Antitumor Effects of Anti-Semaphorin 4D Antibody Unravel a Novel Proinvasive Mechanism of Vascular-Targeting Agents

One of the main consequences of vessel pruning and inhibition of neovessel growth produced by angiogenesis inhibitors is the increased hypoxia levels produced inside the tumors. Cancer cells can live in hypoxic conditions (1), but growing evidence indicates that tumor cells may escape from this hypoxic environment to better nourished locations, presenting hypoxia as a positive stimulus for invasion (2). In fact, a strong correlation among tumor hypoxia and increased invasion, metastasis, and poor patient outcome has been reported (3–5). In this context, alternative angiogenic targets such as semaphorins are being explored (6).

Semaphorins (SEMA) are a superfamily of secreted or membrane-associated glycoproteins implicated in axonal wiring control, angiogenesis, and cancer progression. Semaphorin 4D (Sema4D) is a transmembrane protein of 150 KDa of the IV class of the subfamily of semaphorins involved in the regulation of axon guidance, cell migration in organ development and vascular morphogenesis (7–9). Three receptors are known for Sema4D: high-affinity receptor plexin-B1 (PlxnB1), expressed in a wide variety of cell types, intermediate affinity plexin-B2 (PlxnB2), and low-affinity receptor, CD72, mainly expressed in cells of the immune system (10, 11). Sema4D is highly expressed in the membrane of most frequent solid tumors, including breast, prostate, and colon (12), and also in tumor-associated macrophages (TAM), with a relevant role in tumor invasion, angiogenesis, and metastasis (13). High expression levels of Sema4D have also been reported in tumor stroma (14). Due to a proteolytical cleavage by matrix metalloproteinase type 1 (MT1-MMP, also known as MMP14), a Sema4D-soluble form is released (15, 16), allowing to act through PlxnB1 on endothelial cells and promoting angiogenesis, which permits tumors to be nourished with the necessary nutrients and oxygen to continue its growth (12). In fact, there is a completed phase I clinical trial to evaluate the safety and tolerability of intravenous (i.v.) administration of an antibody anti-Sema4D VX15/2503 (Vaccinex) in patients with advanced solid tumors (ClinicalTrials.gov identifier: NCT01313065; ref. 17).

In this study, using a spontaneous mouse model of pancreatic neuroendocrine cancer (PanNET), RIP1-Tag2 mice, we describe alteration of tumor vascular function by the use of a vascular-targeting agent anti-Sema4D antibody that consequently impairs tumor growth. Unlike VEGF/R blockade, induction of intratumor hypoxia is not observed after anti-Sema4D therapy, but the increase in local invasion and distant metastases is comparable with anti-VEGFR2′s effects. This hypoxia-independent mechanism of increased aggressive phenotype is associated with recruitment of TAMs as mediators of increased invasion and dissemination of tumor cells after anti-Sema4D treatment. Mechanistically, anti-Sema4D antibody induces a Sema4D signaling engagement in the membrane of macrophages not only for their motility and recruitment to the tumor invasive fronts, but also for increased secretion of stromal cell–derived factor 1 (SDF1). In turn, SDF1 enhances tumor cell–invasive behavior via CXCR4 and triggers the malignant PanNET phenotype in anti-Sema4D–treated RIP1-Tag2 mice. Finally, we also present clinical evidence that supports a role for Sema4D and SDF1 overexpression in human macrophages and an association of Sema4D and CXCR4 in PanNET patients' tumor progression.

Transgenic RIP1-Tag2 mice have been previously reported (18). Animal housing, handling, and all procedures were approved by our institution's ethical committee and government committees. Tumor volume, type, invasiveness, and hemorrhagic phenotype were determined as previously described (19). Four-week-long treatments in RIP1-Tag2 mice started at 12 weeks of age with (i) anti-Sema4D Mab67 function blocking murine IgG1 antibody (anti-Sema4D) kindly provided by Vaccinex (20), (ii) anti-VEGFR2 blocking antibody (DC101) purified in our laboratory or (iii) ChromPure Mouse IgG1 whole molecule as isotype control (Jackson Immuno Research Laboratories, Inc). All dosed at 1 mg/animal once a week i.p. except for anti-VEGFR2, which was administered twice a week, as previously described (21).

Frozen or paraffin samples of pancreata and livers were histologically evaluated with primary antibodies: anti-CD31 (550274; 1:50; BD Biosciences); anti-T-antigen (1:10,000; Hanahan Laboratory), anti-Hypoxyprobe (1:50; NPI Inc), anti-GLUT1 (ab652; 1:100; Abcam), anti-type IV collagen (AB756P; 1:200; Millipore), anti-Lyve1 (103-PA50AG; 1:100; ReliaTech), anti-desmin (ab15200; 1:150; Abcam), anti-NG2 (AB5320; 1:50; Millipore), anti-SMA-Cy3 (C6198; 1:200; Sigma-Aldrich), anti-SMA (RB-9010; 1:100; Thermo Fisher Scientific), anti-PlxnB1 (sc-28372; 1:50; Santa Cruz Biotechnology), anti-Sema4D (G3256; 2 μg/mL; Vaccinex), anti-F4/80 (MCA497R; 1:50; AbD Serotec), anti-CD3e (550275; 1:10; BD Biosciences), anti-CD72 (PAB261Mu01; 1:100; Cloud-clone Corp), anti-SDF1 (MAB350, 1:20, R&B System), anti-CXCR4 (C8352, 1:750, Sigma), and anti-insulin (A0564; 1:50; Dako). Microvessel density, pericyte coverage of tumor vessels, macrophage infiltration, CXCR4, SDF1, and Sema4D expression were manually quantified per field. Collagen IV, VE-cadherin, and tumor hypoxia were measured as the mean positive area per field.

The βTC4 cell line was isolated from RIP1-Tag2 tumors in the Hanahan Laboratory and grown in DMEM 20% FBS. To discard undifferentiation events, they were not used beyond 50 passages and their phenotype was authenticated by insulin expression by immunocytofluorescence. The RAW264.7 cell line, donated by E. Ballestar (IDIBELL), was grown in DMEM 10% FBS and THP-1 cells, donated by I. Fabregat (IDIBELL), were grown in RPMI 10% FBS. These had been bought by ATCC and authenticated by STR profiling by the ATCC. HUVEC cells from CellTech (Spain) were grown in EGM/EBM2 10% iFBS. All cell lines were examined for Mycoplasma contamination using PCR analysis every month. For conditioned media, RAW264.7, HUVEC, and THP-1 cells were grown in free-serum DMEM and treated with anti-Sema4D (10 μg/mL), either Vaccinex (Mab67) or Abnova (3B4), isotype-specific anti-IgG1 (10 μg/mL, isotype control), or without treatment (control) during 24 hours. RAW264.7 cells were also treated with recombinant Sema4D (5235-S4-050 and PlxnB2 (6836-PB-050) at 1 μg/mL (R&D Systems). In added conditioned media, the antibodies were added after media collection.

shRNAs designed by The RNAi Consortium cloned into the pLKO.1 lentiviral vector were purchased from Dharmacon (GE Healthcare) for silencing of Sema4D (TRCN0000067493), CD72 (TRCN0000066042), Plxnb2 (TRCN0000078853), and nontargeting shRNA (shNS) as a negative control. shRNA lentivirus were used to transduce RAW264.7 cells with 8 μg/mL polybrene and after 48 hours selected with 1 μg/mL puromycin for 5 days.

Corning migration and invasion assays (#3422 and #354480) were performed following the manufacturer's instructions. RAW264.7 and THP-1 cells were treated with anti-Sema4D (10 μg/mL), either Vaccinex (Mab 67) or Abnova (3B4), isotype-specific IgG1 (10 μg/mL) or without treatment. βTC4 cells in serum-free DMEM were subjected to RAW264.7-conditioned media. For the chemotaxis assay for SDF1, treatments included 1 μg/mL AMD3100 (3299, Tocris) and 100 ng/mL recombinant SDF1 (250-20A, PeproTech).

Tumor samples and βTC4 and RAW264.7 cell lysates were analyzed by Western blot with c-Met (sc-8057; 1:100; Santa Cruz Biotech), phospho-c-Met (3077; 1:750; Cell Signaling Technology), PlxnB2 (AF5329; 1:1000; R&D Systems), Sema4D (MAB5235; 1:250; Novus Biologicals), CD72 (AF1279; 1:500; R&D Systems), α-tubulin (32-2500; 1:2,000; Invitrogen). For mRNA, RNA extraction and High-Capacity RT reaction (Applied Biosystems) produced cDNA for RT-PCR using LDA Arrays for 24 genes (Supplementary Table S1) and SDF1 and CXCR4, HPRT1 (mouse and human), and cMET and β-ACTIN (mouse) TaqMan probes (Applied Biosystems).

Supernatants of RAW264.7-conditioned media were analyzed by mouse cytokine antibody array (#AAM-CYT-1000; RayBiotech, Inc.) according to the manufacturer's instructions. Mouse SDF1 ELISA (MCX120, R&D Systems) was performed after concentrating supernatants with Vivaspin 2 KDa column (Sartorius). Similarly, human SDF1 ELISA (DSA00; R&D Systems) was done in supernatants of HUVEC- and THP-1-conditioned media.

Gene-expression data from different stages of RIP1-Tag2 mice (GEO Omnibus ID GSE73514) and human mRNA transcriptomes from a core independent clinical gene-expression data set of PanNET (GEO Omnibus ID GSE73338) patients were used (22, 23). For RIP1-Tag2 mice data, primary tumor (n = 5) and metastasis (n = 3) samples were compared. For the human study, normal pancreatic islet samples (n = 4), nonfunctional samples (n = 63), which were termed primary tumors, and their corresponding metastases (n = 7) were evaluated. To further study the malignization process, primary tumors were divided into two subcategories, nonmalignant and malignant, according to the clinical history of the patients (23).

Results are presented as mean ± SD, except for transwell assays, which results are presented as mean ± SEM. The statistical tests are noted in each figure and significance is as follows: *, P < 0.05; **, P < 0.005; ***, P < 0.001; ****, P < 0.00001 consensus.

Initially, the presence of Sema4D and its high-affinity receptor PlxnB1 was evaluated. Sema4D was found to be highly expressed in the membrane of scattered single cells inside tumor parenchyma with a pattern compatible with immune cells and weakly expressed in the membrane of tumor cells (Supplementary Fig. S1A), consistent with previous reports (15, 17). PlxnB1 was immunodetected in a 30% of vascular structures (Supplementary Fig. S1B). To assess the effects of anti-Sema4D treatment, we used a specific antibody (anti-Sema4D, Mab67 Vaccinex; ref. 20) in RIP1-Tag2 mice and focused on tumor growth and expansion phase of islet carcinoma. Therapeutical regimes included 2- or 4-week anti-Sema4D treatment along with treatment with DC101, a well-described blocking monoclonal antibody of VEGFR2 (21). We could determine that 4-week therapy produced an inhibition in tumor growth similar to the one observed after anti-VEGFR2 (α-VR2) treatment (Fig. 1A), which promoted an extension of life span of treated mice (Fig. 1B). These results suggest an antitumor benefit of anti-Sema4D therapy in terms of tumor shrinkage and lifespan increase in mice.

A qPCR for angiogenesis-related genes, such as angiopoietins and platelet-derived growth factor receptors, was modified after the treatment (Supplementary Table S1). Surprisingly, in contrast to differences observed after anti-VEGFR2, treatment with anti-Sema4D did not show any differences in the number of vessel structures (Fig. 1C; Supplementary Fig. S1C, top) or CD31 area density (Fig. 1D) nor matrix deposition of endothelial cells determined using type IV collagen (Fig. 1E; Supplementary Fig. S1C, middle). Moreover, there was no difference in the area and structure of endothelial cell–cell junctions, as shown by the VE-cadherin evaluation (Fig. 1F; Supplementary Fig. S1C, bottom), in contrast to the significant alterations observed after anti-VEGFR2 therapy. Other vascular parameters such as the number of branches and empty sleeves did not show any differences either (Supplementary Fig. S1D–S1E). Lymphangiogenesis was also evaluated, observing no lymphangiogenic events neither in the control nor in the anti-Sema4D–treated condition (Supplementary Fig. S2A). Together, these data indicate that anti-Sema4D treatment does not produce a classic antiangiogenic effect at the endothelial level on the RIP1-Tag2 model. To mechanistically understand why anti-Sema4D does not exert a direct antiangiogenic effect, we evaluated the presence of membrane-bound or -soluble Sema4D forms in our model. As shown in Supplementary Fig. S2B, in RIP-Tag2 tumors, we can only detect Sema4D transmembrane full-length form (150 KDa) and not detect any soluble form (110–115 KDa). As the soluble form has been associated with angiogenesis, the lack of this soluble form is consistent with a lack of antiangiogenic effects of anti-Sema4D.

Because many other cellular types such as pericytes play fundamental structural and functional roles in blood vessels, pericyte coverage was evaluated to determine whether they were subjected to a further structural change after anti-Sema4D therapy. Pericytes positive for desmin and NG2 were increased after anti-Sema4D treatment, whereas the number of α-SMA–positive pericytes was decreased (Fig. 1G; Supplementary Fig. S3A). This alteration in the pericyte profile suggests a switch to a more immature vessel type associated with vascular remodeling. In addition, after anti-Sema4D therapy, there is a nearly 2-fold increase in the number of PlxnB1-positive structures (Supplementary Fig. S3B). Next, we checked whether these changes in pericyte coverage occur in PlxnB1-positive vessels (Supplementary Fig. S3C). No difference in pericyte coverage between the PlxnB1-positive or -negative endothelial cells was observed, evidencing that pericyte coverage is independent from the expression of PlxnB1 in endothelial cells. This suggests an indirect cross-talk between endothelial cells and pericytes.

Based on previous work (24), where Sema4D treatment of endothelial cells elicits production of PDGF-BB and promotes differentiation of mesenchymal stem cells into pericytes, thus producing pericyte proliferation, chemotaxis, and association with HUVECs in a capillary network, we checked PDGF-BB expression. ELISA assay of PDGF-BB showed a slight decrease in PDGF-BB levels in α-Sema4D–treated tumors when compared with control tumors (Supplementary Fig. S4). This result was concordant with our RNA analysis, in which a decrease in PDGF-BB levels was also observed (Supplementary Table S1).

To assess if the altered pericyte coverage after anti-Sema4D had further consequences in vascular functionality, we checked vascular integrity. We evaluated a vascular leakage parameter such as extravasation of erythrocytes (microhemorrhaging) in the form of tumor hemorrhagic phenotype. The percentage of hemorrhagic tumors after 2 weeks of anti-Sema4D therapy was significantly reduced when compared with control animals, although this reduction was even stronger with anti-VEGFR2 treatment (Fig. 1H; Supplementary Fig. S5A, left). These effects were maintained in long-term treatment (Supplementary Fig. S5A, right). Aiming at identifying other possible causes of the change in pericyte coverage after anti-Sema4D treatment, we screened for the CD72 low-affinity receptor presence in pericytes from tumor vasculature. We could determine that CD72 is not expressed in vascular nor perivascular cells (Supplementary Fig. S5B). CD72 was rather found to be expressed in single cells, suggestive of its expression in cells of the immune system infiltrated in tumor stroma (Supplementary Fig. S5C). This result is further confirmed by a double costaining of CD72 and F4/80-positive macrophages (Supplementary Fig. S5D).

Anti-Sema4D treatment increases the number of highly invasive tumors progressively, similar to the effect of anti-VEGFR2 (Fig. 2A and B). Although the majority of control tumors were predominantly encapsulated or microinvasive, treated tumors presented wide fronts of invasion encroaching into adjacent acinar tissue. Significantly, this effect was further exacerbated when anti-Sema4D therapy was maintained for 4 continuous weeks (Fig. 2B). Study of livers and peripancreatic lymph nodes (LN) revealed that anti-Sema4D–treated mice more frequently contained enlarged LN containing tumor cells and distant metastases to the liver (Fig. 2C). The incidence of LN metastasis grew from the 30% in untreated controls to more than 70% in treated groups, indicating that similarly to what happens with VEGFR2 inhibition, anti-Sema4D treatment promotes an increase in LN metastasis (Fig. 2D, left). Incidence of liver metastasis was 2-fold higher in mice that had received an antiangiogenic treatment, either anti-VEGFR2 or anti-Sema4D (Fig. 2D, right). When combining anti-Sema4D and anti-VEGFR2, results in tumor burden, survival, invasiveness, and metastasis incidence were identical as in anti-VEGFR2 alone (Supplementary Fig. S6). This lack of additional effect of anti-Sema4D evidences the predominant role of VEGF in triggering angiogenesis over Sema4D in this tumor setting.

Overall, the data presented here demonstrate that anti-Sema4D treatment promotes the acquisition of an adaptive resistance, with similar effects of the complete and lasting inhibition of angiogenesis caused by the use of anti-VEGFR2 or TK inhibitors as sunitinib and sorafenib (19, 25).

Treatments targeting tumor vasculature are described to produce an increase in hypoxia as a consequence of the antiangiogenic effect. Surprisingly, anti-Sema4D treatment did not induce hypoxia in short-term treated tumors, as demonstrated by the presence of pimonidazole adducts or by the increase in the expression of hypoxia-response genes such as Glut1 (Fig. 3A) in anti-VEGFR2–treated tumors. Quantification of this event at longer treatment regimes confirmed this observation (Fig. 3B and C). Taken together, these data suggest that a hypoxia-independent pathway is responsible for the increase in invasion and malignization.

Up to now, the best described mechanism of tumor aggressiveness after antivascular inhibition in RIP1-Tag2 tumors involves hypoxia and c-Met activation (26, 27). RNA analysis of untreated and anti-Sema4D–treated tumors revealed that there were no changes in c-Met expression (Fig. 3D). We then assayed the presence of its precursor protein and its active form, phosphorylated c-Met, by Western blotting, using HGF (c-Met natural ligand) to stimulate cells. No expression of the precursor and any activation of c-Met signaling pathway was observed, neither in the untreated nor in the anti-Sema4D–treated conditions (Fig. 3E). Similarly, even if βTC4 cells express c-Met at low transcriptional levels, there is no pathway activation in response to anti-Sema4D or HGF (Fig. 3F–G). Overall, these data suggest that malignization effects in RIP1-Tag2 mice are restricted to an indirect effect of Sema4D over tumor cells, rather than to a direct action of the proangiogenic molecule upon tumor cell–derived c-Met. Moreover, a retrograde effect of Sema4D over tumor cells was discarded because no changes in cell adhesion, de-adhesion, or proliferation of RIP1-Tag2–derived βTC4 tumor cells were observed (Supplementary Fig. S7).

Among all immune cells expressing Sema4D (28–31), a relevant role in protumorigenic processes has been specifically described for lymphocytes and TAMs (13, 14). CD3e-positive T lymphocytes infiltrated in the RIP-Tag2 tumor parenchyma were very scarce and, although anti-Sema4D treatment produced an increase in their numbers, the absolute amount was too low to consider them functionally relevant (Supplementary Fig. S8). On the other hand, a costaining of macrophage marker F4/80 with Sema4D in our tumors showed that most macrophages did not express Sema4D, few expressed it with high intensity and some only in certain areas of the cell (Fig. 4A). However, we found a visibly higher amount of Sema4D-positive macrophages after anti-Sema4D therapy (Fig. 4A), and the total number of macrophages was significantly increased (Fig. 4B). In fact, although the number of Sema4D-negative macrophages was maintained invariably after the therapy (Fig. 4C), the number and percentage of Sema4D-positive macrophages increased after short-term anti-Sema4D treatment (Fig. 4D). In conjunction, these data demonstrated that, in vivo, there was a change in the number and phenotype of TAMs after anti-Sema4D treatment. In order to functionally validate its consequences, the migration properties of a Sema4D-expressing murine macrophage cell line, RAW264.7 (Supplementary Fig. S9A), were evaluated after anti-Sema4D treatment. As shown in Fig. 4E, there was an increase in migration of RAW264.7 cells after anti-Sema4D therapy, which occurred in a dose-dependent manner (Supplementary Fig. S9B). Moreover, the addition of exogenous recombinant Sema4D did not reduce basal macrophage migration, indicating the requirement for Sema4D expression in cell membrane for the antibody to have an effect (Fig. 4E). To decipher the underlying mechanism, effective knockdowns of the ligand Sema4D and its two receptors expressed in RAW264.7 cells, CD72 and PlxnB2, were generated (Supplementary Fig. S9C). Interestingly, we observed that there was no change in the migratory capacity of RAW264.7 cells in any of the gene knockdowns (Fig. 4F). Moreover, anti-Sema4D treatment continued to produce the same increase of migration in all gene silencing conditions, except for shSema4D cells (Fig. 4G; Supplementary Fig. S9D). Altogether, our results define a receptor-independent and Sema4D requirement for antibody induction of migration, and they demonstrate that Sema4D needs to be expressed in the membrane of the cells for the antibody to have an effect. Thus, all these data define an antibody-induced retrograde signaling engagement of Sema4D, which has already been previously published for this family of transmembrane proteins in different settings (reviewed in refs. 32 and 33).

Most macrophage activity is mediated by cytokines and chemokines that act in autocrine fashion and paracrine fashion, upon other macrophages or even upon other cells from the tumor ecosystem. Aiming to delve into macrophage study, we performed a mass spectrometric analysis (LC-MS/MS) of secreted proteins (secretome) composing RAW264.7-conditioned media previously stimulated with anti-Sema4D. The proteomic approach resulted in the identification of more than a thousand proteins (Supplementary Table S2). Using a Gene Set Enrichment Analysis bioinformatics tool, we showed a statistical enrichment in proteins related to important macrophage functions: cell migration, cell projection, cytoskeleton, and RAC1 pathway (grouped in migration); DNA replication and cell cycle (grouped in proliferation); FCγR-mediated phagocytosis and immunologic synapse (grouped in activation; Supplementary Fig. S10). Taken together, the analysis of the secretome by proteomic profiling suggests a direct effect of Sema4D upon macrophage activity, specially affecting their migration, proliferation, and activation.

To check the behavior of TAMs in tumor periphery, the number of macrophages in the perimeter of the base protrusions of invasive fronts was determined after costaining Sema4D with F4/80 macrophage marker and the RIP1-Tag2 tumor cell marker insulin (Fig. 5A). Contrary to the intratumoral results, the number of macrophages in the invasive fronts remained unaltered after anti-Sema4D treatment (Fig. 5B). The number of peritumoral Sema4D-negative macrophages decreased, whereas the number of Sema4D-positive macrophages and their percentage are strongly increased after treatment (Fig. 5D). The abrupt change in macrophage number and phenotype may indicate a role for these cells in the invasive and malignization process that occurs after the therapy.

To confirm this hypothesis, an in vitro Matrigel invasion assay using βTC4 cells was performed. The addition of the conditioned medium of the RAW264.7 cell line treated with anti-Sema4D significantly increased the invasive properties of βTC4 cells (Fig. 5E). Nevertheless, conditioned media from neither Sema4D, CD72, nor PlxnB2 knockdown RAW264.7 cells did not recapitulate this tumor cell invasion increase (Fig. 5F). On the other hand, conditioned media of anti-Sema4D–treated shSema4D RAW264.7 cells did not induce an increase, but rather a decrease of tumor cell invasion (Supplementary Fig. S11). Therefore, Sema4D retrograde signaling engagement by anti-Sema4D produces a switch of the macrophage phenotype that potentiates tumor cell invasion in RIP1-Tag2, probably by promoting secretion of a proinvasive molecule.

In pursuance of identifying the proinvasive molecule secreted by macrophages and responsible for tumor cell invasion after anti-Sema4D therapy, a mouse cytokine array was performed in supernatants of RAW264.7 conditioned media. Even though not many significant changes were observed between different treatment conditions (Supplementary Table S3), a statistically significant increase in the stromal cell–derived factor 1 (SDF1, also known as CXCL12) molecule was detected in an anti-Sema4D–treated supernatant (Fig. 6A). An ELISA analysis of secreted SDF1 revealed an increase in anti-Sema4D condition, which was not observed neither in control or treated Sema4D knockdown macrophages (Fig. 6B), nor in cells treated with recombinant PlxnB2 (Supplementary Fig. S12A) or receptor-knockdown cells (Supplementary Fig. S12). We validated SDF1 as a possible macrophage secreted candidate responsible for cancer cell invasion in the RIP1-Tag2 model by an invasion assay in the in vitro setting. As expected, βTC4 cells responded to recombinant SDF1 stimulation in vitro by increasing their invasion (Fig. 6C). This phenomenon was inhibited when a CXCR4 receptor was blocked by its antagonist AMD3100. In addition, the increase observed in βTC4 cells' invasion after anti-Sema4D–treated conditioned media addition is comparable with the one produced when exogenous SDF1 was added to IgG1-treated conditioned medium (Fig. 6D). Consistently, when AMD3100 was added to anti-Sema4D–treated conditioned medium, the invasive capability of βTC4 cells dropped to basal levels, confirming that SDF1 is one of the factors secreted by macrophages after anti-Sema4D treatment responsible for tumor cell invasion.

Finally, we sought to check whether in the RIP1-Tag2 model, the SDF1/CXCR4 signaling axis was present and affected by anti-Sema4D treatment. We found an increased trend of both CXCR4 and SDF1 RNA expression in treated tumors (Supplementary Fig. S12C–S12D), which was further confirmed by IHC (Fig. 6E and F). The CXCR4 receptor appears to be expressed homogenously by RIP1-Tag2 tumor cells, albeit at low levels in control samples and highly present in anti-Sema4D–treated mice (Fig. 6E), possibly due to an SDF1-induced–positive feed-forward mechanism (34). Indeed, CXCR4 is naturally present in the tumor progression stages of RIP1-Tag2 mice, showing expression in metastases, in both control-treated and anti-Sema4D–treated mice (Supplementary Fig. S12E–S12F). Therefore, anti-Sema4D treatment seems to exacerbate an already existing CXCR4-mediated metastasis mechanism. As expected, SDF1 was found both in cells with a vascular phenotype and in round-shaped cells compatible with immune infiltrates. The count of the latter showed an increase in SDF1-positive round cells after the treatment (Fig. 6F). A costaining of both Sema4D and SDF1 showed an enrichment in SDF1/Sema4D double-positive cells after the treatment (Supplementary Figs. S6G and S12G). Because endothelial cells also express SDF1, we analyzed the behavior of HUVEC cells in response to anti-Sema4D, observing no changes in gene expression but an increase in SDF1 release after the treatment (Supplementary Fig. S13).

Altogether, the in vivo results may suggest a tumor-independent origin of SDF1 that could bind to its receptor in RIP1-Tag2 tumor cells to exert its activity. Furthermore, a deeper analysis associating SDF1 levels and invasive capacity of the tumor front revealed a relationship between the invasive capacity and ligand concentration in control tumors (Fig. 6H). This relationship was slightly lost after anti-Sema4D treatment.

After demonstrating both in vitro and in vivo the role of Sema4D and SDF1/CXCR4 in tumor malignization of the RIP1Tag2 mouse model, we sought to decipher whether these same mechanisms could also be playing the role in the clinical setting. We found Sema4D expression to be significantly increased in metastatic samples when compared with either primary nonmalignant and malignant tumors or normal pancreatic controls (Fig. 7A). Besides, whereas SDF1 expression remained practically unaltered, we found a significant increase in CXCR4 receptor expression between normal and both primary tumor subtypes and metastases (Fig. 7B and C). In fact, there is a gradual increase of CXCR4 that correlates with malignization, thus implying a role for this protein as a tumor progression driver. Furthermore, we evaluated the correlation between Sema4D and CXCR4 expression in nonmalignant (nonmetastatic) and malignant (metastatic) primary tumor samples of PanNET patients. Contrary to nonmetastatic patients, malignant patients showed higher levels of CXCR4 that showed a correlation with Sema4D (Fig. 7D). We finally validated our results using the human macrophage cell line THP-1 (Fig. 7E and F). After anti-Sema4D treatment, THP-1 cells demonstrated an increased migratory phenotype and SDF1 release (Fig. 7E and F), without alteration of CXCL12 or CXCR4 expression (Supplementary Fig. S14). Overall, the clinical data validate a possible link of SemaD-SDF1/CXCR4 in patient samples of PanNET.

Compared with the canonical VEGF, Sema4D is a molecule with quite a different role in angiogenesis because its binding to PlxnB1 can promote different and sometimes opposing cellular responses including vascular guidance (35). Indeed, these differences in vascular-targeting potential provide an explanation for the negligible effects in endothelial structures after anti-Sema4D treatment without reduction in MVD and no increased levels of tumor hypoxia. Nevertheless, by PDGF-B reduction anti-Sema4D treatment produced a pericyte structural alteration that functionally modified vessel perfusion and hyperpermeability, thus altering tumor growth.

Moreover, it is widely accepted that a partial inhibition of angiogenesis would not produce an increase of hypoxia within tumors and could not trigger the secondary unwanted proinvasive and malignant effects (27). Contrarily, we have observed that although anti-Sema4D therapy produced a partial effect in vessel functionality, without induction of hypoxia, it still produced the same proinvasive effect as anti-VEGF therapy in the PanNET model.

Protumoral roles for Sema4D typically involve tumor cell–derived Sema4D, and there is little evidence about the role of stromal Sema4D (36), Interestingly, in RIP1-Tag2 tumors, the main source of Sema4D, are macrophages infiltrating the tumor stroma. In recent years, a critical role for the tumor microenvironment and particularly TAMs has been demonstrated (37, 38). Their contribution to tumor growth and progression has even been reported in the clinical setting, with correlation between a high intratumor TAM content and a poor prognosis (38). In this study, we observe an increase in Sema4D-positive macrophages inside tumors and in the invasive front, which goes in agreement with previously published data where Sema4D controls immune cell motility (10, 39). Our knockdown and recombinant Sema4D experiments strongly suggest that the antibody mediates a Sema4D-dependent retrograde signaling engagement in the membrane of macrophages, rather than a function blocking effect. This retrograde signaling has previously been published for this family of transmembrane proteins in different settings (reviewed in 32 and 33). Furthermore, we validated these results utilizing another anti-Sema4D antibody clone 3B4, but not with clone SK-3, which demonstrates these are antibody-specific effects over macrophages.

The SDF1/CXCR4 signaling axis has an important role in cancer progression (40). In vitro, we have proven the chemoattractant capacity of SDF1 and its stromal origin, and we have demonstrated that SDF1 release from macrophages is dependent on Sema4D expression and independent of its receptors. In vivo, our results show tumor cells' ability to respond to SDF1 stimulus, in both primary tumors and metastases, and the existence of a receptor–ligand-positive feedback loop. In fact, the correlation between the invasive capacity of the tumor and SDF1 concentration in control tumors, which is lost after anti-Sema4D treatment, suggests that the SDF1/CXCR4 signaling cascade is already activated regardless of the invasive capacity of the treated tumors.

In the clinical setting, VX15/2503, the humanized anti-Sema4D antibody, showed promising results in the first-in-human phase I clinical trial, with a 45% of patients exhibiting the absence of disease progression for at least 8 weeks (17). Consistently, our own data show anti-Sema4D antibody inhibits tumor growth of PanNETs with a tendency to increase life span but also invasion and metastasis. This latter adaptive response to treatment has not been evaluated in patients. Importantly, the combined expression of Sema4D and PlxnB1 is an independent risk factor for disease relapse in colorectal cancer (41). Other tumors where Sema4D overexpression has been reported as a negative prognostic marker include breast, ovary, soft-tissue sarcomas, and pancreas (42–44). Our data demonstrate that both Sema4D and CXCR4 expression increases with tumor progression in PanNETs and also a positive correlation between Sema4D and CXCR4 expression in metastatic PanNET samples, thus implying that Sema4D and CXCR4 expression is related to the malignization process in patients. Taking into account the inexistence of anti-Sema4D–treated PanNET patient samples, these study remark the role of Sema4D as a potential candidate of tumor malignization in this type of tumors. We have proven, using in vitro, in vivo, and in silico approaches, that stromal or immune cells are the primary source of Sema4D, rather than tumor cells.

In conclusion, we describe a hypoxia-independent novel mechanism of tumor malignization in the RIP1-Tag2 model, where the signaling engagement of anti-Sema4D antibody binding to Sema4D in macrophages seems to be responsible for the malignant phenotype via SDF1/CXCR4 signaling axis activation (Fig. 7G). Our study suggests that a combination of anti-Sema4D therapy and small-molecule inhibitors of selected macrophage functions could be a new therapeutically strategy for PanNET patients. Future studies combining nontraditional antiangiogenics and novel immunotherapies would undoubtedly shed light into the role of tumor-associated cells, allowing overcoming the undesired resistance.

No potential conflicts of interest were disclosed.

Conception and design: I. Zuazo-Gaztelu, M. Pàez-Ribes, P. Carrasco, L. Martín, M. Graupera, O. Casanovas

Development of methodology: I. Zuazo-Gaztelu, M. Pàez-Ribes, P. Carrasco, L. Martín, A. Soler, M. Martínez-Lozano

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Zuazo-Gaztelu, P. Carrasco, L. Martín, A. Soler, M. Martínez-Lozano, R. Pons, J. Llena

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Zuazo-Gaztelu, M. Pàez-Ribes, P. Carrasco, L. Martín, A. Soler, R. Pons, L. Palomero, M. Graupera, O. Casanovas

Writing, review, and/or revision of the manuscript: I. Zuazo-Gaztelu, P. Carrasco, L. Martín, O. Casanovas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Zuazo-Gaztelu, P. Carrasco, L. Martín, M. Martínez-Lozano

Study supervision: I. Zuazo-Gaztelu, P. Carrasco, M. Graupera, O. Casanovas

The authors would like to thank Vaccinex Inc. for providing reagents and research money (<20,000 Eur) to support this work, especially to Maurice Zauderer, Ernest S. Smith, and Elizabeth E. Evans for their critical discussions and helpful suggestions in the manuscript. We are also very thankful to Alba López for expert technical support with the animal colony and Álvaro Aytés for critical reading of the manuscript and helpful suggestions. This work is supported by research grants from ERC (ERC-StG-281830) EU-FP7, MinECO (SAF2016-79347-R), ISCIII Spain (AES, DTS17/00194), and AGAUR-Generalitat de Catalunya (2017SGR771). Some of these include European Regional Development Funds (ERDF; “a way to achieve Europe”).

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