VEGF receptor‐2/neuropilin 1 trans‐complex formation between endothelial and tumor cells is an independent predictor of pancreatic cancer survival

Abstract Unstable and dysfunctional tumor vasculature promotes cancer progression and spread. Signal transduction by the pro‐angiogenic vascular endothelial growth factor (VEGF) receptor‐2 (VEGFR2) is modulated by VEGFA‐dependent complex formation with neuropilin 1 (NRP1). NRP1 expressed on tumor cells can form VEGFR2/NRP1 trans‐complexes between tumor cells and endothelial cells which arrests VEGFR2 on the endothelial surface, thus interfering with productive VEGFR2 signaling. In mouse fibrosarcoma, VEGFR2/NRP1 trans‐complexes correlated with reduced tumor vessel branching and reduced tumor cell proliferation. Pancreatic ductal adenocarcinoma (PDAC) strongly expressed NRP1 on both tumor cells and endothelial cells, in contrast to other common cancer forms. Using proximity ligation assay, VEGFR2/NRP1 trans‐complexes were identified in human PDAC tumor tissue, and its presence was associated with reduced tumor vessel branching, reduced tumor cell proliferation, and improved patient survival after adjusting for other known survival predictors. We conclude that VEGFR2/NRP1 trans‐complex formation is an independent predictor of PDAC patient survival. © 2018 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Introduction
Aberrant angiogenesis affects the progression and dedifferentiation of cancer [1]. Anti-angiogenic cancer therapy, given either as a neutralizing anti-vascular endothelial growth factor (VEGFA) antibody (bevacizumab) or as receptor tyrosine kinase inhibitors (sunitinib and sorafenib), increases the progression-free survival or overall survival in several cancer forms including metastatic colon cancer, advanced lung cancer, renal cancer, and hepatocellular carcinoma [2]. In contrast, pancreatic ductal adenocarcinoma (PDAC) remains a dismal diagnosis, with a median overall survival of 6 months and a 5-year survival rate of less than 5% [3], for which anti-angiogenic therapy has shown limited benefits [4]. The poor outcome is influenced by the often late-stage diagnosis of already disseminated disease, and therapy resistance through an abundant desmoplastic stroma and dysfunctional vessels, resulting in poor perfusion and hypoxia [5].
The mechanisms underlying the abnormal vasculature in PDAC have remained unknown.
Neuropilin 1 (NRP1) is a broadly expressed transmembrane glycoprotein lacking intrinsic enzymatic activity. NRP1 influences angiogenesis by forming a ternary complex with VEGFA and VEGFR2, thereby potentiating the bioactivity of VEGFR2, resulting in increased signaling and enhanced endothelial cell migration [6][7][8][9]. NRP1 directs internalization of VEGFR2 via clathrin-dependent endocytosis and guides VEGFR2 intracellular trafficking through binding of the NRP1-PDZ domain to GIPC1/synectin, an intracellular scaffold protein that in turn binds the motor protein myosin VI [10,11]. Additionally, NRP1 has been suggested to induce vascular permeability in a VEGFR2-independent manner, through a pathway relying on the cytoplasmic domain of NRP1 [12].
In a recent study, we used mouse tumor models expressing NRP1 or not, to establish the concept of a distinct, productive signaling pattern induced when 312 E Morin et al VEGFR2 and NRP1 are co-expressed on endothelial cells (denoted cis) compared with arrested signaling when the two molecules are expressed on adjacent endothelial (VEGFR2) and tumor (NRP1) cells (denoted trans). We showed that trans-complex formation in these mouse tumor models results in reduced tumor angiogenesis, correlating with suppressed tumor initiation [13].
The objective of the current study was to translate these findings to human cancer. Here, we studied the pattern of VEGR2 and NRP1 expression and the potential for complex formation in several different human cancers. VEGR2/NRP1 trans-complex formation was identified at low density in human gastric cancer (GAC) and at high density in PDAC. The presence of VEGR2/NRP1 trans-complexes in PDAC, but not in GAC tumors, was associated with decreased vessel formation, reduced cancer cell proliferation, and improved patient survival. Thus, although an overall high level of NRP1 expression in cancer is associated with an unfavorable prognosis [14][15][16][17], the exact distribution of NRP1 on different cell types in the cancer dictates its effect on disease progression. Our findings highlight the clinical significance of VEGFR2/NRP1 trans-complex formation in human PDAC.

Materials and methods
Detailed materials and methods are provided in the supplementary material, Supplementary materials and methods.

Ethics statement on mouse and human tumor samples
Animal experiments were carried out in strict compliance with the ethical permit provided by the Committee on the Ethics of Animal Experiments of the University of Uppsala, permit number C231/9. T241 tumor growth T241:EV (empty vector) or T241:NRP1 cells suspended in Matrigel were injected subcutaneously into the right and left flanks of C57Bl/6 wild-type mice [13,18]. Tumors were dissected and processed after 7 days.

TMAs
The HPA-TMA used for in situ hybridization (ISH), immunohistochemical (IHC) staining, immunofluorescent (IF) staining, RNAscope ® ISH (Advanced Cell Diagnostics, ACD, Newark, CA, USA), and proximity ligation assay (PLA) consisted of 20 different cancer types. Of these, 17 cancer types with duplicate samples from 12 patients and triplicate samples from 44 healthy tissues were analyzed in this study (i.e. thyroid cancer, head and neck cancer, and carcinoids were not included) [19,20]. The TMAs contained formalin-fixed, paraffin-embedded tissue cores and were sectioned to 4 μm thickness and mounted on Superfrost Plus microscope slides (Thermo Fisher Scientific, Waltham, MA, USA).
The Umeå-TMA was constructed using cores of 1 mm diameter selected by an experienced pathologist and placed on recipient blocks using a TMA Grand Master machine (3DHISTECH Ltd, Budapest, Hungary). Three cores were included from each primary tumor and one to three cores from metastatic lymph nodes, if present. The cores were coded and randomly placed on the recipient blocks. Clinical data were retrieved from hospital charts. The observers were blinded for the clinical information during analysis of tissue staining.

Annotation of ISH, IHC, RNAscope, and PLA
Whole-slide scanned images, with 40× objective, of the ISH and IHC arrays were acquired using an Aperio ScanScope CS Slide Scanner system (Aperio Technologies, Vista, CA, USA). ISH signal intensity was scored using a three-grade scale: 0 = no detectable signal; 1 = weak; and 2 = moderate. The distribution of staining in tumor cells and blood vessels was evaluated, and discrimination between structures was based on morphology. RNAscope ISH to detect NRP1 and KDR (gene symbol for human VEGFR2) expression was scored on a four-grade scale: 0 = no detectable signal; 1 = weak; 2 = moderate; and 3 = strong signal. Images were scored by one author, blind to the patient identity and clinical parameters.
VEGFR2/NRP1 PLA complexes in trans were defined as complexes in proximity to the endothelium (one nucleus away). Complexes that were overlapping with the endothelial staining were not included in the scoring. Occurrence of complexes in trans was scored on a four-grade scale of 0-3, where 0 indicates not present; 1, one to two PLA signals per cell; 2, three to four PLA signals per cell; and 3, more than four PLA signals per cell. PLA complexes were scored blindly by two authors independently; in cases of a difference in scoring, images were re-examined to reach consensus.

Neuropilin 1 expression in trans reduces vessel number and branching in vivo
To study the role of VEGFR2/NRP1 interactions between tumor and endothelial cells in vivo, we previously generated a T241 fibrosarcoma cell line stably expressing murine NRP1 (T241:NRP1) by lentivirus transduction. In parallel, T241 tumor cells, which lack endogenous expression of NRP1, were transduced with empty virus (EV) as a control [13]. To facilitate even tumor establishment, T241:EV and T241:NRP1 cells were suspended in Matrigel and injected subcutaneously into the flanks of C57Bl/6 WT mice. In T241:NRP1 tumors, VEGFR2/NRP1 complexes formed both on endothelial cells (cis) and between tumor cells and endothelial cells (trans), thus creating a cis + trans condition. In T241:EV tumors, VEGFR2/NRP1 complexes were established only on endothelial cells (cis condition), which expressed both VEGFR2 and NRP1 endogenously ( Figure 1A).
Tumors were harvested after 7 days, sectioned, and stained for Ki67, CD31, and Hoechst 33342 ( Figure 1B). The cis + trans tumors displayed a marked reduction in vessel number compared with when NRP1 was expressed on endothelial cells alone (cis; Figure 1C). This is in agreement with our previous report on reduced vessel area in the presence of VEGFR2/NRP1 trans-complexes [13]. In contrast, vessel size, measured as individual vessel area, was not affected by the presence of NRP1 in trans ( Figure 1D). The number of branches per vessel area was significantly reduced in the trans condition ( Figure 1E). As tumor progression is dependent on angiogenesis, we also examined the potential consequence of VEGFR2/NRP1 trans-complexes on the tumor compartment. We observed a significant reduction in tumor cell proliferation when NRP1 acted in trans upon endothelial cell-expressed VEGFR2 ( Figure 1F). In summary, in a mouse fibrosarcoma model, NRP1 presentation by tumor cells to the adjacent endothelium had significant effects on vessel numbers and vessel morphology, correlating with reduced tumor proliferation.

Screening for neuropilin 1 expression in tumor cells and tumor vasculature
To investigate if the VEGFR2/NRP1 interaction can be established also in human cancer, we performed an ISH screen using the HPA-TMA (see the Materials and methods section), analyzing duplicate tissue samples from 17 different types of cancer with 12 patients per disease [19,20]. Two independent NRP1 riboprobes of 500-600 nucleotides [21] were used on consecutive tissue sections; a sense probe was used as a negative control (supplementary material, Figure S1A, B). The intensity of the NRP1 signal was scored in blood vessels, in tumor cells or both, based on a graded scale from no to weak and moderate intensity (Table 1).
In the majority of tumor types, vascular NRP1 expression was seen in less than 50% of the samples. Only pancreatic adenocarcinoma (PDAC) and gastric adenocarcinoma (GAC) showed the presence of NRP1 transcripts in endothelial cells in more than 50% of the biopsies (Table 1).
Tumor cell NRP1 expression was more consistently detected; 50-80% of the samples for hepatocellular carcinoma, pancreatic adenocarcinoma, squamous cell carcinoma, and testis and stomach cancers showed expression of NRP1 transcripts in tumor cells ( Table 1).
Expression of NRP1 in both vessels and tumor cells was detected in 11/17 cancer types, but in most cases only in a very limited number of samples, with the exception of pancreatic and gastric tumors. Moreover, only GAC and PDAC showed expression of NRP1 in tumor cells in close proximity to blood vessels ( Figure 2A, B), potentially allowing the formation of VEGFR2/NRP1 complexes in cis and/or trans.

Multiplex mRNA expression of NRP1 and KDR by RNAscope ISH in GAC and PDAC
To further investigate the potential co-expression of NRP1 and VEGFR2 in human GAC and PDAC, we performed multiplex fluorescent ISH using the RNAscope method [22]. Commercial probes against NRP1 and KDR (the HUGO gene symbol for human VEGFR2), were used on the HPA-TMA and on a separate nine-patient PDAC cohort, along with positive (human cyclophilin B) and negative (bacterial DapB) controls (supplementary material, Figure S1C, D). Tissues were counterstained for CD34 to visualize vessels ( Figure 2C). NRP1 and KDR expression was evaluated by manually scoring the frequency of dots in the tumor samples, with 0 for none detected, up to 3 for strong expression ( Figure 2D-F).
NRP1 and KDR transcripts were detected in a majority of tumors (at least one positive biopsy core per tumor) of both GAC (6/11 and 7/11 tumors for NRP1 and KDR, respectively) and PDAC (9/11 and 8/11 tumors, respectively) in the HPA-TMA. Both NRP1 and KDR were more abundantly expressed in PDAC than in GAC tumors. NRP1 expression was present in both endothelial cells and surrounding tumor cells. Most tumors expressing NRP1 also displayed KDR transcripts in endothelial cells ( Figure 2D, E). In the nine-patient PDAC cohort, both NRP1 and KDR were prominently expressed in most tumors ( Figure 2F). In conclusion, NRP1 and KDR transcripts were detected in GAC as well as PDAC, but more frequently and at higher levels in PDAC. In agreement, comparing the mRNA expression of NRP1 in human tumor cell lines of gastric and pancreatic origin from the Broad Institute Cancer Cell Line Encyclopedia confirmed significantly higher NRP1 expression in pancreatic cancer cell lines than in gastric cancer cell lines (supplementary material, Figure S1E) [23].

NRP1 expression in GAC and PDAC
To confirm and characterize NRP1 protein expression in GAC and PDAC, IHC was performed on the HPA-TMA. NRP1 immunoreactivity was detected in endothelial cells based on morphological identification of vessels and by CD34 immunostaining run in parallel ( Figure 3A, B; see magnified insets). Additionally, NRP1 was detected in the surrounding tumor tissue. The overall NRP1-positive area was significantly higher in PDAC than in GAC, in agreement with RNA expression data ( Figure 3C).

Detection of NRP1 and VEGFR2 trans-complexes in human cancer
To identify VEGFR2/NRP1 complexes in human tumors, we performed antibody-mediated in situ PLA on GAC and PDAC tumor samples. In situ PLA allows the detection of molecular interactions on, for example, tissue sections, visualizing the localization of interactions in a complex milieu such as tumor tissues [24]. PLA was performed on GAC and PDAC samples in the HPA-TMA and the nine-patient PDAC cohort. In GAC and PDAC, complex formation was detected within the vasculature as well as between endothelial cells and adjacent tumor cells ( Figure 4A, B). The specificity and reproducibility of antibodies against VEGFR2 and NRP1 were carefully controlled (supplementary material, Figure S2A-H; see supplementary material, Supplementary materials and methods for details). The presence of VEGFR2/NRP1 trans-complexes was scored from 0, not present, up to a maximum of 3. Trans-complexes were identified as PLA signals located outside the CD34-positive vessel area but not more than one nucleus away from the endothelial cell (see Figure 4C for a schematic outline). Of the PDAC tumors, 43% of the samples from the HPA-TMA and the nine-patient cohort displayed VEGFR2/NRP1 trans-complexes, and of these, 21% received the highest score of 3 ( Figure 4D). Forty percent of GAC tumors displayed trans-complexes, although the general frequency of complexes was markedly lower than in PDAC, and none of the tumors received a score of 3 ( Figure 4D).
To correlate the presence of VEGFR2/NRP1 trans-complexes to vessel parameters and tumor proliferation, we immunostained tumor samples for CD34 and Ki67 ( Figure 4E). In congruence with the findings on mouse T241 tumors expressing NRP1 or not on tumor cells (Figure 1), the presence of VEGFR2/NRP1 trans-complexes in PDAC correlated with reduced vascular area, vessel count, vessel branching, and tumor proliferation ( Figure 4F-J), while the individual vessel area was not affected ( Figure 4H). Importantly, GAC biopsies showed the same vessel parameters and tumor proliferation irrespective of the presence of VEGFR2/NRP1 trans-complexes (supplementary material, Figure S3A-E), which occurred at low density.

VEGFR2/NRP1 trans-complexes correlate with improved patient survival in PDAC
To investigate if the presence of VEGFR2/NRP1 trans-complexes is of clinical relevance, PLA for VEGFR2 and NRP1 was performed on a separate cohort (the Umeå-TMA) consisting of 75 PDAC patients with known survival outcome. The PLA result was scored as trans or no-trans (see Figure 4C). Kaplan-Meier analysis revealed that patients with VEGFR2/NRP1 trans-complexes had a significantly improved overall survival compared with patients with no trans-complexes (p = 0.033) ( Figure 5A). To test whether the presence of VEGFR2/NRP1 trans-complexes acts as an independent marker for PDAC survival, multivariable analysis was performed including trans score, sex, age, tumor differentiation, and TNM stage. This analysis demonstrated that the presence of VEGFR2/NRP1 trans-complexes in tumor tissue is an independent marker for overall survival in PDAC patients (HR = 0.3, p = 0.008) ( Table 2).
Moreover, CD34-positive vessel density ( Figure 5B) and vessel count ( Figure 5C) were decreased in the samples containing trans-complexes, while the individual vessel area was unaffected ( Figure 5D). The number of vessel branches per vessel was reduced in trans-complex tumors ( Figure 5E). Importantly, tumor cell proliferation was significantly reduced in the presence of VEGFR2/NRP1 trans-complexes ( Figure 5F). Taken together, the association of trans-complexes with reduced vessel parameters and tumor proliferation was reproduced in this third cohort of PDAC patients. In addition, the increased survival of PDAC patients whose tumors contain trans-complexes shows the clinical relevance of the interaction of VEGFR2 and NRP1 in cancer.

Discussion
High tumor vessel density correlates directly with an increased risk for liver metastasis and decreased overall survival in PDAC patients [25][26][27]. Here, we showed that the tumor vessel density in human PDAC was reduced when NRP1 was expressed on tumor cells adjacent to blood vessels, allowing the formation of VEGFR2/NRP1 complexes in trans ( Figure 6). Such VEGFR2/NRP1 trans-complexes arrest VEGFR2 on the endothelial cell surface, interfering with productive signaling, for example, in the extracellular regulated kinase (ERK) pathway [13], ultimately resulting in suppressed tumor angiogenesis, reduced tumor cell proliferation, and prolonged patient survival.
NRP1 also interacts with ligands and receptors apart from VEGF/VEGFR2, such as semaphorins, fibroblast growth factors, and platelet-derived growth factors, influencing their downstream signaling in different cell types [28,29]. Furthermore, NRP1 binds integrins, either indirectly via semaphorin/plexin interactions with integrins [30] or directly [31], thereby modulating cell adhesion. We cannot exclude that tumor cell-expressed NRP1 may influence tumor angiogenesis through other interactions in addition to that with VEGFR2. However, our previous work utilizing genetic mouse models that allowed probing for NRP1's function specifically in complex formation with VEGFR2 [13] forms the basis for the current work, where we have translated findings from mouse to human cancer.
To identify the expression pattern of NRP1, transcripts (ISH, RNAscope ISH), proteins (IHC), as well as protein complexes (in situ PLA) were studied in the intact tumor tissue. While transcript and protein levels may not strictly correlate [32], we found similar trends in the different analyses, namely that GAC and PDAC both expressed NRP1 transcript and protein in tumor cells, albeit at higher levels in PDAC.
The use of in situ PLA to directly show the presence of VEGFR2/NRP1 complexes in human cancer has challenges, for example in deducing the exact localization of the PLA dots. To corroborate the identification of PLA signals as corresponding to VEGFR2/NRP1 trans-complexes, we studied mouse fibrosarcoma where tumor cells expressed NRP1, allowing trans-complexes to be established (Figure 1). The position of PLA signals in this condition was compared with fibrosarcomas lacking NRP1 expression in tumor cells, negating the formation of trans-complexes. Strikingly, the presence    [33][34][35]. NRP1 is expressed in a wide range of non-tumor cells that may be present in a tumor, including endothelial cells, fibroblasts, and bone marrow-derived macrophages and immune cells [36]; moreover, tumor cells may also express NRP1 [37]. The potential prognostic effect of NRP1 expression in cancer has, however, remained unclear. In colon cancer, tumor expression of NRP1 is associated with a better prognosis [38]. In cutaneous squamous cell carcinoma, NRP1 is expressed in highly differentiated tumors, suggesting that it could function as a reservoir to sequester VEGFA to the epithelial compartment, thereby limiting its bioactivity [39], reducing angiogenesis and tumor progression. On the other hand, in human glioma and breast and prostate cancer, NRP1 expression correlates with higher tumor grade and worse prognosis [14][15][16], and in oral squamous carcinoma, NRP1 expression has been correlated with poor prognosis and disease relapse [17]. Higher levels of overall NRP1 expression correlate with poor prognosis in PDAC as well [40]. Of note, the overall level of NRP1 was increased in trans-complex-containing samples (supplementary material, Figure S2H). However, based on our findings, it is essential to determine whether NRP1 is expressed in tumor cells close to endothelial cells (allowing the establishment of VEGFR2/NRP1 trans-complexes) rather than the overall NRP1 expression level, as the former parameter correlates with reduced vessel density, reduced tumor proliferation, and increased patient survival.
The potential benefit of blocking NRP1 function has been tested using neutralizing anti-NRP1 antibodies on human cancer-xenografted mice. The anti-NRP1 antibodies showed additive effects in slowing tumor growth when combined with anti-VEGFA antibodies [41]. As a follow-up, early clinical trials were performed with an antibody blocking VEGFA and VEGFB binding to NRP1, in combination with VEGFA neutralization using bevacizumab. This therapy combination was, however, discontinued due to the development of proteinuria [42,43].
Anti-angiogenic therapy has been established preferentially in combination with chemotherapy for a number of cancers [44]. Bevacizumab, inhibiting the binding of VEGFA to VEGFR2, has shown positive effects in many cancer types. However, the clinical response is very variable, which we hypothesize in part may depend on the expression pattern of NRP1. Phase III clinical trials of bevacizumab on GAC (AVAGAST) and in a separate study on PDAC patients have failed to show clinical benefit in overall survival, while, in both cases, progression-free survival was significantly prolonged [45,46]. In the GAC AVAGAST trial, bevacizumab/chemo-treated GAC patients with low NRP1 levels showed a significantly increased overall and progression-free survival compared with those with higher NRP1 expression [47]. A trend towards a better response to bevacizumab was also observed in breast cancer tumors with low NRP1 expression [48]. An improved response to anti-angiogenic therapy in cancer types with low NRP1 expression could be dependent on the preferential formation of VEGFR2/NRP1 cis-complexes within the endothelium, promoting angiogenesis, which would be blocked by bevacizumab. In contrast, in tumors with a high degree of trans-complexes, anti-VEGF treatment would potentially interrupt the complexes and unleash VEGFR2 signaling, thereby promoting tumor angiogenesis.
To our knowledge, this is the first time that complexes between VEGFR2 and NRP1, and their correlation with vessel and tumor cell parameters, have been described in human cancer. We have shown that trans-complexes correlate with reduction of several vessel parameters and tumor proliferation, as well as increased patient survival. It would be highly relevant to continue studies of this interaction in a larger cohort of PDAC patients, and also to investigate if the same effect of VEGFR2/NRP1 trans-complexes can be observed in other cancer types not explored in detail here. Another important question is whether the trans-group of PDAC patients shows a therapy response distinct from that of patients lacking trans-complexes. We propose that the VEGFR2/NRP1 complex status could be of significant value as a prognostic marker and a potential predictive marker of anti-angiogenic therapy in PDAC and possibly other cancer types.