential for impeding cancer progression and metastasis. On this basis, we propose a conceptually novel model whereby pro-metastatic collagen linearization is controlled by the relative bioavailability of two secreted factors with opposing function, WISP1 and WISP2.

The 4T1 (ATCC, catalog no. CRL-2539, RRID:CVCL_0125) murine breast cancer cell line was cultivated in RPMI (Thermo Fisher Scientific, catalog no. 11875119), 10% FCS (Thermo Fisher Scientific, catalog no. 10438026), and 1% penicillin/streptomycin (Thermo Fisher Scientific, catalog no. 15140112). The MDA-MB-231 human breast cancer cell line (ATCC, catalog no. HTB-26, RRID:CVCL_0062) and the 293FT cell line (ATCC, catalog no. PTA-5077, RRID:CVCL_6911) were cultivated in DMEM (Thermo Fisher Scientific, catalog no. 11965118), 10% FCS, and 1% penicillin/streptomycin. Cells were negative for Mycoplasma (last tested in June 2021) and used at passage 2 to 15 after thawing.

To prepare collagen solutions, Col I in 0.01 N HCl [PureCol (atelo-type I collagen), Advanced Biomatrix, catalog no. 5005] or telo-Col I [RatCol (telo-type I collagen), Advanced Biomatrix, catalog no. 5153] was neutralized (to pH 7.0 ± 0.2) on ice by adding 0.1 mol/L NaOH and mixed with 10× PBS to yield a stock solution containing 2 mg/mL (∼6.67 μmol/L) collagen in 1× PBS. Recombinant murine WISP1 (R&D Systems, catalog no. 1680-WS) in PBS or recombinant human WISP2 (PeproTech, catalog no. 120-16) were added at a final concentration of 50 and 100 μg/mL, respectively, unless otherwise indicated. The pH of the final collagen + analyte solutions was measured with an Orion PerpHecT ROSS Combination pH Micro Electrode (Thermo Fisher Scientific) and was confirmed to be constant across conditions. A total of 50 μL of collagen solution was transferred to a 16-well glass slide (Thermo Fisher Scientific) and incubated at 37°C for 4 hours (Col I) or at room temperature for 2 hours (telo-Col I) to allow for fibril formation. Collagen lattices were then processed and imaged by scanning electron microscopy as described previously (see also Supplementary Materials and Methods; ref. 14). For each image, the curvature ratio of the collagen fibrils was determined with NIH ImageJ (ImageJ, RRID:SCR_003070) and the numbers of “hairpin- or end-like” structures (fibrils that end or that form sharp bends at angles of 45° or less) and of “knot-like” structures (areas of tangled fibrils) counted in each 72 μm² image as described in ref. 14 and in the Supplementary Materials and Methods.

For detection of binding to Col I, Col I solution was prepared on ice as described above for scanning electron microscopy. A total of 100 μL of Col I solution were coated in 96-well plates for 10 minutes at 4°C. Excess Col I was then removed, and coated plates were incubated at 37°C for 4 hours to allow for fibril formation. For detection of WISP1-WISP2 binding, plates were coated with 50 μL of 500 nmol/L WISP2 in PBS, 0.01% Tween-20 and incubated overnight at 4°C. The wells were blocked with 2% BSA (GoldBio, catalog no. A-420-1) in PBS overnight to prevent nonspecific binding. Different concentrations of WISP1 and/or WISP2 proteins were prepared in PBS, 1% BSA, incubated on a rotator at room temperature for 45 minutes, and added to the collagen-coated wells in a 96-well plate, which was then incubated for 2 hours at 37°C. For the binding assays with prebound WISP1-Col I or WISP2-Col I complexes, 3 washes with 1× wash buffer (from R&D Systems, catalog no. DY008) were performed, different concentrations of WISP1 or WISP2 in PBS, 1% BSA were added, and the plate was incubated for 2 hours at 37°C. After three additional washes, bound proteins were detected by adding horseradish peroxidase (HRP)-conjugated anti-WISP1 antibody (from the mouse WISP1 Quantikine ELISA Kit, R&D Systems, catalog no. MWSP10) or with anti-human WISP2 (Thermo Fisher Scientific, catalog no. PA5-77219, RRID:AB_2720946) followed by HRP-conjugated anti-rabbit IgG (Cell Signaling Technology, catalog no. 7074, RRID:AB_2099233) in 1× reagent diluent (from R&D Systems, catalog no. DY008) and incubated for 2 hours at room temperature. Plates were then washed four times with 1× wash buffer and 100 μL of substrate solution (from R&D Systems, catlog no. DY008) was added. To stop the reaction, 100 μL of stop solution (from R&D Systems, catalog no. DY008) was added to each well and the optical density was read at 450 nm (with wavelength correction at 540 nm) with an Infinite M200Pro plate reader and i-control 1.10 software.

For primary tumor growth and spontaneous metastasis, 8 to 12 weeks old syngeneic wild-type BALB/c (Taconic, catalog no. BALB, RRID:IMSR_TAC:balb) or NSG female mice (The Jackson Laboratory, catalog no. 005557, RRID:IMSR_JAX:005557) were randomized and anesthetized with isoflurane, and 10⁶ tumor cells resuspended in 200 μL HBSS were injected into the 4th mammary fat pad. Mice were euthanized 28 days (4T1) or 70 days (MDA-MB-231) later with CO₂. Tissues were then processed and the tumor weight and numbers of spontaneous lung metastases per mm² of lung tissue determined as described in ref. 14 and in Supplementary Materials and Methods. All mice were housed and handled in accordance with approved St. Jude Children's Research Hospital Institutional Animal Care and Use Committee protocols.

WISP1 and WISP2 gene expression levels in patients with cancer and adjacent normal tissues were from the GSE62944 dataset, which includes RNA sequencing (RNA-seq) data for 9,264 tumor samples and 741 normal samples across 24 cancer types from The Cancer Genome Atlas. Breast cancer subtypes were assigned on the basis of the publicly available clinical data for the samples in GSE62944 (“nationwidechildrens.org_clinical_patient_brca” file from https://tcga-data.nci.nih.gov/publications/tcga). Only solid tumor types with transcripts per million (TPM) data available for at least three samples for both the normal and tumor samples were analyzed. log₂(TPM+1) was plotted for WISP1 and WISP2 gene expression levels. For the WISP2/WISP1 ratio, log₂(TPM_WISP2+1/TPM_WISP1+1) was plotted.

Statistical analysis were performed with the GraphPad Prism 8 software (RRID:SCR_002798). Unpaired two-sided t test with Welch correction was used to compare the means of two independent groups of normally distributed data. If data were not normally distributed, the Mann–Whitney test was used to compare the means of two independent groups. One-way ANOVA followed by Tukey posttest was used to compare the means of more than two independent groups of normally distributed data.

RNA-seq datasets generated and/or analyzed during this study are available at the Gene Expression Omnibus with accession numbers GSE152514, GSE110912, and GSE62944.

Additional methods (generation of stable cell lines, preparation of conditioned medium from stable cell lines, ELISAs, time-lapse migration assay, invasion and migration assays, proliferation assay, RNA-seq, qRT-PCR analysis, immunoblotting, Picrosirius red staining and polarized light microscopy, second harmonic generation microscopy, immunofluorescence staining, atomic force microscopy) were performed essentially as described previously (14) and are also available in the Supplementary Materials and Methods. DNA, primers, and guide RNA (gRNA) sequences used for cloning are also provided (Supplementary Tables S1–S3).

To determine whether WISP2 directly affects collagen remodeling, we examined the influence of recombinant WISP2 on Col I (PureCol atelo-collagen) architecture in vitro by scanning electron microscopy. These experiments revealed that, in contrast to WISP1, which is sufficient to promote Col I linearization (Fig. 1A and B; Supplementary Fig. S1A and S1B; ref. 14), addition of WISP2 to Col I does not induce any changes in collagen architecture or fibril diameter (Fig. 1A and B; Supplementary Fig. S1A–S1C). However, addition of WISP2 in combination with WISP1 (in a 1:3 WISP1:WISP2 molar ratio) results in the formation of Col I lattices that have an architecture similar to control Col I lattices (higher fibril curvature ratio and higher density of knot-like and hairpin- or end-like structures; Fig. 1A and B; Supplementary Fig. S1A–S1C), indicating that WISP2 inhibits the effect of WISP1 on Col I linearization. Similar results were obtained with telo-Col I, where WISP1-induced fibril linearization and organization into larger and more linear bundles (14) is inhibited by WISP2 (Supplementary Fig. S1D–S1F). These findings suggest that the inhibitory effect of WISP2 on WISP1-induced Col I linearization is independent of the presence of collagen telopeptides. On this basis, considering that WISP1 directly binds to Col I (14), WISP2 may inhibit WISP1 by preventing its interaction with Col I.

To test whether WISP2 can disrupt WISP1-Col I interactions, solid-phase binding assays were performed. Results from these assays revealed that both WISP1 and WISP2 have the capacity to bind to fibrillar Col I (Fig. 1C and D). Interestingly, addition of WISP2 to WISP1 (premixed in solution at a 1:3 WISP1:WISP2 molar ratio) inhibits the subsequent binding of WISP1 to Col I (Fig. 1E), but does not interfere with the ability of anti-WISP1 antibodies to bind to WISP1 (Supplementary Fig. S1G), indicating that WISP2 inhibits WISP1's function by preventing WISP1 binding to Col I. Furthermore, combining WISP1 and WISP2 at a 1:1 molar ratio induces a partial inhibition of WISP1-Col I binding (Fig. 1E) and of Col I linearization (Supplementary Fig. S1H–S1L), indicating that the effect of WISP2 is dose dependent. In contrast, addition of an excess of WISP1 or WISP2 to prebound WISP1- or WISP2-Col I complexes does not disrupt these interactions (Fig. 1F and G). Since WISP1 and WISP2 are also able to bind to each other (Fig. 1H), these data suggest that WISP2-WISP1 binding might prevent WISP1-Col I interactions and be a possible mechanism by which WISP2 inhibits WISP1's function. However, this model does not exclude the possibility that free WISP1 and WISP2 also compete for Col I binding sites, and it is conceivable that these two regulatory mechanisms coexist depending on the relative bioavailability of Col I, WISP1, and WISP2 in a specific context. Thus, taken together, these data establish WISP2 as an inhibitor of WISP1 that prevents WISP1-induced Col I linearization. Furthermore, because linearized collagen is more abundant in tumors than in normal tissues (1, 13), these findings suggest that upregulation of WISP1 versus WISP2 expression in tumors may drive Col I linearization.

Mining of publicly available RNA-seq datasets revealed that WISP2 expression is downregulated in a wide range of solid tumors, including breast, colon, lung, prostate, and rectal cancers, in comparison with the corresponding normal tissues (Fig. 2A; Supplementary Fig. S2A). Conversely, consistent with previous studies (19, 20), WISP1 expression is typically higher in tumors than in normal tissues (Fig. 2B; Supplementary Fig. S2B). The WISP2:WISP1 ratio is therefore generally lower in tumors than in normal tissues (Fig. 2C; Supplementary Fig. S2C). Similar results were also obtained across the four breast cancer subtypes (luminal A, luminal B, HER2-enriched, and triple-negative), revealing that WISP2 downregulation and a low WISP2:WISP1 ratio (<1) occur in a majority of breast tumors, regardless of the subtype (Supplementary Fig. S2D–S2G). Furthermore, whereas WISP2 is expressed (TPM > 1) in 100% of normal tissues adjacent to breast cancers, approximately 40% of breast tumors express no or very low WISP2 (TPM < 1; Supplementary Fig. S2H). Taken together, these results suggest that higher levels of functional WISP1 are found in tumors and that restoration of WISP2 to the levels found in normal tissues might provide a therapeutic intervention that inhibits collagen linearization and consequent tumor cell invasion and metastasis of aggressive tumor cells.

Given that both in vivo and in vitro studies have shown that tumor cells migrate more rapidly onto linearized collagen than onto “normal” disorganized fibrils (6–12), we tested whether addition of recombinant WISP2 inhibits the invasion of triple-negative 4T1 breast cancer cells through Col I. For these studies, 4T1 cells were chosen as the main experimental model because Wisp1 or Wisp2 overexpression does not induce any significant gene expression changes in these cells (Supplementary Fig. S2I; ref. 14). Thus, this cell line provides an ideal system to probe directly how WISP1/2-induced changes in Col I architecture impact tumor cell invasion, independently of alterations in gene expression.

To determine the impact of recombinant WISP1 and WISP2 on cell motility on Col I lattices, time-lapse video microscopy was performed. These studies indicated that the migration speed of 4T1 cells is enhanced on Col I lattices formed in the presence of WISP1 in comparison with control lattices but remains unchanged if lattices are formed in the presence of WISP2 (Fig. 2D and E). Interestingly, 4T1 cells plated on Col I lattices formed in the presence of both WISP1 and WISP2 also fail to migrate faster than cells plated on control Col I lattices (Fig. 2D and E). These findings indicate that, consistent with its role in inhibiting WISP1-induced Col I linearization, WISP2 prevents increased cell motility on Col I induced by WISP1. Importantly, addition of neither WISP1 nor WISP2 significantly impacts cell motility on plastic surfaces (Fig. 2F and G), further emphasizing a requirement for Col I in WISP1 and WISP2 function. Consistent with time-lapse video microscopy data, WISP2 inhibited WISP1-induced 4T1 cell invasion through Transwells coated with Col I lattices (either atelo- or telo-Col I; Fig. 2H; Supplementary Fig. S2J), in a dose-dependent manner (Supplementary Fig. S2K). However, neither cell migration through uncoated Transwells (Fig. 2I) nor cell proliferation rates were affected (Fig. 2J; Supplementary Fig. S2L and S2M). Furthermore, WISP1 and WISP2 induced similar effects on the invasion, migration, and proliferation of the human triple-negative breast cancer (TNBC) cell line MDA-MB-231 (Fig. 2K–M; Supplementary Fig. S2N). Taken together, these results indicate that WISP2 inhibits WISP1-induced cell invasion. Given that these results mirror those obtained for collagen remodeling—and because motility on plastic remains unaffected—these data indicate that the effect of WISP1 on cell motility and the inhibitory effect of WISP2 stem from their effects on Col I linearization.

Our data indicate that both WISP1 and WISP2 bind to Col I, but that only WISP1 is able to induce Col I linearization and to enhance cell invasion (Figs. 1 and 2). WISP2 shares 73% sequence similarity with WISP1, but notably lacks a cysteine-rich CT domain (Fig. 3A), raising the possibility that the CT domain of WISP1 is responsible for Col I linearization. To understand whether the inhibitory function of WISP2 is indeed due to its lack of a CT domain, we asked whether adding the CT domain of WISP1 to WISP2 (WISP2+CT) would confer WISP2 the ability to promote Col I linearization and cell invasion and, conversely, whether a form of WISP1 lacking a CT domain (Wisp1ΔCT) would block the function of WISP1. To this end, overexpression constructs were generated (Fig. 3A) and stably transduced into 4T1 cells (Supplementary Fig. S3A–S3C). Solid-phase binding assays performed with conditioned medium from WISP1, WISP1ΔCT, WISP2, and WISP2+CT overexpressing cells (Supplementary Fig. S3C) indicated that these protein variants all interact with Col I (Fig. 3B). At equal concentrations, the binding of WISP1ΔCT to Col I was approximately 75% that of WISP1, suggesting that the CT domain of WISP1, while not required for WISP1-Col I binding, might contribute to facilitating WISP1-Col I interaction (Fig. 3B). However, addition of the CT domain of WISP1 to WISP2 did not further enhance WISP2 binding to Col I (Fig. 3B). Furthermore, coincubation of conditioned media from cells overexpressing non-tagged WISP constructs (i.e., WISP1, WISP1ΔCT, WISP2, and WISP2+CT) together with Myc-tagged WISP1 (WISP1-myc) inhibited WISP1-myc binding to Col I, as measured by detection of Myc-tagged WISP1 (WISP1-myc) interaction with Col I (Fig. 3C; Supplementary Fig. S3D). Thus, all the WISP protein constructs here studied have the ability to bind to Col I and to inhibit WISP1-Col I binding.

We next examined the architecture of Col I lattices formed in the presence of these protein constructs. In contrast to the general capacity of WISP1 and WISP2 constructs to bind to Col I (Fig. 3B), we find that WISP1 and WISP2+CT induce Col I linearization whereas WISP2 and WISP1ΔCT do not (Fig. 3D and E; Supplementary Fig. S3E–S3G). In line with these observations, overexpression of Wisp1 or Wisp2+CT in 4T1 cells enhanced their motility on Col lattices and their ability to invade through Col I-coated Transwells, whereas Wisp2 or Wisp1ΔCT overexpression did not have significant effects in these assays (Fig. 3F and G; Supplementary Fig. S3H). Furthermore, consistent with our previous findings (14), overexpression of these protein constructs affects neither cell motility on plastic nor migration through Transwells (Fig. 3H and I; Supplementary Fig. S3I). Taken together, these data indicate that the CT domain of WISP1 is dispensable for Col I binding but required for the functional effects of WISP1 on Col I linearization and tumor cell invasion. In addition, these data indicate that WISP2 inhibits the function of WISP1 by preventing its binding to Col I.

Having established the inhibitory role of WISP2 on tumor cell invasion, we next tested whether its inhibitory function holds true in a context where Wisp1 is physiologically induced. In 4T1 tumors, Wisp1 and Wisp2 are expressed at comparable levels by tumor cells and cancer-associated fibroblasts (CAF), prominent cell types present in tumors and key modulators of ECM remodeling (Supplementary Fig. S4A and S4B). Furthermore, the expression of Wisp1 is markedly upregulated in both cell types upon TGFβ1 treatment (Supplementary Fig. S4C). Given that TGFβ1 is abundant in primary tumors, that it promotes cell invasion, and that it is known to induce WISP1 expression in a number of cancer cell lines including 4T1 and MDA-MB-231 cells (14, 21–25), we tested whether WISP2 or WISP1ΔCT inhibit TGFβ1-induced cancer cell invasion. To this end, 4T1 cells overexpressing Wisp2 or Wisp1ΔCT were treated with 2 ng/mL recombinant TGFβ1 for 24 hours. As expected, TGFβ1 increased endogenous Wisp1 expression regardless of Wisp2 or Wisp1ΔCT levels (Fig. 4A) and led to increased 4T1-EV cell motility on Col I lattices (Fig. 4B). Interestingly, overexpression of Wisp2 or Wisp1ΔCT was sufficient to block TGFβ1-induced motility of 4T1 cells on Col I lattices (Fig. 4B; Supplementary Fig. S5A), as well as invasion toward a chemoattractant (5% FBS) in Transwell invasion assays (Fig. 4C). Furthermore, similar blockade of TGFβ1-induced 4T1 cell motility on Col I and invasion through Col I in Transwell assays were observed when WISP1 levels were reduced by transducing a pool of 5 Wisp1-targeting gRNAs into 4T1 cells expressing inducible Cas9 (4T1-indCas9; Fig. 4D–F; Supplementary Fig. S5B). However, Wisp1 knockout failed to block TGFβ1-induced cell motility on plastic (Supplementary Fig. S5C and S5D) or Transwell migration (Supplementary Fig. S5E) and did not affect cell proliferation (tested in the presence or absence of TGFβ1; Supplementary Fig. S5F). Thus, inhibition of WISP1 via CRISPR/Cas9 knockout or via overexpression of Wisp2 or Wisp1ΔCT is sufficient to prevent TGFβ1-induced cell motility on Col I, which is one of the most abundant extracellular matrix proteins present in breast tumors (3, 4). On this basis, it is plausible that ectopic expression of WISP2 or WISP1ΔCT would limit tumor cell invasion and prevent metastasis in tumors where the TGFβ1-WISP1 axis is activated.

To assess whether WISP2 and WISP1ΔCT inhibit breast tumor progression and metastasis, and conversely, whether WISP2+CT promotes it, 4T1 cells with stable overexpression of these proteins were implanted into the mammary fat pads of syngeneic BALB/c female mice. No significant differences in tumor growth were noted across the different groups, except for the 4T1-Wisp2 tumors, which grew faster than control 4T1-EV tumors (Fig. 5A). However, mice bearing 4T1-Wisp2 or 4T1-Wisp1ΔCT tumors developed significantly fewer lung metastases than mice bearing 4T1-EV tumors (Fig. 5B), whereas more spontaneous metastases were found in the lungs of mice bearing 4T1-Wisp1 or 4T1-Wisp2+CT tumors (Fig. 5B). Examination of primary tumor sections by second harmonic generation microscopy or by polarized light microscopy following staining with Picrosirius red revealed that Wisp1 or Wisp2+CT overexpression results in an increase in the frequency of linear collagen fibers in primary tumors, whereas overexpression of Wisp2 or Wisp1ΔCT has an opposite effect (Fig. 5C and D; Supplementary Fig. S6A). Moreover, linear collagen fiber abundance positively correlated with the stiffness of tumor tissues (Young's modulus; Fig. 5E), consistent with the previously reported role of fibrillar collagen in increasing tumor stiffness (8, 26). The diameters of collagen fibers were also, on average, slightly larger in the 4T1-Wisp1 and 4T1-Wisp2+CT tumors (Supplementary Fig. S6B), suggesting that WISP1-induce Col I linearization may promote the bundling of collagen fibrils into the larger-diameter collagen fibers present in tumors. However, total type I collagen deposition in tumors remained unchanged across the experimental groups (Supplementary Fig. S6C and S6D), indicating that, as proposed by previous studies (5, 10), the abundance of linear fibers in tumors (rather than total collagen deposition) impacts metastasis. Taken together, these results indicate that WISP1 promotes collagen linearization and spontaneous metastasis in a CT domain–dependent manner, whereas, in contrast, collagen linearization and metastasis are inhibited by WISP2, which lacks a CT domain. Thus, administration of recombinant WISP2 may be envisioned as therapeutic intervention to limit collagen linearization and metastasis.

To test whether treatment with recombinant WISP2 would be a feasible anti-metastatic therapy, 4T1 cells were implanted into the mammary fat pads of syngeneic BALB/c females and mice were treated with PBS or WISP2 (Fig. 5F). These experiments revealed that treatment with recombinant WISP2 does not affect 4T1 primary tumor growth (Fig. 5G) but inhibits spontaneous lung metastasis (Fig. 5H). A decrease in the frequency (Fig. 5I and J; Supplementary Fig. S6E) and in the average diameter (Supplementary Fig. S6F) of linear collagen fibers was also noted in primary tumors of mice treated with WISP2, whereas total type I collagen deposition remained unaffected (Supplementary Fig. S6G and S6H).

In addition, we tested whether a shorter treatment regimen would be sufficient to inhibit metastasis in aggressive tumors characterized by high levels of WISP1. To this end, mice bearing 4T1-Wisp1 tumors were treated with recombinant WISP2 from day 8 to day 19 post tumor inoculation (Supplementary Fig. S7A), a time frame during which 4T1 tumor cells invade and access the bloodstream (27). These experiments revealed that this treatment regimen does not impact tumor growth but is sufficient to inhibit spontaneous lung metastasis (Supplementary Fig. S7B and S7C). However, no difference in linearized collagen fiber abundance was observed (Supplementary Fig. S7D and S7E), likely due to the delay between the last dose of recombinant WISP2 (on day 19) and tumor resection (on day 28). Indeed, an independent experiment with tumor resection on day 19 revealed a lower abundance of linearized collagen fibers in tumors from mice treated with recombinant WISP2 (Supplementary Fig. S7F–S7H), whereas average collagen fiber diameters and total type I collagen deposition in tumors remained unaffected (Supplementary Fig. S7I–S7K).

Taken together, these results thus indicate that targeting WISP1 with recombinant WISP2 may be a valuable therapeutic approach to limit collagen linearization and to impede cancer metastasis.

To further test the relevance of these findings for human disease, we tested whether overexpression of WISP1 or WISP2 in the MDA-MB-231 model of triple-negative human breast cancer xenografts (Supplementary Fig. S8A) affects tumor cell invasion and metastasis. Similar to the results obtained with the 4T1 model, overexpression of WISP1 in MDA-MB-231 cells promoted tumor cell invasion through Col I lattices (Fig. 6A) but did not affect in vitro cell migration (Fig. 6B) or proliferation (Fig. 6C; Supplementary Fig. S8B). Furthermore, upon orthotopic implantation into the mammary fat pad of NSG mice, no differences in tumor sizes were observed when comparing MDA-MB-231 tumors overexpressing WISP1 or WISP2 to control tumors (Fig. 6D). However, WISP1 overexpression in cancer cells promoted collagen linearization in tumors and spontaneous lung metastasis, whereas WISP2 overexpression inhibited these processes (Fig. 6E–G; Supplementary Fig. S8C). WISP1 overexpression also led to a slight but statistically significant increase in collagen fiber diameter (Supplementary Fig. S8D), while no difference in Col I deposition was noted across experimental groups (Supplementary Fig. S8E and S8F). Thus, forced expression of WISP1 promotes metastasis in a preclinical model of human triple-negative breast cancer, whereas WISP2 limits collagen linearization and metastasis. Together with the lower WISP2:WISP1 ratio observed in human tumors in comparison to normal tissues (Fig. 2C), this suggests that therapeutic manipulation of the relative abundance of WISP1 and WISP2 may provide a novel avenue to control collagen linearization and hence limit metastasis of human tumors.

Collagen remodeling is a key event in cancer progression and in other processes such as fibrosis and wound healing. In tumors, aberrant collagen linearization promotes metastasis by providing tracks on which cancer cells can easily migrate (6–8, 12). This mechanism of metastasis is particularly important for breast cancers of the triple-negative subtype (TNBC), which often contain high amounts of aligned fibrillar collagen (28), are very aggressive, and against which no efficient therapy is currently available (29).

The secreted factor WISP1 is by itself able to promote collagen linearization, and knockdown of its expression in cancer cells prevents collagen linearization and inhibits tumor cell invasion and metastasis (14). Thus, in concert with cell-generated mechanical forces, WISP1 plays a pivotal role in promoting collagen linearization in vivo. However, despite the identification of WISP1 as a secreted factor that drives collagen linearization, it remained unknown whether collagen linearization in physiologic and pathologic processes can also be modulated by an endogenous inhibitor.

The data presented here demonstrate that WISP2 is a naturally occurring inhibitor of WISP1 that blocks WISP1-induced collagen linearization, tumor cell invasion, and metastasis. Interestingly, previous data suggest that WISP1 and WISP2 might also play antagonistic roles in other processes that rely on extensive collagen remodeling, such as wound healing and fibrosis. Indeed, Wisp1^−/− mice have a reduced ability to repair wounds, and treatment with anti-WISP1 antibodies prevents pulmonary and liver fibrosis (30–32). In contrast, Wisp2^−/− mice display enhanced cardiac fibrosis, indicating that WISP2 is anti-fibrotic in this context (33).

Mechanistically, our study reveals that WISP2—which is structurally similar to WISP1 but lacks a CT domain—is able to bind to both Col I and to WISP1. Binding of WISP2 to WISP1 may therefore mask or alter the conformation of a Col I binding site on WISP1, and thereby impede WISP1-Col I binding. In addition, WISP2 may directly compete with free WISP1 for available binding sites on Col I. However, WISP2 is unable to displace prebound WISP1 (and vice versa), suggesting that both proteins bind tightly to Col I. Consistent with this, mutant forms of WISP1 and WISP2 (WISP1ΔCT and WISP2+CT) also bind to Col I, indicating that binding to Col I does not require the presence of a CT domain. In contrast, we show that the CT domain of WISP1 is necessary for collagen linearization: WISP1ΔCT and WISP2 (which also lacks a CT domain) do not promote collagen linearization, cell invasion, and metastasis, whereas addition of the CT domain of WISP1 to WISP2 confers pro-invasive and pro-metastatic properties to WISP2. On this basis, we propose a model whereby the CT domain of WISP1 is required for Col I linearization but is dispensable for WISP1-Col I interactions.

CT domains similar to the one found in WISP1 are also present in other proteins such as NGF, TGFβ, PDGF, VEGF, CTGF, vWF, and mucins (34–39). Interestingly, it has been shown that the CT domains of these proteins induce their dimerization, a process that is typically necessary for their bioactivity (37, 38, 40, 41). In addition, analyses of the structure and function of the CT domain of vWF revealed that it not only promotes vWF dimerization, but also allows this large protein to sustain hydrodynamic mechanical forces (38, 40). On this basis, we speculate that the dimerization of Col I-bound WISP1 via its CT domain might stabilize Col I structure and drive its linearization.

The inhibitory effect of WISP2 on the function of WISP1 implies that the relative bioavailability of these two matricellular proteins in tumors or in healthy tissues is a key determinant of Col I architecture. A WISP2/WISP1 balance tipped in favor of WISP1 may therefore promote collagen linearization in tumors. In support of this hypothesis, WISP2 expression is typically downregulated in tumors compared with normal tissues, whereas WISP1 expression is upregulated, leading to a lower WISP2:WISP1 ratio. Therefore, this suggests that restoration of WISP2 levels to those found in normal tissues might inhibit the function of WISP1 in tumors, prevent collagen linearization, and limit metastasis in cancer patients. As a proof of principle of the therapeutic potential of this approach, this study demonstrates that increasing WISP2 levels via genetic overexpression or systemic administration (intraperitoneal injections) of recombinant WISP2 into tumor-bearing mice is sufficient to inhibit TNBC metastasis. Thus, patients with breast cancer with aggressive tumors characterized by high levels of WISP1 and linearized collagen might benefit from WISP2 treatments.

TGFβ1 is frequently enriched in the microenvironment of metastatic TNBC where it is an important driver of cancer progression, metastasis, and fibrosis (4, 22, 24, 42). For example, TGFβ1 enhances Col I expression and matrix remodeling by cancer-associated fibroblasts and tumor cells. It also promotes tumor cell invasion via the induction of a wide array of downstream effectors (22, 24), including WISP1 (14). Notably, our results indicate that inhibiting WISP1 via WISP2 or WISP1 knockout is sufficient to inhibit TGFβ1-induced cell invasion through Col I lattices. Thus, WISP1-driven collagen remodeling is necessary for efficient TGFβ1-induced cell invasion through Col I. Because WISP2 does not affect cell proliferation, it might specifically inhibit the pro-invasive effect of TGFβ1 without affecting its cytostatic function. In line with this hypothesis, an inhibitory role for WISP2 downstream of TGFβ1 signaling has been proposed in the context of fibrotic deformation of the retinal pigment epithelium, but the mechanism involved has not yet been characterized (43). Furthermore, it has been previously reported that, in poorly aggressive MCF-7 breast cancer cells, WISP2 acts as a transcriptional repressor and inhibits Tgfbr2 expression (44). However, WISP2 overexpression did not trigger any significant changes in Tgfbr2 levels in metastatic 4T1 cells, indicating that regulation of Tgfbr2 expression is not a major mechanism by which WISP2 inhibits 4T1 cell invasion and metastasis. In fact, neither WISP2 nor WISP1 overexpression triggers any significant gene expression changes in 4T1 cells (Supplementary Fig. S2I; ref. 14), suggesting that these matricellular proteins do not significantly affect signal transduction pathways that lead to gene expression. On the basis of these findings and of our observation that recombinant WISP2 is sufficient to prevent WISP1-induced collagen linearization in a cell-free system, we propose that WISP1 and WISP2 regulate tumor cell invasion through their direct effect on Col I architecture, in response to TGFβ1.

Overall, this study reveals that the secreted factors WISP1 and WISP2 antagonistically regulate collagen linearization during tumor progression and metastasis. Pharmacologically tilting the WISP2/WISP1 balance might therefore provide therapeutic avenues to impede collagen linearization and limit the progression of metastasis in patients with cancer.

No disclosures were reported.

J. Janjanam: Conceptualization, data curation, formal analysis, investigation, methodology, writing–review and editing. G. Pano: Data curation, formal analysis, investigation, writing–review and editing. R. Wang: Data curation, investigation, writing–review and editing. B.A. Minden-Birkenmaier: Data curation, investigation, writing–original draft, writing–review and editing. H. Breeze-Jones: Data curation, investigation, writing–review and editing. E. Baker: Data curation, investigation, writing–review and editing. C. Garcin: Data curation, investigation, writing–review and editing. G. Clayton: Data curation, formal analysis, investigation, writing–review and editing. A. Shirinifard: Formal analysis. A.M. Zaske: Formal analysis, investigation. D. Finkelstein: Data curation, formal analysis, writing–review and editing. M. Labelle: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

The authors thank Aaron Pitre, Jennifer Peters, and Victoria Frohlich for assistance with time-lapse video microscopy. They thank Camenzind Robinson, Amanda Johnson, and Randall Wakefield for assistance with scanning electron microscopy sample processing and imaging. Images were acquired at the Cell & Tissue Imaging Center, which is supported by SJCRH and NCI P30CA021765. The authors thank Shondra Pruett-Miller and Samuel Peters of St. Jude's Center for Advanced Genome Engineering for designing and assembling the gRNA vectors. This research was supported by the NCI of the NIH under Award Numbers R01CA245301 (to M. Labelle) and P30CA021765 (St. Jude Core Facilities), and by the ALSAC (to M. Labelle), the fundraising and awareness organization for St. Jude Children's Research Hospital. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.