Pancreatic ductal adenocarcinoma (PDAC) has a poor 5-year survival rate and lacks effective therapeutics. Therefore, it is of paramount importance to identify new targets. Using multiplex data from patient tissue, three-dimensional coculturing in vitro assays, and orthotopic murine models, we identified Netrin G1 (NetG1) as a promoter of PDAC tumorigenesis. We found that NetG1+ cancer-associated fibroblasts (CAF) support PDAC survival, through a NetG1-mediated effect on glutamate/glutamine metabolism. Also, NetG1+ CAFs are intrinsically immunosuppressive and inhibit natural killer cell–mediated killing of tumor cells. These protumor functions are controlled by a signaling circuit downstream of NetG1, which is comprised of AKT/4E-BP1, p38/FRA1, vesicular glutamate transporter 1, and glutamine synthetase. Finally, blocking NetG1 with a neutralizing antibody stunts in vivo tumorigenesis, suggesting NetG1 as potential target in PDAC.

Significance:

This study demonstrates the feasibility of targeting a fibroblastic protein, NetG1, which can limit PDAC tumorigenesis in vivo by reverting the protumorigenic properties of CAFs. Moreover, inhibition of metabolic proteins in CAFs altered their immunosuppressive capacity, linking metabolism with immunomodulatory function.

See related commentary by Sherman, p. 230.

This article is highlighted in the In This Issue feature, p. 211

Pancreatic cancer is projected to become the second leading cause of cancer-related death by 2030 (1), due to its abysmal 5-year survival rate (2). The most common form of pancreatic cancer is pancreatic ductal adenocarcinoma (PDAC), to which treatments are often refractory, and where diagnosis is often made at advanced stages of the disease (3), making few patients eligible for surgical interventions (4). Therefore, there is an urgent need to develop new diagnostic and therapeutic approaches.

PDAC has a unique microenvironment that consists of a fibrous expansion known as desmoplasia, characterized by the deposition of an abundant extracellular matrix (ECM) by cancer-associated fibroblasts (CAF; ref. 5). Studies have demonstrated that CAFs limit the efficacy of chemotherapies (6–8), promote PDAC progression (9, 10), and correlate with poor prognosis (11). Although there is a clear role for CAFs in facilitating PDAC progression, there is also evidence that CAFs can be tumor-restrictive, as complete ablation of fibroblasts from the tumor microenvironment accelerated PDAC progression (12, 13) and provided no added patient benefit (14). Thus, the role of CAFs in PDAC development and progression is incompletely understood.

An important consequence of desmoplasia is the generation of substantial interstitial pressure that contributes to the collapse of blood vessels (15), resulting in a nutrient-deprived (16) and hypoxic (17) microenvironment. Further, Kras-driven cancers such as PDAC, in which approximately 90% of tumors have mutant Kras, have been suggested to become “addicted” to exogenous sources of glutamine (Gln) as a main carbon supply (18). As such, PDAC cells make use of metabolic pathways to promote tumor growth and survival (18–20). By taking advantage of CAF-driven metabolic support (21), CAFs supply PDAC cells with key nutrients (22). Therefore, further exploration of these mechanisms, aiming to disrupt tumor–stromal cross-talk, is warranted.

In addition to the metabolic roles, CAFs are also at the center of the immunosuppressive PDAC microenvironment (23–26). CAFs exert immunosuppressive effects through direct modulation of immune cell function via cytokine secretion (27), exclusion of antitumor immune cells from the tumor (28), and/or recruitment of immunosuppressive immune cells to the tumor (29). Intriguingly, recent studies have reported a functional link between metabolism and immune cell function, which has suggested that antitumor immune cells rely on a constant supply of nutrients to perform their functions (30, 31). In the cancer context, key metabolites—glucose, arginine, Gln, and tryptophan—are depleted, with a concomitant increase of immunosuppressive waste products, such as lactate, inhibiting antitumor immune cell function (30, 32, 33). In PDAC, the immunosuppressive environment created by CAFs, combined with the nutrient-poor milieu, builds a hostile environment that inhibits antitumor immune cell function. Thus, means to simultaneously revert these linked features of the PDAC microenvironment would be therapeutically beneficial.

In this study, we sought to clarify the roles of CAFs in PDAC tumorigenesis. We uncovered upregulation of the glutamatergic presynaptic protein Netrin G1 (NetG1; refs. 34, 35) in CAFs. Of note, we also found expression of NetG1 ligand (NGL1), the sole known postsynaptic binding partner of NetG1 (36), in PDAC cells. In fact, we saw that NetG1 was particularly overexpressed in human PDAC tissue compared with samples from normal human pancreas and that fibroblastic NetG1 expression correlated inversely with patient survival. We demonstrated that NetG1 in CAFs and NGL1 in tumor cells enhanced tumorigenesis, as ablation of either protein reduced tumor burden in mice. Functionally, NetG1 was responsible for glutamate (Glu), Gln, and cytokine release by CAFs which allowed PDAC cells to survive in low-nutrient conditions and reduced death induced by natural killer (NK) cells. Mechanistically, NetG1 controlled a signaling circuit comprised of Gln synthetase (GS), vesicular Glu transporter 1 (VGlut1), p38/FRA1, and AKT/4E-BP1, and systematic inhibition of each component of the pathway revealed their roles in the protumor metabolic and immunosuppressive functions of CAFs. Finally, the therapeutic applicability of targeting NetG1 in vivo was demonstrated, as an anti-NetG1 neutralizing mAb inhibited tumorigenesis in a murine model of PDAC. Thus, we demonstrate the importance of considering the stroma as a viable therapeutic option in PDAC.

NetG1 Is Upregulated in CAFs Compared with Patient-Matched Tumor-Adjacent Fibroblasts

To gain insight into the mechanism by which naïve fibroblasts are activated into CAFs, we performed a transcriptomic analysis comparing two sets of patient-matched tumor-adjacent fibroblasts (TA) and CAFs (Fig. 1A; Supplementary Table S1) that were cultured in our cell-derived ECM-based culturing system that assures maintenance of in vivo–like fibroblastic phenotypes (37, 38). We identified the most differentially regulated genes between patient-matched TAs and CAFs (117 genes, P < 0.01, log fold change; Supplementary Table S1). To characterize the underlying molecular pathways and biological processes, we performed gene set enrichment analysis (GSEA) and found significant enrichment (Benjamini–Hochberg FDR < 0.25) of several pathways previously found to be associated with the transition from normal fibroblasts to CAFs, including upregulation of integrin signaling, focal adhesions, actin cytoskeleton, and ECM (Fig. 1B and C; Supplementary Table S1). Interestingly, results also revealed an upregulation of genes involved with axonal guidance and neurite outgrowth in CAFs compared with TAs. The second most upregulated gene overall and top hit among the neuronal genes was NTNG1, encoding NetG1 (34), a synapse-stabilizing protein known for its involvement in axonal communication. We also noticed in CAFs a negative enrichment of pathways associated with antigen presentation [KEGG Systemic Lupus Erythematosus; FDR < 0.25; normalized enrichment score (NES) = –2.66] and IFN signaling (Reactome Interferon Signaling; FDR < 0.25; NES = –2.66; Fig. 1D; Supplementary Table S1). Overall, this transcriptomic data confirmed that our ECM-cultured CAFs indeed upregulate known pathways previously linked to fibroblast activation in cancer, as well as an association with neuronal gene expression.

NetG1 in CAFs and Its Binding Partner NGL1 in Tumor Cells Are Overexpressed in PDAC

NetG1 is present on presynaptic neurons, and it is known to stabilize excitatory glutamatergic synapses by interacting with its only known receptor on postsynaptic cells, NGL1 (encoded by LRRC4C; refs. 35, 36). To assess the clinical relevance of NetG1, we performed simultaneous multichannel immunofluorescence (SMI; ref. 38) on 6 normal, 4 tumor-adjacent, and 15 tumor pancreatic tissue samples to discern NetG1 and NGL1 expression patterns in distinct subsets of cells. NetG1 was upregulated in the stromal compartment (pan-cytokeratin, vimentin+ areas) in tumor tissue compared with normal pancreatic tissue from individuals without cancer (Fig. 1E and F). Moreover, levels of tyrosine 397 phosphorylated focal adhesion kinase (pFAK) followed a similar pattern of increased expression in tumor tissue compared with normal tissue, in line with our established protumor CAF signature (ref. 38; Fig. 1E and F). A strong correlation between stromal NetG1 expression and stromal pFAK expression was observed across all tissues (R2 = 0.8046, P < 0.0001; Supplementary Fig. S1A). Further, NGL1 was upregulated in the epithelial compartment (pan-cytokeratin+ areas) of both tumor-adjacent and tumor tissues (Fig. 1E and F), with a weak correlation between stromal NetG1 and epithelial NGL1 expressions across all tissues (R2 = 0.2359, P < 0.0001; Supplementary Fig. S1A). Comparing four TAs and five CAFs isolated from patient tissue, we confirmed that CAFs upregulated NetG1 protein expression (∼2.5-fold increase; Supplementary Fig. S1B). In addition, we observed the expression of NGL1 in 5 patient-derived cell lines from patient-derived xenograft (PDX) models (Supplementary Fig. S1C), as well as upregulated expression in tumor cell lines derived from an isogenic cell line model, hTERT-HPNE E6/E7 (referred to as E6/E7) and hTERT-HPNE E6/E7/KRASG12D (referred to as PDACc), compared with their immortalized epithelial cell counterpart (hTERT-HPNE, referred to as hTERT; ref. 39; Supplementary Fig. S1D). Collectively, NetG1 and its binding partner NGL1 were upregulated in cancer compared with normal tissue in the pancreas, and segregated into distinct cellular compartments (stromal vs. epithelial).

Fibroblastic NetG1 Expression Correlates with Worse Overall Survival in Patients with PDAC

Using the cBioportal database (40, 41), we queried both NetG1 (NTNG1) and NGL1 (LRRC4C) mRNA expression across 31 cancer types and found that PDAC was among the highest expressers of both genes (Supplementary Fig. S2A, arrows). In addition, we stratified NTNG1 and LRRC4C by clinical subtype according to three independent datasets [refs. 42, 43; The Cancer Genome Atlas (TCGA), ref. 44; Supplementary Fig. S2B–S2E]. We found that in two out of three datasets there was a trend toward elevated NTNG1 and LRRC4C in the basal classification, which presents with the worst prognosis of all PDAC subtypes (43). Moreover, we used the datasets from TCGA and Puleo and colleagues (42, 44); stratified patients into four groups based on NTNG1 and LRRC4C expression: LRRC4C high/NTNG1 high (UP-UP), LRRC4C high/NTNG1 low (UP-DN), LRRC4C low/NTNG1 high (DN-UP), and LRRC4C low/NTNG1 low (DN-DN); and examined if these groups correlated with overall survival (OS; Supplementary Fig. S2F). Although we did not see a significant association between expression and OS, we found a marginal association between relapse-free survival (RFS) and coexpression of LRRC4C and NTNG1 [HR, 9.45 (1.27–70.27); P = 0.0074; FDR > 0.5; Supplementary Fig. S2G]. These inconclusive results are likely due to the intermingling of stromal and epithelial RNA that is inherent in cancers with a high degree of stromal expansion, such as PDAC (44), highlighting the importance of carefully dissecting the contributions of the stromal and epithelial compartments of the tumor microenvironment.

Thus, to further address the clinical significance of NetG1 and NGL1, we performed IHC staining of NetG1 and NGL1 on two independent tissue microarrays (TMA; Fig. 1G; Supplementary Table S2). Protein expressions of fibroblastic NetG1 and tumoral NGL1 were scored in a blinded manner by a pathologist (0–4 scale; Supplementary Table S3). Strikingly, fibroblastic NetG1 expression correlated inversely with patient OS (P < 0.01; median OS NetG1+ vs. NetG1: 20 vs. 47 months; HR: 2.35) in the Fox Chase Cancer Center (FCCC) TMA, whereas there was a similar trend in the MD Anderson (MDA) TMA (P = 0.0793; median OS NetG1Hi vs. NetG1Lo: 19 vs. 57 months; HR: 1.73; Fig. 1H). Of note, tumoral NGL1 expression was not predictive of survival (Supplementary Fig. S3). This was likely due to the fact that nearly every patient stained positive for tumoral expression of NGL1 (72/77 FCCC; 129/135 MDA). Overall, fibroblastic NetG1 expression was associated with worse OS for patients with PDAC, suggesting that NetG1 could be a useful prognostic marker in PDAC.

RNA-seq Analysis Reveals That Ablation of NetG1 in CAFs Results in a Normalized Stromal Gene Expression Signature

CAFs are known to function in an ECM-dependent manner and regulate tumorigenesis in a number of ways (5). Hence, we relied on our well-established three-dimensional cell-derived matrix culturing system, referred to as “3-D” for simplicity, which simulates the physiologic as well as pathologic tumor microenvironment (37, 38, 45), to dissect the functions of NetG1 in more detail in vitro. To this end, we first generated NetG1 knockout (KO) CAFs through CRISPR/Cas9 (Supplementary Fig. S4A), from one of our patient-derived CAF lines that displayed high levels of NetG1 mRNA and protein expression compared with TAs (Supplementary Figs. S1B and S4B). To gain insight into global changes in gene expression, we performed RNA-sequencing (RNA-seq) analysis comparing control (CON) CAFs and NetG1 KO CAFs (Supplementary Fig. S4C; Supplementary Table S4). Pathway and gene enrichment analysis identified downregulation of genes associated with many critical pathways known as being upregulated in CAFs in our microarray analysis (Fig. 1), including axon guidance, focal adhesion, ECM and TGFβ signaling, and inflammatory responses (Supplementary Fig. S4D and S4E; Supplementary Table S4). Thus, taken together, these results suggest a reversion of gene expression profiles linked with known protumor CAF functions upon loss of NetG1 expression.

Ablation of NetG1 Does Not Influence Myofibroblastic Features of CAFs, but Reverses Protumorigenic CAF Markers

Next, we followed NetG1 expression during ECM production in vitro, as this is a major function of CAFs in the in vivo tumor microenvironment (5). Importantly, NetG1 protein expression in CAFs was significantly elevated in 3-D compared with two-dimensional (2-D) culturing conditions (Supplementary Fig. S5A). We then assessed the desmoplastic phenotype of CON or NetG1 KO CAFs during ECM production, according to our previous study (38). Interestingly, NetG1 KO CAFs did not have altered canonical myofibroblastic features, as determined by ECM fiber (fibronectin) alignment and levels/localization of alpha smooth muscle actin (α-SMA; Supplementary Fig. S5B and S5C). However, NetG1 KO CAFs displayed reduced levels of active α5β1-integrin and pFAK, indicative of a CAF phenotype that was associated with delayed recurrence following surgery in patients with PDAC (Supplementary Fig. S5D and S5E; ref. 38), consistent with the correlation seen in human tissue samples (Supplementary Fig. S1A). Moreover, we segregated six patient tumor tissues into two groups, stromal NetG1 high (N = 3) and low (N = 3), and questioned if the pan-CAF marker podoplanin (PDPN; refs. 46–48) and the myofibroblastic marker α-SMA correlated with NetG1 expression in vivo. Interestingly, PDPN had a significantly higher stromal IHC score in NetG1 high tissue compared with NetG1 low tissue, whereas α-SMA levels remained high in both tissues (Supplementary Fig. S5F and S5G), similar to the in vitro data (Supplementary Fig. S5B–S5E). Thus, deletion of NetG1 does not alter canonical myofibroblastic features of CAFs, but attenuates the expression of known protumor CAF markers. Therefore, we hypothesized that deletion of NetG1 would revert protumorigenic functions of CAFs and potentially stunt tumorigenesis in vivo.

Ablation of NetG1 in CAFs and NGL1 in PDAC Cells Significantly Stunts Tumorigenesis in Orthotopic Murine Models of PDAC

To test the impact of NetG1 and NGL1 in pancreatic cancer in vivo, we used multiple murine models of PDAC, in which we orthotopically injected CAFs and PDAC cells, and observed tumor progression and burden. In the first model, we coinjected human CON or NetG1 KO CAFs with RFP+ CON or NGL1 KO PDACc cells (NGL1 knockout confirmed; Supplementary Fig. S6A), at a CAF:PDAC ratio of 3:1, into the pancreas of SCID mice and followed tumorigenesis over 1 month. Animals coinjected with CON CAFs and CON PDACc cells had a substantially worse tumor burden, with more obvious macroscopic RFP+ coverage of the pancreas (Fig. 2A), as well as greater tumor weight (∼2–3-fold; Fig. 2B), pancreas weight (∼2–3-fold; Fig. 2C), and tumor area (Fig. 2D), than all other conditions. Importantly, mice coinjected with NetG1 KO CAFs and CON or NGL1 KO PDACc cells developed tumors to a similar extent as mice injected with CON or NGL1 KO tumor cells alone, suggesting that CAFs lacking NetG1 do not confer any benefit for tumorigenesis. As a control, mice were injected with the CON or NetG1 KO CAF lines alone, and those animals did not develop tumors. Histologically, tumors that developed from CON CAF and CON PDACc coinjections developed significantly more sarcomatous PDAC morphology (∼3.5–5-fold more) compared with all other groups (Fig. 2E; Supplementary Fig. S6B), which has been associated with an extremely poor patient prognosis (49–51). We also observed that mice injected with NGL1 KO PDACc cells alone displayed a tendency for less tumor burden than CON PDACc cells; thus, we proceeded to investigate this in more detail.

We first explored NetG1 expression in the stroma of LSL-KrasG12D/+; Pdx-1-Cre (KC) mice, which readily develop precancerous lesions known as pancreatic intraepithelial neoplasia (PanIN) and eventually full-blown invasive PDAC (52). Using SMI, we observed that as KC mice age and develop PanINs, stromal NetG1 expression was detected in pancreatic tissue as early as 12 and 16 weeks, while undetectable in non–tumor-bearing mice (Supplementary Fig. S7A). We generated two murine PDAC cell lines, isolated and cloned from the more aggressive LSL-KRASG12D;TP53−/−;PDX1-Cre (KPC) model (53), one with low NGL1 expression that resulted in an average survival time of 7.5 weeks after orthotopic injection into C57BL/6 mice (KPC4B), and one with high expression of NGL1 that resulted in an average survival time of 3.5 weeks after orthotopic injection (KPC3; Supplementary Fig. S7B). These data indicated that, similarly to human PDAC, NetG1/NGL1 are present in PDAC in mice. Using CRISPR/Cas9, we generated two NGL1 KO murine PDAC cell lines and one control line from the aggressive KPC3 cells (Supplementary Fig. S7C) and injected them orthotopically into syngeneic C57BL/6 mice (CON, KO1, KO2). Mice injected with CON KPC3 cells (CON group) had a tumor incidence of 100% (7/7), whereas the groups injected with NGL1 KO1 and KO2 KPC3 cells (NGL1 KO group) had an incidence of 22.2% (2/9) and 40% (4/10) with KO tumors displaying significantly less tumor weight (Supplementary Fig. S7D), pancreas weight (Supplementary Fig. S7E), and percent tumor area (Supplementary Fig. S7F and S7G) compared with the CON group. In addition, representative MRI images revealed the expansion of the pancreas in CON group over 3 weeks, whereas the pancreas from the NGL1 KO group seemingly remained at normal size over that time span (Supplementary Fig. S7H). Moreover, tumors from the KO group had a trend toward being more differentiated than those from the CON group. An average of 79% of the cells from CON tumors were classified as poorly differentiated, compared with 48% in KO tumors (Supplementary Fig. S7I), indicative of less aggressive tumors overall. Consequently, NGL1 KO tumors also proliferated slower, as measured by Ki-67 staining, and had higher rates of apoptosis, as measured by terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) staining (Supplementary Fig. S7J). These results were in line with less Ki-67 expression in human NGL1 KO PDACc compared with CON PDACc cultured in 3-D in vitro (Supplementary Fig. S7K).

In order to evaluate the impact of the immune system in this model, orthotopic injections of CON or NGL1 KO KPC3 cells into SCID (lacks T and B cells) and NSG (lacks T, B, and NK cells) mice were performed and compared with immunocompetent mice (B6). All animals injected with CON cells presented a greater tumor burden than those injected with NGL1 KO cells, independently of the immune background (Supplementary Fig. S7L). This suggests that tumorigenesis, in relation to tumoral NGL1, is not significantly affected by the immune system. Overall, the results demonstrated that both NetG1 in CAFs and NGL1 in tumor cells promote PDAC tumorigenesis in vivo.

NetG1 Increases Heterotypic Cell–Cell Interactions between CAFs and PDAC

Because of NetG1's known role in cell–cell adhesion in the brain, namely through its sole known receptor NGL1, we tested the potential function of NetG1/NGL1 in CAF–PDAC cell–cell interactions. Ablation of NetG1 in CAFs reduced cell engagement with CON or NGL1 KO PDACc by 72% and 79%, respectively (identified by the yellow regions that resulted from the interaction between GFP+ CAFs and RFP+ PDACc; Supplementary Fig. S8A–S8C). Accordingly, CON or NGL1 KO PDACc had increased motility (81% and 74%, respectively) when cocultured with NetG1 KO CAFs in 3-D (Supplementary Fig. S8D and S8E). Unexpectedly, deletion of NGL1 from PDACc was not sufficient to alter PDACc–CAF cell engagement or PDACc motility, signifying that there may be a redundant mechanism controlling PDAC–CAF engagement in the absence of NGL1 in tumor cells and that NetG1 in CAFs is the driver of PDAC–CAF heterotypic interactions.

NetG1 in CAFs Enhances Direct Material Transfer to PDAC Cells, and NGL1 Regulates Macropinocytosis in PDAC Cells

Because NetG1 in CAFs is needed for PDAC–CAF interactions, we questioned if there was a further functional consequence of the increased heterotypic cell–cell interactions observed during coculture. Because PDAC cells have been suggested to use numerous extracellular macromolecules as a source of energy (16, 54) and receive nutrients from CAFs (55, 56), we hypothesized that CAFs may provide resources to PDAC cells in a NetG1-dependent manner. To test this, we followed the exchange of GFP from GFP+ CAFs and RFP from RFP+ PDACc. Indeed, after 36 hours of coculture, GFP was detected intracellularly in PDACc (Supplementary Fig. S9A), demonstrating that PDACc received CAF-derived material. Knockout of NetG1 in CAFs, or NGL1 in PDACc, resulted in a 49% to 61% reduction in GFP transfer to PDACc, compared with CON PDACc/CON CAF group (Supplementary Fig. S9B). Interestingly, when performing the same material transfer assay with conditioned media (CM) derived from GFP+ CAFs, instead of direct coculture, only NGL1 loss in PDAC cells resulted in an approximately 55% to 65% reduction in GFP transfer from either CON or NetG1 KO CM (Supplementary Fig. S9C). This suggested that NGL1 in PDAC cells could control uptake of material from an extracellular source, and thus we sought to understand this uptake mechanism in PDAC cells.

KRAS-driven cancers, such as PDAC, have been shown to rely on macropinocytosis to overcome nutrient deprivation (57), and therefore we decided to determine the intrinsic ability of PDACc and PANC-1 cells (NGL1 knockout confirmed; Supplementary Fig. S9D) to perform macropinocytosis in 3-D. First, we performed a dose–response curve, using the alamarBlue assay, to find the optimal nontoxic concentration to treat cells with the macropinocytosis inhibitor 5-(N-ethyl-N-isopropyl) amiloride (EIPA), which was determined to be approximately 25 μmol/L for all cell lines (Supplementary Fig. S9E). Next, we plated RFP+ CON or NGL1 KO PDAC cells in 3-D in the presence of DMSO or EIPA for 24 hours and tested their ability to uptake FITC-dextran (58). Both CON PDACc and PANC-1 cells could effectively uptake FITC-dextran, whereas all EIPA-inhibited cells displayed an approximately 50% decrease in uptake (Supplementary Fig. S9F and S9G). Strikingly, NGL1 KO PDACc or PANC-1 cells had a significant impairment in their ability to uptake FITC-dextran (∼50% and ∼30%, respectively), similar to the levels of EIPA-treated cells. We next tested whether macropinocytosis was the mechanism PDACc cells used to uptake GFP from CAF CM. Indeed, although CON PDACc cells could efficiently uptake extracellular GFP, NGL1 KO PDACc had approximately 50% less uptake of GFP, at similar levels to EIPA-treated PDACc cells (Supplementary Fig. S9H).

Therefore, although NGL1 was dispensable for CAF–PDAC engagement, both NGL1 and NetG1 were necessary for PDACc to directly receive material from CAFs. Conversely, NGL1 regulated the macropinocytic ability of PDAC cells to uptake extracellular material, independent of NetG1 expression in CAFs.

NetG1/NGL1 Axis Protects PDAC Cells from Death under Nutrient Restriction

Because the PDAC microenvironment contains limited nutrients and CAFs can transfer material in a NetG1/NGL1–dependent manner (Supplementary Fig. S9), we hypothesized that CAFs serve as a critical nutrient conduit and could rescue PDAC cells from nutrient deprivation. To this end, we cocultured GFP+ CAFs with RFP+ PDACc in 3-D under serum-free/Gln-free conditions and measured the survival of RFP+ PDACc over a 4-day time span. We observed that disruption of NetG1/NGL1, by deletion of either NetG1 in CAFs or NGL1 in PDAC cells, resulted in a 30% to 37% decrease in PDACc survival, compared with CON PDACc/CON CAF group, similar to PDACc cultured alone (Fig. 2F). These results were replicated using the pancreatic cancer cell line PANC-1 (CON or NGL1 KO), with a 40% to 59% reduction in PANC-1 survival compared with CON/CON group (Fig. 2G). To determine if CAF-secreted factors alone were important for tumor cell survival, PDACc cells were cultured in 3-D with CM obtained from CON and NetG1 KO CAFs. We found that NetG1 KO CAF CM was 76% to 82% less efficient at supporting PDACc survival when compared with CON CAF CM, similar to the CM from TAs (81%–83% of CON CM; Fig. 2H). To further assess the effect of NetG1 on the microenvironment generated by CAFs, we performed survival assays using ECMs produced by CON and NetG1 KO CAFs as substrates for cancer cells, and compared them with the ECM produced by TAs, which express less NetG1 than CAFs. Accordingly, the microenvironment generated by NetG1 KO CAFs was less supportive compared with the CON ECM, with a 30% decrease in PDACc survival (Fig. 2I). The TA-generated ECM was even less supportive, resulting in a decrease in PDACc survival of 65%. For experimental rigor, we also generated NetG1 knockdown (KD) CAFs, using CRISPR interference (CRISPRi), in two CAF lines (CAF and CAF2) and replicated the survival assay results (Supplementary Fig. S10A–S10C). Collectively, these results suggest that direct physical contact with both CAFs and the factors secreted by CAFs, including CM and ECM, support PDAC survival under nutritional deprivation, in a NetG1/NGL1–dependent fashion.

Alterations in Glutamate/Glutamine Generation in NetG1 KO CAFs Reduce PDAC Cell Survival

In an effort to gain insight into key factors that could support PDAC survival under nutritional deprivation, we performed an amino acid screen comparing TAs, CON CAFs, and NetG1 KO CAFs, during ECM production (Supplementary Tables S5–S7). The greatest net change in amino acid secretion between TAs and CAFs was found to be in Glu and Gln, as CAFs secreted 2-fold more of these amino acids (Fig. 2J). These results are in line with recent studies that suggest that the catabolism of Glu plays a role in PDAC and other cancers (18, 19, 59, 60). Strikingly, we found that NetG1 KO CAF CM contained 64% less Glu and 81% less Gln, compared with CON CAFs (Fig. 2J), and this was comparable in CAF2 (Supplementary Fig. S10D). To confirm that patient-derived CAFs could produce Glu and Gln de novo, we performed an isotope-tracing experiment by providing CAFs with 13C6-glucose and 13C3-pyruvate, and followed labeled carbons of Glu and Gln in CAFs. Indeed, in both 3-D and 2-D, 70% to 80% of Glu and approximately 35% Gln were labeled (Supplementary Fig. S10E), with all M+1–5 populations receiving a fraction of the label (Supplementary Fig. S10F), indicating that CAFs in our system could produce these metabolites de novo.

In order to determine the relative contributions of Glu and Gln to PDAC survival conferred by CON CAFs CM, we performed a Glu/Gln rescue assay, where the amounts of Glu/Gln produced by CON CAFs (150 and 25 μmol/L respectively) were added to the depleted media and compared with the rescue provided by CON CAFs CM. Interestingly, although Glu or Gln alone or in combination promoted survival by 40%, 49%, or 55% compared with nutrient-depleted media alone, CON CAF CM increased CON PDACc survival by 3.5-fold, indicating that there are additional factors in the CM, besides Glu and Gln, contributing to PDAC survival (Fig. 2K). Importantly, Glu/Gln were not able to rescue NGL1 KO PDACc survival, suggesting a potential defect in Glu/Gln utilization in PDAC cells lacking NGL1. However, supplementing back Glu/Gln at CON CAF CM levels to NetG1 KO CM was able to significantly rescue PDACc survival, although to a lesser degree in NGL1 KO cells. These results suggest that adding Glu/Gln to factors secreted by NetG1 KO CAFs suffices to mimic CON CAF CM effects in rescuing NGL1+ PDACc survival under nutrient-deprived conditions.

Guided by the clear differences in Glu/Gln levels upon KO of NetG1 in CAFs, we explored if glutaminase (GLS) and GS protein levels were altered in these cells. We detected expression of GLS and GS in TAs and CAFs (Fig. 2L). Although GLS levels were generally higher in CON and NetG1 KO CAFs than TAs, GS was significantly reduced in NetG1 KO CAFs, suggesting that NetG1 KO CAFs are compromised in their ability to generate Gln from Glu. Interestingly, NetG1 KO CAFs had reduced levels of VGlut1, a major protein responsible for loading presynaptic vesicles with Glu, assuring synaptic Glu secretion, during neuronal communication (ref. 61; Fig. 2L). Thus, although CON and NetG1 KO CAFs expressed similar levels of GLS, NetG1 KO CAFs displayed decreased levels of VGlut1 and GS (also confirmed at the mRNA level; Supplementary Fig. S10G), which could account for the reduced amounts of Gln produced and Glu released by NetG1 KO CAFs. These results were also observed in the NetG1 KD CAFs (Supplementary Fig. S10A) and CAF2 (Supplementary Fig. S11A). As expected from KRAS-mutant cells, PDACc expressed lower levels of GS compared with CON CAFs, while maintaining GLS expression (Supplementary Fig. S11B). Interestingly, knockout of NGL1 in PDACc resulted in a reduction in GLS protein levels, suggesting an even greater metabolic dependence on extracellular nutrients in the KO cells compared with CON PDACc. In addition, NGL1 expression in PDAC cells appears to be critical for the utilization of some extracellular factors, such as Gln and Glu, perhaps due to downregulation of Glu receptor–binding pathways, as suggested by the TCGA NGL1 stratification (Supplementary Fig. S12). Collectively, these results suggest that Gln and Glu derived from CAFs help support PDAC survival, under nutrient-deprived conditions, and that this support depends on NetG1 expression in CAFs.

Knockout of NetG1 in CAFs Abrogates Their Immunosuppressive Phenotype and Permits NK-Cell Antitumor Function, Partially through IL15

Thus far, we noticed that NetG1 expression in CAFs was critical for the ability of CAFs to provide tumor cells with essential nutrients for their survival (Fig. 2). Moreover, the RNA-seq analysis suggested that KO of NetG1 led to a normalization of CAFs, through the downregulation of protumor fibroblastic pathways, including inflammatory signaling (Supplementary Fig. S4). This led us to question if NetG1 is also a functional mediator of immunosuppression, another key CAF function (5). Hence, we performed both traditional and multiplex (U-Plex) ELISAs with TAs and CAFs cultured in 3-D, to identify a cytokine profile that could define the immunosuppressive milieu that CAFs generate. As expected, we found that CON CAFs produce increased levels of GM-CSF, IL1β, IL8, CCL20/MIP3α, and TGFβ compared with TAs (Fig. 3A). Strikingly, ablation of NetG1 in CAFs significantly decreased protein levels of GM-CSF (40.6-fold), IL6 (1.4-fold), IL8 (13.3-fold), IL1β (5.2-fold), CCL20/MIP3α (80-fold), and TGFβ (1.5-fold) compared with CON CAFs (Fig. 3A), with a similar downregulation in NetG1 KD CAFs (Supplementary Fig. S13A) and CAF2 (Supplementary Fig. S13B). At the mRNA level, GM-CSF and TGFβ were correspondingly downregulated in NetG1 KO CAFs, with increases in IL6 and IL8 mRNA expression (Supplementary Fig. S13C), suggesting posttranslational regulation of IL6 and IL8. On the other hand, expression of IL2, IL12 p70, IL10, IFNγ, TNFα, and IFNβ were all below the limit of detection in all three fibroblastic populations tested (data not shown). Interestingly, we observed that IL15 protein levels, one of the most potent known NK-cell activators, were upregulated in both CON and NetG1 KO CAFs compared with TAs (4- and 2.9-fold, respectively; Fig. 3A). Also, no differences were noted in mRNA expression of IL15 when comparing CON CAFs with NetG1 KO CAFs (Supplementary Fig. S13C). This suggested that NetG1 KO CAFs may generate a less immunosuppressive microenvironment than the one from CON CAFs, possibly allowing NK cells to more effectively kill PDAC cells, which could represent an important antitumor mechanism. Indeed, in our two TMA cohorts, the majority of patients presented with low NK-cell infiltrates (represented by IHC of NKp46; Supplementary Fig. S14A; FCCC: 59/79 or 75%; MDA: 80/137 or 58%). Importantly, similar to the median OS times observed in stromal NetG1–high patients (Fig. 1H), patients with low NK-cell infiltrate numbers had a median OS of 20.5 months and 21 months in the FCCC and MDA TMAs, respectively (Supplementary Fig. S14B; Supplementary Table S3). However, in patients who had more NK-cell infiltration, there was a trend for a noticeable increase in survival, 46 and 33 months in the FCCC and MDA TMAs, respectively (P values: 0.1213, 0.0546; HR: 0.65, 0.68).

To test the influence of CAF NetG1 in NK cell activation and function, we first assayed the activation of an NK cell line, NK-92, in our in vitro 3-D culturing system. After IL2 treatment alone, 68.5% and 72.3% of the NK cells became positive for IFNγ and Granzyme B (i.e., activated), respectively (Fig. 3B). Conversely, direct coculture of activated NK cells with CON CAFs or with CM from CON CAFs abolished NK-cell activation marker expression, illustrating the potent immunosuppressive potential of CON CAFs. NetG1 KO CAFs, however, were less immunosuppressive, as 12% to 23% of NK cells maintained IFNγ or Granzyme B expression. This result is consistent with the decreased levels of immunosuppressive cytokines generated by the NetG1 KO CAFs (Fig. 3A). A similar experiment was performed using primary NK cells isolated from healthy human donor blood, with IFNγ and CD69 as markers of activation. As with NK-92 cells, we observed that although CM from CON CAFs inactivates primary NK cells in a concentration-dependent manner (i.e., 10% CM prevented NK activation by ∼50%), NetG1 KO CAF CM was significantly less immunosuppressive at the same dilutions (Fig. 3C). We next tested if the activation status of NK-92 cells correlated with their functional ability to kill PDAC cells. To this end, we performed an experiment coculturing resting and active NK cells with CON or NGL1 KO RFP+ PDACc, using complete media (to avoid stress due to nutrient deprivation) in 3-D. Accordingly, active NK cells were twice as effective in killing both CON and NGL1 KO PDACc as resting NK cells (Supplementary Fig. S15A and S15B), suggesting that the ability of CAFs to control antitumor NK function is independent of PDACc NGL1 status. To probe how CAFs could directly affect the ability of NK cells to kill PDACc, we cocultured CON or NetG1 KO CAFs with CON or NGL1 KO RFP+ PDACc in the presence of resting or active NK cells, again using complete media in 3-D. In agreement with the observed immunosuppressive profile of CON CAFs (Fig. 3A and B), active NK-92 cells were prevented from killing PDACc in the presence of CON CAFs, and this protection was decreased by 38% to 39% when cultured with NetG1 KO CAFs (Fig. 3-D; Supplementary Fig. S15A, bottom). These results were effectively replicated when using PANC-1 and CAF2 (Supplementary Fig. S15C). Intriguingly, even when resting NK cells were cultured in the presence of NetG1 KO CAFs, they were 45% to 65% more effective at eliminating PDACc compared with coculture with CON CAFs, suggesting that NetG1 KO CAFs could partially activate NK-92 cells (Fig. 3E; Supplementary Fig. S15A, bottom).

The NK-92 killing assay was replicated in 2-D culturing conditions. The KO of NetG1 in CAFs resulted in a significant decrease in CON PDACc or CON PANC-1 survival against NK cell–induced death compared with CON CAF conditions (84% and 66% less, respectively; Supplementary Fig. S15D–S15F). Interestingly, in 2-D conditions, NGL1 KO cells (KO1 and KO2) were more sensitive to NK-cell killing even when cocultured with CON CAFs (88% and 78% less survival compared with CON PDACc, respectively), suggesting that NetG1/NGL1 heterotypic cell–cell contacts were important for protection against NK cell–induced death in the absence of ECM (Supplementary Fig. S15D–S15F). Importantly, neither resting nor active NK cells affected CON or NetG1 KO CAF survival (Supplementary Fig. S15G). To question whether tumor to stromal cell ratios are important for the observed effect, we cocultured various ratios of PDACc to CAFs (5:1, 3:1, 1:3) and again performed the NK-cell killing assay. As the number of CAFs increased, PDACc survival increased as well, indicating a direct effect of CAF numbers on NK-cell killing function (Supplementary Fig. S15H).

Overall, NetG1 expression in CAFs creates an immunosuppressive microenvironment that inactivates NK cells and protects PDAC cells from NK cell–mediated death. Loss of NetG1 expression in CAFs partially reverts the immunosuppressive capacity of CAFs, allowing NK cells to maintain their activity and eliminate PDAC cells. Moreover, the microenvironment generated by CAFs (i.e., ECM) also plays an important role in the support of PDAC survival, highlighting the differential effects of 2-D culture versus 3-D culture.

Although deletion of NetG1 in CAFs led to a decrease in the expression of immunosuppressive cytokines, levels of IL15, a key activator of NK cells, remained significantly higher than in TAs (Fig. 3A). Thus, we hypothesized that the downregulation of immunosuppressive cytokines, coupled with the maintenance of IL15 expression, was partially responsible for the observed antitumor phenotype of the NetG1 KO CAFs. We repeated the NK-cell killing assay in 2-D and 3-D, coculturing CON or NetG1 KO CAFs with CON or NGL1 KO PDACc, in the presence of a neutralizing IL15 antibody or IgG isotype control. Indeed, neutralization of IL15 resulted in an approximately 35% and 25% (2-D and 3-D, respectively) increase in PDACc survival compared with IgG-treated conditions, independent of PDACc NGL1 status, suggesting that CAF-expressed IL15 was partially responsible for the antitumor microenvironment created by NetG1 KO CAFs (Fig. 3F and G). Thus, collectively, the in vitro NK assays and patient data suggest CAFs lacking NetG1 improve NK-cell function, which is better for OS in patients with PDAC, in a mechanism partially dependent on IL15 expression by CAFs that is accompanied by limiting immunosuppressive cytokine secretion.

p38 and AKT Pathways Mediate NetG1-Dependent Protumor CAF Functions

Having identified key functions regulated by NetG1, we sought to uncover the downstream mediators of NetG1 that are responsible for protumor functions of CAFs. We identified the downregulation of pAKT and p-p38 in NetG1 KO CAFs (Fig. 4A), which were confirmed in CAF2 (Supplementary Fig. S11A), and postulated that these two pathways were largely mediating the effects of NetG1. Therefore, we employed the use of pharmacologic inhibitors of p38 (p38i) and AKT (AKTi) on CAFs in our 3-D system and assessed PDAC survival against metabolic stress and NK-induced death, as well as Glu/Gln and cytokine production. Interestingly, we observed a greater decrease in PDACc survival under serum/Gln-free conditions when CAFs were pretreated with AKTi (∼70%) versus p38i (∼30%) compared with DMSO-pretreated CAFs (Fig. 4B). Accordingly, AKTi reduced Glu/Gln production in CAFs (∼30%), whereas p38i inhibited only Gln production (Fig. 4C). Together, these findings suggest that AKT regulated the metabolic parameters of CAFs to a greater degree than p38. On the other hand, p38i CAFs displayed a significant decrease in the secretion of more cytokines than AKTi, with a reduction in GM-CSF, IL6, and TGFβ levels, and a maintenance of IL15 (akin to NetG1-deficient CAFs), compared with DMSO-treated CAFs (Fig. 4D). In contrast, AKTi CAFs had increased GM-CSF production compared with DMSO-treated CAFs, with decreases in IL8, TGFβ, and IL15 production. Based on these profiles, we hypothesized that p38i CAFs would be less immunosuppressive than AKTi CAFs. To test this, we performed the NK killing assay and saw that indeed p38i CAFs lost their immunosuppressive capacity compared with DMSO CAFs, whereas AKTi CAFs did not (Fig. 4E), thus confirming our hypothesis.

To further dissect the p38 and AKT signaling pathways in CON and NetG1 KO CAFs, we tested the expression of numerous known downstream factors of these two pathways. We found substantial modulations of both pathways in NetG1 KO CAFs compared with CON CAFs, with increased levels of FosB and cFos and reduced levels of c-Jun, JunB, and FRA1 downstream to p38, and reduced protein levels of pFOXO1, pFOXO3, pGSK3β, and p4E-BP1 downstream to AKT (Supplementary Fig. S16A and S16B; Fig. 4F). To complement our loss-of-function approaches, we decided to further explore the modulation of the two most downregulated proteins observed in each pathway, FRA1 and 4E-BP1, due to their previous association with inflammation or fibrosis (62, 63). The downregulation of FRA1 and 4E-BP1 was also confirmed in CAF2 NetG1 KDs (Supplementary Fig. S11A). Next, we measured the expression of VGlut1, NetG1, GS, as well as FRA1 and 4E-BP1 in p38i and AKTi CAFs to observe how the newly identified signaling circuit changes in response to the kinase inhibitors (Fig. 4G). In accordance with Glu/Glu production, p38i CAFs had reduced protein levels of NetG1 and GS, but not VGlut1, whereas AKTi CAFs exhibited decreases in all three proteins. In addition, FRA1 and 4E-BP1 were only downregulated in response to the corresponding pathway inhibitor (p38 or AKT, respectively). Overall, the data pointed toward a regulation of p38 and AKT downstream to NetG1, with a greater role for p38 in the immunosuppressive functions of CAFs, whereas AKT regulated metabolic parameters to a greater extent.

FRA1 and 4E-BP1 Partially Regulate Metabolic and Immunosuppressive Characteristics of CAFs

Because there was a large downregulation of FRA1 and 4E-BP1 in NetG1 KO CAFs, we decided to explore the functional roles of these proteins by knocking them down in CAFs and observing how they fit into the NetG1 signaling circuit. We obtained one effective and one ineffective KD of FRA1, as well as two effective KDs for 4E-BP1 (Fig. 5A). Intriguingly, the effects of FRA1 and 4E-BP1 KDs in CAFs on the signaling circuit were quite different; FRA1 KD in CAFs resulted in a loss in expression of VGlut1, NetG1, and GS, whereas 4E-BP1 KD CAFs upregulated these three proteins. In terms of the upstream kinases, pAKT remained unchanged in both KDs, whereas p-p38 increased in FRA1 KD CAFs and remained largely unchanged in 4E-BP1 KD CAFs. Strikingly, although loss of 4E-BP1 in CAFs had no effect on FRA1 expression levels, KD of FRA1 resulted in a downregulation of p4E-BP1 in CAFs. These results illustrated a cross-talk between the p38/FRA1 and AKT/4E-BP1 pathways, as well as a potential compensatory feedback loop that upregulated factors upstream in the circuit. Importantly, both FRA1 and 4E-BP1 KD CAFs failed to support PDACc and PANC-1 survival under metabolic stress compared with CON CAFs (Fig. 5B). Consequently, Gln levels were lower in the CM of both FRA1 and 4E-BP1 KD CAFs compared with CON CAFs, and Glu was lower in 4E-BP1 KD CAFs (Fig. 5C), which could account for the decreased PDAC cell survival observed in Fig. 5B. On the other hand, when measuring cytokine secretion, FRA1 KD CAFs displayed markedly reduced cytokine levels compared with CON CAFs, and 4E-BP1 KDs had increased cytokine levels, which were in line with the results seen in p38i and AKTi CAFs (Fig. 5D). Therefore, we hypothesized that the FRA1 KD CAFs would allow NK cell–mediated killing, whereas 4E-BP1 KD CAFs would retain their immunosuppressive capacity. As expected, in the NK killing assay, PDACc and PANC-1 cells cocultured with FRA1 KD CAFs had significantly reduced survival compared with CON CAFs (Fig. 5E). In contrast, more PDACc and PANC-1 cells survived when cocultured with 4E-BP1 KDs compared with FRA1 KD CAFs, at levels similar to CON CAFs (Fig. 5E). Collectively, these results suggest that FRA1 and 4E-BP1 contribute to the metabolic support of PDAC cells by CAFs, whereas FRA1 more effectively mediates the immunosuppressive function of CAFs. In addition, FRA1 can modulate the 4E-BP1 side of the signaling circuit, illustrating a cross-talk between the downstream effectors of the p38 and AKT pathways (Fig. 5F).

VGlut1 and GS Inhibition in CAFs Modulates Metabolic and Immunosuppressive Functions in CAFs, in a Manner Similar to NetG1 Loss

Because of the consistent joint modulation of VGlut1, GS, and NetG1 across nearly all of the genetically and pharmacologically inhibited CAFs, we evaluated if the absence of VGlut1 or GS in CAFs could phenocopy the effects of NetG1 KO in CAFs. First, we produced KDs for VGlut1 (VGlut1 KD) using CRISPRi (Fig. 6A). Interestingly, VGlut1 KD CAFs lost the expression of NetG1 and GS (Fig. 6B), suggesting a coregulatory mechanism for these proteins. Functionally, VGlut1 KD CAFs had a similar effect on PDACc survival as NetG1 KD CAFs, with PDACc survival reduced by approximately 50% compared with CON CAFs (Fig. 6C). Moreover, VGlut1 KD CAFs produced similar amounts of Glu and Gln, as was observed for NetG1 KD CAFs, both producing significantly less Glu/Gln than CON CAFs (Fig. 6D; Supplementary Table S8). In terms of their cytokine profile, VGlut1 KD CAFs produced significantly less GM-CSF, IL6, and TGFβ than CON CAFs, while maintaining IL15 levels (Fig. 6E), indicative of less immunosuppressive capacity. Indeed, in the NK killing assay, PDACc cells had approximately 30% to 40% decreased survival when cocultured with VGlut1 KD CAFs compared with CON CAFs (Fig. 6F). Related to downstream signaling, VGlut1 KD CAFs displayed reduced levels of pAKT, p-p38, FRA1, and p4E-BP1, phenocopying NetG1 KO CAFs (Fig. 6G). Lastly, we stained tissue of patients with PDAC for VGlut1 expression and detected high fibroblastic expression in 4 of 14 cases and low expression in 10 of 14 cases (Fig. 6H). In sum, VGlut1 is expressed in human CAFs and regulates expression of proteins and functions in CAFs involved in supporting PDAC cell survival, phenocopying NetG1-dependent functions (Fig. 5F).

To address the contribution of GS to the support provided by CAFs for PDACc survival, we genetically and pharmacologically inhibited GS in CAFs, using CRISPRi and increasing concentrations of methionine sulfoximine (MSO), respectively. GS was knocked down with high efficiency (Fig. 6I), and MSO effectively inhibited GS expression at concentrations greater than 0.1 mmol/L (Supplementary Fig. S17A). GS KD CAFs exhibited decreased levels of VGlut1, FRA1, and p4E-BP1, while maintaining expression of NetG1 and pAKT, and increasing p-p38 (Fig. 6I). Transient inhibition with MSO resulted in decreases of VGlut1 and NetG1, as well as downstream mediators p-p38 FRA1, and p4E-BP1, with pAKT increasing in expression (Supplementary Fig. S17B). We attribute the differences in signaling molecules to the nature of the treatment (transient vs. permanent), which may have resulted in compensatory upregulation of pAKT, for example. Nevertheless, blockade of GS led to an overall downregulation of numerous proteins identified in the novel NetG1 signaling circuit. Therefore, due to the similar profile, we predicted that loss of GS would functionally mimic the loss of NetG1 and VGlut1 in CAFs and reduce PDAC survival in our metabolic stress assay and when challenged with NK cells. Accordingly, PDAC cells that were cocultured with either GS KD– or MSO-pretreated CAFs under serum/Gln-free conditions survived significantly less than when cultured with CON CAFs, similar to coculture with NetG1 KO CAFs (Fig. 6J; Supplementary Fig. S17C). This was likely due to the decreased levels of Glu/Gln detected in the CM of GS KD CAFs (Fig. 6K), comparable to other inhibited CAFs in the NetG1 signaling hub that failed to support PDAC survival. The immunosuppressive activity was determined in both GS KD– and MSO-treated CAFs by measuring cytokine secretion and protection of PDAC cells from NK cell–induced death. Both GS KD– and MSO-pretreated CAFs showed a decrease in secretion of most cytokines tested (Fig. 6L; Supplementary Fig. S17D), and this resulted in a corresponding decrease in PDAC cell survival when cocultured with GS KD– or MSO-treated CAFs, similar to coculture with NetG1 KO CAFs (Fig. 6M; Supplementary Fig. S17E). Finally, we performed IHC on patient PDAC tissue to detect fibroblastic expression of GS and found that 7 of 15 patients expressed high levels of GS in fibroblasts, and 8 of 15 had low expression (Fig. 6N). Collectively, these data suggested that NetG1, VGlut1, and GS may be all linked in a new signaling circuit in CAFs, driving the metabolic and immunosuppressive functions of CAFs (Fig. 5F).

Therapeutic Inhibition of NetG1 Ablates the Protumor Functions of CAFs, Stunting Tumorigenesis In Vivo

Having established the roles that NetG1 played in driving protumor CAF functions, we next explored the therapeutic potential of targeting NetG1 in vitro and in vivo. We identified a neutralizing mAb against NetG1, by testing its ability to modulate NetG1-mediated signaling, Glu/Gln/cytokine production, and functions in CAFs in vitro. We measured the changes in expression of the proteins in the signaling hub by dose and time responses. Anti-NetG1 mAb began to decrease expression of the majority of the signaling circuit at a concentration of 1.25 μg/mL, with maximal efficacy at 10 μg/mL (Fig. 7A). We proceeded with the minimal dose that inhibited all proteins (2.5 μg/mL) and explored the changes in protein expression over time. Interestingly, most proteins lost substantial expression by 24 hours of treatment, with only GS having a transient increase in expression between 30 and 60 minutes of treatment (Fig. 7B). pAKT/p4E-EP1 remained downregulated after 60 minutes of mAb treatment, whereas p-p38 levels rapidly decreased starting at 5 minutes, and rebounded by 120 minutes of treatment. These results largely confirm the putative signaling circuit uncovered in CAFs that we observed through the systematic inhibition of each individual component. Functionally, anti–NetG1 mAb-pretreated CAFs cocultured with PDACc or PANC-1 cells resulted in an approximately 25% to 40% reduction in PDAC cell survival under metabolic stress compared with IgG-pretreated CAFs, similar to coculture with NetG1 KO CAFs (Fig. 7C), owed likely to the reduced Glu/Gln production in mAb-treated CAFs (Fig. 7D). In addition, anti-NetG1 mAb-treated CAFs produced less protumor cytokines compared with IgG-treated CAFs, showing a dose-dependent decrease in IL6, IL8, and TGFβ, suggesting less immunosuppressive capacity (Fig. 7E). Thus, inhibition of NetG1 by mAb was extremely effective at modulating all protumor signaling and functions of CAFs, serving as the impetus to move the treatment in vivo.

Next, we decided to test the efficacy of mAb in vivo, relying upon our previously performed syngeneic orthotopic injection model of PDAC in mice. Mice were injected with KPC3 PDAC cells and allowed to recover for 3 days, and were randomized into two groups, one receiving 5 mg/kg isotype control IgG and the other receiving 5 mg/kg anti-NetG1 mAb, 3 times a week for the duration of the experiment (22 days; Fig. 7F). Excitingly, anti-NetG1 mAb treatment was effective at limiting tumorigenesis, evident in the representative images of the pancreata isolated from IgG- or mAb-treated animals (Fig. 7G). Mice treated with anti-NetG1 mAb had lower tumor weight, a trend toward reduced pancreas weight, diminished tumor area percentage, and greater necrosis, all indicative of the efficacy of the therapy (Fig. 7HJ; Supplementary Fig. S18A). Importantly, there was greater infiltration of NK cells into the tumors of mAb-treated mice compared with IgG (Fig. 7K; Supplementary Fig. S18B), and fibroblasts in tumor tissue expressed equal levels of IL15 (Supplementary Fig. S18C), evident from the IHC staining (Supplementary Fig. S18D). Moreover, analysis of the tumor culture of isolated pieces of the pancreas from each group demonstrated a significant decrease in Glu, GM-CSF, and IL6 protein levels, with a trend toward less Gln, after mAb treatment, supporting our in vitro findings (Fig. 7L and M). Overall, in an extremely aggressive orthotopic model, anti-NetG1 mAb treatment was successful at limiting PDAC tumorigenesis, as well as at reducing key CAF-generated metabolites and cytokines, while increasing NK-cell infiltrates, highlighting NetG1 as an important druggable target in PDAC.

The present study has uncovered a previously unknown target for the potential treatment of pancreatic cancer: the neural glutamatergic synaptic stabilizing protein NetG1. NetG1 (also known as Laminet-1 and not to be confused with Netrin-1) is a unique member of the netrin family of proteins, because it is anchored to the plasma membrane through a glycosylphosphatidylinositol linkage and is not secreted like the majority of netrin family members (34). NetG1 is a synapse-stabilizing protein, located presynaptically on neurons, and its only known receptor is NGL1, localized on postsynaptic neurons (35, 36) and on microglial cells where it plays a role in neuronal survival (64). Dysregulation of NetG1 has been associated with Rett syndrome, schizophrenia, bipolar disorder, and anxiety (65–67). In cancer, NetG1 displays a low mutation rate (68), and NetG1 mRNA expression, methylation, or mutational status has been linked to colon and uterine cancers (68–70). Nevertheless, no studies have ascribed a functional role to NetG1 in the context of cancer.

PDAC is the most common form of pancreatic cancer and has a dismal prognosis (2). CAFs are a major cell type present in the PDAC microenvironment, whose role is incompletely understood. CAFs regulate at least three critical hallmarks of PDAC: (i) the deposition of desmoplastic ECM; (ii) overcoming nutrient deprivation; and (iii) vast immunosuppression of antitumor immune cells (3, 25). This study aimed to characterize the functions of NetG1 in PDAC, with an emphasis on the contribution of the desmoplastic stroma. Our results confirm that all three hallmarks are interconnected and that NetG1 expression in CAFs is a central driver of all of them in the PDAC microenvironment.

The PDAC microenvironment is nutrient-deprived due to the collapse of blood vessels enabled by desmoplasia, and this drives PDAC cells to use an alternative nutritional supply (16). Recent studies have demonstrated that CAFs supply PDAC cells with important metabolites, through exosomes (56) and amino acid secretion (55). Moreover, the ECM has been shown to be utilized as fuel for cancer cells (54), which potentially positions the scaffolding produced by CAFs in the PDAC microenvironment as a vital source of energy for PDAC cells. Here, we found that CAFs lacking NetG1 were less supportive of PDAC cell survival in direct coculture with their CM, or ECM (Fig. 2). These findings prompted us to perform a screen comparing amino acid production and secretion and revealed a NetG1-dependent increase in extracellular Glu/Gln in CAFs.

Interestingly, it has long been recognized that patients with pancreatic cancer have elevated serum levels of Glu (71), with a corresponding depletion of serum Gln levels (72). PDAC cells have a high capacity to utilize Glu, by catabolizing Gln through the enzyme GLS, in a process known as glutaminolysis (21). Although it is known that inhibition of GLS can decrease cancer cell growth in vitro and in vivo in a number of cancer models (19, 59), a recent study using a potent small-molecule inhibitor of GLS displayed little efficacy in preclinical mouse models, in part through upregulation of compensatory Gln-procuring pathways in PDAC cells (73). In addition, it has also been demonstrated that the conditions used in cell culture provide an overabundance of nutrients that does not accurately mimic the tumor microenvironment in vivo, and culturing cells in a medium with nutrients at levels more representative of tumors in vivo resulted in tumor cells with a decreased dependence on Gln, due to increased utilization of cystine (74). Intriguingly, an elegant recent study suggested that in mouse interstitial fluid of PDAC tumors, Gln levels were surprisingly higher than previously thought (75). It is tempting to speculate that CAFs may be responsible for the maintenance of Gln levels, by utilizing the high levels of Glu produced and secreted by tumor cells. On the other hand, it has been recently shown that PDAC cells can upregulate expression of GS, and its tumor-specific inhibition stunts tumorigenesis (76). In this study, the PDAC cells used did not acquire significant expression of GS, whereas CAFs expressed abundant GS, evident by their direct comparison (Supplementary Fig. S11B). Thus, our findings that CAFs can assist PDAC cells in overcoming nutritional stress by supplying key metabolites, such as Glu/Gln, add to the ways in which cancer cells in PDAC can acquire Gln from the tumor microenvironment.

Our results demonstrated that ablation of NetG1 in CAFs reduced GS (catalyzing conversion of Glu to Gln), but not GLS protein levels (Fig. 2). Indeed, GS is upregulated in CAFs in ovarian cancer compared with their normal counterparts, and inhibition of stromal GS resulted in decreased tumor development (60). In addition, we found that deletion of NGL1 from PDAC cells resulted in a decrease in cancer cell GLS expression (Supplementary Fig. S11), signifying an even greater dependence of these cells on the microenvironment, evident by the lack of rescue during Glu/Gln addition in NGL1-deficient PDAC cells (Fig. 2). An additional interpretation of the data could be that CAF-driven metabolism of Glu/Gln is required to first process these amino acids into other metabolites that are then used by PDAC cells. Analysis of TCGA data demonstrated that patients expressing the highest NGL1 levels have enriched pathway signatures associated with increased Glu receptor binding genes (Supplementary Fig. S12), suggesting a link between NGL1 and Glu uptake. Indeed, there are several studies demonstrating the importance of amino acid transporters in pancreatic cancer by which PDAC cells can overcome Gln and nutrient deprivation, such as Glu/cystine exchanger xCT (74, 77), as well as the neutral amino acid transporters SLC6A14 (78) and SLC38A2 (79). Pharmacologic or genetic inhibition of any of the aforementioned transporters leads to reduced tumor burden in mice, as xCT is required to overcome Gln deprivation through the utilization of cystine, SLC6A14 for the bulk uptake of amino acids, and SLC38A2 for maintenance of alanine levels, specifically in PDAC cells. Further analysis into the expression and activity of receptors that mediate uptake of extracellular amino acids in PDAC, such as Glu/Gln, as well as tricarboxylic acid cycle metabolites produced by CAFs, will be necessary to gain more insight into the general uptake ability of PDAC cells.

Macropinocytosis is a major nonspecific mechanism by which cancer cells harboring oncogenic KRAS mutations uptake bulk extracellular material (such as amino acids) to overcome nutrient stress (80), and was recently demonstrated to be mediated through sydecan-1 in a large panel of established and primary PDAC cell lines (81). We observed uptake of CAF-derived GFP by NGL1+ PDAC cells in direct coculture with CAFs or CAF CM, and direct transfer of GFP was mediated through fibroblastic NetG1/tumoral NGL1, whereas only tumoral NGL1 was required for the uptake of extracellular GFP (Supplementary Fig. S9). Moreover, we found that NGL1 regulated tumor-intrinsic uptake of extracellular material, largely through macropinocytosis (Supplementary Fig. S9). Combined with the survival assays (Fig. 2; Supplementary Fig. S10), we believe that although tumor-intrinsic bulk uptake of material is an essential PDAC cell survival mechanism, we would also argue that the quality of the content also plays a role in PDAC cell survival, as direct coculture with NetG1-deficient CAFs or their CM resulted in a significant decrease in survival for NGL1+ PDAC cells, despite the maintenance of their bulk uptake ability. Collectively, the data support both PDAC cell-autonomous and reciprocal mechanisms to survive nutritional stress.

We also observed a dramatically reduced expression of VGlut1 upon NetG1 loss in CAFs (Fig. 2). Interestingly, VGlut1 is present in glutamatergic synapses in neurons, where it loads Glu into presynaptic vesicles (61), and VGlut1 expression in neurons was shown to be dependent on NetG1 expression (35). These studies agree with our data in CAFs, where the KO of NetG1 resulted in downregulation of VGlut1 expression, and provide a potential mechanistic explanation for the decreased levels of Glu in NetG1 KO CAF CM. Interestingly, VGlut1, 2, and 3 have been implicated in pancreatic neuroendocrine tumors, and Glu signaling in these tumor cells drives tumorigenesis (82). Glu is a key neurotransmitter, but to maintain homeostasis and avoid excitotoxic stress, excess Glu-loaded vesicles are then taken up by neighboring astrocytes, which express high levels of GS. Glu is then recycled into Gln, and subsequently secreted back to neurons, where the cycle can continue (83). Our results suggest that NetG1+ CAFs may serve both as “hyperexcited presynaptic neurons” and “astrocyte-like cells,” thus providing a constant supply of microenvironmental Gln and Glu, fueling tumorigenesis.

Another hallmark of PDAC tumors is an extremely immunosuppressive microenvironment. Numerous studies have confirmed a major role for CAFs in generating factors, such as secreted cytokines and chemokines (26, 27), that contribute to this environment, as well as physically sequestering antitumor immune cells (28). Furthermore, compared with many other tumor types, PDAC has one of the lowest mutational burdens and display of neoantigens (84, 85), which correlates directly with low response rates to immune checkpoint inhibitors, such as PD-1 (86). Collectively, these findings suggest that even if cytotoxic T cells could penetrate the tumor and CAF-driven immunosuppression was inhibited, the lack of tumor antigens would render them unable to kill PDAC cells effectively. However, the innate cytotoxic immune cell population—NK cells—can destroy PDAC cells independent of neoantigen presentation, and thus represents an attractive antitumor cell type for PDAC therapy (87). In addition, NK cells play a natural role in the pancreas, which is to remove activated stromal cells during acute fibrosis in wound healing at the end of a process known as nemosis (88). Importantly, they are unable to accomplish this in the setting of PDAC, due to the overwhelming numbers of CAFs and immunosuppressive signals, resulting in a situation akin to unresolved chronic wound healing (88, 89). Thus, reverting the immunosuppressive phenotype of CAFs could allow for NK cell–stimulating therapies (e.g., IL15 supplementation; ref. 90) to more efficiently clear protumoral CAFs as well as PDAC cells. In the present work, we confirmed that CAFs produce the immunosuppressive cytokines TGFβ and GM-CSF, which can expand immunosuppressive myeloid cells (91, 92), as well as chemokines known to attract immunosuppressive immune cells (IL8 and CCL20; Fig. 3). GM-CSF in particular has been demonstrated to play an important role in PDAC tumorigenesis, where mesenchymal cells or tumor cells secrete and respond to this protumor cytokine, and its inhibition results in decreased tumor burden (93–95), in agreement with our in vivo inhibition of NetG1, where less GM-CSF was produced in mAb-treated tumor cultures (Fig. 7). Intriguingly, we showed that CAFs, compared with TAs, upregulate IL15, a potent stimulator of NK-cell activity. Functionally, however, IL15 activity is likely overwhelmed by the greater number of immunosuppressive factors secreted by CAFs, especially TGFβ, as CAFs significantly inhibited NK-cell activation and function. Our results support this hypothesis, as CAFs lacking NetG1 failed to protect PDAC cells from NK cell–induced death, and that IL15 was responsible for activation of NK cells. Accordingly, IL15 expression was detected in fibroblasts within tumor tissue treated with mAb, matching our in vitro results. Moreover, NK-cell infiltration increased into murine PDAC tumors treated with mAb compared with IgG control, likely contributing to the observed increase in necrosis. Collectively, the data demonstrate that NetG1 in CAFs shapes the immunosuppressive microenvironment in PDAC, and its ablation reprograms CAFs to an antitumor phenotype allowing NK cells to eliminate cancer cells.

In recent years, there has been growing interest in linking microenvironmental metabolites with immune cell function (30), and CAFs in the PDAC microenvironment are important players in this connection (21). Thus, we postulated that CAF metabolism, driven by NetG1 and, by extension, VGlut1 and GS could also be partially responsible for the immunosuppressive phenotype of CAFs. Accordingly, our results demonstrate a link between the metabolic circuit NetG1–VGlut1–GS and suppression of antitumor immunity. Taking lessons from the brain and synaptic biology, it is known that cytokines, such as TNFα, IFNγ, IL6, TGFβ, and IL1β, regulate glutamatergic synaptic activity and subsequently Glu release and uptake by neurons and astrocytes (96–98). Our data suggest that a signaling mechanism may exist in CAFs, whereby disruption of synaptic proteins like NetG1 and VGlut1 results in modulation of cytokine secretion from CAFs, and Glu/Gln cycling may also play a role, as GS inhibition also affected cytokine secretion in CAFs. In sum, our findings illustrate that simply modulating NetG1 expression in CAFs can simultaneously “disarm” two major protumorigenic features of CAFs: metabolic support of PDAC cells and protection from immune cell killing (Fig. 5F). These data suggest that synaptic biology in CAFs might be a missing link that controls metabolic and immunosuppressive programs in pancreatic cancer. Indeed, PDAC has been found to have a general dysregulation of axonal guidance genes (99), and neoneurogenesis and perineural invasion have been shown to play roles in promoting tumorigenesis (100). Thus, PDAC could have a general neural reprograming that is hijacked for the benefit of the tumor.

Mechanistically, little is known about NetG1-mediated signaling. Thus far, studies have shown that NetG1 physically interacts with its receptor, NGL1, in trans on postsynaptic neurons (36) and microglia (64) and in cis to leukocyte antigen receptor, a tyrosine protein phosphatase (35), on presynaptic neurons. In addition, NetG1 has also been shown to modulate expression levels of the potassium channel KATP (101), but no downstream signaling pathways have been associated with any of these findings. Our results revealed that NetG1 controls the activity of the kinases p38 and AKT, as well as known mediators of these pathways (Figs. 4 and 5). In particular, FRA1 and 4E-BP1 were drastically downregulated, and collectively these pathways could largely account for the protumor functions in CAFs regulated by NetG1. Previously, p38 MAP kinase activity in CAFs has been shown to promote breast and KRAS-driven lung carcinogenesis by enhancing the secretion of protumor factors (akin to senescence-associated secretory proteins) and activating CAFs to produce hyaluronan networks (102, 103). In addition, AKT activity in CAFs, mediated through the actin-binding protein Girdin, has been shown to enhance lung tumorigenesis (104). Downstream of p38 is the AP1 transcription factor complex, which is a dimer that can be comprised of members of the FOS, JUN, ATF, and MAF families, with FRA1 belonging to the FOS family (105). AP1 has long been associated with cellular responses to stress (activated by MAP kinases such as p38) and inflammatory signals (106). FRA1 has been linked to activated CAF methylation signatures in PDAC (107), as well as in certain CAF subtypes in breast cancer, as determined by single-cell RNA-seq (108). FRA1 is upregulated in a variety of tumors, including pancreatic cancer (105). Our results demonstrate that FRA1 regulated cytokine production in CAFs and subsequently NK cell–induced death of PDAC cells, supporting the notion that FRA1 acts as an AP1 subunit. We also found that 4E-BP1, a critical regulator of cap-dependent translation (109), was highly expressed in CAFs and lost upon ablation of NetG1 (Fig. 5). Previous work has shown that 4E-BP1 plays a major role in fibrogenesis (62), and a protein synthesis inhibitor that blocks mTOR/4E-BP1 in pancreatic CAFs reverts chemotherapeutic resistance in mice, likely by blocking secretory factors from CAFs (110). This supports our findings that CAFs lacking 4E-BP1 display a less tumorigenic profile, with the downregulation of Glu/Gln and cytokines. Of note, the identification of a neutralizing antibody allowed us to transiently block NetG1, and we were able to further confirm the signaling network downstream to NetG1 (Fig. 7). This positions NetG1 as an attractive target, as the blockade of a single protein inhibits multiple pathways known to sustain protumor functions in CAFs. This was further confirmed by the antitumor effects seen in vivo, when genetically targeting NetG1 in CAFs (Fig. 2) or blocking it with a neutralizing antibody during tumor development (Fig. 7). The developing NetG1 signaling circuit is presented in Fig. 5F.

Overall, we have identified a new potential desmoplastic target, NetG1, for the treatment of pancreatic cancer, for which there are no effective therapeutic interventions. By targeting NetG1 in CAFs, we can revert tumor-promoting CAFs back into tumor-restrictive CAFs, ultimately limiting tumorigenesis in vivo. Finally, NetG1 was established as a viable therapeutic target in preclinical murine models. It is our hope that this study provides strong rationale to consider effects of the stroma on cancer development and progression, and that specific normalization of the stroma is a reliable therapeutic option.

Ethical Collection of Human Samples

The authors declare that they obtained written informed consent from patients who donated samples for research purposes. The studies conducted herein were in accordance with recognized ethical guidelines. FCCC, through the Bio Sample Repository Facility, adheres to the International Society for Biological and Environmental Repositories and National Cancer Institute Best Practices for Biospecimen Resources and participates in the NCI Office of Biorepositories and Biospecimen Resources and Biospecimen Research Network. Further, human sample collections were approved by FCCC's and MDA's Institutional Review Boards.

Reproducibility: Key Resources Table

All information (manufacturer/source, RRID, website) regarding key antibodies, mouse models, chemical and biological reagents, software, and instruments has been compiled into a master “Key Resources” Table (see supplementary material).

Cell Culture

The PDAC cells used for the majority of the assays, termed “PDACc” in this study (39), were maintained in media containing 1 part M3 Base F (INCELL) to 4 parts DMEM low glucose (1 g/mL) + sodium pyruvate (110 mg/mL) + 1% penicillin–streptomycin (P/S; 10,000 U/mL; Thermo Fisher Scientific), supplemented with 5% FBS (Atlanta Biologicals). PANC-1 cells were maintained in DMEM supplemented with 5% FBS and 1% P/S (10,000 U/mL). Patient-matched TAs and CAFs (38, 111) were maintained in DMEM containing 15% FBS, 1% P/S, and 4 mmol/L Gln. For individual experiments, cells were cultured as described in the figure legends or Methods, either with the maintenance conditions outlined above, or in DMEM without serum (serum free), or DMEM without serum and without Gln (–). NK-92 cells were maintained in Alpha Minimum Essential medium without ribonucleosides and deoxyribonucleosides supplemented with the following components: 2 mmol/L l-glutamine, 1.5 g/L sodium bicarbonate, 0.2 mmol/L inositol, 0.1 mmol/L 2-mercaptoethanol, 0.02 mmol/L folic acid, 400 units/mL IL2, 12.5% horse serum, and 12.5% FBS. All cell lines used in experiments were within 7 to 15 passages after being thawed. Cell line authentication was performed by IDEXX BioResearch, and genetic analysis was performed to determine (i) species of origin and (ii) genetic profile–based 16 different allelic markers (AMEL, CSF1PO, D13S317, D16S539, D18S51, D21S11, D3S1358, D5S818, D7S820, D8S1179, FGA, Penta_D, Penta_E, TH01, and vWA). All cell lines were regularly tested for Mycoplasma, as determined by PCR detection methods, and all lines tested negative. Date of last test: July 20, 2020.

Isolation and Immortalization of Patient-Derived Fibroblasts

Fresh patient pancreatic tissue collected after surgery was digested enzymatically, using our established protocol (38, 111). Briefly, tissue was initially stored in PBS with antibiotics (P/S, Fungizone) Next, tissue was physically minced with a sterile scalpel and was digested overnight in collagenase (2 mg/mL). After a series of filtration and centrifugation steps, cells were plated on gelatin-coated plates for subsequent expansion. Cells were characterized as done previously (37, 38), and fibroblasts were confirmed by a lack of cytokeratins, expression of vimentin, cell shape, ability to create substantive ECM, and the presence of lipid droplets (TAs) or acquisition of lipid droplets after TGFβ signaling inhibition (CAFs). For the immortalization, cells were retrovirally transduced as previously described (38, 111), with the pBABE-neo-hTERT vector, which was a gift from Dr. Robert Weinberg (Addgene plasmid # 1774). In this study, the procedure above generated the following lines: TA and CAFs (CAF1 and CAF2), which were subsequently used in the 3-D system (outlined below) or in vivo (CAF1 CON and NetG1 KO).

Generation of Patient-Derived PDAC Cell Lines from PDX

Procedure was performed with an approved protocol from FCCC's Institutional Review Board and was done as previously described (112). Briefly, tumor fragments were obtained after surgery and washed in RPMI media. The washed fragments were then resuspended in 1:1 mixture of RPMI and Matrigel on ice and were implanted s.c. into 5- to 8-week-old C-B17.scid mice (Taconic Bioscience). Once tumors developed to a size of at least 150 mm3, but no larger than 1,500 mm3, mice were sacrificed, and the tumors were collected. These tumors were then digested with a solution containing collagenase/hyaluronidase and dispase (StemCell Technologies). After several washes, dissociated PDX cells were plated in DMEM supplemented with 10% FBS, 1% P/S, and 2 mmol/L Gln. All of the following reagents were from Thermo Fisher Scientific and supplemented into the media: F12 nutrient mix, 5 μg/mL insulin, 25 μg/mL hydrocortisone, 125 ng/mL EGF, 250 μg/mL fungizone, 10 μg/mL gentamycin, 10 mg/mL nystatin, 11.7 mmol/L cholera toxin, and 5 mmol/L Rho kinase inhibitor Y27632 (Sigma-Aldrich). After 1 month of passages, cells were lysed and subjected to Western blot analysis for the detection of NGL1 expression.

3-D Microenvironment Preparation (Cell-Derived ECM Production)

Production of cell-derived ECM (often referred to as CDM) was conducted as previously described (38, 111). Briefly, CON or NetG1 KO CAFs (1.25 × 105/mL) were seeded onto cross-linked [1% (v/v) glutaraldehyde and then 1 mol/L ethanolamine] 0.2% gelatin-coated wells of a 24-well or 6-well plate. These cells were grown in DMEM supplemented with 15% FBS, 1% P/S (10,000 U/mL), 4 mmol/L Gln, and 50 μg/mL l-ascorbic acid (Sigma-Aldrich), provided fresh daily. After 5 days of ECM production, cultures were used to collect secreted factors (described below for CM preparation), or cells were lysed and extracted from ECMs using 0.5% Triton X-100 (Sigma-Aldrich) supplemented with 20 mmol/L Ammonia Hydroxide (Sigma-Aldrich), followed by extensive washes with PBS. The resulting cell-free ECMs were used as the 3-D scaffold environment for subsequent functional assays.

CM Preparation

Fibroblast CM was prepared by growing cells during 3-D CDM production (see above) for 5 days and then replacing media with Gln/serum-free (−) media. Cells were allowed to condition the media for 2 days. The CM were then collected and used immediately or aliquoted and frozen at −80°C until use (for up to 1 month).

mRNA Microarray

RNA was isolated from two sets of patient-matched TAs and tumor-associated fibroblasts (CAFs) at the end of CDM production (see above), which were generated in two technical repetitions. The RNA was processed by the Genomics Facility at FCCC to render differentially expressed genes between the two sets of matching TAs and CAFs. Briefly, an Agilent Technologies Gene Chip (∼44,000 genes) was used, and RNA was hybridized to the chip and run through the Agilent Technologies Instrument (S/N: US22502680). Raw data were processed using the Agilent Feature Extraction 9.5.1.1 Software, outputting processed data into excel spreadsheets. Quantile normalization was applied to microarray data. We applied limma (113) to identify differentially expressed genes between TA and CAFs. Top differentially expressed genes were selected based on P value < 0.01 and depicted as a heat map (z-scores). In order to identify pathways that are enriched between these two cell types, we applied GSEA (preranked with default parameters) on significantly differentially expressed genes (P value < 0.05) with no fold-change cutoff. To depict the resulting enriched pathways (FDR < 0.25), we applied enrichment map method (114). The full data set, the top hits with P < 0.01, and GSEA list with FDR <0.25, and complete GSEA for positive and negative enrichment, are all available in Supplementary Table S1, organized by tabs. Data were also deposited into the Gene Expression Omnibus database (accession #GSE156962) as part of the superseries “To determine the differentially expressed genes in cancer associated fibroblasts upon ablation of the glutamatergic synapse protein NetrinG1” (accession #GSE156964).

Quantitative Reverse Transcriptase PCR

The Ambion PureLink Kit (Life Technologies) was used, following the manufacturer's instructions, to extract total RNA cells from the various experimental conditions and tested for RNA quality using a Bioanalyzer (Agilent). Contaminating genomic DNA was removed using Turbo DNA free from Ambion. RNA concentrations were determined with a spectrophotometer (NanoDrop; Thermo Fisher Scientific). RNA was reverse transcribed using Moloney murine leukemia virus reverse transcriptase (Ambion, Life Technologies) and a mixture of anchored oligo-dT and random decamers (Integrated DNA Technologies). Two reverse-transcription reactions were performed for each biological replicate using 100 and 25 ng of input RNA. Taqman assays were used in combination with Taqman Universal Master Mix and run on a 7900 HT sequence detection system (Applied Biosystems). Cycling conditions were as follows: 95°C, 15 minutes, followed by 40 (two-step) cycles (95°C, 15 seconds; 60°C, 60 seconds).

Cycle threshold (Ct) values were converted to quantities (in arbitrary units) using a standard curve (four points, 4-fold dilutions) established with a calibrator sample. Values were averaged per sample, and SDs were from a minimum of two independent PCRs. Polymerase (RNA) II (DNA directed) polypeptide F (POLR2F) was used as an internal control. Identification numbers of commercial assays (Life Technologies) and sequences of the NTNG1 primers (FW and RV) and probe for the POLR2F assay are as follows:

  • GGAAATGCAAGAAGAATTATCAGG (forward),

  • GTTGTCGCAGACATTCGTACC (reverse),

  • 6fam-CCCCATCATCATTCGCCGTTACC-bqh1 (probe)

For other genes, the same RNA extraction was performed, but reverse transcriptase reaction was performed using High Capacity cDNA RT Kit, and qRT-PCR reactions were performed in a barcoded 96-well plate using Power SYBR Green PCR master mix, gene-specific primers, and template DNA, using the StepOnePlus Real Time-PCR system. All reagents and instruments were from Applied Biosystems/Thermo Fisher Scientific. All primer sequences were obtained from PrimerBank (115) and are available in the Key Resources Table. Relative mRNA expression for each gene was calculated by the ΔΔCt method, with 18s used as a housekeeping control, and then normalized to CON CAFs to determine fold-change differences in gene expression.

SDS-PAGE and Western Blot Analysis

Cell lysates were prepared by adding 150 μL standard RIPA buffer to 6-well plates containing 106 cells. The lysates were homogenized using a sonicator. After centrifugation, the supernatant was collected and 30 μg of lysate (determined by the Bradford Assay) was diluted in 2x loading buffer (Bio-Rad) and was loaded onto 4% to 20% polyacrylamide gels (Bio-Rad). The gels were run at 70 volts for 2 hours and transferred to PVDF membranes (Millipore) using the semi-dry transfer system (Bio-Rad). Membranes were blocked in 5% nonfat milk in 0.1% Tween in tris-buffered saline (TBST) for 1 hour at room temperature (RT), and primary antibodies were incubated with the membranes overnight at 4°C. Next, membranes were washed 5 times with 0.1% TBST, and then secondary antibodies linked to horseradish peroxidase (HRP) were diluted in 5% nonfat milk in 0.1% TBST and were applied for 2 hours at RT. Membranes were washed again as before and incubated for 5 minutes in Immobilon Western Chemiluminescent HRP substrate (Millipore). Membranes were developed using film and scanned to create digital images. Primary and secondary antibodies are listed in the Reagents table.

Indirect Immunofluorescence

CAFs were allowed to generate ECM (3-D microenvironment) as outlined above (CDM production). After 5 days of ECM production, cells on coverslips were fixed and simultaneously permeabilized for 5 minutes at RT in 4% paraformaldehyde containing 0.5% TritonX-100 and 150 mmol/L sucrose and allowed to continue fixation, using the same preparation lacking triton, for an additional 20 minutes at RT. Fixed/permeabilized cells were then blocked for 1 hour at RT with Odyssey Blocking Buffer (PBS base; Li-Cor). Cells were stained for 1 hour at RT with the following antibodies diluted in blocking buffer: fibronectin (Sigma-Aldrich) and α-SMA (Sigma-Aldrich). After 1-hour incubation, cells were washed 3 times with 0.05% Tween in phosphate-buffered saline (PBST), and secondary antibodies diluted in blocking buffer were applied for 1 hour at RT (TRITC for α-SMA; Cy5 for fibronectin, Jackson ImmunoResearch). Next, cells were washed 3 times with 0.05% PBST, and nuclei were counterstained with SYBR green (1:10,000; Invitrogen). Stained coverslips were imaged using Spinning disk confocal microscopy (PerkinElmer), captured by a CoolSNAP CCD Camera (Photometrics) attached to an Eclipse 2000-S inverted microscope (Nikon). Images were exported from Volocity 3-D Image Analysis Software (PerkinElmer). The integrated intensity of α-SMA was quantified using Metamorph Image Analysis Software (Molecular Devices). Fibronectin alignment was quantified as described below.

ECM Fiber Orientation Analysis

ECM fiber orientation, indicative of alignment, was performed as previously described (38, 111). Briefly, confocal images of fibronectin in CAF-generated ECMs were analyzed for fiber orientation using the OrientationJ plugin (116) for ImageJ. Fiber angles generated by the plugin were normalized to fit within the angle range of −90° to +90°. Next, normalized angles were plotted in Excel to generate orientation curves (alignment output at every angle). Finally, the percentage of fibers between −15° and +15° was plotted as a measurement indicative of ECM fiber alignment. In general, TAs tend to have disorganized −15° to +15° angle fibers (<50%), whereas CAFs tend to have fibers with angles >50%.

PDAC Cell Proliferation Assay

CON or NGL1 KO PDACc cells (2 × 104) were seeded in CAF-derived 3-D ECMs and allowed to proliferate overnight. Ki-67, a nuclear marker of proliferation, and nuclei (DAPI) were stained by immunofluorescence (as outlined above). The number of Ki-67-positive cells/nuclei per field was calculated using Metamorph image analysis software.

CRISPR/Cas9-Mediated Knockout of NetG1 and NGL1

Knockout of NetG1 in CAFs, or NGL1 in PDACc or PANC-1 cells, was achieved by introducing a frameshift mutation in coding regions of each gene, induced by cuts made by CRISPR/Cas9 and gene-specific gRNAs introduced into cells by lentiviral transduction.

gRNA Design.

Using the MIT Optimized CRISPR Design website (http://crispr.mit.edu/), the first 200 bp of exon 5 for NetG1 (first common exon for all splice variants) and first 220 bp of the NGL1 gene were used to generate guide RNAs (gRNA). The top two hits for each gene were selected and designed to include BsmBI/Esp3I overhangs (bold, underline, italics), the specific gRNA sequence (plain text). G (italics, underline) was added to the 1 gRNAs (and corresponding C on the 2 gRNA) to optimize its expression for the human U6 promoter. The gRNA sequences for NetrinG1 were: 1.1, CACCGGCGTCCAGACCAAATGATCC; 1.2, AAACGGATCATTTGGTCTGGACGCC; 2.1, CACCGTGATGCGAGTACCCCTGAGC; and 2.2, AAACGCTCAGGGGTACTCGCATCAC. The gRNA sequences for NGL1 were: 1.1, CACCGAACCTGCGTGAGGTTCCGGA;1.2, AAACTCCGGAACCTCACGCAGGTTC; 2.1 CACCGTGCCATCCGGAACCTCACGC; and 2.2 AAACGCGTGAGGTTCCGGATGGCAC. Empty vector or nontargeting gRNA against eGFP were used as a control, and the eGFP gRNA sequences usedwere: 2.1, CACCGCATGTGATCGCGCTTCTCGT and 2.2, AAACACGAGAAGCGCGATCACATGC.

Generation of Lentiviral Vectors.

Designed oligos were ordered from Integrated DNA Technologies and cloned into the LentiCRISPR v2 vector, as previously published (a gift from Feng Zhang; Addgene plasmid # 52961; refs. 38, 117). Briefly, oligos were phosphorylated using T4 PNK (New England Biolabs) and annealed in a thermal cycler. Annealed oligos were cloned into a gel-purified (GeneJet Gel Extraction Kit; Thermo Fisher Scientific), dephosphorylated (Fast AP; Thermo Fisher Scientific), and digested vector (Fast Digest Esp3I; Thermo Fisher Scientific). Annealed oligos were ligated into the digested vector using quick ligase (New England Biolabs). The ligated vectors were transformed into Stbl3 strain of Escherichia coli (Thermo Fisher Scientific) and selected with 100 μg/mL ampicillin on LB/agar plates. The following day, single colonies of bacteria were screened by colony PCR using a forward primer for the U6 promoter (GAGGGCCTATTTCCCATGATT) and the reverse primer sequence (gRNA x.2) for each gene. Positive clones were subjected to sequencing and subsequently expanded for plasmid purification and lentivirus production.

Lentiviral Transduction.

To package the LentiCRISPRv2 plasmid into functional virus particles, 10 μg LentiCRISPRv2, 5 μg psPAX2 (psPAX2 was a gift from Didier Trono; Addgene plasmid # 12260), and 2 μg VSVg were mixed with 30 μL X-tremeGene9 transfection reagent (Sigma-Aldrich) in 1 mL of serum-free/antibiotic-free DMEM and held at RT for 30 minutes. The transfection mixture was added dropwise to 293T cells containing 5 mL serum-free/antibiotic-free DMEM, and the cells were incubated at 37°C overnight. The next day, the media were replaced with complete DMEM. On days 2 and 4 after transfection, the media with viral particles were collected and syringe filtered through a 0.45 μm filter (Millipore). The filtered viral supernatant was used immediately to transduce target cells (PDACc/PANC-1 or fibroblasts) with 10 μg/mL polybrene (Santa Cruz Biotechnology) or stored at −80°C for use in the future. Target cells were selected with puromycin (2 μg/mL for CAFs, 5 μg/mL PANC-1, and 12 μg/mL KRAS) for 2 weeks. Cells that survived selection were used for future experiments.

Generation of RFP+ and GFP+ CRISPR/Cas9-Transduced Cells.

eGFP and mCherry (RFP) were PCR amplified using Phusion HF polymerase master mix (Thermo Fisher Scientific) to add overhangs compatible with restriction enzymes. Then cloned into XbaI/XhoI (New England Biolabs) digested, Fast AP dephosphorylated pLV-CMV-H4-puro vector (kindly provided by Dr. Alexey Ivanov, West Virginia University School of Medicine, Morgantown, WV). Empty vector (CON) or NetG1 KO CAF were transduced with pLV-CMV-H4-puro-GFP and the CON or NGL1 KO PDACc were transduced with pLV-CMV-H4-puro-RFP, with 10 μg/mL polybrene. Cells were selected by flow cytometry. The selected cells were used for future experiments.

CRISPRi-Mediated Knockdown of Target Genes

We generated a lentiviral vector named CRISPRi-Puro (modified from pLV hU6-sgRNA hUbC-dCas9-KRAB-T2a-Puro; Addgene Plasmid #71236; a gift from Charles Gersbach; ref. 118) containing dead Cas9 (dCas9) fused to Krueppel-associated box (KRAB) domain, gRNA cloning site, and puromycin resistance. The gRNA cloning site was modified to contain a “stuffer” (cut from LentiCRISPRv2 plasmid using the restriction enzyme BsmBI and ligated into our vector) that resulted in a higher gRNA cloning efficiency. Upon insertion of gene-specific gRNA, dCas9-KRAB–mediated KD of target genes could be achieved. Three gRNA oligos were selected from the top ten gRNA sequences generated from the study “Compact and highly active next-generation libraries for CRISPR-mediated gene repression and activation” (119). Generation of the vector and transduction of the target cells was performed as described above for the generation and transduction of the LentiCRISPRv2 vector. The best two KD cell lines generated were subsequently used for all experiments. Cells transduced with vector alone were used as a control. The NetrinG1 CRISPRi gRNA sequences were: 1.1, CACCGGCGCCCCGAGGTCGTGGAG; 1.2, AAACCTCCACGACCTCGGGGCGCCC; 2.1, CACCGACAGCAACAGCGAGCGGGA; 2.2, AAAC TCCCGCTCGCTGTTGCTGTCC; 3.1, CACCGCGAGAGCCGGAAAGAGGAG; and 3.2, AAACCTCCTCTTTCCGGCTCTCGCC. The VGlut1 CRISPRi gRNA sequences were: 1.1, CACCGAGAGAGAGTGGAGCCGGGT; 1.2, AAACACCCGGCTCCACTCTCTCTCC; 2.1, CACC GGCTCCGCTCGGGGGGAAGG; 2.2, AAACCCTTCCCCCCGAGCG GAGCCC; 3.1, CACCGAGAGACCCAGAAGTGGTGG; and 3.2, AAACCCACCACTTCTGGGTCTCTCC. The GS CRISPRi selected gRNA sequences were: 1.1, CACCGGCTCTGCAGAGTCGAGAGT; 1.2,AAACACTCTCGACTCTGCAGAGCCC; 2.1, CACCGGAGCGTGTGAGCAGTACTG; 2.2, AAACCAGTACTGCTCACACGCTCCC; 3.1, CACCGGACGGGTCCAAGCCACCAG; and 3.2, AAACCTGGTGGCTTGGACCCGTCCC. For 4E-BP1, the selected gRNA sequences were: 1.1,CACCGCACAGGAGACCATGTCCGG; 1.2, AAACCCGGACATGGTCTCCTGTGC; 2.1, CACCGAGGGCAGCGAGAGGTTCGC; 2.2, AAACGCGAACCTCTCGCTGCCCTC; 3.1, CACCGACCGGGTGTCCAGGCTCAA; and 3.2, AAACTTGAGCCTGGACACCCGGTC. The FRA1 CRISPRi sequences were: 1.1, CACCGAAGTCTCGGAACATGCCCG; 1.2, AAACCGGGCATGTTCCGAGACTTC; 2.1, CACCGGGCGGCTCTGCGGGGTACA; 2.2, AAACTGTACCCCGCAGAGCCGCCC; 3.1, CACCGGGCATGTTCCGAGACTTCG; and 3.2, AAACCGAAGTCTCGGAACA TGCCC.

Cell Engagement/Motility Assays

RFP+ CON or NGL1 KO PDACc (5 × 104) were plated with GFP+ CON or NetG1 KO CAFs (5 × 104) in 3-D and allowed to attach for 2 hours. Following cell attachment, three regions per well, where there was obvious clustering of red PDACc and green CAFs, were selected to be imaged every 15 minutes for 24 hours, using a motorized XYZ stage (Optical Apparatus Co.) controlled by MetaMorph 6.3r7 (Molecular Devices) software. Time-lapse movies were built with images from each corresponding fluorescence channel acquired every 15 minutes for a period of 24 hours using either 4X or 20X Pan Fluor objectives mounted on a Nikon TE-2000U wide-field inverted microscope (Optical Apparatus Co.) with a Cool Snap HQ camera rendering three time-lapse movies per condition for a total of nine videos generated over three independent experimental repetitions. The resulting time-lapse videos were analyzed to quantify heterotypic cell engagement (yellow interacting areas using both channels simultaneously in merged file stacks) and PDACc cell motility (using only the monochromatic images corresponding to the red channel), as outlined below.

Cell Engagement.

For image analysis, MetaMorph offline 7.8.1.0 software was used. To achieve sequence coherence, out of focus or blank images from particular fluorescence channels were removed, together with their fluorescence counterpart images. All movies were normalized (trimmed) to 18-hour data for analysis. Images were scaled and pseudocolored (RFP in red, GFP in green) similarly for all conditions. After merging of fluorescence channels, cell engagement areas obtained from regions of fluorescence colocalization (in yellow) were digitally detected, isolated, and measured for each condition over time.

PDAC Cell Motility.

Monochromatic channel of RFP+ PDAC cells was tracked using the Track Object function in MetaMorph 7.8.1.0 offline software (Molecular Devices). A minimum of 5 cells per region were chosen to carefully follow. Red PDACc that were touching other red PDACc, moving out of the imaging plane, or that began to round up prior to mitosis were not counted in this experiment. Motility was analyzed according to the total distance (μm) traveled over the timespan of the experiment (24 hours), expressed as velocity (μm/hr).

AlamarBlue Assay

In order to determine the optimal (noncytotoxic) concentration of EIPA (TOCRIS Bioscience) to inhibit macropinocytosis in PDAC cells, 5 × 103 CON or NGL1 PDAC cells (PDACc or PANC-1) were plated in a 96-well plate and cultured in 100 μL PDAC media (M3 base, DMEM low glucose, 5% FBS, and P/S) overnight in order for cells to attach. The following day, cells were allowed to proliferate for 24 hours in the presence of increasing concentrations of EIPA (0–500 μmol/L). DMSO was used as a vehicle control. Next, 10 μL alamarBlue reagent (Bio-Rad) was added to the wells, and PDAC cells metabolized the alamarBlue reagent for 6 hours (reduction of alamarBlue reagent). For controls, one set of 6 wells contained only media with no cells or reagent (media blank), whereas another set of 6 wells contained only media and alamarBlue reagent (filter correction factor). The 96-well plate was measured for absorbance at 570 and 600 nm using the Spark Multimode Microplate Reader (Tecan). The reduction of alamarBlue reagent was calculated according to the following formula:

% reduction alamarBlue = [(A570 nm – media blank) – [(A600 nm – media blank) * (filter correction factor at 570 nm/filter correction factor at 600 nm)]] *100

The percent reduction of alamarBlue was normalized to the DMSO control and graphed to determine the highest concentration possible without affecting cell proliferation. This concentration was determined to be 25 μmol/L for all PDAC cells tested and was used in subsequent macropinocytosis assays.

Macropinocytosis Assay

Uptake of the Macropinocytosis Substrate FITC-Dextran.

The macropinocytic activity of PDAC cells was assessed as previously described (58), with minor modifications. To determine the baseline macropinocytosis capacity of PDAC cells cultured in 3-D, 2 × 104 RFP+ CON or NGL1 KO PDAC cells (PDACc or PANC-1) were plated onto coverslips containing CAF ECM and cultured for 24 hours in Gln-free/serum-free DMEM, treated either with DMSO control or with EIPA (25 μmol/L) to block macropinocytosis. The following day, lysine-fixable FITC-Dextran (D1845; Thermo Fisher Scientific) was added to the cells at a final concentration of 1 mg/mL, and the cells were incubated at 37°C for 30 minutes. Next, cells were placed on ice and washed 5 times with ice-cold PBS. Then cells were fixed in a 4% paraformaldehyde PBS solution for 30 minutes at RT. Cells were then washed 3 times with PBS, and nuclei were counterstained with a DAPI solution (1:10,000) for 15 minutes. Finally, samples were washed 3 times with PBS, and coverslips were mounted onto slides using the ProLong Gold Antifade Mountant (P10144; Thermo Fisher Scientific). Slides were then imaged, and macropinocytotic ability of cells was quantified as outlined below.

Imaging and Quantification.

A minimum of 30 confocal images (60x objective; 0.5 μm slices) per condition were captured by a Nikon A1 camera (Nikon) linked to a Nikon Eclipse Ti2-E Inverted Microscope Imaging System (Nikon) for each channel (performed in triplicate, 2 independent experiments; N = 6 coverslips), which included RFP+ PDAC cells (red channel), FITC-labeled dextran (green channel), and DAPI-labeled nuclei (blue channel). First, each channel of stacked images was merged by max intensity using FIJI/Image J plugin “Z project.” Next, MetaMorph 7.8.1.0 offline software (Molecular Devices) was used to quantify the amount of FITC-dextran taken up by RFP+ PDAC cells. RFP was used to generate a mask, and the amount of FITC under that mask was queried (integrated density) and recorded. The integrated density of FITC under RFP was normalized to the DMSO-treated CON PDAC cells (PDACc or PANC-1) for each experiment.

Material Transfer Assay

2 × 104 RFP+ CON or NGL1 KO PDACc and/or 2 × 104 GFP+ CON or NetG1 KO CAFs were seeded in 3-D and were cocultured for 36 hours, in serum-free DMEM. Spinning disk confocal images (PerkinElmer) were captured by a CoolSNAP CCD Camera (Photometrics) attached to an Eclipse 2000-S inverted microscope (Nikon). Images were exported from Volocity 3-D Image Analysis Software (PerkinElmer) and quantified for the integrated intensity of GFP+ pixels under RFP+ pixels, using Metamorph Image Analysis Software (Molecular Devices), which was indicative of CAF-derived GFP material transfer to RFP+ PDACc.

For the material transfer assays involving CAF CM and EIPA, 2 × 104 RFP+ PDACc (CON or NGL1 KO) were seeded in 3-D onto coverslips and cocultured for 24 hours with 50% GFP+ CON CAF CM in the presence of DMSO or EIPA (25 μmol/L). Next, coverslips were then washed 3 times with PBS and fixed in 4% paraformaldehyde PBS solution for 30 minutes at RT. Cells were then washed 3 times, and nuclei were labeled with DAPI (1:10,000) for 15 minutes at RT. Finally, cells were washed 3 times and mounted onto slides using ProLong Gold Antifade Mountant. Imaging and quantification were performed exactly as described above in the macropinocytosis assay, with the amount of GFP+ under RFP+ mask indicative of CAF-derived material transfer (GFP+) to PDACc cells (RFP+).

PDAC Survival Assay

2 × 104 RFP+ CON or NGL1 KO PDAC cells (PDACc or PANC-1) and/or 2 × 104 GFP+ CON or NetG1 KO CAFs were seeded in 3-D (CON CAF, NetG1 KO, or TA ECMs, as indicated) in 24-well plates. Cells were cocultured for 96 hours in Gln-free/serum-free DMEM. Next, at least 10 images were taken of red PDAC cells and green CAFs using a Cool Snap 1HQ camera (Roper Scientific) on an Eclipse 2000-U inverted microscope (Nikon). PDAC cell survival was quantified as the number of RFP+ per field of view, using Image J. Dead cells were excluded by Sytox Blue (1 μmol/L; Thermo Fisher Scientific) positivity or cell size. Briefly, identical thresholds were established for each color channel for each batch of images per experiment. Then the images were batch processed using a macro that measures cell counts and % area coverage of cells in each image, at a given threshold. The mean RFP+/Sytox Blue–negative cells (live) from each condition were calculated (counts or % area, depending on the experiment) and normalized to the CON PDAC/CON CAF condition. Cell counts were chosen when cells were clearly isolated (accurate cell counts), and percent area was chosen when cells had more overlap (could not accurately determine cell counts). The 96-hour timepoint was determined empirically, when cell death became evident by eye. Other PDAC survival assays were performed with CAF CM, VGlut1, GS, 4E-BP1, and FRA1 KD CAFs, or CAFs pretreated 24 hours with 1 mmol/L l-Methionine sulfoximine (GS inhibitor; Sigma-Aldrich), 10 nmol/L SB 202190 (P38 inhibitor; Tocris Bioscience), 10 μmol/L MK-2206 (AKT inhibitor; Cayman Chemical), or 2.5 μg/mL NetG1 mAb (Santa Cruz Biotechnology), as indicated. Note that all small-molecule inhibitors were used to pretreat CAFs overnight, and CAFs were washed 3 times with PBS before being cocultured with PDAC cells.

NK-92 Cell Killing Assay

2 × 104 RFP+ CON or NGL1 KO PDAC cells and/or 2 × 104 GFP+ CON or NetG1 KO CAFs were seeded in 3-D in 24-well plates and were cocultured in 500 μL NK-92 media without IL2 to establish their niche. The following day, wells were seeded with 8 × 104 resting or active NK-92 cells, in 500 μL of NK-92 media without IL2. Cells were cocultured for 48 hours. and PDAC survival was assessed identically to the PDAC cell survival assay. To produce active NK-92 cells, cells were maintained with 400 units/mL IL2, and for resting NK-92 cells, cells were maintained for 1 week in 100 units/mL IL2 and switched overnight to NK culture media containing no IL2 the day before the experiment. For 2-D conditions, all experimental conditions were the same, except the endpoint of the experiment was at 6 to 8 hours, as lack of CAF-derived ECM (3-D microenvironment) allowed NK-92 cells to kill at a faster rate. For the IL15-neutralizing experiments, an IL15-neutralizing antibody (10 μg/mL; R&D Systems) or nonspecific IgG isotype control (10 μg/mL; Jackson ImmunoResearch) was added to the cocultures where specified. Whenever CON or NetG1 KO CAF CM was used, it was diluted 1:1 in NK-92 media without IL2. Images were quantified for live PDAC cells as in the “PDAC Survival Assay” section above. Note that all CAFs that were pretreated with a small-molecule inhibitor were washed 3 times with PBS before subsequent coculture with PDAC and/or NK-92 cells.

NK-92 Cell Activation Assay

8 × 104 IL2-activated NK-92 cells were directly cocultured with CAFs or with CM derived from CAFs for 2 hours. After the coculture period, NK-92 activation status was determined by staining Granzyme B intracellularly or by using the IFNγ secretion assay detection kit. Stained cells were subjected to flow cytometry using by LSRII (Becton Dickinson) and analyzed with FlowJo 9.9 software (Tree Star). IL2-activated NK-92 cells cultured alone were used as the positive control. Percentage of cells that were positive for either Granzyme B or IFNγ was considered activated.

ELISA

ELISAs for IL15, IFNβ, TGFβ, IL6, IL8, GM-CSF, mTGFβ, mGM-CSF, and mIL6 (R&D Systems), as well as Glu (Abcam) and Gln (MyBioSource), were performed as recommended by the manufacturers. Briefly, buffers for the IL15, IFNβ, IL6, IL8, GM-CSF, mGM-CSF, and mIL6 ELISAs were prepared using the DuoSet ELISA Ancillary Reagent Kit 2, and buffers were prepared using the DuoSet ELISA Ancillary Reagent Kit 1 for the TGFβ and mTGFβ ELISAs (R&D Systems). Standards, CM, or lysates from various fibroblastic lines or murine tumor cultures were incubated for 2 hours on plates precoated overnight with capture antibody. For the TGFβ ELISAs, samples first underwent activation using the Sample Activation Kit 1 (R&D Systems) before sample incubation. Next, detection antibody linked to biotin was added to the wells and incubated for 2 hours. Streptavidin-HRP was then added to the wells and incubated for 20 minutes. Samples were rinsed 3 times with wash buffer in between all steps. Substrate was added for 20 minutes and was neutralized with stop buffer. Plates were read using Spark Multimode Microplate Reader (Tecan) at 450 nm to measure the specified protein present in sample. Plate imperfections, measured at 570 nm, were subtracted from the readings at 450 nm, to produce corrected readings. Finally, media alone (BLANK) was subtracted from the corrected readings for each sample. To determine the concentration of the indicated protein in each sample, a standard curve was generated using known concentrations of analyte, and the blank-corrected values determined for each sample were compared with the standard curve. The extrapolated values were normalized to cell number or total protein content.

U-Plex Assay

U-Plex Assay (multiplex ELISA) was carried out as recommended by the manufacturer (Meso Scale Discoveries). Briefly, TA and CON or NetG1 KO CAF CM samples were generated and added to a precoated plate containing 10 capture antibodies per well for the detection of the following proteins: IL1β, IL2, IFNγ, GM-CSF, IL6, IL8, IL10, IL12p70, MIP3α, and TNFα. After 3 washes, sample, standards, or calibrator were added to the wells and incubated for 1 hour. Next, after 3 washes, detection antibody was added to the wells and incubated for 1 hour. After a final set of 3 washes, read buffer was added to the wells and the plate was read using the MSD SECTOR Imager 2400 (Meso Scale Discoveries), and the data were processed using the MSD Workbench 4.0 software (Meso Scale Discoveries).

Amino Acid Screen: Biochrom Method

Media.

Twenty microliters of media was mixed with 80 μL of water and 100 μL of 10% sulfosalicylic acid (Sigma-Aldrich) followed by incubation for 1 hour at 4°C and then centrifugation at 13,000 rpm (15 minutes; 4°C). No visible pellet was seen. Supernatant was filtered using Ultrafree-MC-GV centrifugal filters (Millipore) and used for amino acid analysis employing Biochrom 30 amino acid analyzer. Results are in terms of μM and were normalized to total protein content.

Cell Lysates.

14.55 μg of cell lysate protein in RIPA buffer was brought to a total volume of 50 μL in water and was mixed with 50 μL of 10% sulfosalicylic acid (Sigma-Aldrich) followed by incubation for 1 hour at 4°C and then centrifugation at 13,000 rpm (15 minutes; 4°C). Supernatant was filtered using Ultrafree-MC-GV centrifugal filters (Millipore) and used for amino acid analysis employing the Biochrom 30 amino acid analyzer (Biochrom US). Results were provided in nmol/mg protein units (normalized to total protein content).

These results are available in Supplementary Tables S5–S7.

Amino Acid Screen: Mass Spectrometry

Metabolite Extraction and Quantitation.

Metabolites were extracted from media and quantitated as described previously (74). Ten microliters of media was mixed with 10 μL of a mixture of isotopically labeled metabolites and amino acids of known concentrations (Cambridge Isotope Laboratories) and extracted with 600 μL ice-cold high performance liquid chromatography (HPLC) grade methanol. Samples were vortexed for 10 minutes at 4°C, centrifuged at maximum speed on a tabletop centrifuge for 10 minutes at 4°C, and then 450 μL of supernatant was dried under nitrogen gas and frozen at −80°C until analyzed by GC-MS.

Gas Chromatography–Mass Spectrometry Analysis.

Polar metabolites were analyzed by GC-MS as described previously (74). Dried metabolite extracts were derivatized with 16 μL MOX reagent (Thermo Fisher Scientific) for 90 minutes at 37°C, followed by 20 μL of N-tertbutyldimethylsilyl-N-methyltrifluoroacetamide with 1% tert-butyldimethylchlorosilane (Sigma Aldrich) for 60 minutes at 60°C. After derivatization, samples were analyzed by GC-MS using a DB-35MS column (Agilent Technologies) installed in an Agilent 7890A gas chromatograph coupled to an Agilent 5997B mass spectrometer. Helium was used as the carrier gas at a flow rate of 1.2 mL/min. One microliter of sample was injected in split mode (all samples were split 1:1) at 270°C. After injection, the GC oven was held at 100°C for 1 minute and increased to 300°C at 3.5°C/min. The oven was then ramped to 320°C at 20°C/min and held for 5 minutes at this 320°C. The MS system operated under electron impact ionization at 70 eV, and the MS source and quadrupole were held at 230°C and 150°C, respectively. The detector was used in scanning mode, and the scanned ion range was 100 to 650 m/z.

All results were normalized to cell number and are available in Supplementary Table S8.

Stable Isotope Tracing of Cultured Fibroblasts.

To measure steady-state incorporation of glucose and pyruvate carbon into polar metabolites, fibroblasts were plated at 200,000 cells/well in 2 mL of DMEM with 10% FBS into 6-well dishes with or without ECM (termed 2-D without fibroblast-derived ECM or 3-D with fibroblast-derived ECM). Cells were then incubated and allowed to adhere overnight. After attachment, the cells were washed twice with 2 mL of PBS. Cells were then incubated for 48 hours in 2 mL of Glu/Gln-free DMEM supplemented with 10% dialyzed FBS where glucose and pyruvate were replaced with 25 mmol/L 13C6-glucose (Cambridge Isotopes Laboratories; #CLM-1396-PK) and 1 mmol/L 13C3-pyruvate (Cambridge Isotopes Laboratories; #CLM-2440-PK). Following the labeling period, the cells were placed on ice, and all subsequent steps took place with the cells on ice. First, the media were aspirated. Each well was then washed rapidly with approximately 8 mL of ice-cold blood bank saline (Thermo Fisher Scientific; #p-4527840). The saline was then aspirated, and 600 μL of ice-cold methanol:water (4:1) was added. Cells were then scraped, and the resulting extracts were transferred to microcentrifuge tubes. The extracts were then vortexed for 10 minutes and centrifuged to pellet-insoluble material at 21,000 RCF for 10 minutes. 450 μL of each extract was carefully removed to avoid contamination with insoluble material, transferred to new microcentrifuge tubes, dried under nitrogen gas, and stored at −80°C until further analysis. GC-MS analysis was then performed on these extracts as described in the previous section.

Mass isotopomer distributions were then determined by integrating appropriate ion fragments for each metabolite (120) using in-house software (121) that corrects for natural abundance using previously described methods (122).

Human NK-Cell Isolation and Activation Experiments

Isolation of Primary NK Cells.

Blood was collected into heparinized tubes from healthy volunteer donors. The authors received written informed consent using HIPAA-compliant procedures approved by the Institutional Review Board of FCCC. NK cells were enriched from healthy donor PBMCs using the EasySep NK-cell negative selection kit (StemCell Technologies) and confirmed to be >95% purity using flow cytometry.

Primary NK-Cell Activation Assay.

Purified NK cells were incubated overnight at 37°C in indicated percentages of CON or NetG1 CAF CM diluted in NK cell media (10%–65%) in the presence of IL2 (500 units/mL) and IL12 (10 ng/mL). NK cells activated with IL2 and IL12 were used as a positive control. After overnight incubation, the population of IFNγ-secreting cells was identified using the IFNγ Secretion assay detection kit (PE; Miltenyi Biotec) according to the manufacturer's protocol. After a 2-hour incubation with IFNγ catch reagent, cells were washed twice. Next, cells were stained with IFNγ detection antibody and antibodies against NKp80 (anti–NKp80-APC, Biolegend), CD3 (anti–CD3-Cy7APC, BD Pharmigen), CD56 (anti–CD56-BUV395, BD Horizon), and CD69 (anti–CD69-Pac Blue). Cells were centrifuged and rinsed twice, with propidium iodide in the second wash. NK cells were defined as NKp80+CD56+CD3 cells.

Flow Cytometry and Data Analysis.

Stained cells were analyzed on a BD ARIA II flow cytometer (Becton Dickinson). Data were collected with BD FACS Diva software (v6) and analyzed with FlowJo (v9.9; Tree Star Inc.). IFNγ and CD69 expressions were identified as the percentage of IFNγ+ cells or mean fluorescence intensity of CD69+ cells. The data were normalized to the positive control (NK cells with IL2/IL12 treatment alone).

Mouse Models

Tissue for stromal analysis was obtained from the LSL-KRASG12D, PDX1-Cre (KC) model (52), and KPC4B and KPC3 murine PDAC cell lines were derived from tumors generated from a variant of LSL-KRASG12D, TP53−/−, PDX1-Cre (KPC) mice (53).

PDAC Cell Orthotopic Injections

All animal work was performed under protocols approved by Institutional Animal Care and Use Committee at FCCC. A small incision was made in the abdominal skin of the mice and the spleen was gently moved aside to reveal the pancreas. Next, 106 murine CON or NGL1 KO #1 or #2 PDAC cells (derived from KPC tumors) were injected (100 μL) orthotopically into the pancreas of syngeneic C57BL/6 mice (source: FCCC, 6–8 weeks old), and tumors were allowed to develop for 3.5 weeks. Mice were monitored by observation, palpation, and weekly MRIs and sacrificed at 3.5 weeks. Pancreas weights were obtained, and pancreatic tissue was further analyzed as described below.

To determine the effect of the immune system on tumorigenesis in the orthotopic model, the injections were performed as above, with the following mice used as hosts: C57BL/B6 (immunocompetent; FCCC), C.B17 SCID (lacks T and B cells; Taconic Bioscience), and NOD.Cg-Prkdcscid IL2rgtm1Wjl/SzJ: NSG (lacks NK, T, and B cells; Jackson Laboratories). Mice were 6 to 8 weeks old.

For the human tumor cell/CAF coinjections, RFP+ PDACc cells (CON or NGL1 KO) and CAFs (CON or NetG1 KO) were coinjected at a 1:3 ratio (2.5 × 105 tumor: 7.5 × 105 CAF) into SCID mice (6–8 weeks old). CAFs and tumor cells injected alone were used as controls. Tumors were allowed to form for 4 weeks. Subsequently, mice were sacrificed, and tissue was analyzed as outline below.

For the NetG1-neutralizing mAb experiment, after mice (C57BL/B6, 6–8 weeks old) recovered from the tumor cell (106) orthotopic injections for 3 days, mice were divided into two groups (IgG, n = 9; mAb, n = 10) and were treated with 5 mg/kg isotype-specific IgG (IgG1) or mAb (Santa Cruz Biotechnology) 3 times per week, until the completion of the model (3.5 weeks, as above).

MRI

Animals were imaged in a 7 Tesla vertical wide-bore magnet, using a Bruker DRX 300 spectrometer with a microimaging accessory, which included a micro 2.5 gradient set, a 30 cm radiofrequency coil, and Paravision software (Bruker). Animals were anesthetized with a mixture of oxygen and isoflurane (2%–3%). 0.2 mL of 10:1 diluted Magnevist (gadopentetate dimeglumine; Bayer) was injected into the shoulder region immediately preceding the scan. Scout scans were performed in the axial and sagittal orientations, permitting us to accurately prescribe an oblique data set in coronal and sagittal orientations that included the organs of interest. A two-dimensional spin echo pulse sequence was employed with echo time of 15 msec, repetition time of 630 msec, field of view = 2.56 cm, acquisition matrix = 256 × 256, slice thickness = 0.75 mm, 2 averages, and scan time = 5 minutes. Fat suppression (standard on Bruker DRX systems) was used for all scans. Raw data files were converted to Tiff format image files using the Bruker plugin on ImageJ.

Murine Tissue Preparation and Histopathologic Analysis

Tissue Preparation.

Tissues were collected and fixed in 10% phosphate-buffered formaldehyde (formalin) 24 to 48 hours, dehydrated, and embedded in paraffin. Tissues were processed by dehydration in a series of ethanol followed by xylene (70% ethanol 3 hours, 95% ethanol 2 hours, 100% ethanol 2 hours, 100% ethanol xylene mixed 1 hour, xylene 3 hours, then immersed in paraffin). Paraffin blocks were cut into 5-μm sections, mounted on microscope slides, and stored at RT until used. Sections were stained with hematoxylin and eosin (H&E) for pathologic evaluation.

All H&E slides were viewed with a Nikon Eclipse 50i microscope, and photomicrographs were taken with an attached Nikon DS-Fi1 camera.

Tumor Area Analysis.

H&E-stained slides were also scanned using Aperio ScanScope CS scanner (Aperio). Scanned images were then viewed with Aperio's image viewer software (ImageScope). Selected regions of interest for PDAC, necrotic, and normal tissue were outlined respectively by a pathologist (Dr. K.Q. Cai), who was blinded to the treatment groups. The ratio of PDAC, necrotic, and normal tissue to the total pancreatic areas scanned was calculated for each animal.

Differentiation Status of Tumor Cells.

H&E images of pancreata injected with CON or NGL1 KO cells were scored for the percentage of cells that were well differentiated, moderately differentiated, or poorly differentiated, as determined by a pathologist, who was blinded to the treatment groups. For the coinjection model, sarcomatoid/sarcomatous PDAC was also scored, as it represented a significant tumor cell type in this model.

TUNEL Staining.

Apoptotic cells were quantified using the DeadEnd Fluorometric TUNEL System (Promega), per the manufacturer's instructions. Briefly, tissue was deparaffinized and fixed in 4% formaldehyde for 15 minutes at RT. After two PBS washes, tissue was treated with 20 μg/mL proteinase K solution for 10 minutes at RT. After two PBS washes, slides were fixed again in 4% formaldehyde for 5 minutes at RT. Tissue was then submerged in equilibration buffer at RT for 5 minutes. Next, apoptotic cells were labeled with TdT reaction mixture for 1 hour at 37°C. After 1 hour, the reaction was stopped by adding side scatter (SSC) buffer for 15 minutes at RT. Tissue was then washed 3x in PBS and then stained with DAPI (1:10,000) for 15 minutes at RT. Finally, after an additional wash, tissue was mounted and imaged using the Eclipse 2000-U inverted microscope. Metamorph software was used to quantify dead cells (green)/nuclei (blue) in each field of view. At least 10 images (20X) per sample were acquired.

Ki-67 Staining (Proliferation).

Immunofluorescence on tissue sections was done as previously published (38). Briefly, deparaffinized tissue underwent antigen retrieval by boiling in buffer containing 10 mmol/L sodium citrate and 0.05% Tween 20, pH 6.0, for 1 hour. Next, tissue was blocked in Odyssey Blocking Buffer (PBS base; LI-COR Biosciences) for 1 hour at RT. Tissue was then stained with primary antibody against Ki-67 (Abcam) overnight at 4°C. The following day, tissue was washed 3 times in 0.05% PBST, and secondary antibody conjugated to TRITC (Jackson ImmunoResearch) was applied to the samples for 2 hours at RT. After three additional washes, nuclei were stained with DAPI (1:10,000) for 15 minutes at RT. After one final wash, samples were mounted and images (20X) were captured using an Eclipse 2000-U inverted microscope. Proliferating cells (red)/nuclei (blue) in each field of view were quantified using Metamorph Software.

Tissue Preparation, Histology, and IHC for NetG1 mAb Mouse Model (NK1.1 and IL15)

Tissue Preparation.

Mouse tissues were collected and fixed in 10% phosphate-buffered formaldehyde (formalin) for 24 to 48 hours, dehydrated, and embedded in paraffin. H&E-stained sections were used for morphologic evaluation purposes and unstained sections for IHC studies.

IHC Staining of NK1.1 and IL15.

IHC staining was carried out according to standard methods. Briefly, 5-μm formalin-fixed, paraffin-embedded (FFPE) sections were deparaffinized and hydrated. Sections were then subjected to heat-induced epitope retrieval with EDTA (pH 9.0) for NK1.1 and with citrate buffer (pH 6.0) for IL15. Endogenous peroxidases were quenched by the immersion of slides in 3% hydrogen peroxide solution. The sections were incubated overnight with primary antibodies to anti-NK1.1 (Mouse, 1:50, MA1-70100; Invitrogen-Thermo Fisher Scientific) and IL15 (Rabbit, 1:50, LS-B10080; Lifespan Biosciences) at 4°C in a humidified slide chamber. Immunodetection was performed using the Dako Envision+ polymer system (#K4000 and #K4002, for mouse and rabbit, respectively; Dako), and immunostaining was visualized with the chromogen 3, 3′-diaminobenzidine.

The sections were then washed, counterstained with hematoxylin, dehydrated with ethanol series, cleared in xylene, and mounted.

All slides were viewed with a Nikon Eclipse 50i microscope, and photomicrographs were taken with an attached Nikon DS-Fi1 camera.

NK1.1 and IL15 IHC Scoring.

NK1.1 and IL15 IHC of mouse tissue was scored in a blinded manner by a pathologist. For NK1.1, the number of clusters (foci) of NK1.1+ cells was counted. Each focus consisted of at least 20 NK1.1-positive cells. Foci were normalized to tumor area for each mouse and then normalized to the IgG group to compute fold changes between IgG- and mAb-treated groups. For IL15, fibroblastic cells were scored on a 0 to 4 scale, based on intensity and coverage of the staining.

Supernatant Collection of Tumor Cultures

Small pieces of tissue (30–60 mg) were excised from the pancreata isolated from mice treated with IgG or mAb (described above) and were cultured in serum/Gln-free media overnight. The following day, the media were centrifuged for 15 minutes at 15,000 rpm to remove cellular debris. The resultant supernatant was collected and subjected to ELISAs to measure levels of Glu, Gln, mTGFβ, mGM-CSF, and mIL6. All data were normalized to mg of tissue cultured.

Acquisition and Use of Human Tissue (Used in TMA)

All human tissues used in this study, for the purpose of generating TMAs, were collected using HIPAA-approved protocols and exemption approval of the FCCC's Institutional Review Board, after patients signed a written informed consent in which they agreed to donate surgical specimens to research. To protect patients' identity, samples were classified, coded, and distributed by the Institutional Biosample Repository Facility. Normal pancreatic tissue was obtained through US Biomax Inc. following their rigorous ethical standards for collecting tissue from healthy expired donors.

TMA

Acquisition of the TMAs.

The FCCC TMA was generated by the Biosample Repository at FCCC. There were 80 patient samples generated from 3 independent TMAs. The MDA TMA was generated at The University of Texas MD Anderson Cancer Center. There were 143 patient samples generated from 3 independent TMAs. The Union for International Cancer Control (UICC) system was used for all of the samples from both TMAs for tumor grading and staging. The clinical parameters for each TMA are available in Supplementary Table S2.

IHC.

TMA sections were cut from FFPE blocks. After deparaffinizing in xylene, sections were hydrated in graded alcohol and quenched in 3% H2O2. Slides were subjected to heat-induced epitope retrieval in citrate buffer (pH 6.0) for 1 hour. After cooling to RT, sections were permeabilized in Triton X-100 for 20 minutes, followed by a 1-hour incubation in blocking buffer (1% BSA, 2% normal goat sera, or 0.2% cold water fish skin gelatin in PBS). Antibodies against NGL1 (GTX 121508), NetG1 (GTX 115637), GS (Abcam), VGlut1 (Thermo Fisher Scientific), PDPN (Biolegend, 916606), α-SMA (Sigma), and NKp46 (R&D Systems, AF1850) were used at 1:100 dilution, and the incubation was carried out overnight at 4°C. The sections were washed and incubated with anti-rabbit HRP (Vector Lab MP-7451) and then treated with DAB (ImmPACT SK-4105) for 2.5 minutes. Sections were counterstained in Mayer's Hematoxylin to show nuclei, dehydrated in graded alcohol, and mounted in Cytoseal-60 after clarifying in toluene.

Blind Pathologic Scoring of TMAs.

A pathologist (Dr. A.J. Klein-Szanto) blindly scored the 80 and 143 stained TMA cores for fibroblastic NetG1 or tumoral NGL1 on a scale of 0 to 4, which is based on staining intensity and coverage. NKp46 was also scored on a 0 to 4 scale, but with each value representing number of cells staining positive for NKp46, per core (0 = 0 cells, 1 = 1–10 cells, 2 = 11–20 cells, 3 = 21–30 cells, and 4 = >30 cells). Each patient had two cores stained per marker, and these were averaged together to generate the IHC score. These values, along with OS data, are available in Supplementary Table S3.

Generation of Kaplan–Meier Plots.

Kaplan–Meier (KM) plots were generated for OS correlated with IHC staining of fibroblastic NetG1, tumoral NGL1, and NKp46.

SMI

Qdot Antibody Conjugation.

Corresponding antibodies were linked to Qdot molecules using SiteClick Qdot labeling from Molecular Probes-Thermo Fisher Scientific. In brief, approximately 100 mg of specific IgG was prepurified to remove excessive carrier proteins (if needed), and each conjugation was performed following manufacturer's three-step protocol requiring antibody's carbohydrate domain modifications to allowing an azide molecule attachment to facilitate linking DIBO-modified nanocrystals to the antibodies.

FFPE Tissue Processing.

FFPE sections were deparaffinized using xylene and rehydrated in progressive ethanol to water dilutions. Tissues then were processed through heat-induced epitope retrieval using pH 6.0 citrate buffer, followed by permeabilization with 0.5% TritonX-100. After treating the specimens with Odyssey (PBS base) blocking buffer (LI-COR Biosciences), samples were labeled by a 3-day indirect immunofluorescence scheme. Initially, primary antibodies against mouse monoclonal NetG1 (D-2, Santa Cruz Biotechnology) and rabbit polyclonal NGL1 (N1N3, Genetex Inc.) were incubated at 4°C overnight, followed by anti-mouse Q625 and anti-rabbit Q655 secondary antibodies, together with primary rabbit polyclonal Y397P-FAK (Genetex Inc.) prelinked to Q605, for an overnight incubation (see Qdot antibody conjugation details above). After washes, specimens were treated with corresponding monovalent Fab IgG fragments (Jackson ImmunoResearch Inc.) to avoid cross-reaction during the incubation of the following cocktail of primary antibodies: mouse monoclonal anti–pan-cytokeratin (clones AE1/AE3, DAKO) to detect the epithelial compartment; rabbit monoclonal anti-vimentin (EPR3776, Abcam) antibodies for mesenchymal (stromal) components. Secondary antibodies used were donkey anti-mouse Cy3 and donkey anti-rabbit Cy2 (Jackson ImmunoResearch Inc.). Nuclei were labeled by DRAQ5 (Invitrogen-Thermo Fisher Scientific). Lastly, sections were dehydrated in progressive ethanol concentrations and clarified in Toluene before mounting using Cytoseal-60 (Thermo Fisher Scientific). Slides were cured overnight in dark at RT before the imaging.

Image Acquisitions for SMI Analysis.

Imaging of fluorescent-labeled tissue sections was performed utilizing Nuance-FX multispectral imaging system (Caliper LifeSciences, PerkinElmer), using full capabilities of its Tunable Liquid Crystal imaging module, attached to a Nikon Eclipse TE-2000-S epifluorescence microscope. Images were acquired using a Plan Fluor 40X/0.75 objective. A specific wavelength-based spectral library for each system was built using control pancreatic tissues (i.e., murine and human), individually labeled for each Qdot and fluorophore used. Unstained specimens were also included to digitally discard specific autofluorescence spectra. Each particular spectral library was used for the subsequent image acquisition and analysis. All excitations were achieved via a high-intensity mercury lamp using the following filters for emission–excitation, for Nuance-FX, DAPI (450–720), FITC (500–720), TRITC (580–720), and CY5 (680–720). Excitation was conducted from the infrared to ultraviolet spectrum, attempting to avoid energy transfer and excitation of nearby fluorophores. Emission was collected using “DAPI” filter (wavelength range 450–720) for all Qdot-labeled markers, whereas masks used the conventional FITC, TRITC, and CY5 filters.

Collected image cubes (multispectral data files) were unmixed to obtain 16-bit (gray scale) individual monochromatic files. Next, the images were processed in bulk for similar levels adjustment and 8-bit monochromatic image conversion for each channel, using Photoshop's “Levels” and “Batch processing” automated functions. The resulting images were sampled to set identical threshold values for each marker (NetG1, Y397P-FAK, and NGL1) and masks (vimentin and nuclei). Threshold values distinguishing cytokeratin in normal versus tumor samples needed to be independently adjusted for each tissue set to assure variations in intensity staining patterns between the two types of epithelium (i.e., normal vs. tumoral) were accounted for, as these were needed to be used as “gating masks” identifying epithelial/tumoral area pixels. Processed images were used to analyze distribution, intensity, and colocalization of positive-selected pixels using SMIA-CUKIE (SCR_014795) algorithm, available online at https://github.com/cukie/SMIA (38).

SMIA-CUKIE Usage and Outputs.

As described previously (38), SMIA-CUKIE algorithm (SCR_014795) was used to analyze high-throughput monochromatic images acquired from SMI-processed FFPE tissue sections. In brief, the algorithm is based on the usage of digital masks, constructed from pixel-positive area distribution, resulting in tissue compartmentalization (i.e., cytokeratin-positive/vimentin-negative area: epithelium/tumor compartment vs. vimentin-positive/cytokeratin-negative area: stromal compartment). Next, we queried mean, median, coverage area, and integrated intensity of the markers of interest (i.e., NetG1, NGL1, Y397P-FAK), under each mask. Each mask and marker required a numeric threshold (0–255), indicating the value of pixel intensity considered positive for each channel (corresponding to each mask and marker). These were kept identical for each marker and mask throughout the study. Finally, the analysis generated an Excel file containing numerical values, which were scrutinized for statistical significance as described in the corresponding statistical analysis section. Black lines through a row indicate an unquantifiable image. The algorithm is available in the public domain: https://github.com/cukie/SMIA_CUKIE.

RNA-seq

Stranded mRNA-seq library: 1,000 ng total RNAs from each sample were used to make library according to the product guide of Truseq-stranded mRNA library kit (Illumina, Cat# 20020595). In short, mRNAs were enriched twice via poly-T–based RNA purification beads and subjected to fragmentation at 94°C for 8 minutes via divalent cation method. The first-strand cDNA was synthesized by Superscript II reverse transcriptase (Thermo Fisher Scientific, Cat# 18064014) and random primers at 42°C for 15 minutes, followed by second-strand synthesis at 16°C for 1 hour. During second-strand synthesis, the dUTP was used to replace dTTP, thereby the second strand was quenched during amplification. A single “A” nucleotide is added to the 3′ ends of the blunt fragments at 37°C for 30 minutes. Adapters with illuminaP5, P7 sequences as well as indices were ligated to the cDNA fragment at 30°C for 10 minutes. After SPRIselect beads (Beckman Coulter, Cat# B23318) purification, a 15-cycle PCR reaction was used to enrich the fragments. PCR was set at 98°C for 10 seconds, 60°C for 30 seconds, and extended at 72°C for 30 seconds. Libraries were again purified using SPRIselect beads, had a quality check on Agilent 2100 bioanalyzer (serial # DE34903146) using Agilent high sensitivity DNA kit (Cat# 5067-4626), and quantified with Qubit 3.0 flurometer (Thermo Fisher Scientific, Cat#Q33216) using Qubit 1x dsDNA HS assay kit (Cat#Q33230). Sample libraries were subsequently pooled and loaded to the Hiseq 2500 (Illumina, serial number SN930). Single end reads at 50 bp were generated by using Hiseq rapid SR cluster kit V2 (Illumina, Cat# GD-402-4002) and Hiseq rapid SBS kit v2 50 cycles (Illumina, Cat# FC-402-4022). Faseq files were obtained at Illumina base space (https://basespace.illumina.com) and aligned to Hg38 using Tophat2 (123), and resulting BAM files were used to estimate the gene counts using HT-Seq (124). Subsequently, to identify differentially expressed genes between CON and NetG1 KO CAFs, we applied DESeq2 (125). The resulting differentially expressed genes were used as input to GSEA (ref. 126; default parameters with enrichment statistic set as classic). Heat map depicting differentially expressed genes using mean-centered rlog-transformed gene counts was plotted using pheatmap package available in R. Data are organized in tabs (gene list P < 0.001, GSEA list with FDR <0.25, and complete GSEA for positive and negative enrichment), available in Supplementary Table S4. Data were also deposited into the Gene Expression Omnibus database (accession #GSE156963) as part of the superseries “To determine the differentially expressed genes in cancer associated fibroblasts upon ablation of the glutamatergic synapse protein NetrinG1” (accession #GSE156964).

Pancreatic Cancer Datasets

We obtained the gene expression data from ref. 42 (processed data available at ArrayExpress ID: E-MTAB-6134), ref. 43 (processed expression data available at Gene Expression Omnibus ID GSE21501), and TCGA (rsem-normalized gene expression data available at GDAC Firehose repository for 76 high tumor purity samples). Clinical outcome data indicating patient OS were obtained from the respective data repositories. In the case of Moffitt and TCGA studies, we used Moffitt and colleagues' (43) classification to stratify PDAC tumors into basal and classic subtypes. In the case of Puleo and colleagues (42), we applied five different classes as described by that study. In order to assess the association between expression of LRRC4C and NTNG1 in PDAC tumors to OS, we used expression and clinical data for TCGA and Puleo and colleagues (42, 44). We stratified the expression data into lower (DN) and upper quartiles (UP) for both LRRC4C and NTNG1 and categorized them into four different groups (DN-DN, DN-UP, UP-DN, and UP-UP). We then estimated the distributions of OS and RFS using KM methods and compared them using log-rank tests. Analyses to compare the OS among the basal and classic subtypes were performed by comparing survival curves with log-rank tests. These were calculated using the R “survival” package Therneau, T. M. & Grambsch, P. M. Modeling Survival Data: Extending the Cox Model (Springer-Verlag, 2010). We applied gene set variation analysis (127) with mx.diff parameter set to true on all the 76 PDAC tumor cases in TCGA using 12,913 gene sets available from MSigDB (128). To classify NGL1 into low- and high expressors, we selected cases on Z-scores (85th confidence interval for mean). Wilcoxon rank-sum test was applied to identify significantly enriched MSigDB datasets (P < 0.005). Select enriched pathways that are significantly different between low and high expressors of NGL1 were plotted as waterfall plots with low (green color) and high expressors (red). Cases that did not satisfy the Z-score cutoff were plotted as gray bars.

KM Plots Generated with KM Plotter

To generate KM plots correlating with OS and RFS data with NTNG1 and LRRC4C mRNA expression from the TCGA dataset, we used the online tool KM Plotter (kmplot.com; (129)). We selected the pancreatic adenocarcinoma dataset (N = 177) from the Pan-Cancer RNA-seq section of the website for OS. For the RFS analysis, we selected the RFS dataset, for which there were data available for 69 patients (130). We used the default auto cutoff algorithm to establish the high and low expressors of NTNG1 and LRRC4C (cutoff = 22) combined. The FDRs for the cutoff values for NTNG1 and LRRC4C were >0.5. KM plots were generated, and HR and P values (log-rank test) were obtained.

Statistical Analysis

For all in vitro and animal experiments, Prism 7.0 (Graph Pad Software) was used for all statistical analysis. For comparison between two groups, a two-tailed Student t test was performed. For comparisons between more than two groups, a one-way ANOVA was performed, using either a Dunnett multiple comparisons test (compared with control condition) or Tukey multiple comparisons test (comparing all conditions to one another), depending on the experiment. Groups were deemed statistically significantly different from one another if the P value was less than or equal to 0.05. On graphs, significance was defined as follows: *, P < 0.05; **, P < 0.01; ***, < 0.001; ****, P < 0.0001. For the correlation of the SMI analysis, linear regression was applied to the SMIA-generated values of two markers at the designated components, and the R2 value was calculated. All in vitro experiments were performed at least 3 times (unless explicitly stated in the figure legend), and the orthotopic injections were performed in two cohorts each time, with the following resultant sample sizes:

  • —Human coinjection SCID model: N = 6 for all coinjections; N = 4 and 5 for CON or NGL1 KO PDAC cells alone; N = 3 and 4 for CON or NetG1 KO CAFs alone.

  • —Murine cancer cell model: N = 7 for Control, and N = 19 for combined NGL1 KO–injected mice.

  • —Murine immunocompromised mouse models: N = 5 for all groups

  • —mAb model: N = 9 for IgG group; N = 10 for mAb group.

R. Francescone reports grants from NIH/NCI during the conduct of the study; in addition, R. Francescone has a patent for US62/570,694 pending to E. Cukierman, R. Francescone, J. Franco-Barraza, and K. Raghavan, and a patent for US62/838,593 pending to E. Cukierman, R. Francescone, J. Franco-Barraza, and K. Raghavan, R. Dunbrack, and D. Barbosa Vendramini-Costa. D. Barbosa Vendramini-Costa reports grants from NIH/NCI during the conduct of the study; in addition, D. Barbosa Vendramini-Costa has a patent for US62/838,593 pending to E. Cukierman, R. Francescone, J. Franco-Barraza, and K. Raghavan, R. Dunbrack, and D. Barbosa Vendramini-Costa. J. Franco-Barraza reports a patent for E. Cukierman, R. Francescone, J. Franco-Barraza, and K. Raghavan. Published patent application WO2019074915A1 “NetrinG1 as a biomarker for enhancing tumor treatment efficacy.” Priority to provisional patent application “Treatment of Pancreatic Cancer” US62/570,694 field on 10-11-2017; International PCT/US18/58994 “Netrin GI As a Biomarker for Enhancing Tumor Treatment Efficacy” filed 10/09/2018 and pending a patent for Cukierman, E., Francescone, R., Franco-Barraza J., Raghavan, K. Dunbrack, R. and D. Vendramini Costa, D. Published patent application WO2020219854A1 “Netrin G1 and Netrin G1 ligand peptides and antibodies and uses thereof.” Priority to provisional patent application # 62/838,593 “Netrin G1 and Netrin G1 Ligand Peptides and Antibodies And Uses Thereof” filed 04/25/2019; International PCT/US20/29781 “Netrin G1 and Netrin G1Ligand Peptides and Antibodies And Uses Thereof” filed 04/24/2020 and pending a patent for E. Cukierman, R. Francescone, J. Franco-Barraza, and K. Raghavan, R. Dunbrack, and D. Barbosa Vendramini-Costa. Applications filed by Institute For Cancer Research D/B/A The Research Institute Of Fox Chase Cancer Center. K. Devarajan reports grants from NIH during the conduct of the study. M.G. Vander Heiden reports personal fees from Agios Pharmaceuticals, personal fees from Aeglea Biotherapeutics, personal fees from Auron Therapeutics, personal fees from iTeos Therapeutics, and personal fees from Faeth Therapeutics outside the submitted work. K.S. Campbell reports grants from NCI and grants from the PA Department of Health CURE fund during the conduct of the study; grants from Janssen Pharma, Bristol-Myers Squibb, NantKwest, Horizon Pharma, and Immune Oncology Biosciences outside the submitted work; in addition, K.S. Campbell has a patent for Genetically Modified Human Natural Killer Cell Lines issued, licensed, and with royalties paid from NantKwest. E. Cukierman reports grants from NCI/NIH, grants from DOD, grants from Worldwide Cancer Research, grants from 5th AHEPA Cancer Research Foundation, and grants from Pennsylvania's DOH Health Research Formula Funds during the conduct of the study; in addition, E. Cukierman has a patent for US62/570,694 pending and a patent for US62/838,593 pending; and wishes to disclose that she is currently a consultant and SAB member for Phenomic AI. No disclosures were reported by the other authors.

R. Francescone: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. D. Barbosa Vendramini-Costa: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. J. Franco-Barraza: Data curation, formal analysis, investigation, visualization, methodology. J. Wagner: Formal analysis, validation, methodology. A. Muir: Formal analysis, validation, investigation, methodology. A.N. Lau: Formal analysis, investigation, methodology. L. Gabitova: Investigation, methodology. T. Pazina: Formal analysis, visualization, methodology, writing-review and editing. S. Gupta: Investigation, methodology.T. Luong: Formal analysis, validation, investigation, visualization. D. Rollins: Data curation, formal analysis, validation, visualization. R. Malik: Formal analysis. R.J. Thapa: Methodology.D. Restifo: Validation. Y. Zhou: Data curation, formal analysis.K.Q. Cai: Formal analysis, methodology. H.H. Hensley: Investigation, methodology. Y. Tan: Data curation, validation, methodology. W.D. Kruger: Resources. K. Devarajan: Formal analysis. S. Balachandran: Resources, methodology, writing-review and editing. A.J. Klein-Szanto: Formal analysis. H. Wang: Resources, formal analysis. W.S. El-Deiry: Resources. M.G. Vander Heiden: Funding acquisition, methodology, writing-review and editing. S. Peri: Data curation, formal analysis, methodology, writing-review and editing.K.S. Campbell: Supervision, funding acquisition, methodology, writing-review and editing. I. Astsaturov: Resources, funding acquisition, methodology, writing-review and editing. E. Cukierman: Conceptualization, resources, formal analysis, supervision, funding acquisition, visualization, methodology, writing-review and editing.

This study is dedicated to the memories of the late Dr. Patricia Keely (pioneer in the field of ECM biology and an amazing mentor) and Neelima Shah (a noteworthy coauthor on this study, with over 30 years in research and the most beautiful soul), who continue to inspire our work. We would like to thank Dr. Martin Humphries for the SNAKA 51 antibody. We appreciate the advice from Dr. Alexander Macfarlane on NK cells and flow cytometry. We thank Jim Oesterling for his feedback and technical assistance with flow cytometry. We greatly appreciate the NetG1/NGL1 discussions with Dr. Shigeyoshi Itohara. We would also like to acknowledge Dr. Stephen M. Sykes, Dr. Daniela Di Marcantonio, and Esteban Martinez for their time and expertise with qRT-PCR. We are grateful for the kind gifts provided by Robert Weinberg (pBABE-neo-hTERT vector; Addgene plasmid # 1774), Dr. Feng Zhang (LentiCRISPRv2; Addgene plasmid # 52961), Dr. Didier Trono (psPAX2; Addgene plasmid # 12260), Charles Gersbach (pLV hU6-sgRNA hUbC-dCas9-KRAB-T2a-Puro; Addgene plasmid #71236), and Dr. Alexey Ivanov (pLV-CMV-H4-puro vector). This work was supported in part by gifts donated to the memory of Judy Costin, funds from the Marianne DiNofrio Pancreatic Research Foundation, funds from the Martin and Concetta Greenberg Pancreatic Cancer Institute, Pennsylvania's DOH Health Research Formula Funds, the 5th AHEPA Cancer Research Foundation, Inc., a grant by Worldwide Cancer Research (E. Cukierman), DOD grant WX81XWH-15-1-0170 (E. Cukierman, R. Francescone, and J. Franco-Barraza), as well as NIH/NCI grants R01CA113451-10 (E. Cukierman), T32CA009035 (R. Francescone and D. Barbosa Vendramini-Costa, independently), R01CA113451-09S1 (J. Franco-Barraza and E. Cukierman), R21 CA231252 (E. Cukierman and I. Astsaturov), R01CA232256 (E. Cukierman), R01CA188430 (I. Astsaturov), R03CA212949 (I. Astsaturov), R01GM116911 (I. Astsaturov), R01CA194263 (K.S. Campbell), and the Core Comprehensive Cancer Center Grant CA06927 in support to FCCC's facilities including: Bio Sample Repository, Light Microscopy, Small Animal Imaging, Biostatistics and Bioinformatics, Flow Cytometry, Laboratory Animal, Cell Culture, DNA sequencing, Genotyping and Real-time PCR, Histopathology, Immune Monitoring Facility, and Talbot Library. The authors also acknowledge support from F32CA213810 (A. Muir); the Damon Runyon Cancer Research Foundation (A.N. Lau); R01CA168653, R01CA201276, and P30CA1405141 (M.G. Vander Heiden); the Lustgarten Foundation (M.G. Vander Heiden); the MIT Center for Precision Medicine (M.G. Vander Heiden); SU2C (M.G. Vander Heiden); and the Ludwig Center at MIT (M.G. Vander Heiden). M.G. Vander Heiden is a Howard Hughes Medical Institute Faculty Scholar.

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.

1.
Rahib
L
,
Smith
BD
,
Aizenberg
R
,
Rosenzweig
AB
,
Fleshman
JM
,
Matrisian
LM
. 
Projecting cancer incidence and deaths to 2030: the unexpected burden of thyroid, liver, and pancreas cancers in the United States
.
Cancer Res
2014
;
74
:
2913
21
.
2.
Siegel
RL
,
Miller
KD
,
Jemal
A
. 
Cancer statistics, 2020
.
CA Cancer J Clin
2020
;
70
:
7
30
.
3.
Ryan
DP
,
Hong
TS
,
Bardeesy
N
. 
Pancreatic adenocarcinoma
.
N Engl J Med
2014
;
371
:
2140
1
.
4.
Hidalgo
M
. 
Pancreatic cancer
.
N Engl J Med
2010
;
362
:
1605
17
.
5.
Sahai
E
,
Astsaturov
I
,
Cukierman
E
,
DeNardo
DG
,
Egeblad
M
,
Evans
RM
, et al
A framework for advancing our understanding of cancer-associated fibroblasts
.
Nat Rev Cancer
2020
;
20
:
174
86
.
6.
Olive
KP
,
Jacobetz
MA
,
Davidson
CJ
,
Gopinathan
A
,
McIntyre
D
,
Honess
D
, et al
Inhibition of hedgehog signaling enhances delivery of chemotherapy in a mouse model of pancreatic cancer
.
Science
2009
;
324
:
1457
61
.
7.
McCarroll
JA
,
Naim
S
,
Sharbeen
G
,
Russia
N
,
Lee
J
,
Kavallaris
M
, et al
Role of pancreatic stellate cells in chemoresistance in pancreatic cancer
.
Front Physiol
2014
;
5
:
141
.
8.
Chan
TS
,
Hsu
CC
,
Pai
VC
,
Liao
WY
,
Huang
SS
,
Tan
KT
, et al
Metronomic chemotherapy prevents therapy-induced stromal activation and induction of tumor-initiating cells
.
J Exp Med
2016
;
213
:
2967
88
.
9.
Hwang
RF
,
Moore
T
,
Arumugam
T
,
Ramachandran
V
,
Amos
KD
,
Rivera
A
, et al
Cancer-associated stromal fibroblasts promote pancreatic tumor progression
.
Cancer Res
2008
;
68
:
918
26
.
10.
Djurec
M
,
Grana
O
,
Lee
A
,
Troule
K
,
Espinet
E
,
Cabras
L
, et al
Saa3 is a key mediator of the protumorigenic properties of cancer-associated fibroblasts in pancreatic tumors
.
Proc Natl Acad Sci U S A
2018
;
115
:
E1147
56
.
11.
Erkan
M
,
Michalski
CW
,
Rieder
S
,
Reiser-Erkan
C
,
Abiatari
I
,
Kolb
A
, et al
The activated stroma index is a novel and independent prognostic marker in pancreatic ductal adenocarcinoma
.
Clin Gastroenterol Hepatol
2008
;
6
:
1155
61
.
12.
Ozdemir
BC
,
Pentcheva-Hoang
T
,
Carstens
JL
,
Zheng
X
,
Wu
CC
,
Simpson
TR
, et al
Depletion of carcinoma-associated fibroblasts and fibrosis induces immunosuppression and accelerates pancreas cancer with reduced survival
.
Cancer Cell
2014
;
25
:
719
34
.
13.
Rhim
AD
,
Oberstein
PE
,
Thomas
DH
,
Mirek
ET
,
Palermo
CF
,
Sastra
SA
, et al
Stromal elements act to restrain, rather than support, pancreatic ductal adenocarcinoma
.
Cancer Cell
2014
;
25
:
735
47
.
14.
Kim
EJ
,
Sahai
V
,
Abel
EV
,
Griffith
KA
,
Greenson
JK
,
Takebe
N
, et al
Pilot clinical trial of hedgehog pathway inhibitor gdc-0449 (vismodegib) in combination with gemcitabine in patients with metastatic pancreatic adenocarcinoma
.
Clin Cancer Res
2014
;
20
:
5937
45
.
15.
Provenzano
PP
,
Cuevas
C
,
Chang
AE
,
Goel
VK
,
Von Hoff
DD
,
Hingorani
SR
. 
Enzymatic targeting of the stroma ablates physical barriers to treatment of pancreatic ductal adenocarcinoma
.
Cancer Cell
2012
;
21
:
418
29
.
16.
Kamphorst
JJ
,
Nofal
M
,
Commisso
C
,
Hackett
SR
,
Lu
W
,
Grabocka
E
, et al
Human pancreatic cancer tumors are nutrient poor and tumor cells actively scavenge extracellular protein
.
Cancer Res
2015
;
75
:
544
53
.
17.
Erkan
M
,
Kurtoglu
M
,
Kleeff
J
. 
The role of hypoxia in pancreatic cancer: a potential therapeutic target?
Expert Rev Gastroenterol Hepatol
2016
;
10
:
301
16
.
18.
Son
J
,
Lyssiotis
CA
,
Ying
H
,
Wang
X
,
Hua
S
,
Ligorio
M
, et al
Glutamine supports pancreatic cancer growth through a kras-regulated metabolic pathway
.
Nature
2013
;
496
:
101
5
.
19.
Jin
L
,
Alesi
GN
,
Kang
S
. 
Glutaminolysis as a target for cancer therapy
.
Oncogene
2016
;
35
:
3619
25
.
20.
Seo
JW
,
Choi
J
,
Lee
SY
,
Sung
S
,
Yoo
HJ
,
Kang
MJ
, et al
Autophagy is required for pdac glutamine metabolism
.
Sci Rep
2016
;
6
:
37594
.
21.
Lyssiotis
CA
,
Kimmelman
AC
. 
Metabolic interactions in the tumor microenvironment
.
Trends Cell Biol
2017
;
27
:
863
75
.
22.
von Ahrens
D
,
Bhagat
TD
,
Nagrath
D
,
Maitra
A
,
Verma
A
. 
The role of stromal cancer-associated fibroblasts in pancreatic cancer
.
J Hematol Oncol
2017
;
10
:
76
.
23.
Basso
D
,
Gnatta
E
,
Plebani
M
. 
Pancreatic cancer fostered immunosuppression privileges tumor growth and progression
.
J Clin Cell Immunol
2014
:
16
.
24.
Clark
CE
,
Hingorani
SR
,
Mick
R
,
Combs
C
,
Tuveson
DA
,
Vonderheide
RH
. 
Dynamics of the immune reaction to pancreatic cancer from inception to invasion
.
Cancer Res
2007
;
67
:
9518
27
.
25.
Fu
Y
,
Liu
S
,
Zeng
S
,
Shen
H
. 
The critical roles of activated stellate cells-mediated paracrine signaling, metabolism and onco-immunology in pancreatic ductal adenocarcinoma
.
Mol Cancer
2018
;
17
:
62
.
26.
Harper
J
,
Sainson
RC
. 
Regulation of the anti-tumour immune response by cancer-associated fibroblasts
.
Semin Cancer Biol
2014
;
25
:
69
77
.
27.
Wu
Q
,
Tian
Y
,
Zhang
J
,
Zhang
H
,
Gu
F
,
Lu
Y
, et al
Functions of pancreatic stellate cell-derived soluble factors in the microenvironment of pancreatic ductal carcinoma
.
Oncotarget
2017
;
8
:
102721
38
.
28.
Ene-Obong
A
,
Clear
AJ
,
Watt
J
,
Wang
J
,
Fatah
R
,
Riches
JC
, et al
Activated pancreatic stellate cells sequester cd8+ t cells to reduce their infiltration of the juxtatumoral compartment of pancreatic ductal adenocarcinoma
.
Gastroenterology
2013
;
145
:
1121
32
.
29.
Kumar
V
,
Donthireddy
L
,
Marvel
D
,
Condamine
T
,
Wang
F
,
Lavilla-Alonso
S
, et al
Cancer-associated fibroblasts neutralize the anti-tumor effect of csf1 receptor blockade by inducing pmn-mdsc infiltration of tumors
.
Cancer Cell
2017
;
32
:
654
68
.
30.
Buck
MD
,
Sowell
RT
,
Kaech
SM
,
Pearce
EL
. 
Metabolic instruction of immunity
.
Cell
2017
;
169
:
570
86
.
31.
Kouidhi
S
,
Ben Ayed
F
,
Benammar Elgaaied
A
. 
Targeting tumor metabolism: a new challenge to improve immunotherapy
.
Front Immunol
2018
;
9
:
353
.
32.
Renner
K
,
Singer
K
,
Koehl
GE
,
Geissler
EK
,
Peter
K
,
Siska
PJ
, et al
Metabolic hallmarks of tumor and immune cells in the tumor microenvironment
.
Front Immunol
2017
;
8
:
248
.
33.
Gardiner
CM
,
Finlay
DK
. 
What fuels natural killers? Metabolism and nk cell responses
.
Front Immunol
2017
;
8
:
367
.
34.
Nakashiba
T
,
Ikeda
T
,
Nishimura
S
,
Tashiro
K
,
Honjo
T
,
Culotti
JG
, et al
Netrin-g1: a novel glycosyl phosphatidylinositol-linked mammalian netrin that is functionally divergent from classical netrins
.
J Neurosci
2000
;
20
:
6540
50
.
35.
Song
YS
,
Lee
HJ
,
Prosselkov
P
,
Itohara
S
,
Kim
E
. 
Trans-induced cis interaction in the tripartite ngl-1, netrin-g1 and lar adhesion complex promotes development of excitatory synapses
.
J Cell Sci
2013
;
126
:
4926
38
.
36.
Lin
JC
,
Ho
WH
,
Gurney
A
,
Rosenthal
A
. 
The netrin-g1 ligand ngl-1 promotes the outgrowth of thalamocortical axons
.
Nat Neurosci
2003
;
6
:
1270
6
.
37.
Amatangelo
MD
,
Bassi
DE
,
Klein-Szanto
AJ
,
Cukierman
E
. 
Stroma-derived three-dimensional matrices are necessary and sufficient to promote desmoplastic differentiation of normal fibroblasts
.
Am J Pathol
2005
;
167
:
475
88
.
38.
Franco-Barraza
J
,
Francescone
R
,
Luong
T
,
Shah
N
,
Madhani
R
,
Cukierman
G
, et al
Matrix-regulated integrin alphavbeta5 maintains alpha5beta1-dependent desmoplastic traits prognostic of neoplastic recurrence
.
Elife
2017
;
6
:
e20600
.
39.
Campbell
PM
,
Groehler
AL
,
Lee
KM
,
Ouellette
MM
,
Khazak
V
,
Der
CJ
. 
K-ras promotes growth transformation and invasion of immortalized human pancreatic cells by raf and phosphatidylinositol 3-kinase signaling
.
Cancer Res
2007
;
67
:
2098
106
.
40.
Cerami
E
,
Gao
J
,
Dogrusoz
U
,
Gross
BE
,
Sumer
SO
,
Aksoy
BA
, et al
The cbio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data
.
Cancer Discov
2012
;
2
:
401
4
.
41.
Gao
J
,
Aksoy
BA
,
Dogrusoz
U
,
Dresdner
G
,
Gross
B
,
Sumer
SO
, et al
Integrative analysis of complex cancer genomics and clinical profiles using the cbioportal
.
Sci Signal
2013
;
6
:
pl1
.
42.
Puleo
F
,
Nicolle
R
,
Blum
Y
,
Cros
J
,
Marisa
L
,
Demetter
P
, et al
Stratification of pancreatic ductal adenocarcinomas based on tumor and microenvironment features
.
Gastroenterology
2018
;
155
:
1999
2013
.
43.
Moffitt
RA
,
Marayati
R
,
Flate
EL
,
Volmar
KE
,
Loeza
SG
,
Hoadley
KA
, et al
Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma
.
Nat Genet
2015
;
47
:
1168
78
.
44.
Cancer Genome Atlas Research Network
. 
Integrated genomic characterization of pancreatic ductal adenocarcinoma
.
Cancer Cell
2017
;
32
:
185
203
.
45.
Cukierman
E
,
Pankov
R
,
Stevens
DR
,
Yamada
KM
. 
Taking cell-matrix adhesions to the third dimension
.
Science
2001
;
294
:
1708
12
.
46.
Friedman
G
,
Levi-Galibov
O
,
David
E
,
Bornstein
C
,
Giladi
A
,
Dadiani
M
, et al
Cancer-associated fibroblast compositions change with breast cancer progression linking the ratio of s100a4+ and pdpn+ cafs to clinical outcome
.
Nature Cancer
2020
;
1
:
692
708
.
47.
Hirayama
K
,
Kono
H
,
Nakata
Y
,
Akazawa
Y
,
Wakana
H
,
Fukushima
H
, et al
Expression of podoplanin in stromal fibroblasts plays a pivotal role in the prognosis of patients with pancreatic cancer
.
Surg Today
2018
;
48
:
110
8
.
48.
Shindo
K
,
Aishima
S
,
Ohuchida
K
,
Fujiwara
K
,
Fujino
M
,
Mizuuchi
Y
, et al
Podoplanin expression in cancer-associated fibroblasts enhances tumor progression of invasive ductal carcinoma of the pancreas
.
Mol Cancer
2013
;
12
:
168
.
49.
Chiarelli
M
,
Burati
M
,
Tagliabue
F
,
Terragni
S
,
Guttadauro
A
,
Cioffi
G
, et al
Undifferentiated pancreatic carcinoma: presentation, classification and prognosis
.
J Pancreas
2019
;
21
:
1
6
.
50.
Strobel
O
,
Hartwig
W
,
Bergmann
F
,
Hinz
U
,
Hackert
T
,
Grenacher
L
, et al
Anaplastic pancreatic cancer: presentation, surgical management, and outcome
.
Surgery
2011
;
149
:
200
8
.
51.
Xie
Y
,
Xiang
Y
,
Zhang
D
,
Yao
X
,
Sheng
J
,
Yang
Y
, et al
Sarcomatoid carcinoma of the pancreas: a case report and review of the literature
.
Mol Med Rep
2018
;
18
:
4716
24
.
52.
Hingorani
SR
,
Petricoin
EF
,
Maitra
A
,
Rajapakse
V
,
King
C
,
Jacobetz
MA
, et al
Preinvasive and invasive ductal pancreatic cancer and its early detection in the mouse
.
Cancer Cell
2003
;
4
:
437
50
.
53.
Hingorani
SR
,
Wang
L
,
Multani
AS
,
Combs
C
,
Deramaudt
TB
,
Hruban
RH
, et al
Trp53r172h and krasg12d cooperate to promote chromosomal instability and widely metastatic pancreatic ductal adenocarcinoma in mice
.
Cancer Cell
2005
;
7
:
469
83
.
54.
Muranen
T
,
Iwanicki
MP
,
Curry
NL
,
Hwang
J
,
DuBois
CD
,
Coloff
JL
, et al
Starved epithelial cells uptake extracellular matrix for survival
.
Nat Commun
2017
;
8
:
13989
.
55.
Sousa
CM
,
Biancur
DE
,
Wang
X
,
Halbrook
CJ
,
Sherman
MH
,
Zhang
L
, et al
Pancreatic stellate cells support tumour metabolism through autophagic alanine secretion
.
Nature
2016
;
536
:
479
83
.
56.
Zhao
H
,
Yang
L
,
Baddour
J
,
Achreja
A
,
Bernard
V
,
Moss
T
, et al
Tumor microenvironment derived exosomes pleiotropically modulate cancer cell metabolism
.
Elife
2016
;
5
:
e10250
.
57.
Recouvreux
MV
,
Commisso
C
. 
Macropinocytosis: a metabolic adaptation to nutrient stress in cancer
.
Front Endocrinol
2017
;
8
:
261
.
58.
Commisso
C
,
Flinn
RJ
,
Bar-Sagi
D
. 
Determining the macropinocytic index of cells through a quantitative image-based assay
.
Nat Protoc
2014
;
9
:
182
92
.
59.
Xiang
Y
,
Stine
ZE
,
Xia
J
,
Lu
Y
,
O'Connor
RS
,
Altman
BJ
, et al
Targeted inhibition of tumor-specific glutaminase diminishes cell-autonomous tumorigenesis
.
J Clin Invest
2015
;
125
:
2293
306
.
60.
Yang
L
,
Achreja
A
,
Yeung
TL
,
Mangala
LS
,
Jiang
D
,
Han
C
, et al
Targeting stromal glutamine synthetase in tumors disrupts tumor microenvironment-regulated cancer cell growth
.
Cell Metab
2016
;
24
:
685
700
.
61.
Liguz-Lecznar
M
,
Skangiel-Kramska
J
. 
Vesicular glutamate transporters (vgluts): the three musketeers of glutamatergic system
.
Acta Neurobiol Exp
2007
;
67
:
207
18
.
62.
Woodcock
HV
,
Eley
JD
,
Guillotin
D
,
Plate
M
,
Nanthakumar
CB
,
Martufi
M
, et al
The mtorc1/4e-bp1 axis represents a critical signaling node during fibrogenesis
.
Nat Commun
2019
;
10
:
6
.
63.
Rajasekaran
S
,
Reddy
NM
,
Zhang
W
,
Reddy
SP
. 
Expression profiling of genes regulated by fra-1/ap-1 transcription factor during bleomycin-induced pulmonary fibrosis
.
BMC Genomics
2013
;
14
:
381
.
64.
Fujita
Y
,
Nakanishi
T
,
Ueno
M
,
Itohara
S
,
Yamashita
T
. 
Netrin-g1 regulates microglial accumulation along axons and supports the survival of layer v neurons in the postnatal mouse brain
.
Cell Rep
2020
;
31
:
107580
.
65.
Zhang
Q
,
Sano
C
,
Masuda
A
,
Ando
R
,
Tanaka
M
,
Itohara
S
. 
Netrin-g1 regulates fear-like and anxiety-like behaviors in dissociable neural circuits
.
Sci Rep
2016
;
6
:
28750
.
66.
Eastwood
SL
,
Harrison
PJ
. 
Decreased mRNA expression of netrin-G1 and netrin-G2 in the temporal lobe in schizophrenia and bipolar disorder
.
Neuropsychopharmacology
2008
;
33
:
933
45
.
67.
Nectoux
J
,
Girard
B
,
Bahi-Buisson
N
,
Prieur
F
,
Afenjar
A
,
Rosas-Vargas
H
, et al
Netrin g1 mutations are an uncommon cause of atypical Rett syndrome with or without epilepsy
.
Pediatr Neurol
2007
;
37
:
270
4
.
68.
Hao
W
,
Yu
M
,
Lin
J
,
Liu
B
,
Xing
H
,
Yang
J
, et al
The pan-cancer landscape of netrin family reveals potential oncogenic biomarkers
.
Sci Rep
2020
;
10
:
5224
.
69.
Sho
S
,
Court
CM
,
Winograd
P
,
Russell
MM
,
Tomlinson
JS
. 
A prognostic mutation panel for predicting cancer recurrence in stages ii and iii colorectal cancer
.
J Surg Oncol
2017
;
116
:
996
1004
.
70.
Andrew
AS
,
Baron
JA
,
Butterly
LF
,
Suriawinata
AA
,
Tsongalis
GJ
,
Robinson
CM
, et al
Hyper-methylated loci persisting from sessile serrated polyps to serrated cancers
.
Int J Mol Sci
2017
;
18
:
535
.
71.
Castro-Bello
F
,
Ramos
F
,
Vivanco
F
,
Marina-Fiol
C
. 
High serum glutamic acid levels in patients with carcinoma of the pancreas
.
Digestion
1976
;
14
:
360
3
.
72.
Roux
C
,
Riganti
C
,
Borgogno
SF
,
Curto
R
,
Curcio
C
,
Catanzaro
V
, et al
Endogenous glutamine decrease is associated with pancreatic cancer progression
.
Oncotarget
2017
;
8
:
95361
76
.
73.
Biancur
DE
,
Paulo
JA
,
Malachowska
B
,
Quiles Del Rey
M
,
Sousa
CM
,
Wang
X
, et al
Compensatory metabolic networks in pancreatic cancers upon perturbation of glutamine metabolism
.
Nat Commun
2017
;
8
:
15965
.
74.
Muir
A
,
Danai
LV
,
Gui
DY
,
Waingarten
CY
,
Lewis
CA
,
Vander Heiden
MG
. 
Environmental cystine drives glutamine anaplerosis and sensitizes cancer cells to glutaminase inhibition
.
Elife
2017
;
6
:
e27713
.
75.
Sullivan
MR
,
Danai
LV
,
Lewis
CA
,
Chan
SH
,
Gui
DY
,
Kunchok
T
, et al
Quantification of microenvironmental metabolites in murine cancers reveals determinants of tumor nutrient availability
.
Elife
2019
;
8
:
e44235
.
76.
Bott
AJ
,
Shen
J
,
Tonelli
C
,
Zhan
L
,
Sivaram
N
,
Jiang
YP
, et al
Glutamine anabolism plays a critical role in pancreatic cancer by coupling carbon and nitrogen metabolism
.
Cell Rep
2019
;
29
:
1287
98
.
77.
Daher
B
,
Parks
SK
,
Durivault
J
,
Cormerais
Y
,
Baidarjad
H
,
Tambutte
E
, et al
Genetic ablation of the cystine transporter xct in pdac cells inhibits mtorc1, growth, survival, and tumor formation via nutrient and oxidative stresses
.
Cancer Res
2019
;
79
:
3877
90
.
78.
Coothankandaswamy
V
,
Cao
S
,
Xu
Y
,
Prasad
PD
,
Singh
PK
,
Reynolds
CP
, et al
Amino acid transporter slc6a14 is a novel and effective drug target for pancreatic cancer
.
Br J Pharmacol
2016
;
173
:
3292
306
.
79.
Parker
SJ
,
Amendola
CR
,
Hollinshead
KER
,
Yu
Q
,
Yamamoto
K
,
Encarnacion-Rosado
J
, et al
Selective alanine transporter utilization creates a targetable metabolic niche in pancreatic cancer
.
Cancer Discov
2020
;
10
:
1018
37
.
80.
Commisso
C
,
Davidson
SM
,
Soydaner-Azeloglu
RG
,
Parker
SJ
,
Kamphorst
JJ
,
Hackett
S
, et al
Macropinocytosis of protein is an amino acid supply route in ras-transformed cells
.
Nature
2013
;
497
:
633
7
.
81.
Yao
W
,
Rose
JL
,
Wang
W
,
Seth
S
,
Jiang
H
,
Taguchi
A
, et al
Syndecan 1 is a critical mediator of macropinocytosis in pancreatic cancer
.
Nature
2019
;
568
:
410
4
.
82.
Li
L
,
Hanahan
D
. 
Hijacking the neuronal nmdar signaling circuit to promote tumor growth and invasion
.
Cell
2013
;
153
:
86
100
.
83.
Popoli
M
,
Yan
Z
,
McEwen
BS
,
Sanacora
G
. 
The stressed synapse: the impact of stress and glucocorticoids on glutamate transmission
.
Nat Rev Neurosci
2011
;
13
:
22
37
.
84.
Chalmers
ZR
,
Connelly
CF
,
Fabrizio
D
,
Gay
L
,
Ali
SM
,
Ennis
R
, et al
Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden
.
Genome Med
2017
;
9
:
34
.
85.
Evans
RA
,
Diamond
MS
,
Rech
AJ
,
Chao
T
,
Richardson
MW
,
Lin
JH
, et al
Lack of immunoediting in murine pancreatic cancer reversed with neoantigen
.
JCI Insight
2016
;
1
:
e88328
.
86.
Yarchoan
M
,
Hopkins
A
,
Jaffee
EM
. 
Tumor mutational burden and response rate to pd-1 inhibition
.
N Engl J Med
2017
;
377
:
2500
1
.
87.
Van Audenaerde
JRM
,
Roeyen
G
,
Darcy
PK
,
Kershaw
MH
,
Peeters
M
,
Smits
ELJ
. 
Natural killer cells and their therapeutic role in pancreatic cancer: a systematic review
.
Pharmacol Ther
2018
;
189
:
31
44
.
88.
Vaheri
A
,
Enzerink
A
,
Rasanen
K
,
Salmenpera
P
. 
Nemosis, a novel way of fibroblast activation, in inflammation and cancer
.
Exp Cell Res
2009
;
315
:
1633
8
.
89.
Heneberg
P
. 
Paracrine tumor signaling induces transdifferentiation of surrounding fibroblasts
.
Crit Rev Oncol Hematol
2016
;
97
:
303
11
.
90.
Van Audenaerde
JRM
,
De Waele
J
,
Marcq
E
,
Van Loenhout
J
,
Lion
E
,
Van den Bergh
JMJ
, et al
Interleukin-15 stimulates natural killer cell-mediated killing of both human pancreatic cancer and stellate cells
.
Oncotarget
2017
;
8
:
56968
79
.
91.
Trotta
R
,
Dal Col
J
,
Yu
J
,
Ciarlariello
D
,
Thomas
B
,
Zhang
X
, et al
Tgf-beta utilizes smad3 to inhibit cd16-mediated ifn-gamma production and antibody-dependent cellular cytotoxicity in human nk cells
.
J Immunol
2008
;
181
:
3784
92
.
92.
Takeuchi
S
,
Baghdadi
M
,
Tsuchikawa
T
,
Wada
H
,
Nakamura
T
,
Abe
H
, et al
Chemotherapy-derived inflammatory responses accelerate the formation of immunosuppressive myeloid cells in the tissue microenvironment of human pancreatic cancer
.
Cancer Res
2015
;
75
:
2629
40
.
93.
Bayne
LJ
,
Beatty
GL
,
Jhala
N
,
Clark
CE
,
Rhim
AD
,
Stanger
BZ
, et al
Tumor-derived granulocyte-macrophage colony-stimulating factor regulates myeloid inflammation and t cell immunity in pancreatic cancer
.
Cancer Cell
2012
;
21
:
822
35
.
94.
Pylayeva-Gupta
Y
,
Lee
KE
,
Hajdu
CH
,
Miller
G
,
Bar-Sagi
D
. 
Oncogenic kras-induced GM-CSF production promotes the development of pancreatic neoplasia
.
Cancer Cell
2012
;
21
:
836
47
.
95.
Waghray
M
,
Yalamanchili
M
,
Dziubinski
M
,
Zeinali
M
,
Erkkinen
M
,
Yang
H
, et al
GM-CSF mediates mesenchymal-epithelial cross-talk in pancreatic cancer
.
Cancer Discov
2016
;
6
:
886
99
.
96.
Galic
MA
,
Riazi
K
,
Pittman
QJ
. 
Cytokines and brain excitability
.
Front Neuroendocrinol
2012
;
33
:
116
25
.
97.
Haroon
E
,
Miller
AH
,
Sanacora
G
. 
Inflammation, glutamate, and glia: a trio of trouble in mood disorders
.
Neuropsychopharmacology
2017
;
42
:
193
215
.
98.
Vesce
S
,
Rossi
D
,
Brambilla
L
,
Volterra
A
. 
Glutamate release from astrocytes in physiological conditions and in neurodegenerative disorders characterized by neuroinflammation
.
Int Rev Neurobiol
2007
;
82
:
57
71
.
99.
Biankin
AV
,
Waddell
N
,
Kassahn
KS
,
Gingras
MC
,
Muthuswamy
LB
,
Johns
AL
, et al
Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes
.
Nature
2012
;
491
:
399
405
.
100.
Demir
IE
,
Friess
H
,
Ceyhan
GO
. 
Neural plasticity in pancreatitis and pancreatic cancer
.
Nat Rev Gastroenterol Hepatol
2015
;
12
:
649
59
.
101.
Long
Y
,
Zhuang
K
,
Ji
Z
,
Han
Y
,
Fei
Y
,
Zheng
W
, et al
2-deoxy-d-glucose exhibits anti-seizure effects by mediating the netrin-g1-katp signaling pathway in epilepsy
.
Neurochem Res
2019
;
44
:
994
1004
.
102.
Brichkina
A
,
Bertero
T
,
Loh
HM
,
Nguyen
NT
,
Emelyanov
A
,
Rigade
S
, et al
P38mapk builds a hyaluronan cancer niche to drive lung tumorigenesis
.
Genes Dev
2016
;
30
:
2623
36
.
103.
Alspach
E
,
Flanagan
KC
,
Luo
X
,
Ruhland
MK
,
Huang
H
,
Pazolli
E
, et al
P38mapk plays a crucial role in stromal-mediated tumorigenesis
.
Cancer Discov
2014
;
4
:
716
29
.
104.
Yamamura
Y
,
Asai
N
,
Enomoto
A
,
Kato
T
,
Mii
S
,
Kondo
Y
, et al
Akt-girdin signaling in cancer-associated fibroblasts contributes to tumor progression
.
Cancer Res
2015
;
75
:
813
23
.
105.
Jiang
X
,
Xie
H
,
Dou
Y
,
Yuan
J
,
Zeng
D
,
Xiao
S
. 
Expression and function of fra1 protein in tumors
.
Mol Biol Rep
2020
;
47
:
737
52
.
106.
Atsaves
V
,
Leventaki
V
,
Rassidakis
GZ
,
Claret
FX
. 
Ap-1 transcription factors as regulators of immune responses in cancer
.
Cancers
2019
;
11
:
1037
.
107.
Bhagat
TD
,
Von Ahrens
D
,
Dawlaty
M
,
Zou
Y
,
Baddour
J
,
Achreja
A
, et al
Lactate-mediated epigenetic reprogramming regulates formation of human pancreatic cancer-associated fibroblasts
.
Elife
2019
;
8
:
e50663
.
108.
Busch
S
,
Andersson
D
,
Bom
E
,
Walsh
C
,
Stahlberg
A
,
Landberg
G
. 
Cellular organization and molecular differentiation model of breast cancer-associated fibroblasts
.
Mol Cancer
2017
;
16
:
73
.
109.
Qin
X
,
Jiang
B
,
Zhang
Y
. 
4e-bp1, a multifactor regulated multifunctional protein
.
Cell Cycle
2016
;
15
:
781
6
.
110.
Duluc
C
,
Moatassim-Billah
S
,
Chalabi-Dchar
M
,
Perraud
A
,
Samain
R
,
Breibach
F
, et al
Pharmacological targeting of the protein synthesis mtor/4e-bp1 pathway in cancer-associated fibroblasts abrogates pancreatic tumour chemoresistance
.
EMBO Mol Med
2015
;
7
:
735
53
.
111.
Franco-Barraza
J
,
Beacham
DA
,
Amatangelo
MD
,
Cukierman
E
. 
Preparation of extracellular matrices produced by cultured and primary fibroblasts
.
Curr Protoc Cell Biol
2016
;
71
:
10 9 1–10 9 34
.
112.
Beglyarova
N
,
Banina
E
,
Zhou
Y
,
Mukhamadeeva
R
,
Andrianov
G
,
Bobrov
E
, et al
Screening of conditionally reprogrammed patient-derived carcinoma cells identifies ercc3-myc interactions as a target in pancreatic cancer
.
Clin Cancer Res
2016
;
22
:
6153
63
.
113.
Ritchie
ME
,
Phipson
B
,
Wu
D
,
Hu
Y
,
Law
CW
,
Shi
W
, et al
Limma powers differential expression analyses for rna-sequencing and microarray studies
.
Nucleic Acids Res
2015
;
43
:
e47
.
114.
Merico
D
,
Isserlin
R
,
Stueker
O
,
Emili
A
,
Bader
GD
. 
Enrichment map: a network-based method for gene-set enrichment visualization and interpretation
.
PLoS One
2010
;
5
:
e13984
.
115.
Wang
X
,
Spandidos
A
,
Wang
H
,
Seed
B
. 
Primerbank: a pcr primer database for quantitative gene expression analysis, 2012 update
.
Nucleic Acids Res
2012
;
40
:
D1144
9
.
116.
Rezakhaniha
R
,
Agianniotis
A
,
Schrauwen
JT
,
Griffa
A
,
Sage
D
,
Bouten
CV
, et al
Experimental investigation of collagen waviness and orientation in the arterial adventitia using confocal laser scanning microscopy
.
Biomech Model Mechanobiol
2012
;
11
:
461
73
.
117.
Sanjana
NE
,
Shalem
O
,
Zhang
F
. 
Improved vectors and genome-wide libraries for crispr screening
.
Nat Methods
2014
;
11
:
783
84
.
118.
Thakore
PI
,
D'Ippolito
AM
,
Song
L
,
Safi
A
,
Shivakumar
NK
,
Kabadi
AM
, et al
Highly specific epigenome editing by crispr-cas9 repressors for silencing of distal regulatory elements
.
Nat Methods
2015
;
12
:
1143
9
.
119.
Horlbeck
MA
,
Gilbert
LA
,
Villalta
JE
,
Adamson
B
,
Pak
RA
,
Chen
Y
, et al
Compact and highly active next-generation libraries for crispr-mediated gene repression and activation
.
Elife
2016
;
5
:
e19760
.
120.
Lewis
CA
,
Parker
SJ
,
Fiske
BP
,
McCloskey
D
,
Gui
DY
,
Green
CR
, et al
Tracing compartmentalized nadph metabolism in the cytosol and mitochondria of mammalian cells
.
Mol Cell
2014
;
55
:
253
63
.
121.
Young
JD
,
Walther
JL
,
Antoniewicz
MR
,
Yoo
H
,
Stephanopoulos
G
. 
An elementary metabolite unit (emu) based method of isotopically nonstationary flux analysis
.
Biotechnol Bioeng
2008
;
99
:
686
99
.
122.
Fernandez
CA
,
Des Rosiers
C
,
Previs
SF
,
David
F
,
Brunengraber
H
. 
Correction of 13c mass isotopomer distributions for natural stable isotope abundance
.
J Mass Spectrom
1996
;
31
:
255
62
.
123.
Kim
D
,
Pertea
G
,
Trapnell
C
,
Pimentel
H
,
Kelley
R
,
Salzberg
SL
. 
Tophat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions
.
Genome Biol
2013
;
14
:
R36
.
124.
Anders
S
,
Pyl
PT
,
Huber
W
. 
Htseq–a python framework to work with high-throughput sequencing data
.
Bioinformatics
2015
;
31
:
166
9
.
125.
Love
MI
,
Huber
W
,
Anders
S
. 
Moderated estimation of fold change and dispersion for rna-seq data with deseq2
.
Genome Biol
2014
;
15
:
550
.
126.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
127.
Hanzelmann
S
,
Castelo
R
,
Guinney
J
. 
Gsva: gene set variation analysis for microarray and rna-seq data
.
BMC Bioinformatics
2013
;
14
:
7
.
128.
Liberzon
A
,
Birger
C
,
Thorvaldsdottir
H
,
Ghandi
M
,
Mesirov
JP
,
Tamayo
P
. 
The molecular signatures database (msigdb) hallmark gene set collection
.
Cell Syst
2015
;
1
:
417
25
.
129.
Nagy
A
,
Lanczky
A
,
Menyhart
O
,
Gyorffy
B
. 
Validation of mirna prognostic power in hepatocellular carcinoma using expression data of independent datasets
.
Sci Rep
2018
;
8
:
9227
.
130.
Liu
J
,
Lichtenberg
T
,
Hoadley
KA
,
Poisson
LM
,
Lazar
AJ
,
Cherniack
AD
, et al
An integrated tcga pan-cancer clinical data resource to drive high-quality survival outcome analytics
.
Cell
2018
;
173
:
400
16
.