Highlights
- •Netrin-1 is upregulated in metastatic pancreatic cancer
- •Hepatic stellate cell-secreted retinoic acid upregulates Netrin-1 in the liver
- •Netrin-1 on extracellular vesicles pre-conditions the metastatic niche
- •Anti-Netrin-1 therapy inhibits metastasis and improves survival in murine PDAC
Summary
The biology of metastatic pancreatic ductal adenocarcinoma (PDAC) is distinct from that of the primary tumor due to changes in cell plasticity governed by a distinct transcriptome. Therapeutic strategies that target this distinct biology are needed. We detect an upregulation of the neuronal axon guidance molecule Netrin-1 in PDAC liver metastases that signals through its dependence receptor (DR), uncoordinated-5b (Unc5b), to facilitate metastasis in vitro and in vivo. The mechanism of Netrin-1 induction involves a feedforward loop whereby Netrin-1 on the surface of PDAC-secreted extracellular vesicles prepares the metastatic niche by inducing hepatic stellate cell activation and retinoic acid secretion that in turn upregulates Netrin-1 in disseminated tumor cells via RAR/RXR and Elf3 signaling. While this mechanism promotes PDAC liver metastasis, it also identifies a therapeutic vulnerability, as it can be targeted using anti-Netrin-1 therapy to inhibit metastasis using the Unc5b DR cell death mechanism.
Introduction
One of the defining features of pancreatic adenocarcinoma (PDAC) is its proclivity for early metastatic dissemination. This is underscored by two facts: (1) the majority of patients (85%) are stage IV at diagnosis; and (2) the overwhelming majority (>80%) of patients with early-stage resectable disease succumb to metastatic recurrence. This highlights the need to better understand the biology of metastatic PDAC to identify therapeutic targets. Recent studies have found that metastatic pancreatic cancer is driven more by global epigenetic reprogramming than by metastasis-specific gene mutations, with some exceptions such as parathyroid hormone-related protein. This reprogramming leads to the selection of genetic programs that confer survival for disseminated tumor cells. One such category in advanced PDAC is that of axonal guidance genes, which include members of four canonical families (netrins, semaphorins, slits, and ephrins). These gene families were originally described for their involvement in neuronal development but have recently been implicated in cancer biology, particularly the later stages of tumor progression and metastasis.
Netrin-1 is a secreted, laminin-like protein that was initially described for its role as both an axon attractant and repellent across a diverse number of organisms. The Netrin-1 receptors DCC (deleted in colon cancer) and uncoordinated-5 (Unc5; in mammals there are four: Unc5a, Unc5b, Unc5c, Unc5d) belong to the family of dependence receptors (DRs) that mediate different functions depending on whether they are ligand bound or unbound. In the presence of ligand, cells receive positive signals (survival, proliferation, migration) while in the absence of ligand, cells receive negative signals (apoptosis). Through their death-inducing pathway, DRs function as tumor suppressors with three described means of inactivating this pathway for tumor progression and metastasis: (1) upregulation of the ligand (i.e., Netrin-1); (2) loss of expression of its receptor (through loss of heterozygosity or epigenetic silencing); and (3) loss of downstream death signaling partners. In metastasis, upregulation of Netrin-1 in breast cancer has been reported as a survival mechanism, although the mechanism of this upregulation is unknown.
While Netrin-1 has been found to be upregulated in PDAC, its role in PDAC tumor progression is poorly understood. Here we describe that Netrin-1 becomes upregulated to drive PDAC metastasis through the DR Unc5b. For the first time we show that Netrin-1 is detected on the surface of extracellular vesicles (EVs), which function to facilitate liver metastasis through the activation of hepatic stellate cells (HSCs). The mechanism of Netrin-1 upregulation in disseminated tumor cells is mediated by a feedforward mechanism whereby EVs containing Netrin-1 activate HSCs to secrete retinoic acid, which upregulates Netrin-1 by both RAR/RXR- and Elf3-mediated mechanisms. Disruption of Netrin-1/Unc5b signaling both genetically and pharmacologically prevents metastasis and increases survival in murine and human PDAC models, validating Netrin-1 as a therapeutic target for metastatic PDAC.
Results
Netrin-1 is upregulated in metastatic PDAC
We identified Netrin-1 as a gene upregulated in metastatic tumors from a murine PDAC model (INK4.1syn_Luc) whereby immunocompetent FVB mice were orthotopically injected with a primary pancreatic cancer cell line (Ink4a.1) derived from the Pdx-1-Cre;KrasG12D/+;p16−/−;p19−/− murine model. This spontaneously disseminating model was used to harvest liver and peritoneal metastases for RNA sequencing (RNA-seq) and cell-line establishment. When we compared the liver metastases to the primary tumors in this model, we observed marked differences in morphology. The histology of the primary tumors was poorly differentiated, with scattered glandular-appearing cells consistent with the quasi-mesenchymal subtype that this cell line was previously classified (Figure 1Ai). E-cadherin staining of primary tumors exhibited a lack of staining in the mesenchymal-like tumor cells and positive staining in the well-differentiated cells which were sparse (Figure 1Aii). Collagen deposition was observed at the stromal border along the tumor edge (Figure 1Aiii), similar to other orthotopic models of pancreatic cancer. While the primary tumors resembled the quasi-mesenchymal subtype of pancreatic cancer, liver metastases displayed hallmarks consistent with the “classic” molecular subtype of pancreatic cancer, including well-differentiated areas with more tumor cells forming glandular structures (Figure 1Aiv) with high expression of E-cadherin (Figure 1Av) and robust collagen stroma (Figure 1Avi).
We hypothesized that the observed plasticity in morphology in the metastatic tumors is a consequence of epigenetic reprogramming. We compared the transcriptomes from metastatic and primary tumor tissue (n = 5 for both) by total RNA-seq and found the transcriptomes to be substantially different, as evident from the volcano plot showing p value and fold change of gene expression between groups (Figure S1A). We then examined a cell line derived from a liver metastasis (Met38) for differential open chromatin regions in comparison to the parental cells using the assay for transposase accessible chromatin (ATAC-seq). While genome-wide higher-order chromatin accessibility patterns were largely comparable (see example of chromosome 1, Figure S1B), a total of 238 loci showed significant changes in chromatin accessibility. The primary tumor cell line, Ink4a.1, had 51 regions of open chromatin that were not observed in the metastatic line, Met38, whereas the latter had gain of open chromatin in 187 regions (Figure S1C). Taken together, these data indicate that this model recapitulates the cellular plasticity characteristic of epithelial-to-mesenchymal transition (EMT) and mesenchymal-to-epithelial transition (MET) in human metastatic biology.
We performed a differential gene expression analysis comparing hepatic metastases to primary tumors. Pathway enrichment analysis for Biological Processes showed top upregulated pathways included Developmental Biology, Nervous System Development, Axon Guidance, Extracellular Matrix Organization, Regulation of IGF uptake, and Signaling by PDGF (Figures 1B [left] and S1D [left]). Top downregulated enriched pathways include Cell-Cell Signaling, Biological and Cellular Adhesion, Small Molecule Biosynthetics Process, and Epithelial Cell Differentiation (Figures 1B [right] and S1D [right]). Taken together, these pathways have been shown by others as part of the metastatic process for pancreatic cancer. Differential gene expression involved in pancreatic cancer metastasis included upregulation of Pdgfrb and Cldn9 (and downregulation of Cldn1). Interestingly, metastatic tumors were high in Col6a3 (type VI collagen), which has previously been shown to be highly expressed in human PDAC as part of the desmoplastic stroma, consistent with what we observed morphologically (Figure 1Avi). We also observed that Netrin-1 was among the most upregulated genes (Figure 1B). Previous reports have shown expression of multiple axon guidance genes in PDAC, including Slit, Robo, and Sema3, and their importance regulating perineural invasion, a common feature of PDAC. Homing in on the expression of this pathway from our dataset shows that Ntn1 was the most upregulated (Figure 1B, bottom). We validated this result on a small subset of these genes by quantitative PCR (qPCR) using four cell lines derived from separate hepatic metastases (Met25, Met35, Met36, and Met38). Compared to the parental Ink4a.1 cell line, Ntn1, Selp, and Pdgfb were all significantly upregulated while expression of Basp1 trended downward, validating the RNA-seq results (Figure S1E). To confirm this result at the protein level, we examined Netrin-1 expression by immunohistochemistry of primary and metastatic tumor tissue. We found greater Netrin-1 expression in hepatic metastasis compared to primary tumors in the orthotopic model (Figure S1F). We created primary cell lines from several metastatic liver tumors and found Netrin-1 to be upregulated when compared to the Ink4a.1 primary tumor cell line (Figure 1C). To validate this in human PDAC, we performed Netrin-1 immunohistochemical staining on a tissue microarray (TMA) composed of 93 PDAC samples containing 18 metastatic and 75 primary tumors and found that Netrin-1 was significantly greater in the metastatic tumors (p = 0.03) (Figure 1D). We found a similar result in a separate TMA composed of 25 PDAC specimens taken from patients with stage II primary cancers and stage IV metastatic disease where Netrin-1 was greatest in metastatic tumors (Figure S1G). Owing to the quasi-mesenchymal (QM) status of Ink4a.1, we looked to see whether high Netrin-1 expression differs among PDAC molecular subtypes. Interestingly, in the Collisson et al. dataset, NTN1 expression is significantly greater in both the QM and exocrine as compared to classical subtypes (Figure 1E).
Unc5b is the dominant Netrin-1 receptor in PDAC and functions as a dependence receptor to promote metastasis
We examined the Ink4a.1 RNA-seq dataset and found the Netrin-1 receptor that was primarily expressed in PDAC was Unc5b. We found no detectable levels of DCC and minimal to no detectable levels of Unc5a, Unc5c, and Unc5d (Figure 2A). We confirmed this by western blot in the Ink4a.1, Met25, and Met38 metastatic clones (Figure S2A). We also detected Unc5b in all cell lines of a larger panel of murine and human PDAC cells that also lacked DCC expression (Figure S2B). Using The Cancer Genome Atlas dataset in human PDAC, we also found that the dominant Netrin-1 receptor is Unc5b (Figure S2C). We then determined the impact of Netrin-1/Unc5b signaling on PDAC metastasis both in vitro and in vivo. We first confirmed the cell death mechanism of the Unc5b receptor was intact, whereby downregulation of the ligand using small interfering RNA (siRNA) in both the Ink4a.1 and Met38 cell lines led to increased apoptosis by annexin V staining (Figure 2B). To demonstrate that Unc5b functioned positively when ligand bound, we performed in vitro migration and invasion assays in the presence of recombinant Netrin-1 using the Ink4a.1 line and found that Netrin-1 significantly increased both migration and invasion (Figure 2C). Conversely, when we knocked down Netrin-1 using siNtn1, we observed a significant decrease (Figure 2D). We validated Ntn1 knockdown (KD) in these cell lines by western blot (Figure S2D). We created a stable KD of Ntn1 in the Met38 cell line by shNtn1 (KD verified by western blot, Figure S2E) whose proliferation was unchanged by Ntn1 KD (Figure S2F). We evaluated this cell line’s ability to form pseudopodia using a 3D tumor spheroid invasion assay. We found that in the control short hairpin RNA (shRNA) Met38 cells, the cells form abundant pseudopods in 7 days that were significantly attenuated in the Ntn1 KD cells (Figure 2E). When we deleted the Unc5b gene from another mouse PDAC cell line, A1925, and from the Met38 cells, we also observed a significant reduction in both pseudopodia formation and cell invasion (Figures 2E and 2F, respectively). We validated these in vitro findings in vivo using a liver metastasis assay in which Met38 Ntn1 wild type (WT) and Ntn1 KD cells were injected into the spleen followed by splenectomy. After 10 days, mice were sacrificed and liver metastases quantitated from H&E-stained tissue sections. We found that Ntn1 KD significantly decreased the percent tumor area of metastatic tumor growth (Figure 2G). We found a similar situation when we evaluated the Unc5b knockout (KO) cells by the same assay (Figure 2H). Taken together, these experiments demonstrate that Unc5b is the dominant DR in PDAC and facilitates metastasis when ligand bound and cell death when ligand unbound.
We hypothesized that the observed plasticity in morphology in the metastatic tumors is a consequence of epigenetic reprogramming. We compared the transcriptomes from metastatic and primary tumor tissue (n = 5 for both) by total RNA-seq and found the transcriptomes to be substantially different, as evident from the volcano plot showing p value and fold change of gene expression between groups (Figure S1A). We then examined a cell line derived from a liver metastasis (Met38) for differential open chromatin regions in comparison to the parental cells using the assay for transposase accessible chromatin (ATAC-seq). While genome-wide higher-order chromatin accessibility patterns were largely comparable (see example of chromosome 1, Figure S1B), a total of 238 loci showed significant changes in chromatin accessibility. The primary tumor cell line, Ink4a.1, had 51 regions of open chromatin that were not observed in the metastatic line, Met38, whereas the latter had gain of open chromatin in 187 regions (Figure S1C). Taken together, these data indicate that this model recapitulates the cellular plasticity characteristic of epithelial-to-mesenchymal transition (EMT) and mesenchymal-to-epithelial transition (MET) in human metastatic biology.
We performed a differential gene expression analysis comparing hepatic metastases to primary tumors. Pathway enrichment analysis for Biological Processes showed top upregulated pathways included Developmental Biology, Nervous System Development, Axon Guidance, Extracellular Matrix Organization, Regulation of IGF uptake, and Signaling by PDGF (Figures 1B [left] and S1D [left]). Top downregulated enriched pathways include Cell-Cell Signaling, Biological and Cellular Adhesion, Small Molecule Biosynthetics Process, and Epithelial Cell Differentiation (Figures 1B [right] and S1D [right]). Taken together, these pathways have been shown by others as part of the metastatic process for pancreatic cancer. Differential gene expression involved in pancreatic cancer metastasis included upregulation of Pdgfrb and Cldn9 (and downregulation of Cldn1). Interestingly, metastatic tumors were high in Col6a3 (type VI collagen), which has previously been shown to be highly expressed in human PDAC as part of the desmoplastic stroma, consistent with what we observed morphologically (Figure 1Avi). We also observed that Netrin-1 was among the most upregulated genes (Figure 1B). Previous reports have shown expression of multiple axon guidance genes in PDAC, including Slit, Robo, and Sema3, and their importance regulating perineural invasion, a common feature of PDAC.
Homing in on the expression of this pathway from our dataset shows that Ntn1 was the most upregulated (Figure 1B, bottom). We validated this result on a small subset of these genes by quantitative PCR (qPCR) using four cell lines derived from separate hepatic metastases (Met25, Met35, Met36, and Met38). Compared to the parental Ink4a.1 cell line, Ntn1, Selp, and Pdgfb were all significantly upregulated while expression of Basp1 trended downward, validating the RNA-seq results (Figure S1E). To confirm this result at the protein level, we examined Netrin-1 expression by immunohistochemistry of primary and metastatic tumor tissue. We found greater Netrin-1 expression in hepatic metastasis compared to primary tumors in the orthotopic model (Figure S1F). We created primary cell lines from several metastatic liver tumors and found Netrin-1 to be upregulated when compared to the Ink4a.1 primary tumor cell line (Figure 1C). To validate this in human PDAC, we performed Netrin-1 immunohistochemical staining on a tissue microarray (TMA) composed of 93 PDAC samples containing 18 metastatic and 75 primary tumors and found that Netrin-1 was significantly greater in the metastatic tumors (p = 0.03) (Figure 1D). We found a similar result in a separate TMA composed of 25 PDAC specimens taken from patients with stage II primary cancers and stage IV metastatic disease where Netrin-1 was greatest in metastatic tumors (Figure S1G). Owing to the quasi-mesenchymal (QM) status of Ink4a.1, we looked to see whether high Netrin-1 expression differs among PDAC molecular subtypes. Interestingly, in the Collisson et al. dataset, NTN1 expression is significantly greater in both the QM and exocrine as compared to classical subtypes (Figure 1E).
Unc5b is the dominant Netrin-1 receptor in PDAC and functions as a dependence receptor to promote metastasis
We examined the Ink4a.1 RNA-seq dataset and found the Netrin-1 receptor that was primarily expressed in PDAC was Unc5b. We found no detectable levels of DCC and minimal to no detectable levels of Unc5a, Unc5c, and Unc5d (Figure 2A). We confirmed this by western blot in the Ink4a.1, Met25, and Met38 metastatic clones (Figure S2A). We also detected Unc5b in all cell lines of a larger panel of murine and human PDAC cells that also lacked DCC expression (Figure S2B). Using The Cancer Genome Atlas dataset in human PDAC, we also found that the dominant Netrin-1 receptor is Unc5b (Figure S2C). We then determined the impact of Netrin-1/Unc5b signaling on PDAC metastasis both in vitro and in vivo. We first confirmed the cell death mechanism of the Unc5b receptor was intact, whereby downregulation of the ligand using small interfering RNA (siRNA) in both the Ink4a.1 and Met38 cell lines led to increased apoptosis by annexin V staining (Figure 2B). To demonstrate that Unc5b functioned positively when ligand bound, we performed in vitro migration and invasion assays in the presence of recombinant Netrin-1 using the Ink4a.1 line and found that Netrin-1 significantly increased both migration and invasion (Figure 2C). Conversely, when we knocked down Netrin-1 using siNtn1, we observed a significant decrease (Figure 2D). We validated Ntn1 knockdown (KD) in these cell lines by western blot (Figure S2D). We created a stable KD of Ntn1 in the Met38 cell line by shNtn1 (KD verified by western blot, Figure S2E) whose proliferation was unchanged by Ntn1 KD (Figure S2F). We evaluated this cell line’s ability to form pseudopodia using a 3D tumor spheroid invasion assay. We found that in the control short hairpin RNA (shRNA) Met38 cells, the cells form abundant pseudopods in 7 days that were significantly attenuated in the Ntn1 KD cells (Figure 2E). When we deleted the Unc5b gene from another mouse PDAC cell line, A1925, and from the Met38 cells, we also observed a significant reduction in both pseudopodia formation and cell invasion (Figures 2E and 2F, respectively). We validated these in vitro findings in vivo using a liver metastasis assay in which Met38 Ntn1 wild type (WT) and Ntn1 KD cells were injected into the spleen followed by splenectomy. After 10 days, mice were sacrificed and liver metastases quantitated from H&E-stained tissue sections. We found that Ntn1 KD significantly decreased the percent tumor area of metastatic tumor growth (Figure 2G). We found a similar situation when we evaluated the Unc5b knockout (KO) cells by the same assay (Figure 2H). Taken together, these experiments demonstrate that Unc5b is the dominant DR in PDAC and facilitates metastasis when ligand bound and cell death when ligand unbound.
