CXCL14 is an autocrine growth factor for fibroblasts and acts as a multi-modal stimulator of prostate tumor growth

Abstract

This study explored the role of secreted fibroblast-derived factors in prostate cancer growth. Analyses of matched normal and tumor tissue revealed up-regulation of CXCL14 in cancer-associated fibroblasts of a majority of prostate cancer. Fibroblasts over-expressing CXCL14 promoted the growth of prostate cancer xenografts, and increased tumor angiogenesis and macrophage infiltration. Mechanistic studies demonstrated that autocrine CXCL14-stimulation of fibroblasts stimulate migration and ERK-dependent proliferation of fibroblasts. CXCL14-stimulation of monocyte migration was also demonstrated. Furthermore, CXCL14-producing fibroblasts, but not recombinant CXCL14, enhanced in vitro proliferation and migration of prostate cancer cells and in vivo angiogenesis. These studies thus identify CXCL14 as a novel autocrine stimulator of fibroblast growth and migration, with multi-modal tumor-stimulatory activities. In more general terms, our findings suggest autocrine stimulation of fibroblasts as a previously unrecognized mechanism for chemokine-mediated stimulation of tumor growth, and suggest a novel mechanism whereby cancer-associated fibroblasts achieve their pro-tumorigenic phenotype.

Chemokines are a family of secreted proteins that stimulate chemotaxis and cell growth. More than 50 chemokines have been identified and are divided into the CXC, CC, C, and CX3C groups (1). Chemokines exert their cellular effect by activation of cell surface receptors belonging to the G protein-coupled-receptor family. Approximately 20 chemokine receptors have been identified (1, 2). Many receptors bind multiple ligands, whereas others are highly specific such as CXCR4 and CXCR6, which bind only CXCL12 and CXCL16, respectively. For some of the chemokines, such as CXCL14, the receptor has not yet been identified.

Recent studies have suggested important functions of chemokines in various aspects of tumor growth (2). Chemokines contribute to leukocyte infiltration in tumors, and some, such as IL-8, CXCL1–3, and CXCL5, also have direct proangiogenic effects (3, 4). Concerning chemokines acting on malignant cells, most attention has been paid to CXCL12, which stimulates tumor cell proliferation and migration through CXCR4 and CXCR7 (5, 6). CXCL12 also contributes to tumor growth by recruitment of bone marrow–derived endothelial precursor cells (7). Additionally, mesenchymal stem cell–derived CCL5/RANTES was recently shown to confer prometastatic effects on breast cancer cells (8). Most recently, CCL3 was implicated as a prometastatic agent acting through multiple mechanisms including stimulation of fibroblasts (9).

Within the tumor microenvironment, multiple cell types have been identified as sources of chemokine production, including the malignant cells and inflammatory leukocytes. Recent characterization of the expression profiles of cancer-associated fibroblasts (CAFs) has also identified this cell type as an important producer of chemokines (7, 10–12). In breast cancer, both CXCL12 and CXCL14 were found to be up-regulated in the tumor stroma (10, 13). Furthermore, CAFs of ovarian cancer over-express the chemokine CXCL1/GRO1 (11).

In this study, characterization of human prostate CAFs led to the identification of CXCL14 as a novel CAF-derived tumor-stimulatory factor.

Results

CXCL14 Is Up-Regulated in CAFs of Human Prostate Cancer.

Fibroblast-enriched stroma was isolated from 4 matched sets of normal and tumor tissue by laser capture microdissection. Array-based comparison of amplified mRNA identified 266 transcripts that were at least 2-fold up-regulated in tumor stroma, 15 of which encoded secreted proteins. Among them were the chemokine CXCL14 which was ≈40-fold up-regulated in tumor stroma.

qRT-PCR analyses of 8 matched normal and tumor tissues confirmed stromal up-regulation of CXCL14 in 6 out of 8 cases (Fig. 1A). qRT-PCR with primers specific for endothelial cells (CD31), macrophages (CD163) and leukocytes (CD45) indicated a moderate, 1.5- to 3-fold increase in relative content of these cell types in the tumor stroma preparations. Since these changes were too small to explain the increased CXCL14 levels, it was concluded that the up-regulation of CXCL14 occurred in the prostate CAFs.

Fig. 1.

CXCL14 is up-regulated in fibroblast-enriched prostate cancer stroma. (A) Stroma from normal prostate (open bars) and prostate cancer tissue (filled bars) was microdissected from 8 different patients and the expression of CXCL14 was analyzed with qRT-PCR. (B) Examples of CXCL14 immunohistochemistry analyses of paired tissues with up-regulation of CXCL14 in tumor stroma, together with unchanged (Top), up-regulated (Middle), or down-regulated (Bottom) expression in the epithelial cells. (Scale bar, 100 μm.)

Immunohistochemical analyses of 27 matched pairs of prostate cancer and normal prostate were performed to further compare CXCL14 expression in normal and tumor stroma. These analyses revealed a statistically significant (P < 0.05) increase in stromal expression of CXCL14 in 15/27 (56%) cancer samples (Fig. 1B). Variable CXCL14 expression in the epithelial compartment of normal and cancer tissue was also observed. Double immunofluorescent staining with antibodies against CXCL14 and the PDGF β-receptor, used as a fibroblast marker, confirmed that fibroblasts of human prostate cancer produced CXCL14 (Fig. S1). Stromal and epithelial CXCL14 status were not significantly associated with Gleason grade.

These analyses demonstrate that up-regulation of CXCL14 in CAFs is a common feature of human prostate cancer.

NIH-CXCL14 Cells Promote Tumor Growth and Proliferation Without Affecting Epithelia-Stroma Ratio.

Subcutaneous tumor growth of human prostate cancer LNCaP cells is enhanced by co-injection of prostate cancer CAFs or NIH 3T3 fibroblasts (14). This tumor model was therefore used to investigate the functional significance of CXCL14 up-regulation in CAFs. Two NIH 3T3 cell lines were generated by infection of cells with control vectors (NIH-ctr) or vectors encoding human CXCL14 (NIH-CXCL14). CXCL14 overexpression in transfected cells was confirmed by qRT-PCR, and using a CXCL14-specific ELISA (Fig. S2).

The mixture of LNCaP cells and NIH-CXCL14 fibroblasts formed tumors that appeared much earlier and grew faster than the LNCaP/NIH-ctr tumors (Figs. 2 A and B). Injection of LNCaP cells alone did not yield detectable tumors throughout the experiment. Injection of only NIH-ctr or NIH-CXCL14 cells did not lead to tumor formation during the first 30 days after injection.

Fig. 2.

NIH-CXCL14 cells accelerate growth of prostate cancer xenografts. (A) The tumor growth of injected LNCaP cells (filled triangles) or LNCaP cells together with NIH-ctr (open squares) or NIH-CXCL14 cells (filled squares) was followed over time after s.c. injection into SCID mice. (# n = 6; ^§ n = 5 remaining animals) (B) Tumor incidence at days 28 and 49, corresponding to the time when the first animal of the LNCaP/NIH-CXCL14 and LNCaP/NIH-ctr groups were killed, was calculated. Tumors from both groups were collected and analyzed for cell density (C), cell proliferation (D), and the abundance of vessels (E) and macrophages (G), using H&E staining or immunohistochemistry as indicated. Spearman correlation analysis was performed to analyze correlations between intratumor CXCL14 expression and CD31 (F) or CSF1R (H) expression. Error bars in B–D indicate SEM. (Scale bar, 100 μm.) *, P < 0.05, by unpaired t test.

No significant differences in cell density between the 2 tumor types were observed (Fig. 2C). qRT-PCR analyses with the epithelial marker cytokeratin 18 did not indicate any differences with regard to relative content of tumor epithelial cells. Analyses with the fibroblast markers puromycin (present in the pBABE vector of NIH-ctr and NIH-CXCL14 cells) and vimentin indicated no significant differences, although a tendency toward an increased fibroblast content in LNCaP/NIH-CXCL14 tumors was observed (Fig. S3). LNCaP/NIH-CXCL14 tumors displayed a significantly higher proliferation rate, as compared to control tumors (Fig. 2D). Because no major differences in epithelia-stroma ratio were observed, it was concluded that that the increased proliferation was occurring in both these compartments.

These analyses indicate protumorigenic effects of NIH-CXCL14 cells, which involve increased cell proliferation in tumors, occurring in the absence of major alterations of the epithelia-stroma ratio.

LNCaP/NIH-CXCL14 Tumors Are Characterized by Increased Macrophage Infiltration and Increased Angiogenesis.

Analysis of angiogenesis in the 2 tumor types, based on CD31 staining of tumor sections, revealed an increased vessel content in NIH-CXCL14 tumors (Fig. 2E). qRT-PCR analyses indicated an ≈2-fold increase in CD31-postive cells along with reduced αSMA levels in the LNCaP/NIH-CXCL14 tumors (Fig. S3). Moreover, a highly significant association (P = 0.0039) was noted between CXCL14 and CD31 levels (Fig. 2F). Analyses of sections double-stained with fluorescent labeled CD31 and αSMA antibodies demonstrated a reduced relative content of pericytes in the vasculature of LNCaP/NIH-CXCL14 tumors (Fig. S4).

Macrophage staining using a CD68 antibody revealed a locally enriched staining pattern, which was found more frequent in the LNCaP/NIH-CXCL14 tumors (Fig. 2G). qRT-PCR analyses with another macrophage-marker, CSF-1R, confirmed higher macrophage content in LNCaP/NIH-CXCL14 tumors (Fig. S3). Also, the levels of CXCL14 and CSF1R mRNA displayed a highly significant correlation (P = 0.0008) (Fig. 2H). However, analysis of NK-cell content (NK1.1) did not indicate any difference between the 2 tumor types (Fig. S3).

Taken together, the data show that LNCaP/NIH-CXCL14 tumors are characterized by a higher density of immature vessels and macrophages.

CXCL14 Stimulates Growth and Migration of Fibroblasts.

To obtain a mechanistic understanding of the observed tumor phenotypes, several in vitro studies were performed. When cultured in 1% FCS, NIH-CXCL14 cells grew significantly faster than control cells (Fig. 3A). Measurements of BrdU incorporation, confirmed increased DNA synthesis in NIH-CXCL14 under low serum conditions (Fig. S5). Importantly, neither NIH-CXCL14 nor NIH-ctr cells were able to form colonies in soft agar under full or serum-reduced conditions.

Fig. 3.

CXCL14 promotes in vitro growth and migration of fibroblasts. (A) The growth of NIH-ctr (open squares) and NIH-CXCL14 cells (filled squares) in 1% FCS culture medium was determined by cell counting. (B) Western blot analysis of CXCL14-induced ERK1/2 phosphorylation in NIH-ctr cells. (C) The cell density of NIH-ctr and NIH-CXCL14 cells was determined by crystal violet staining after culture in low serum in the presence of the MEK inhibitor UO126. (D) For migration studies, fibroblasts were seeded in the upper compartment of a 2-compartment chamber and allowed to migrate for 20 h, after which migrated cells in the lower compartment were collected and counted using a hemocytometer. Error bars indicate SEM. *, P < 0.05, by unpaired t test (A), ANOVA (C), and paired t test (D).

Chemokine signaling involves phosphorylation of, for example, MAPKs [reviewed in Thelen (15)]. In agreement with this, CXCL14 induced a time- and concentration-dependent increase of ERK1/2 phosphorylation in NIH-ctr cells (Figs. 3B and S5). A detectable, but much weaker phosphorylation of AKT was also detected. To investigate the importance of ERK and AKT signaling for the growth of NIH-CXCL14 cells under reduced serum conditions, growth experiments were performed in the presence of the MEK inhibitor UO126 or the PI3K inhibitor Wortmannin. UO126, but not Wortmannin, significantly reduced the growth in 1% FCS of the NIH-CXCL14 cells (Figs. 3C and S5).

Next, the intrinsic migration capacity of CXCL14-expressing fibroblasts was studied in a 2-compartment migration chamber. This assay revealed an increased intrinsic migration capacity of the NIH-CXCL14 cells, as compared to the control cells (Fig. 3D).

Finally, the effects of exogenous recombinant CXCL14 on the growth and migration of immortalized human fibroblasts were analyzed. CXCL14-stimulated responses were observed in both assays (Fig. S6).

These experiments demonstrate that autocrine CXCL14-stimulation of mouse fibroblasts is associated with strong migratory and ERK-dependent proliferative responses. Furthermore, human fibroblasts were also shown to respond to exogenously added CXCL14. Collectively, these data thus identify previously unrecognized growth- and migration-stimulatory effects of CXCL14 on fibroblasts.

NIH-CXCL14 Cells Stimulate in Vitro Migration and Proliferation of LNCaP Cells.

Next, it was investigated whether NIH-CXCL14 cells could promote the growth and migration of LNCaP cells in a paracrine manner.

In an in vitro co-culture assay, NIH-CXCL14 cells were more potent than control cells in stimulating growth of LNCaP cells (Fig. 4A). This growth-promoting effect of CXCL14-producing fibroblasts was also observed in cocultures with prostate cancer PC-3 cells and breast cancer MCF-7 cells (Fig. S7). Also, NIH-CXCL14-derived medium was more potent in stimulating LNCaP migration than control medium (Fig. 4B). In contrast, recombinant CXCL14 did not stimulate LNCaP migration, suggesting that the effects occurred through other factors induced in fibroblasts by CXCL14 (Fig. 4B). This notion was further substantiated by the finding that recombinant CXCL14 failed to induce ERK phosphorylation in LNCaP cells (Fig. 4C) or PC3 cells. However, MCF-7 breast cancer cells, which previously were found to respond to CXCL14 (10), showed a robust ERK phosphorylation after CXCL14 stimulation (Fig. 4C).

Fig. 4.

Conditioned medium from NIH-CXCL14 cells, but not recombinant CXCL14, promote LNCaP proliferation and migration in vitro. (A) The effects of NIH-ctr or NIH-CXCL14 fibroblasts on LNCaP cell growth was determined in a co-culture assay by Rhodanile staining after 10 days. (B) LNCaP cells were allowed to migrate toward control medium alone or supplemented with 100 ng/mL CXCL14 or medium conditioned by NIH-ctr or NIH-CXCL14 cells for 20 h. The number of migrated cells was determined using a hemocytometer. (C) Effects on ERK1/2 phosphorylation in LNCaP cells and MCF cells treated with 100 ng/mL CXCL14 were analyzed by immunoblotting of total cell lysates. Error bars indicate SEM. *, P < 0.05, by paired t test (A) and ANOVA (B).

In summary, these experiments demonstrate that NIH-CXCL14 cells exert paracrine stimulatory effects on LNCaP cells, and these effects are mediated by factors other than CXCL14.

CXCL14 Stimulates Monocyte Migration, and NIH-CXCL14 Cells Display an Enhanced Ability to Simulate in Vivo Angiogenesis.

In agreement with previous studies, CD14⁺ monocytes displayed a migratory response to CXCL14. Conditioned medium from NIH-CXCL14 cells was also more potent than medium from control cells in stimulating migration of CD14⁺ cells (Fig. 5A). This difference was even more prominent when analysis was restricted to the CD14⁺/CD16⁺ subset of monocytes, indicating differential effects on various monocyte subsets (Fig. S8).

Fig. 5.

NIH-CXCL14 cells promote monocyte migration in vitro and angiogenesis in vivo. (A) Freshly isolated monocytes were allowed to migrate toward 10 ng/mL CXCL14, or conditioned medium derived from NIH-ctr or NIH-CXCL14 fibroblasts for 4 h. The number of migrated monocytes present in the lower compartment was determined by FACS analysis of CD14⁺ cells. (B) Effects on in vivo angiogenesis were analyzed by monitoring vessel in-growth in Matrigel plugs mixed with 1% FCS culture medium alone, or medium containing 100 ng/mL CXCL14, NIH-ctr or NIH-CXCL14 cells. Error bars indicate SEM. *, P < 0.05, by ANOVA.

To analyze the angiogenic capacity of the 2 fibroblast types, Matrigel-plugs supplemented with the 2 cell types were implanted subcutaneously, and in-growth of vessels was determined. Matrigel-plugs containing NIH-CXCL14 cells showed a higher vessel density (Fig. 5B). Conditioned media from NIH-CXCL14 cells also induced a more potent proangiogenic effect than control media (Fig. S9). Importantly, qRT-PCR analyses revealed an up-regulation of FGF-2 and, to a lesser extent, VEGF-A, -B, and -C in NIH-CXCL14 cells (Fig. S10), suggesting a possible mechanism for this phenomena. Consistent with a previous study (16), recombinant CXCL14 failed to stimulate angiogenesis (Fig. 5B).

The in vitro analyses of monocyte migration thus suggest that the tumor-promoting effect of NIH-CXCL14 cells involves an increased recruitment of macrophages into the tumor, mediated directly by CXCL14. Furthermore, the Matrigel-plug-assay demonstrates that CXCL14-expression in fibroblasts leads to the induction of factors, including FGF-2, that stimulate angiogenesis.

Discussion

Based on the characterization of human prostate CAFs, this study identifies fibroblast-derived CXCL14 as a novel potential prostate cancer-stimulatory protein (Figs. 1 and 2). Multiple mechanisms whereby fibroblast-derived CXCL14 promote tumor growth were also revealed. These include a direct stimulation by CXCL14 on growth and migration of fibroblasts (Fig. 3 and Figs. S5 and S6), and on attraction of monocytes (Fig. 5 and Fig. S8). Additionally, CXCL14-producing fibroblasts exert more potent paracrine effects, mediated by other factors than CXCL14 itself, on malignant cells and on angiogenesis (Figs. 4 and 5 and Figs. S7 and S9).

CXCL14, also designated BRAK, MIP-2γ, BMAC, or KS1, is an orphan member of the CXC chemokine family, belonging to the subfamily that lacks the amino-terminal ELR motif. Many studies suggest a broad chemotactic activity, as demonstrated for NK cells, dendritic cells, monocytes and macrophages (16–19). However, CXCL14 was found to be dispensable for dendritic-cell and macrophage function under in vivo inflammatory conditions (20). CXCL14 also has a role in insulin signaling (21, 22). A general association between activation of mesenchymal cells and CXCL14 up-regulation, in agreement with the present and earlier findings, are also supported by the finding of CXCL14 up-regulation in activated stellate cells (23).

The functional role(s) of CXCL14 in tumor biology has been addressed in a few previous reports. In apparent contradiction to our present study, these reports claim antitumoral and anti-angiogenic effects of CXCL14 (16, 24, 25). However, 2 of these studies report the effects on tumor formation after overexpression in malignant LAPC4 prostate cancer cells or HSC-3 squamous carcinoma cells (24, 25). The CXCL14 effects described in the present study are predominantly derived from CXCL14-activated fibroblasts. It is thus possible that the absence of protumorigenic effects in the LAPC4 or HSC-3 tumor models is caused by a fibroblast-independence of these tumor models. Concerning the effects on angiogenesis, our studies suggest that the proangiogenic effects of NIH-CXCL14 cells occur through factors that are induced in fibroblasts by CXCL14 (Fig. 5), whereas the anti-angiogenic effects were observed in experiments in which the effects of purified recombinant CXCL14 were analyzed (16). However, the combined results from the present study and from previous publications call for continued analyses of possible tumor type- and stage-specific effects of CXCL14 in additional tumor models. In this context it should also be noted that malignant cells of some tumors might respond directly to CXCL14, as indicated by the CXCL14 responsiveness of MCF7 cells (Fig. 4).

A series of topics for continued mechanistic studies are suggested by the present findings. The identification of the receptor for CXCL14 is highly warranted. Analyses of the effects of GPCR inhibitors with known target profiles will possibly aid in receptor identification. It should also be explored to what extent the increased macrophage recruitment contributes to the increased growth and angiogenesis of LNCaP/NIH-CXCL14 tumors. Identification of the factor(s) that cause the up-regulation of CXCL14 in CAFs is another relevant issue for further analyses.

As a secreted protein, CXCL14 has high “drugability,” and it should be possible to generate neutralizing antibodies, aptamers, and possibly small molecule inhibitors. Based on the limited information of the physiological functions of CXCL14, it is difficult to predict possible adverse effects of CXCL14 antagonists. However, the normal development and viability of CXCL14 knock-out mice (20) are promising with respect to potential use of CXCL14-targeting compounds.

The findings of the present study were achieved through an integrated procedure allowing the identification of novel CAF-derived potential drug targets. This procedure includes expression-profiling of human CAFs followed by mechanistic in vivo and in vitro studies. We hope that the findings of this study will encourage the continued use of this strategy in other tumor types.

In more generalized terms our findings suggest autocrine stimulation of fibroblasts as a previously unrecognized mechanism for chemokine-mediated stimulation of tumor growth, and thereby identify a novel mechanism through which CAFs achieve their protumorigenic phenotype. It is predicted that future analyses of the roles of chemokines in cancer, and of the properties of CAFs, will continue to uncover novel potential therapeutic opportunities.

Methods

Human Tissue Samples and Microdissection.

Tissue samples were collected and used in accordance with the ethical rules of the Department of Pathology, Uppsala University Hospital, and the Department of Oncology-Pathology, Karolinska Institutet, and in agreement with the Swedish biobank legislation. For details of microdissection see SI Methods.

RNA Isolation, cDNA Synthesis and qRT-PCR Analyses.

The PicoPure RNA Isolation Kit (Arcturus Engineering), TRIzol (Invitrogen) and GeneElute (Sigma–Aldrich) were used for RNA isolation from microdissected material, xenograft tumors, and cell lines, respectively. cDNA synthesis and qRT-PCR assays were performed according to instructions of the manufacturer (see SI Methods). For primer sequences, see Table S1.

Histological Analyses.

Immunohistochemical analyses of CD31, CD68, and CXCL14 were performed on frozen tissue sections. PCNA and hematoxylin stainings were done on paraffin embedded tissue. For further details see SI Methods.

Generation of NIH-CXCL14 Cells.

Details are presented in SI Methods. Briefly, human CXCL14 cDNA was cloned into a pBabe puromycin vector, which was used for viral transfection of NIH 3T3 cells. Expression of CXCL14 was confirmed by qRT-PCR and a CXCL14-specific ELISA (R&D Systems).

Xenograft Experiments.

The animal experiments were conducted in accordance with national guidelines and approved by the Stockholm North Ethical Committee on Animal Experiments. The generation of xenograft tumors was performed as described in ref. 14 and in SI Methods.

In Vitro Growth and Migration Assays.

Growth rate of NIH 3T3 and LNCaP cells.

A total of 1 × 10⁴ fibroblasts and 2 × 10⁴ LNCaP cells were seeded in 12-well plates (Sarstedt). Fibroblasts were grown in DMEM supplemented with 1% FCS, whereas LNCaP cells were cultured in conditioned medium from NIH-ctr or NIH-CXCL14 cells. Cell numbers were determined with a Buerker chamber.

Cocultures of LNCaP and NIH 3T3 cells.

A total of 1 × 10⁴ NIH-ctr or NIH-CXCL14 cells were seeded in 96-well plates (Sarstedt). The following day the medium was replaced with 2 × 104 LNCaP cells, resuspended in DMEM with 1% FCS. The number of LNCaP cells after 10 days was determined using Rhodanile dye (Sigma–Aldrich), as described in ref. 26.

Growth of fibroblasts in the presence of MEK inhibitor.

A total of 4 × 10³ fibroblasts were seeded into 96-well plates and treated with DMSO or different concentrations of UO126 (Promega). Cell density after 4 days was determined with crystal violet, as described in ref. 27.

Migration of LNCaP and NIH 3T3 cells.

Migration assays of 4 × 10⁵ fibroblasts and LNCaP cells were performed in Transwell chambers containing 8-μm pore-sized inserts (Costar, Corning).

Monocyte migration.

Monocytes were isolated by centrifugation through a Ficoll gradient (Ficoll-Paque Plus, GE Healthcare) and with CD14-microbeads (Miltenyi Biotec). Their migration through a polycarbonate 5-μm-pore filter toward conditioned media or medium supplemented with different concentrations of recombinant CXCL14 was assayed in 96-well chemotaxis chambers (Neuroprobe). The number of migrated CD14⁺ cells after 4 h was determined by FACS counting.

All in vitro growth and migration experiments were performed at least 3 times.

Analyses of CXCL-14-induced ERK Phosphorylation.

After stimulation of serum-starved cells with CXCL14, cell lysates were prepared and subjected to immunoblotting. Further details are presented in SI Methods.

Matrigel Assay.

A total of 2.4 × 10⁴ NIH-ctr cells or NIH-CXCL14 cells were resuspended in 100 μL of medium and 200 μL of matrigel (BD Biosciences). Matrigel supplemented with only medium or medium containing 100 ng/mL CXCL14 was also prepared. The suspensions were injected into mice (300 μL s.c., n = 6). Plugs were isolated after 1 week, placed in 4% PFA overnight and then kept in 30% sucrose before embedding in Tissue-Tec (Sakura) and sectioning (Microm HM 560 Cryo-Star Cryostat). CD31 immunohistochemsitry was performed as described in SI Methods.