Research ArticleCancer

Pleiotrophin promotes vascular abnormalization in gliomas and correlates with poor survival in patients with astrocytomas

See allHide authors and affiliations

Science Signaling  08 Dec 2015:
Vol. 8, Issue 406, pp. ra125
DOI: 10.1126/scisignal.aaa1690

Growing blood vessels in gliomas

Aggressive gliomas are particularly lethal, in part, because of increased density of blood vessels and abnormal vasculature that enables tumor growth and damages the brain. Various secreted factors, including VEGF and pleiotrophin, act on endothelial cells to promote blood vessel formation. By analyzing patient data, Zhang et al. correlated increased pleiotrophin abundance to more aggressive grades of glioma and decreased survival. When implanted in mice, glioma cells that released pleiotrophin formed larger tumors with more blood vessels and increased VEGF concentrations near the blood vessels. Mice had smaller gliomas and survived longer when treated with inhibitors of ALK, a receptor for pleiotrophin, or an inhibitor of the VEGF receptor. These results suggest that blocking the signals that promote the abnormal blood vessel growth may be beneficial to patients with aggressive gliomas.

Abstract

Glioblastomas are aggressive astrocytomas characterized by endothelial cell proliferation and abnormal vasculature, which can cause brain edema and increase patient morbidity. We identified the heparin-binding cytokine pleiotrophin as a driver of vascular abnormalization in glioma. Pleiotrophin abundance was greater in high-grade human astrocytomas and correlated with poor survival. Anaplastic lymphoma kinase (ALK), which is a receptor that is activated by pleiotrophin, was present in mural cells associated with abnormal vessels. Orthotopically implanted gliomas formed from GL261 cells that were engineered to produce pleiotrophin showed increased microvessel density and enhanced tumor growth compared with gliomas formed from control GL261 cells. The survival of mice with pleiotrophin-producing gliomas was shorter than that of mice with gliomas that did not produce pleiotrophin. Vessels in pleiotrophin-producing gliomas were poorly perfused and abnormal, a phenotype that was associated with increased deposition of vascular endothelial growth factor (VEGF) in direct proximity to the vasculature. The growth of pleiotrophin-producing GL261 gliomas was inhibited by treatment with the ALK inhibitor crizotinib, the ALK inhibitor ceritinib, or the VEGF receptor inhibitor cediranib, whereas control GL261 tumors did not respond to either inhibitor. Our findings link pleiotrophin abundance in gliomas with survival in humans and mice, and show that pleiotrophin promotes glioma progression through increased VEGF deposition and vascular abnormalization.

INTRODUCTION

Gliomas, a heterogeneous group of central nervous system tumors with variable grades and histopathology, are the most common primary malignant brain tumors. Despite new therapeutic approaches, overall survival has not improved substantially over the past 30 years (1). Increased microvascular density, endothelial cell proliferation, and abnormal vascular morphology are hallmarks of glioblastoma (2). The vasculature is hyperpermeable and contributes to patient morbidity by causing vasogenic edema. Therefore, targeting molecules and pathways that regulate glioma angiogenesis and vascular abnormalization represents an attractive therapeutic approach. Several proangiogenic factors have been implicated in glioblastoma angiogenesis, including vascular endothelial growth factor (VEGF) and transforming growth factor–β (3, 4). Antibodies that neutralize VEGF have been approved as a second-line treatment for glioblastoma but have not led to improved overall survival for glioblastoma patients (5). An increased understanding of the molecular regulation of glioma angiogenesis may lead to novel vascular targeting strategies for glioblastoma therapy.

Pleiotrophin (PTN) is a small heparin-binding cytokine that is abundant in the brain during embryonic development and is induced during certain pathological conditions (6). PTN binds to and inactivates receptor-type protein tyrosine phosphatase receptor ζ (PTPRζ), increasing tyrosine phosphorylation of its downstream substrates including β-catenin, β-adducin, Fyn, and P190RhoGAP (7, 8). Another transmembrane receptor activated by PTN is anaplastic lymphoma kinase (ALK) (9). Activation of ALK occurs downstream of PTN-induced inactivation of PTPRζ (10). PTN abundance is increased in several cancer cell lines and human primary tumors, including glioma, breast cancer, lung cancer, melanoma, neuroblastoma, pancreatic cancer, and prostate cancer, and may increase tumor growth either through direct effects on tumor cells or through stimulation of angiogenesis and remodeling of the tumor microenvironment (1120). The abundance of PTN in human astrocytomas correlates to malignancy grade (21). Studies using human glioblastoma cell lines have implicated PTN, PTPRζ, and ALK in the regulation of glioma cell proliferation and migration in vitro and tumor growth in vivo (9, 2227), supporting an important role for PTN in promoting proliferation of glioma cells that endogenously express PTPRζ and ALK. We sought to understand the contribution of PTN in stimulating tumor angiogenesis and establishment of the abnormal vasculature that is a hallmark of glioblastoma. Here, we showed that high PTN abundance correlated with poor survival of patients with astrocytomas. Using an orthotopic, syngeneic model of glioma, we provided evidence that PTN enhanced the growth of intracranial tumors through stimulation of the tumor vasculature. Our results indicate that PTN has an important role in establishing the characteristic abnormal tumor vasculature in glioblastoma and identify PTN as a target for antiangiogenic therapy.

RESULTS

PTN abundance is increased in high-grade gliomas

PTN abundance in human gliomas was assessed by immunohistochemical staining of tissue microarrays (TMAs) representing low- and high-grade gliomas collected retrospectively at Uppsala University Hospital (28, 29). PTN abundance was semiquantitatively scored according to the stained area fraction on a scale from 0 to 3 (Fig. 1A). Consistent with a previous study of a small patient cohort demonstrating increased PTN abundance in glioblastomas as compared to diffuse astrocytomas (21), we found that PTN abundance was increased in high-grade gliomas in this larger cohort (Table 1). PTN abundance was similar when comparing oligodendrogliomas to astrocytomas within each malignancy grade. A significantly higher proportion of high-grade tumors [World Health Organization (WHO) grade III to IV] had PTN compared to low-grade tumors (WHO grade II) and normal brain. Furthermore, high PTN abundance (score, 2 to 3) was found in 40.3% of high-grade tumors compared with only 8.1% in low-grade tumors. We conclude that high-grade gliomas were more likely to be PTN-positive and have higher PTN abundance as compared to low-grade gliomas and nonmalignant control samples (Fig. 1B).

Fig. 1 PTN and ALK abundance is increased in human high-grade gliomas.

(A) Immunohistochemical staining of human brain tumor TMAs showing the semiquantitative scoring of PTN in representative tumors (0, no staining; 1, <25% staining; 2, 25 to 75% staining; and 3, >75% staining) (scale bar, 50 μm). GBM, glioblastoma. (B) Summary of PTN scoring. (C) Immunohistochemical staining of human brain tumor TMAs showing the semiquantitative scoring of PTPRζ in representative tumors (0, no staining; 1, weak staining; 2, moderate staining; and 3, strong staining) (scale bar, 100 μm). (D) Summary of PTPRζ scoring. (E) Immunohistochemical staining of human brain tumor TMAs using an antibody recognizing ALK. ALK staining was scored according to the presence of vascular staining (0, no vessels stained; 1, minority of vessels stained; and 2, majority of vessels stained) (scale bar, 50 μm). (F) Summary of scoring of ALK. Complete scoring data are in Table 1. (G) Immunostaining of a human glioblastoma cryosection with Ulex europaeus agglutinin I (UEA-1) lectin (white) and antibodies recognizing α-smooth muscle actin (α-SMA) (green) and ALK (red) (scale bar, 20 μm). The image is representative of tissue sections from three patients.

Table 1 Summary of TMA quantification of PTN and ALK staining in human glioma.

Higher proportion of grade III and IV tumor samples were PTN-positive in comparison to the grade II tumors (P < 0.0001, χ2 test). The ALK vascular staining was correlated with PTN abundance in tumor cells in the glioma samples of all grades (P < 0.0001, correlation test). Control, nonmalignant brain tissue; AII, grade II astrocytoma; OII, grade II oligodendroglioma; OAII, grade II oligoastrocytoma; AIII, grade III astrocytoma; OIII, grade III oligodendroglioma; OAIII, grade III oligoastrocytoma; GB, grade IV glioblastoma.

View this table:

ALK abundance is increased in the vasculature of high-grade gliomas

To identify cells that can respond to PTN in the tumor microenvironment, we determined the abundance of PTPRζ and ALK with the glioma TMAs. PTPRζ was equally distributed in normal and tumor tissues (Fig. 1, C and D). In contrast, ALK was preferentially detected in the tumor vasculature. ALK abundance was semiquantitatively scored according to the prevalence of vascular staining in a scale from 0 to 2 (Fig. 1E and Table 1). ALK was found in the vasculature of 60.9% of high-grade tumors compared with 21.3% of low-grade tumors (Fig. 1F and Table 1). ALK was not detected in normal brain vessels. Exclusively, grade III to IV gliomas were positive for ALK staining in most vessels (Fig. 1F and Table 1). The presence of ALK in mural cells was significantly correlated to that of PTN in tumor cells when comparing gliomas of all grades (Table 1). To validate the vascular staining of ALK, we stained cryosections from human glioblastoma using an independent antibody directed against ALK in combination with antibodies directed against the vascular mural cell marker α-SMA and the endothelial cell–binding lectin UEA-1. ALK colocalized with α-SMA but not with UEA-1, indicating that ALK is present in vascular mural cells in human gliomas (Fig. 1G).

High PTN abundance is associated with poor patient survival in high- and low-grade astrocytomas

Next, we evaluated whether PTN abundance correlated with patient survival. Survival data, pathological assessment, and PTN staining were available for 31 of the patients represented in the TMA with low-grade (grade II) astrocytoma. PTN abundance was significantly associated with shorter survival of patients with low-grade astrocytomas (Fig. 2A). The median survival time was 1620 days for patients with tumors that lacked PTN compared to 1058 days for patients with PTN-positive tumors. Survival analysis was performed in a similar manner for high-grade astrocytomas (grade III to IV), where survival data, pathological assessment, and PTN staining were available for 79 patients (15 grade III and 64 grade IV). Patients with high-grade astrocytomas with high PTN abundance (staining score, 2 to 3) showed significantly decreased survival as compared to patients with tumors with low PTN abundance (staining score, 0 to 1) (Fig. 2B). The median survival time was 337 days for patients with tumors with undetectable or low PTN abundance compared to 292 days for patients with tumors having high PTN abundance.

Fig. 2 High PTN abundance is associated with shorter patient survival in grade II to III astrocytoma and glioblastoma in univariate survival analysis.

(A) Survival plot for patients with grade II astrocytoma with low PTN abundance (score, 0) or high PTN abundance (score, 1 to 3) (P = 0.0076, log-rank test). (B) Survival plot for patients with grade III astrocytomas and glioblastomas with low PTN abundance (score, 0 to 1) or high PTN abundance (score, 2 to 3) (P = 0.0377, log-rank test). (C and D) Kaplan-Meier plot for patients from The Cancer Genome Atlas (TCGA) patient cohort dichotomized by PTN mRNA abundance. (C) Plot for TCGA patients with grade II and III astrocytomas (P = 0.0007, log-rank test). (D) Plot for TCGA patients with glioblastomas (P = 0.0263, log-rank test). (E) Box plots showing PTN mRNA abundance in patients with grade II and III astrocytoma from the TCGA with wild-type (WT) isocitrate dehydrogenase 1 (IDH1) or with mutated IDH1. (F) PTN mRNA abundance in patients with grade II and III astrocytoma from the TCGA with WT α-thalassemia/mental retardation syndrome X-linked (ATRX) or with mutated ATRX. (G) PTN mRNA abundance in glioblastomas from the TCGA with G-CIMP (glioma–CpG island methylator phenotype–negative) signature or with G-CIMP+ signature. (H) PTN mRNA abundance in samples from glioblastomas from the TCGA with WT IDH1 or with mutated IDH1.

The relationship between tumor PTN mRNA expression and patient survival was further analyzed in independent patient cohorts using data from TCGA database (http://cancergenome.nih.gov). Ninety-eight patients with astrocytomas (grade II to III) from the TCGA database were evaluated. PTN gene expression data were extracted from the RNA sequencing array in the cBioPortal database. Survival curves (Fig. 2C) showed a median survival of 1762 days for patients with astrocytomas having low PTN abundance and 605 days for patients with astrocytomas with high PTN abundance (Table 2). Analysis of two additional publicly available glioma data sets [Gravendeel (30) and Rembrandt (31)] supported these results and revealed significantly shorter survival of grade II to III astrocytoma patients with tumors with high PTN abundance (fig. S1, A and B).

Table 2 Univariate analysis of survival of patients with grade II and III astrocytomas included in the TCGA database.

CI, confidence interval.

View this table:

To analyze the impact of PTN mRNA expression on survival of patients with glioblastomas, PTN gene expression data were extracted from the Affymetrix array (U133 platform) in the cBioPortal database. Survival curves (Fig. 2D) showed a median survival of 394 days for patients with glioblastomas having low PTN abundance and 378 days for the patient group with glioblastomas with high PTN abundance (Table 3). This result was supported by the Gravendeel data set (fig. S1C), in which glioblastoma patients with tumors with high PTN abundance had significantly shorter survival, whereas no significant difference in survival was found in the Rembrandt data set (fig. S1D).

Table 3

Univariate analysis of survival of patients with glioblastoma included in the TCGA database.

View this table:

The G-CIMP, the IDH1 gene mutation, and the ATRX gene mutation are used as molecular markers for glioma diagnosis and prognosis, and patients with G-CIMP, IDH1 mutation, or ATRX mutation have better prognoses (3234). By correlating PTN mRNA expression in tumors from the TCGA database with the mutation status of IDH1 and ATRX and the reported CIMP of these tumors, we found that high abundance of PTN mRNA significantly correlated to wild-type IDH1 and ATRX in grade II and III astrocytomas and was associated with wild-type IDH1 and G-CIMP–negative glioblastomas (Fig. 2, E to H).

PTN does not affect cell proliferation of GL261 gliomas cells in vitro

PTN increases the proliferation and migration of glioma cells bearing its receptors (9, 11). To investigate the paracrine effects of PTN in gliomas, we used a syngeneic, orthotopic glioma model using GL261 cells, which lack endogenous PTN or ALK and have only minimal PTPRζ protein abundance (fig. S2, A to D) (35). We used a lentivirus-based system to generate a subline of GL261 cells, designated GL261PTN, that constitutively expressed PTN, luciferase (to enable in vivo monitoring of tumor growth), and green fluorescent protein (GFP) (to enable detection of tumor cells in brain sections). In parallel, GL261 control cells expressing only luciferase and GFP were generated (GL261con). PTN mRNA and protein were robustly detected in GL261PTN cells but undetectable in parental GL261 or GL261con cells (fig. S2, C and D). Similar amounts of luciferase activity were detected in GL261con and GL261PTN cells (fig. S2E). Consistent with low PTPRζ abundance and lack of ALK in GL261 cells (fig. S2, A and B), parental GL261wt, GL261con, and GL261PTN cells showed similar growth curves, demonstrating that PTN overexpression does not affect GL261 cell proliferation in vitro (fig. S2F).

PTN promotes GL261 tumor growth and causes poor survival in vivo

To examine whether PTN overexpression affects growth of GL261 glioma in vivo, we orthotopically inoculated GL261con and GL261PTN cells into the cerebral cortex of syngeneic C57BL/6 mice. Analysis of PTN mRNA expression confirmed that PTN was robustly expressed in GL261PTN tumors (Fig. 3A). Tumor growth was accelerated in PTN-overexpressing tumors as compared to control tumors (Fig. 3, B and C). Consistent with this finding, survival was significantly decreased in mice inoculated with PTN-overexpressing tumors as compared to those inoculated with GL261con tumors (Fig. 3D). To further investigate the role of PTN in affecting glioma growth through stimulation of the tumor microenvironment, we used the CT-2A astrocytic glioma cell line, which has endogenous PTN but not PTPRζ or ALK (fig. S1, A, B, and G). In luciferase-expressing CT-2A sublines, we found that knockdown of PTN did not affect proliferation of CT-2A cells in vitro (fig. S1, G to I). In accordance with the increase in GL261 glioma growth associated with PTN overexpression, knockdown of PTN significantly decreased the growth of intracranial CT-2A tumors (Fig. 3, E and F). We evaluated proliferation and apoptosis in GL261PTN and GL261con tumors by immunostaining for phosphorylated histone H3 and cleaved caspase-3, respectively (Fig. 3G and fig. S3A). PTN-overexpressing tumors had more proliferating cells than control tumors but no change in apoptotic cells (Fig. 3H and fig. S3B). Our data show that PTN accelerates tumor growth and decreases mouse survival in orthotopic GL261 gliomas.

Fig. 3 Expression of PTN promotes tumor growth in vivo and leads to poor survival.

(A) Mice were inoculated with GL261con or GL261PTN cells, and hPTN mRNA expression of GL261con or GL261PTN tumor tissue was measured by quantitative polymerase chain reaction (qPCR). HPRT, hypoxanthine phosphoribosyltransferase. Data are means ± SEM of at least four mice per group, and the experiment was performed twice. (B and C) Luciferase signals (B) and growth curve (C) in mice bearing GL261con and GL261PTN tumors at day 23. Data are means ± SEM; representative growth curve and image from two independent experiments using 6 to 10 mice per group each time. (D) Survival curve of control (con) GL261 tumor–bearing mice and PTN-producing GL261 tumor–bearing mice. Median survival time: control group, 30.5 days; PTN group, 28 days (n = 8 to 10 mice per group). (E and F) Growth curve (E) and luciferase signals (F) in mice bearing CT-2A control [CT2A NS; nonsense short hairpin RNA (NS shRNA)] tumors and CT-2A PTN shRNA–expressing (CT2A SH) tumors at day 23 (n = 8 mice per group). (G) Immunohistochemical staining of GL261con and GL261PTN tumors with an antibody directed against phosphorylated histone H3 (P-H3) (scale bar, 50 μm). (H) Quantification of phosphorylated histone H3–positive cells normalized to total cells (mean ± SEM; n = 4 mice per group). (I) Quantification of CD31/phosphorylated histone H3 double-positive cells normalized to total endothelial cells (ECs) (mean ± SEM; n = 4 mice per group). (J) Double staining of CD31 (red) and phosphorylated histone H3 (green) in GL261con and GL261PTN tumors. Arrows indicate double-positive cells (scale bar, 100 μm).

PTN overexpression increases microvascular proliferation and modulates tumor vessel morphology and function

We evaluated endothelial cell proliferation in GL261con and GL261PTN tumor vessels by immunostaining for CD31 and phosphorylated histone H3. PTN-overexpressing tumors had more proliferating endothelial cells than control tumors (Fig. 3, I and J). To investigate whether PTN regulates vascular density and/or vessel morphology, we measured microvessel density and blood vessel diameters in GL261PTN and GL261con tumors using stereological quantification (36). Consistent with an increased number of proliferating endothelial cells, microvascular area was greater in GL261PTN tumors than in control tumors (Fig. 4, A and B). Moreover, blood vessels in PTN-overexpressing tumors had larger lumens (Fig. 4, A and C). The abnormal vascular morphology was accompanied by decreased coverage by desmin-positive pericytes in PTN-expressing tumors (Fig. 4, D and E). ALK was present in perivascular cells in mouse GL261 gliomas (fig. S4), consistent with the presence of ALK in vascular mural cells in human gliomas. Mutations that result in constitutively active ALK fusion proteins increase VEGF production in anaplastic large-cell lymphomas (37), and high VEGF abundance reduces pericyte coverage and destabilizes vessels by suppressing platelet-derived growth factor receptor β (PDGFRβ) signaling (38). Therefore, we compared VEGF abundance in GL261PTN and GL261con tumors using an Flt1–immunoglobulin G 1 (IgG1) fusion protein to detect VEGF in the tissue (39). VEGF was mainly deposited in the perivascular area surrounding blood vessels, and the amounts were significantly higher in PTN-overexpressing tumors, coinciding with reduced pericyte coverage and increased angiogenesis in these tumors (Fig. 4, F to I). VEGF staining partially overlapped with desmin-positive pericytes and, to a lesser extent, with CD31-positive endothelial cells, and was also found in the rim of the vasculature (Fig. 4G).

Fig. 4 PTN expression in GL261 tumors modulates tumor vessel morphology and function.

(A) Immunostaining of tumor sections with a CD31 antibody (scale bar, 100 μm). (B and C) Stereological quantification of vessel area (B) and vessel diameter (C) in GL261con and GL261PTN tumors (mean ± SEM; n = 4 mice per group). (D) Quantification of the proportion of CD31-positive vessel area that is covered by desmin-positive pericytes (mean ± SEM; n = 6 to 7 mice per group). (E) Immunostaining of CD31 (gray) and desmin (red) in GL261con and GL261PTN tumors (scale bar, 100 μm). (F) Immunodetection of VEGF in GL261con and GL261PTN tumors using an sFlt1 domain (red) and antibodies against CD31 (green). Hoechst 33342 stain shows nuclei (blue) (scale bar, 100 μm). (G) Immunostaining of CD31 (gray), desmin (green), and VEGF (red) in GL261 tumors. Hoechst 33342 stain shows the nuclei (blue) (scale bar, 20 μm). (H) Negative control for sFlt1staining. Blood vessels shown in green with CD31 antibody. Hoechst 33342 stained the nuclei (blue) (scale bar, 100 μm). (I) Quantification of VEGF staining shown in (F) (means ± SEM; n = 6 mice per group). (J and K) Quantification of CD31 staining and lectin-perfused vessels (means ± SEM; n = 6 to 7 mice per group). (L) Immunostaining of CD31 (red); perfused vessels are indicated by the presence of lectin (green) (scale bar, 50 μm).

Next, we analyzed perfusion and vascular permeability by injecting biotinylated lectin, which labels endothelial cells in perfused vessels (40), and Alexa 555–labeled cadaverine, a tracer for vascular leakage (41), before sacrificing mice bearing GL261PTN or GL261con tumors. The experiment was terminated 17 days after tumor cell injection, when the mean tumor size was similar between the two groups. Vascular permeability was evaluated by determining the area of cadaverine leakage and by analyzing the area of endogenous mouse IgG leakage. Substantial leakage of cadaverine and IgG was detected within both GL261con and GL261PTN tumors and at the tumor borders (fig. S5, A and B). There was no significant difference in vascular permeability between the two groups (fig. S5, C and D). Although PTN-expressing tumors had more vascular area (Fig. 4, J to L), they had significantly reduced vessel functionality as indicated by a reduced proportion of lectin-perfused vessel (Fig. 4, K and L). Our data show that PTN increases tumor angiogenesis and modulates vascular morphology, leading to increasingly abnormal vessels with reduced vascular function.

To determine whether PTN abundance was associated with vascular abnormalization in human glioblastomas, we performed stereological quantifications of CD34-stained vessels of glioblastomas from the glioma TMAs (Fig. 5A). The presence of PTN in human glioblastoma tissue did not change the vessel density but was significantly correlated to increased vascular area and larger vessel diameter (Fig. 5, B to D).

Fig. 5 PTN detection in human glioblastoma correlates with increased vascular area and enlarged diameter of vessels.

(A) Representative photomicrographs of CD34-stained human glioblastoma TMA samples corresponding to PTN-negative (left panel) or PTN-positive (right panel) immunohistochemical stainings (scale bar, 200 μm). (B to D) Box plots of microvascular density defined as number of vessels per area (B), vessel diameter (C), and vascular area (D) both determined by stereological quantification of the human glioblastoma TMA samples (n = 8 samples for PTN score 0 group and n = 22 samples for PTN score 1 to 3 group). ns, not significant.

PTN and VEGF cooperate during angiogenic sprouting in embryoid bodies

PTN expression in GL261 tumors was associated with increased amounts of VEGF deposited proximal to the vasculature. To determine whether PTN impinges on VEGF-induced angiogenesis, we used the embryoid body (EB) model of vascular development because of the high abundance of ALK and PTPRζ in endothelial cells in this model (fig. S1, A and B). PTN was not detected in R1 embryonic stem (ES) cells but was induced upon formation of EBs (fig. S6). ES cells were transduced with lentiviruses expressing two different shRNAs designed to knock down Ptn (R1-SH1 and R1-SH2), nonsense shRNA (R1-NS), or lentiviruses overexpressing human PTN (R1-PTN) (Fig. 6A). EBs were induced to undergo VEGF-induced angiogenesis either in a three-dimensional (3D) collagen matrix or on glass slides. Angiogenesis did not occur in the absence of VEGF in either condition, and PTN was not sufficient to induce vessel formation in the absence of VEGF. VEGF-induced sprouting of endothelial cells into a 3D matrix and vascular ring formation in EBs cultured on glass slides were impaired when PTN was knocked down (R1-SH1 and R1-SH2 EBs; Fig. 6, B to E). PTN overexpression did not affect the magnitude of VEGF-induced sprouting into a 3D matrix (Fig. 6, B and C) but enhanced vascular ring formation and increased vessel density when EBs were cultured on glass slides (Fig. 6, D and E).

Fig. 6 PTN modulates angiogenesis in EBs.

(A) qPCR analysis of expression of mouse or human PTN expression in control-transfected (R1-NS), PTN-overexpressing (R1-PTN), and PTN shRNA–expressing (R1-SH1 and R2-SH2) EBs. Data are means ± SEM from three experiments. (B) Immunostaining of endothelial cells (CD31; red) in EBs cultured in 3D collagen gels with VEGF (20 ng/ml) until day 14. Hoechst 33342 stain shows nuclei (blue) (scale bar, 500 μm). (C) Quantification of CD31-positive sprout area per EB in 3D culture (data are means ± SEM; n ≥ 13 EBs per group, and the experiment was performed twice). (D) Immunostaining of endothelial cells (CD31) in 2D EBs cultured with VEGF (20 ng/ml) until day 9 (scale bar, 500 μm). (E) Quantification of vascular area per EB in 2D culture (data are means ± SEM; n ≥ 8 EBs per group, and experiment was performed twice). (F) Immunostaining of CD31 (red) and α-SMA (green) in control Matrigel plugs, PTN-containing Matrigel plugs, VEGF-containing Matrigel plugs, and Matrigel plugs containing PTN in combination with VEGF (scale bar, 100 μm). (G) Quantification of vascular branch points in the Matrigel plugs (mean ± SEM; n = 5 mice per group).

PTN does not induce angiogenesis in the Matrigel plug assay

To further investigate cooperation of PTN and VEGF during angiogenesis, we injected Matrigel alone or supplemented with PTN, VEGF, or a combination of PTN and VEGF into the flanks of nude mice. VEGF significantly enhanced angiogenesis in the plugs, whereas PTN failed to induce vessel formation above control conditions and did not alter the magnitude of the angiogenic response to VEGF in this model (Fig. 6, F and G).

PTN enhancement of GL261 tumor growth is mediated by an ALK-mediated increase in VEGF

To determine whether PTN-induced activation of ALK enhanced GL261 tumor growth, we treated mice bearing PTN-expressing or control GL261 gliomas with the ALK inhibitor crizotinib. Crizotinib reversed the growth-promoting effect of PTN expression but did not affect the growth of control GL261 gliomas (Fig. 7, A and B). Similar results were obtained when treating tumor-bearing mice with the second-generation ALK inhibitor ceritinib (Fig. 7, C and D). Tumor growth inhibition was associated with decreased abundance of VEGF in PTN-expressing GL261 gliomas treated with crizotinib (Fig. 7, E and F). To determine whether the ALK-induced increase in VEGF was responsible for the enhanced tumor growth, we used the VEGF receptor (VEGFR) inhibitor cediranib to target VEGF signaling pathways in PTN-expressing and control GL261 tumors. Whereas GL261 control tumors did not respond to cediranib, the growth of PTN-expressing GL261 tumors was repressed to that of control tumors (Fig. 7, G and H). Treatment of PTN-expressing GL261 tumors with either the ALK inhibitor crizotinib or the VEGFR inhibitor cediranib reversed the increases in vessel area and vascular diameter associated with PTN overexpression (Fig. 7, I and K).

Fig. 7 PTN tumor growth– and angiogenesis-promoting effects are abolished by inhibition of ALK or VEGFR signaling.

(A and B) Graph showing bioluminescence signal corresponding to the relative tumor size (A) and luciferase imaging at day 21 (B) in crizotinib-treated and untreated GL261PTN and GL261con group. n = 3 to 8 mice per group; mean ± SEM are shown. (C and D) Graph showing bioluminescence signal corresponding to the relative tumor size (C) and luciferase imaging at day 23 (D) in ceritinib-treated and untreated GL261PTN and GL261con group (n = 5 to 8 mice per group). (E) Immunofluorescence staining of VEGF, blood vessels, and cell nuclei using an sFlt1 domain (red), anti-CD31 antibody (green), and Hoechst 33342 (blue), respectively, in the GL261tumors (scale bar, 50 μm). (F) Quantification of VEGF staining in the perivascular area in the tumors. n = 4 to 7 mice per group; data are means ± SEM. (G and H) Luciferase signals (G) and growth curves (H) of mice with GL261PTN and GL261con tumors treated with or without cediranib (n = 3 to 6 mice per group; means ± SEM are shown). (I) Immunofluorescence staining of CD31 (red) in GL261con or GL261PTN tumors from mice left untreated or treated with crizotinib or cediranib (scale bar, 200 μm). (J and K) Stereological quantification of vessel area (J) and vessel diameter (K) in GL261con and GL261PTN tumors from mice left untreated or treated with crizotinib or cediranib (n = 3 to 8 mice per group). Data are means ± SEM.

DISCUSSION

PTN stimulates proliferation and migration of glioma cells bearing its receptors (11). Accordingly, antagonistic antibody blocking or knocking down PTN and/or its receptors PTPRζ or ALK in glioma cells through RNA interference or ribozyme targeting leads to decreased growth of tumor xenografts in immunodeficient mice (23, 25, 27, 42, 43). The abundance of PTN in human gliomas and glioma cell lines is variable, but only glioma cell lines with high abundance of PTN and its receptors have been used to study the role of PTN in glioma growth in vivo. These studies convincingly show the presence of a PTN autocrine loop in a subset of gliomas, but the intrinsic proliferative response of the tumor cells to PTN signaling precludes conclusions regarding the role of PTN in modulating the tumor microenvironment. We used the GL261 murine glioma cell line with minimal amounts of endogenous PTN receptors. Overexpression of PTN in GL261 cells did not affect proliferation in vitro but led to increased tumor growth, enhanced angiogenesis, and changes in vascular morphology when transplanted orthotopically in syngeneic mice. Our results demonstrated that PTN aggravated glioma development through modulation of the tumor vasculature. In the GL261 model, PTN expression was associated with increased vascular abnormalities with larger vessel diameter, decreased pericyte coverage, and poor perfusion. PTN expression did not affect vascular permeability in this model, possibly because of the increased permeability of vessels in control GL261 tumors and large intratumoral variation. We cannot exclude that PTN enhances permeability in human glioma. Consistent with the GL261 study, the presence of PTN in the human glioblastoma tissue correlated with increased vascular area and diameter, supporting a role of PTN in vascular abnormalization.

PTN overexpression in GL261 tumors led to increased microvascular area and endothelial cell proliferation, demonstrating that PTN acted as a proangiogenic factor in glioma. PTN has previously been assigned both pro- and antiangiogenic roles in different model systems. PTN increases tumor angiogenesis in breast cancer (19, 44), choriocarcinoma (45), and melanoma (14), consistent with correlation of PTN abundance to vascular density in human tumors, including gliomas (24). In contrast, expression of PTN decreases tumor vascularization in neuroblastoma (46) and inhibits angiogenesis induced by C6 rat gliomas cells cultured in the chicken embryo chorioallantoic membrane (47). The contradictory effects of PTN on regulating angiogenesis in different systems may depend on cues in the microenvironment, such as interaction with other growth factors or the presence of co-receptors. Indeed, αvβ3 integrin is required for PTN to stimulate endothelial and glioma cell migration in vitro, and the absence of αvβ3 integrin instead leads to PTN inhibiting migration (48). Crosstalk between PTN and proangiogenic VEGF signaling has been assigned both inhibitory and stimulatory effects. The angiostatic effect of PTN has been associated with inhibition of VEGF signaling through direct binding to VEGF (47). Conversely, PTN has been shown to increase VEGF release through inactivation of PTPRζ and subsequent activation of β-catenin signaling pathways (49). We showed that knockdown of PTN inhibited VEGF-induced angiogenesis in EBs and that PTN overexpression enhanced formation of vascular structures in EBs cultured on glass slides, indicating functional synergy between PTN and VEGF in the EB model of developmental angiogenesis. In contrast, PTN did not affect VEGF-induced angiogenesis in the Matrigel plug assay or induce angiogenesis when administrated in Matrigel plugs alone. These findings supported the notion that the interplay between PTN and VEGF during angiogenesis was model-dependent and that PTN may cooperate with other proangiogenic growth factors to induce its effects. Our results indicate that PTN-induced angiogenesis and enhancement of tumor growth occurs through a VEGF/VEGFR-dependent mechanism in GL261 gliomas as discussed below.

Chromosomal rearrangements, gene amplifications, or mutations that constitutively activate ALK have been found in many types of human cancer, and ALK inhibitors such as crizotinib and ceritinib have been approved or are undergoing clinical trials as cancer therapies (50). ALK abundance is increased in some glioma cell lines and human tumors (9, 11). Here, we only detected diffuse ALK staining in tumor cells in a few gliomas included in the TMA analysis. The most intense staining was found in the tumor vessels, suggesting that ALK could have a direct role in regulating vascular function. Notably, ALK was present in mural cells and PTN in tumor cells in most glioblastomas in our study. Constitutively activated ALK induces VEGF secretion in anaplastic large-cell lymphoma (37). Similar to this, expression of PTN in GL261 tumors was associated with increased VEGF amounts predominantly deposited in close proximity to the tumor vessels. Treating tumor-bearing mice with either of the two ALK kinase inhibitors crizotinib or ceritinib led to decreased growth of PTN-expressing GL261 tumors. In PTN-expressing GL261 tumors treated by crizotinib, the perivascular deposition of VEGF was reduced. This finding is consistent with a role of ALK in inducing VEGF production in mural cells, leading to a high local concentration of VEGF in direct proximity to the endothelial cells. Moreover, tumor growth was inhibited only in PTN-expressing GL261 tumors after treatment with the VEGFR inhibitor cediranib, emphasizing an important role of PTN-induced VEGF increase in enhancing tumor growth. It is tempting to speculate that intratumoral PTN abundance may be useful as a biomarker to identify glioblastoma patients who are likely to respond to VEGF-targeting therapies. This possibility warrants further investigation.

We showed that high intratumoral PTN abundance correlated to poor survival in a clinical cohort of patients with low- and high-grade astrocytomas. Although we demonstrated that PTN was increased with malignancy grade and that PTN correlated to shorter survival of astrocytoma patients, it is important to bear in mind that this study was performed through assessment of retrospectively collected cohorts of glioma patients using a univariate analysis. Furthermore, high abundance of PTN was associated with wild-type IDH1 and ATRX in grade II and III astrocytomas, and correlated to the non–G-CIMP phenotype and wild-type IDH1 in glioblastomas. By correlating clinical outcomes with integrative genomic analysis, most low-grade gliomas with wild-type IDH have recently been found to be similar to glioblastoma (51). This may explain the poor outcome of individuals with PTN-expressing, low-grade tumors in our study. We therefore conclude that our data do not support a role for PTN as an independent prognostic marker but rather suggest that high PTN abundance is a feature of tumor groups previously found to have a poor survival.

Our results predict that inhibition of PTN signaling in the tumor microenvironment would reduce perivascular amounts of VEGF and normalize the tumor vasculature. Consistent with this notion, treatment with the ALK inhibitor crizotinib or the VEGFR inhibitor cediranib reduced vascular density and decreased vascular diameter specifically in PTN-expressing GL261 tumors. Because PTN abundance is limited in the adult, targeting PTN would be expected to specifically affect the tumor growth and lead to less adverse events than targeting VEGF itself. The role of PTN in driving vascular abnormalization in glioma, together with its previously established role in stimulating glioma cells bearing its receptors, suggests that PTN or its downstream signaling molecules may serve as new therapeutic targets to achieve vascular normalization. The development of efficient and safe strategies to reduce PTN signaling in glioma is an important area of further investigation.

MATERIALS AND METHODS

TMAs and image analysis

TMAs of human brain tumors representing low- and high-grade gliomas collected retrospectively at Uppsala University Hospital contains duplicate tissue cores (1 mm in diameter) representing characteristic areas from tumors and nonmalignant brain tissue as summarized in Table 1 (28). The abundance of PTN, PTPRζ, and ALK was analyzed by immunohistochemistry-based protein profiling using antibodies listed in table S1. Immunohistochemistry and evaluation of immunostaining were performed as previously described (4). Quantification was done by two individual scientists in a blinded fashion. For statistical analysis, tumors were dichotomized into low or high PTN expression according to immunostaining scores. Survival curves were plotted according to the Kaplan-Meier method.

Survival and PTN expression analysis of patients in the TCGA, Gravendeel, and Rembrandt data set

Patient information and mRNA expression data from glioblastoma samples were collected as described (TCGA Research Network 2008). IDH1 mutation status, ATRX mutation status, and G-CIMP status have been published or are available in the public access data portal (www.cbioportal.org/public-portal/) (52). Processed data sets were obtained from the public access data portal (www.cbioportal.org/public-portal/). Data sets were cross-referenced using tumor-specific numbers. Patient information and mRNA expression data for Gravendeel and Rembrandt data sets were downloaded from the Gliovis database (http://gliovis.bioinfo.cnio.es/) (30, 31).

In all three data sets, the grade II and III astrocytoma patients with top 25% or bottom 75% PTN abundance were dichotomized into high PTN– or low PTN–expressing subgroups, respectively. The glioblastoma patients with top 75% or bottom 25% PTN abundance were dichotomized into high PTN– or low PTN–expressing subgroups, respectively. Survival curves were plotted by Kaplan-Meier method using the date of histological diagnosis to the date of death or last follow-up.

Cell culture and generation of cell lines

The GL261 murine glioma cell line was a gift from G. Safrany, National Research Institute for Radiobiology and Radiohygiene (NRIRR), Budapest. Cells were cultured in Dulbecco’s modified Eagle’s medium (61695, Life Technologies Gibco) with 10% heat-inactivated fetal bovine serum (F7524, Sigma). The cells were transfected with lentiviruses, and GFP-positive cells were isolated by fluorescence-activated cell sorting to generate stable cell lines GL261-GFP-LUC (GL261con) and GL261-GFP-LUC-PTN (GL261PTN).

R1 cells (mouse ES cells) were transduced with lentiviruses with PTN shRNA or PTN overexpression constructs, and GFP-positive cells were isolated using a fluorescence-activated cell sorter to generate stable cell lines. Transduced R1 cells were aggregated in hanging droplets to create EBs, cultured in eight-well chamber glass slide (BD) or a 3D collagen matrix as previously described (53), and treated with mVEGF-A 164 (20 ng/ml; PeproTech). Endothelial cells were isolated from EBs as described (54).

The CT-2A cell line was a gift from T. Seyfried, Boston College. The cells were cultured in RPMI-1640 (Life Technologies Gibco) with 10% heat-inactivated fetal bovine serum (F7524, Sigma). CT-2A cells were transfected with lentiviruses, and GFP-positive cells were isolated by fluorescence-activated cell sorting to generate the stable cell lines CT-2A–GFP–LUC (CT-2A NS) and CT-2A–GFP–LUC–SH2 (CT-2A shRNA).

In vitro proliferation assay

GL261 cells (5 × 104) or CT2A cells (1 × 105) were seeded in 10-cm dishes and counted daily for up to 6 days using a cell counter (Z1 counter, Beckman).

Orthotopic GL261 and CT-2A glioma models

Six- to 8-week-old C57BL/6 female mice were purchased from Taconic M&B. Mice were anesthetized with 2.5% isoflurane, and a midline incision was made on the scalp. At stereotactic coordinates of bregma, −1 mm anteroposterior and +1.5 mm mediolateral, a small hole was drilled in the skull. GL261 cells (1 × 104) or CT-2A cells (5 × 104) were delivered in 2 μl of Dulbecco’s phosphate-buffered saline (14040, Life Technologies Gibco) at depth of 2.7 mm over 2 min (55). After the needle was withdrawn, the incision was sutured and the mice were placed on a heated surface until fully recovered from anesthesia. Mice were imaged every second day from day 13 by in vivo bioluminescence imaging as described below and were sacrificed 23 days after injection of tumor cells. For survival studies, mice were sacrificed when the tumor induced symptoms such as hunched posture, lethargy, persistent recumbency, or weight loss of more than 10%. For inhibition studies, crizotinib, ceritinib, and cediranib were purchased from MedChemExpress. Crizotinib and ceritinib were formulated in 0.5% methylcellulose/0.5% Tween 80 and cediranib was 1% Tween 80. Mice received daily treatments of crizotinib (25 mg/kg), ceritinib (50 mg/kg), or cediranib (6 mg/kg) by oral gavage starting from 5 days after tumor inoculation until the mice were sacrificed at day 21 in the crizotinib and cediranib studies or day 23 in the ceritinib study.

Statistical analysis

Data were analyzed using GraphPad Prism 5.0. Mann-Whitney test or Student’s t test was used for comparison between two groups. For TMA analysis, Kruskal-Wallis test with Dunnett’s posttest was used. Survival curves were plotted according to the Kaplan-Meier method (product-limit method), using JMP version 10.0 (SAS Institute Inc.). Univariate survival analysis was performed with log-rank test. SEM was indicated as an error bar in the figures. All statistical tests were two-sided, and P values smaller than 0.05 were considered statistically significant.

Lentivirus production

For overexpression of PTN, a complementary DNA (cDNA) sequence encoding human PTN was subcloned into a third-generation self-inactivating lentiviral vector together with coding sequences for GFP and firefly luciferase (Luc) to generate pBMN(GFP-Luc-PTN). The coding sequences were under control of the immediate-early cytomegalovirus promoter and separated by viral 2A self-cleaving peptides.

For knockdown of Ptn, shRNA sequences (SH1 or SH2) targeting mouse Ptn or a nonsense shRNA sequence was cloned under the control of the H1 promoter into pBMN-based vectors. The shRNA sequences are presented in table S2 with underlined sequences indicating the self-complementary parts. The vectors used for knockdown also contained GFP-Luc expression driven by the elongation factor 1α (EF1α) promoter. Lentivirus particles were produced as previously described (47).

Western blot

Cells were lysed in LDS sample buffer and reducing agent (Life Technologies). After vigorous pipetting and incubation at 70°C for 15 min, samples were separated on NuPAGE 4–12% Bis-Tris gels (Life Technologies) using MES running buffer (Life Technologies) and transferred to a Hybond-C Extra filter (GE Healthcare). Membranes were blocked with 5% milk in tris-buffered saline + 0.01% Tween (TBS-T). Primary antibodies (goat anti-PTN; PL187, Millipore) diluted in blocking solution were incubated overnight at 4°C. Membranes were washed in TBS-T and incubated with horseradish peroxidase (HRP)–conjugated secondary antibodies (mouse anti-goat HRP; clone GT-34, Sigma-Aldrich). Membranes were washed in TBS-T and TBS before detection using ECL Plus substrate and ECL film (GE Healthcare).

cDNA synthesis and qPCR

cDNA from total RNA was synthesized using random hexamer primers and SuperScript III reverse transcriptase (Life Technologies), according to the manufacturer’s instructions. qPCRs were performed using 2× SYBR Green PCR Master Mix (Life Technologies) with 0.25 μM sense and antisense primer per well. Gene expression relative to HPRT was calculated according to the following formula: relative expression gene X = 2−(CT HPRT – CT gene X). Primer sequences are listed in table S3.

Immunofluorescence and immunohistochemistry

Immunohistochemistry was performed on 6-μm sections of snap-frozen tissue embedded in optimal cutting temperature (OCT; Tissue-Tek Sakura) fixed in acetone (Sigma-Aldrich). Endogenous peroxidase was blocked with 1% H2O2 for 20 min. Slides were blocked with 3% bovine serum albumin (Roche Diagnostics) in phosphate-buffered saline (PBS) and incubated with primary antibody diluted in blocking solution for 2 hours, followed by incubation with biotinylated secondary antibody and streptavidin conjugated to HRP (Vector Laboratories). The staining was developed with the AEC kit (Vector Laboratories) according to the manufacturer’s protocol. Slides were counterstained with hematoxylin (Histolab) and rinsed with tap water before mounting with permanent aqueous mounting solution (Dako).

Immunofluorescence was performed on 6-μm sections of snap-frozen tissue embedded in OCT (Tissue-Tek Sakura) or 70-μm section of paraformaldehyde (PFA)–fixed brain cut by a vibratome. Sections were fixed in acetone (Sigma-Aldrich) and incubated with primary antibody diluted in blocking solution, followed by incubation with secondary antibody and nuclear staining with Hoechst 33342 (2 μg/ml; Sigma-Aldrich). The slides were mounted with Fluoromount-G (SouthernBiotech).

Immunohistochemistry of 2D EBs and immunofluorescence staining of 3D EBs were done as previously described (44, 48). Antibodies and reagents used are listed in table S1. Specimens were analyzed using a Zeiss LSM 700 (20×/0.8 objective) or Nikon Eclipse E800 (10×/0.45 or 20×/0.75) microscopes. The images were acquired with the ZEN 2010 (Carl Zeiss) or NIS software (Nikon), and analysis was performed using ImageJ.

Lectin perfusion and cadaverine leakage

Biotinylated tomato lectin (Lycopersicon esculentum; Vector Laboratories) and Alexa Fluor 555 cadaverine (A-30677, Life Technologies) were administered by tail vein injection of a 150-μl solution containing lectin (0.5 mg/ml) and cadaverine (1 mg/ml) in anesthetized mice. PFA (4%) transcardial perfusion and tumor dissection were performed 15 min after injection.

Stereological quantification

Cryosections from GL261 tumors, stained with CD31 by immunohistochemistry, were analyzed using a Nikon Eclipse E100 microscope and an eyepiece grid as described by Wassberg et al. (36). For the human TMA-based blood vessel quantification, manual stereological quantification method was applied with a modification of creating a counting frame grid in a computer using a calibrator sample. The vessel count and morphology were determined using the microphotographs of the TMA samples.

In vivo bioluminescence imaging

In vivo bioluminescence imaging was performed after intraperitoneal injection of d-luciferin (10 μl/g) (15 mg/ml in PBS; Caliper Life Sciences). Mice were anesthetized 3 min after injection by inhalation of 2% isoflurane gas in O2. Sedated animals were placed on the temperature-controlled imaging platform of an IVIS machine (IVIS Imaging System, Caliper). The image was acquired at high resolution for 5 min. The intensity of the signal was quantified by measuring luminescence signal by Living Image software (IVIS LIS, Caliper).

Matrigel plug assay

Six-week-old female nude mice were purchased from Taconic M&B. Mice were anesthetized with 3% isoflurane and subcutaneously injected with 400 μl of Matrigel (vWR International) supplemented with sphingosine-1-phosphate (Avanti Polar Lipids). Mice were injected with control Matrigel or with Matrigel supplemented with either PTN (2 μg/ml; R&D Systems), VEGF-A165 (5 μg/ml; PeproTech), or the combination of PTN (2 μg/ml) and VEGF-A165 (5 μg/ml). Mice were sacrificed 7 days later and stained as previously described (56).

Study approval

Mouse studies were approved by the Uppsala Ethical Committee on Animal Experiments. The use of human samples was approved by the Regional Ethical Review Board in Uppsala. Human tissue was obtained and used in a manner compliant with the Declaration of Helsinki. Patients participated after giving informed consent.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/8/406/ra125/DC1

Fig. S1. Survival of grade II to III astrocytoma and glioblastoma patients with low and high PTN mRNA abundance.

Fig. S2. Expression of PTN in GL261 and CT-2A murine glioma cells does not affect tumor cell proliferation in vitro.

Fig. S3. PTN expression does not affect tumor cell apoptosis in orthotopic GL261 tumors.

Fig. S4. ALK is found in GL261 tumor vasculature.

Fig. S5. PTN overexpression does not affect vascular permeability in GL261 tumors.

Fig. S6. PTN is not detectable in R1 ES cells until EBs are formed.

Table S1. Antibodies and staining reagents list.

Table S2. shRNA sequences.

Table S3. qPCR primers.

REFERENCES AND NOTES

Acknowledgments: The GL261 murine glioma cell line was a gift from G. Safrany, NRIRR, Budapest. The CT-2A cell line was a gift from T. Seyfried, Boston College. We thank L. Dimberg, University of Colorado, for critical reading of the manuscript. Funding: This work was supported by grants from the Swedish Cancer Society (CAN 2011/862), the Swedish Childhood Cancer Society (PR2013-0107 and PROJ11/083), and the Swedish Research Council (2013-3797 and 2008-2853). Author contributions: L.Z. designed and performed the research; collected, analyzed, and interpreted the data; and wrote the manuscript. S.K. contributed to development of methodology and performed the experiment. K.F.N., K.E.O., A.S., F.P., and M.E. contributed to development of the methodology, analyzed the data, and wrote the paper. T.F., C.J., L.L., D.Y., T.O., and F.P. performed the experiment. P.U.M., A.-K.O., X.L., T.E.A.E.H., A.S., and L.C.D. analyzed and interpreted the data and wrote the paper. A.D. designed and performed the research, analyzed and interpreted the data, wrote the manuscript, and supervised the study. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Signaling

Navigate This Article