A Stromal Lysolipid–Autotaxin Signaling Axis Promotes Pancreatic Tumor Progression

Pancreatic ductal adenocarcinoma (PDAC) is an aggressive cancer that, despite a low incidence, is projected to be the second highest contributor to cancer deaths by 2030 (1). PDAC progression typically features a dramatic desmoplastic reaction, including fibroblasts, immune cells, and a dense extracellular matrix (2). The profound matrix deposition in the PDAC microenvironment physically impedes the vasculature and blocks delivery of intravenously injected molecules or of chemotherapeutic agents (3–5), raising the possibility that serum nutrients are also limited in this context. Consistent with this possibility, our previous metabolomic analysis of surgically resected human PDAC demonstrated that these tumors are nutrient-poor (6), leaving in question how PDAC cells maintain proliferative capacity within this nutrient-restricted microenvironment. PDAC cells were previously shown to scavenge extracellular protein and lipid to support proliferation (7–9). Under the hypovascular conditions of the PDAC microenvironment, we reasoned that PDAC cells may stimulate paracrine metabolite transfer from neighboring nonmalignant cells to support their metabolism and growth, analogous to a recently described axis in metastatic melanoma (10).

The PDAC desmoplastic reaction is established by a prominent population of cancer-associated fibroblasts (CAF). This CAF population derives in part from tissue-resident pancreatic stellate cells (PSC), which differentiate to a CAF phenotype upon activating signals from the neoplastic compartment during pancreatic tumorigenesis (11, 12). In healthy pancreatic tissue, PSCs are in a quiescent state, characterized by a limited secretome and cytoplasmic lipid droplets (13). Malignant cells signal in a paracrine manner to induce PSC activation to a CAF phenotype, and this is associated with loss of characteristic lipid droplets. Though this lipid droplet loss during stellate cell activation has long been appreciated, in-depth, lipidome-wide investigations of intracellular remodeling and potential associated roles in the PDAC microenvironment have not yet been addressed. We aimed to understand the significance of this stromal lipid remodeling during PDAC progression, and to address the possibility that stroma-derived lipids support PDAC metabolism and growth. In the present study, we describe a novel role for PSC-derived CAFs as a source of specific, growth-permissive lipids in the PDAC microenvironment.

Mining of previously published RNA-sequencing data revealed that PSCs undergo significant transcriptional alterations in genes involved in lipid metabolism (14), including a broad downregulation of genes implicated in lipid storage (Supplementary Fig. S1A), coincident with acquisition of an activated, fibroblastic phenotype (Supplementary Fig. S1B). This includes reduced expression of the fatty acid carrier FABP4 and the lipid droplet coat proteins PLIN1 and PLIN2 (Fig. 1A), as well as a depletion of the lipid droplet-resident triglycerides (Supplementary Fig. S1C). PSCs drastically reduce their capacity to store neutral lipids upon activation, as a near-complete depletion of lipid droplets in activated PSCs was directly confirmed by imaging (Fig. 1B). Loss of lipid droplets during PSC activation required autophagy rather than neutral lipases (Supplementary Fig. S1D), consistent with previous findings in other contexts (15, 16). To more broadly analyze lipid metabolic changes, we performed intracellular lipidomics and found that PSCs undergo extensive lipidomic remodeling during activation, beyond loss of neutral lipids. Significant changes included a sharp rise in intracellular levels of lysophospholipids, particularly lysophosphatidylcholines (LPC; Fig. 1C; Supplementary Table S1). RAS-transformed cells were previously shown to take up extracellular lipids to support fatty acid scavenging and proliferative capacity, and of all lipids measured, this precise lipid class—LPCs—was the preferred scavenging substrate (9). Serum lipids were probed in this prior study, but we reasoned that PSC-derived CAFs may serve as a cellular source for these lipids in the PDAC microenvironment, which would align with the metabolite and protein secretory phenotype that these cells adopt. We thus asked whether activated PSCs secrete lysophospholipids and other lipid species as a potential fuel source for PDAC cells. Conditioned media (CM) from primary, activated mouse PSCs was analyzed by LC/MS, which revealed the presence of 230 unique secreted lipids including various (lyso)phospholipids, fatty acids, sphingolipids, and neutral lipids (Fig. 1D; Supplementary Table S1). Similar lipid secretomes were observed for permanently activated, immortalized mouse PSCs (ImPSC) and immortalized human PDAC-derived CAFs (0082T; Fig. 1E; Supplementary Fig. S1E; Supplementary Table S1) with respect to lysophospholipid species of interest and their levels, though some differences were noted. Differentiating primary PSCs into either inflammatory CAFs (iCAF) or myofibroblastic CAFs (myCAF) as previously described (12) yielded similar levels of LPC production (Supplementary Fig. S1F and S1G). To determine whether secreted lipids act in a paracrine manner to promote the growth of PDAC cells, we incubated mouse or human PDAC cells with CM from immortalized mouse PSCs or from primary CAFs. Consistent with earlier results, stromal cell CM elicited a concentration-dependent increase in PDAC cell proliferation (refs. 6, 17, 18; Supplementary Fig. S1H). This effect was specific to stromal cells, as PDAC CM failed to induce growth (Supplementary Fig. S1I). Notably, PSC CM or the CM lipid fraction alone normalized PDAC cell growth in fatty acid–limiting conditions, characterized by low serum lipid availability and fatty acid desaturation inhibition to model conditions in which PDAC cells cannot make all the lipids needed for proliferation (Supplementary Fig. S1J and S1K), strongly suggesting that stroma-derived lipids can support PDAC cell metabolism and growth.

To conclusively demonstrate a paracrine lipid flux from PSCs to PDAC cells, and to determine their metabolic fate, we performed a qualitative stable isotope tracing experiment by incubating PSCs with ¹³C-labeled palmitate and oleate to label secreted lipids (Fig. 2A; Supplementary Fig. S2A). Lipidomic analysis showed significant accumulation of ¹³C-labeled, stroma-derived fatty acids in PDAC cells (Fig. 2B), in both the phospholipid and triglyceride pools. This demonstrates that PSC-derived lipids are taken up by PDAC cells and channeled to various lipid pools, including phospholipids for membrane synthesis and growth. We next investigated specific lipid classes that may support PDAC growth and focused on LPCs, as they are avidly consumed by tumor cells (9), and because they are abundantly secreted by PSCs (Fig. 2C; Supplementary Fig. S2A). Although PSCs release LPCs, PDAC cells do not, consistent with PDAC cell avidity for these lipids. Tracing experiments demonstrated that PSCs can produce LPCs from glucose and glutamine; however, incorporation of glucose- and glutamine-derived carbons into LPCs was suppressed in the presence of free fatty acids, suggesting that fatty acids are readily used for LPC synthesis when available (Supplementary Fig. S2B). To investigate the fate of LPCs upon uptake by PDAC cells, LPC 17:1 was used as a tracer, which resulted in significant 17:1 incorporation into phosphatidylcholine species which comprise cell membranes (PC 16:0/17:1 and PC 18:1/17:1), supporting the notion that LPCs are used by PDAC cells for membrane synthesis (Supplementary Fig. S2C). To determine whether activated PSCs or CAFs serve as the principal cellular source of LPCs in the PDAC tumor microenvironment, we isolated CAFs, leukocytes, or remaining cell types (PDAC cells, endothelial cells, other minor cell populations) by FACS, subjected these three populations to brief ex vivo culture, harvested supernatant, and analyzed LPC levels. LC/MS revealed that CAFs are the major producers of LPCs on a per-cell basis within the PDAC microenvironment (Fig. 2D; Supplementary Fig. S2D). The abundance of PDAC CAFs suggests that they are a significant source of these lysophospholipids in vivo, though we note that they are likely not the exclusive source.

In addition to uptake, LPCs can be hydrolyzed in the extracellular space by the secreted enzyme autotaxin to give rise to lysophosphatidic acid (LPA; Fig. 2E). LPAs function as potent extracellular proliferation- and migration-inducing signals with established roles in cancer (19), and we noticed both LPA and autotaxin in the CM of cultured PSCs (Fig. 1D and E; Fig. 2F). PDAC cells also released autotaxin into their CM, and autotaxin secretion by PDAC cells was markedly increased in a paracrine manner by PSCs (Fig. 2G); this induction was similar by PSCs differentiated into either iCAFs or myCAFs (Supplementary Fig. S2E). Although the lipid fraction of PSC CM was not sufficient to induce autotaxin, boiled CM was (Supplementary Fig. S2F), raising the possibility that a metabolite or small peptide is responsible for paracrine regulation of autotaxin. Autotaxin inhibition with HA130 led to a drastic reduction in CM LPA levels (Supplementary Fig. S2G). Western blot results agreed with autotaxin activity assays (Supplementary Fig. S2H). Thus, lipids secreted by PSCs are directly used by PDAC cells to support membrane synthesis and act as substrates for the production of signaling lipids, in part through autotaxin-mediated hydrolysis of stroma-derived LPC.

Although LPA levels in PSC CM were in the nmol/L range, this is likely underrepresentative of actual LPA-producing capacity in the tumor microenvironment, in part due to the limited autotaxin production in homotypic PSC cultures (Fig. 2G). In addition, as the k_d for LPAR1 is ∼10 nmol/L (20), we considered that even low LPA concentrations might have potent biological effects. We therefore continued with determining the effect of LPAs on PDAC cells. LPA receptors LPAR1 and LPAR2 were found to be consistently expressed in PDAC cells (Supplementary Fig. S3A), whereas other LPARs were not detected. LPA–LPAR engagement has been shown to activate mitogenic signaling, including the PI3K/AKT pathway (19), and PSC CM led to potent activation of AKT in PDAC cells (Fig. 3A; Supplementary Fig. S3B). Notably, engagement of the AKT pathway by PSC CM has also been reported by others (21), although the underlying molecular mechanism remains ill-defined. We found here that paracrine AKT induction was largely abolished when inhibiting autotaxin (Fig. 3A; Supplementary Fig. S3B) whereas this effect was rescued with the addition of LPA, suggesting a central role for the stroma-derived LPC–autotaxin–LPA axis in paracrine regulation of AKT signaling. Similar effects were observed for the mitogenic MAPK pathway (Supplementary Fig. S3C).

Exogenous LPC or LPA was sufficient to induce proliferation, even in the absence of stromal growth factors (Fig. 3B; Supplementary Fig. S3D), implicating the stroma-derived LPC–autotaxin–LPA axis in PDAC cell proliferation. Consistent with this, paracrine induction of PDAC cell proliferation was abolished upon autotaxin inhibition (Fig. 3C; Supplementary Fig. S3E and S3F), an effect rescued by LPA, whereas autotaxin overexpression augmented stroma-inducible proliferation. In addition, proliferation induced by the extracted lipid fraction of PSC CM was also significantly reduced upon autotaxin inhibition (Fig. 3D). To further probe the significance of this axis in growth-permissive tumor–stroma cross-talk, PDAC cells and PSCs were directly cocultured (Fig. 3E). Direct coculture of PDAC cells with two separate PSC lines increased PDAC cell proliferation by nearly 2-fold compared with monoculture (Supplementary Fig. S3G). Strikingly, ablating LPA production with the autotaxin inhibitor HA130 was sufficient to cause a near 50% drop in this induced proliferation. This indicates that stromal LPA signaling is maintained in coculture, and highlights the important role of this pathway in the proliferation-inducing effects of PSCs (Fig. 3F).

In addition to proliferation, LPAs can promote migration (19). Accordingly, PSC CM promoted migration of PDAC cells, whereas CM from PDAC cells did not (Fig. 3G; Supplementary Fig. S3H and S3I). The stromal LPA axis was the primary contributor to this effect, as HA130 or LPAR inhibitor Ki16425 blocked this effect (Fig. 3H; Supplementary Fig. S3J). Autophagy-dependent alanine secretion by PSCs was recently shown to support PDAC cell mitochondrial metabolism (17). Autophagy inhibition in PSCs did not impair paracrine induction of PDAC cell migration (Supplementary Fig. S4A and S4B), and exogenous alanine did not stimulate PDAC cell migration or proliferation under our experimental conditions in these cells (Supplementary Fig. S4C and S4D), highlighting distinct functions of stroma-derived alanine and the lysophospholipids under present investigation. Collectively, these data show that stromal LPA potently induces proliferation and migration of PDAC cells.

We next established the occurrence and importance of this mechanism in vivo. We noted that autotaxin expression was limited to low expression in ducts in healthy pancreas, but highly abundant in mouse and human pancreatic tumors (Fig. 4A and B; Supplementary Fig. S5A). Interestingly, most autotaxin appeared to be on the outside of the PDAC cells rather than the stromal cells, and this may constitute a mechanism for generating high LPA levels in the vicinity of PDAC cell LPARs (Supplementary Fig. S5B). Consistently, we have found that, although both cell types can produce autotaxin in homotypic cell cultures, PDAC cells produce more autotaxin than stromal cells in vivo (Supplementary Fig. S5C). Together with the results in Fig. 2G supporting stroma-inducible autotaxin expression by PDAC cells, these results suggest that LPA measurements in PSC or CAF CM are likely artificially low. We performed subcutaneous cotransplantation assays in immune-compromised mice with both PDAC cells and PSCs and found that PSCs increased both tumor growth (Supplementary Fig. S5D) and, remarkably, intratumoral LPA levels (Fig. 4C; Supplementary Table S2). To further assess PSCs as the source of autotaxin substrate in the PDAC microenvironment, we performed subcutaneous transplantation of PDAC cells and PSCs in immune-competent hosts. KPC-derived PDAC cells were previously shown to recruit a robust innate and adaptive immune response in syngeneic hosts in the subcutaneous setting (22, 23), allowing us to address the role of pancreas-resident PSCs in complex tumor microenvironments. Autotaxin inhibition with the bioavailable inhibitor ONO-8430506 (24) suppressed PDAC growth to a greater extent in the presence of PSCs (Fig. 4D); together with Fig. 2D, these results implicate PSCs as a key source of autotaxin substrate in vivo. We next assessed the significance of the LPC–autotaxin–LPA axis for PDAC growth by orthotopically injecting the pancreata of immune-competent mice with PDAC cells, which generates a spontaneous stromal response. We then administered autotaxin inhibitor or vehicle control and monitored tumor growth. Autotaxin inhibition reduced tumor growth by ∼2-fold following orthotopic injection of two different mouse PDAC cell lines (Fig. 4E; Supplementary Fig. S5E), and significantly reduced proliferation (Supplementary Fig. S5F and S5G) and AKT phosphorylation (Fig. 4F) while significantly increasing apoptosis (Supplementary Fig. S5H). To further assess the role of autotaxin in PDAC growth, two CRISPR-mediated Enpp2 (the gene encoding autotaxin) knockdown lines were generated (Supplementary Fig. S5I). Genetic disruption of Enpp2 in PDAC cells significantly reduced stroma-inducible proliferation in vitro (Supplementary Fig. S5J), and though these Enpp2 knockdown lines show proliferation defects of 36% (A11) or 59% (A18) under standard culture conditions in vitro, transplantation of these cells into pancreata of immune-competent hosts showed severely compromised tumor growth (Fig. 4G). Importantly, proliferation was rescued with the addition of autotaxin (Supplementary Fig. S5K). These PDAC cell–restricted genetic inhibition results highlight the importance of local autotaxin function and LPA production for pancreatic tumor growth. Together, these results support a role for stromal lysophospholipid secretion in PDAC progression.

Here we show that PSCs undergo a dramatic lipid metabolic shift during differentiation in the context of pancreatic tumorigenesis, including remodeling of the intracellular lipidome. Particularly interesting in the context of this study is the observation that intracellular LPCs are strongly elevated upon activation, and PSC-derived CAFs secrete abundant LPCs in the activated, fibroblastic state. Lipid release by PSCs promotes multiple hallmarks of aggressive PDAC progression, including PDAC cell proliferation and migration. Our work demonstrates that PSCs contribute to PDAC progression through the release of not only growth factors and alanine (11, 17, 25), but also specific lipid species. These lipids act as a source for uptake and biomass production and as precursors for signaling lipids, particularly LPA (Fig. 4H). The origin of intratumoral LPA in PDAC has thus far remained unclear, but it was presumed to come directly from the circulation or to be made locally from circulating LPC. Although these are likely sources, poorly perfused tumors, such as PDAC (3–6), may experience limited exposure to this signaling mediator via these routes. Instead, we find that the fibroblast-like PSCs profusely release the LPA precursor LPC, which is metabolized in the direct vicinity of the PDAC cells, and our results highlight the importance of this local LPC generation and hydrolysis as an important source of LPA (Fig. 4C, D and G). Moreover, in a tumor setting, PDAC cells appear to actively co-opt this signaling axis by activating neighboring PSCs, leading to lipidomic remodeling coincident with LPC release, and by expressing both autotaxin and LPA receptors. LPA production by autotaxin may also contribute to the metabolic symbiosis between tumor and stroma, and an improved understanding of lipid uptake mechanisms will be helpful in parsing out the relative contributions of lipid uptake versus signaling in paracrine regulation of tumor growth by stromal lipid cues. Our model fits the paradigm that PDAC tumors deregulate and exploit normal wound-healing processes for their benefit.

Previous studies using genetic or pharmacologic means to ablate CAFs during pancreatic tumorigenesis revealed that PDAC that evolves in a CAF-deficient microenvironment is highly aggressive and more lethal than CAF-replete controls (26–28). Together with reports of tumor-supportive functions for PDAC CAFs, these papers suggest that CAFs act to suppress tumor growth by creating a hypovascular and nutrient-poor microenvironment that is challenging for proliferation, but at the same time support tumor growth by secreting factors that enable PDAC cells to proliferate and survive in this nutrient-poor, wound-like context. Although stromal ablation therapy for PDAC remains controversial, our results suggest a potential therapeutic opportunity in maintaining the stromal reaction, while identifying and targeting the specific pathways that enable growth under the nutrient-deprived conditions of an intact tumor microenvironment.

Cell lines were from ATCC with the following exceptions: hPSC-T cells were kindly provided by R.F. Hwang and have been previously described (18). The 0082T cell line was derived from primary CAF sample 0082, provided by Dr. Andrew Lowy (University of California, San Diego) and prepared from human PDAC surgical specimens by the outgrowth method (29, 30). The cells were confirmed to lack KRAS exon 2 mutations and were immortalized with SV40 large T antigen (pLenti-SV40-T, Applied Biological Materials Inc.) at a multiplicity of infection of 2. Primary CAF lines CAF 4414, CAF 4586, and CAF 4442 were initiated from disaggregated primary PDAC tumor tissue and isolated from epithelial cells by differential trypsinization. IHC and RNA sequencing were used to confirm minimal expression of cytokeratin and abundant expression of αSMA in both CAF lines. Wild-type KRAS codon 12 sequence was confirmed for both CAF lines, and mutant KRAS-G12 was confirmed in both parent tumors. Primary CAF lines were generated in accordance with protocols approved by the Institutional Review Boards of the University of California, San Diego and Oregon Health & Science University. FC1199 (provided by Dr. David Tuveson, Cold Spring Harbor Laboratory, New York) and 4662 (provided by Dr. Robert Vonderheide, University of Pennsylvania, Philadelphia) PDAC cell lines were derived from primary tumors in KPC mice. The p53 2.1.1 PDAC cell line was described previously (31). Mouse primary PSC isolation and immortalization was performed as described below. All cell lines were routinely passaged in Dulbecco's Modified Eagle Medium (DMEM; Thermo Fisher Scientific) containing 10% fetal bovine serum (FBS; HyClone), 25 mmol/L glucose, and 2 mmol/L glutamine for no more than 25 to 30 passages, whereas primary CAFs were used for a maximum of 10 passages, and cells were routinely tested for Mycoplasma at least monthly (MycoAlert Detection Kit, Lonza; latest test before submission January 28, 2019). Cell line authentication was not performed.

Primary PSCs were isolated from mouse pancreata as previously described (14). Briefly, healthy pancreata were harvested from 8-week-old male C57BL/6J mice. Tissues were minced with scissors and digested with 0.02% Pronase (Roche), 0.05% Collagenase P (Roche), and 0.1% DNase (Roche) in Gey's balanced salt solution (GBSS; Sigma-Aldrich) at 37°C for 20 minutes. After dissociation, tissue was triturated until large pieces were no longer visible, and the resulting cell suspension was filtered through a 100-μm nylon mesh. Cells were then washed with GBSS, pelleted, and resuspended in 9.5 mL GBSS containing 0.3% bovine serum albumin (BSA) and 8 mL 28.7% Nycodenz solution (Sigma-Aldrich). The cell suspension was layered beneath GBSS containing 0.3% BSA, and centrifuged at 1,400 × g for 20 minutes at 4°C. Stellate cells were harvested from the interface of the Nycodenz solution at the bottom and the aqueous solution at the top. Isolated PSCs were washed with GBSS and resuspended in DMEM containing 10% characterized FBS (HyClone) and antibiotics (100 U/mL penicillin and 100 μg/mL streptomycin; Invitrogen). Cells were maintained at 37°C in a standard tissue culture incubator. After reaching 80% confluence, cells were trypsinized (0.25% Trypsin-EDTA, Invitrogen) and passaged. ImPSC1 and ImPSC2 cell lines were obtained by immortalizing primary PSCs with the pRetro.Super.shARF retroviral plasmid (kindly provided by the Karen Vousden lab) and then selected with blasticidin at a concentration of 4 μg/mL.

For cellular lipids, at the time of collection primary PSCs were washed twice with ice-cold PBS (1 mL), cold (−20°C) PBS/MeOH (1:1; 0.75 mL) was added to the dish, and the cells were scraped and transferred to a microfuge tube and incubated at −20°C for 20 minutes. A mix of the following internal standards (Avanti Polar Lipids) was added to the final indicated concentrations, prior to cell scraping: FA(19:0) at 6 μmol/L, TG(19:0/19:0/19:0) at 80 nmol/L, CE(19:0) at 3 μmol/L, PC(17:0/14:1) at 112 nmol/L, PE(17:0/14:1) at 112 nmol/L, and LPC(17:1) at 4 μmol/L. Cold (−20°C) chloroform (0.5 mL) was added, and the samples vortexed for 1 minute and centrifuged for 10 minutes at 18,000 × g. The lower chloroform phase was transferred to glass vials using a Hamilton syringe and dried under nitrogen gas. Dried samples were reconstituted in CHCl₃/MeOH (1:1) at 1 mL per 1.5 × 10⁶ cells.

For medium apolar lipids, 0.5 mL samples were extracted by adding an equal volume of methanol containing 0.2 mol/L HCl, adding the same internal standards as above and then mixing with an equal volume of cold (−20°C) chloroform. Samples were vortexed for 1 minute and centrifuged for 10 minutes at 18,000 × g. The lower chloroform phase was transferred to glass vials using a Hamilton syringe and dried under nitrogen gas. Dried samples were reconstituted in 100 μL CHCl₃/MeOH (1:1). For specific accurate quantitation of CM lysophosphatidylcholines (LPC), extraction was performed as for apolar lipids, but only the internal standard LPC (17:1) at a final concentration of 4 μmol/L was added, and the samples were reconstituted in 1 mL of methanol. For medium polar lipids, 1 mL of butanol was added to 750 μL of CM from PSCs containing the internal standards LPC(17:1) at 4 μmol/L, FA(19:0) at 6 μmol/L, LPA(17:0) at 10 nmol/L. Samples were vortexed for 10 minutes and centrifuged at 20,000 × g for 5 minutes. The top butanol layer was transferred to glass vials, and the extraction was repeated once. The pooled extracts were dried under nitrogen gas and reconstituted in 100 μL of methanol.

For tumor tissue LPA analysis, tissue samples were extracted as described earlier (32) but with minor modifications. Briefly, 30 to 50 mg of the tissue sample was lysed in 0.5 mL of citrate phosphate buffer (15 mmol/L citrate, 20 mmol/L Na₂HPO₄, pH 4) containing the internal standards LPC(17:1) at 4 μmol/L, FA(19:0) at 6 μmol/L, LPA(17:0) at 10 nmol/L, using a Precellys evolution tissue homogenizer (Bertin instruments). Following this, 1 mL of butanol was added to the lysate, vortexed for 10 minutes, and centrifuged at high speed for 10 minutes to recover the butanol layer. This extraction was repeated twice, and the extracts were pooled and dried under nitrogen gas. The dried extracts were reconstituted in 100 μL of methanol.

Apolar lipids in reconstituted samples from both cells and medium were analyzed using a Q Exactive Orbitrap mass spectrometer coupled to a Dionex UltiMate 3000 LC system (Thermo Scientific). The LC parameters were as follows: 5 μL of sample was injected onto a 1.7-μm particle 100 × 2.1 mm ID Waters Acquity CSH C18 column (Waters), which was kept at 50°C. A gradient of (A) water/acetonitrile (40:60, v/v) with 10 mmol/L ammonium formate and (B) acetonitrile/2-propanol (10:90, v/v) with 10 mmol/L ammonium formate at a flow rate of 0.3 mL/minute was used. The gradient ran from 0% to 40% B over 6 minutes, then from 40% to 100% B in the next 24 minutes, followed by 100% B for 4 minutes, and then returned to 0% B in 2 minutes where it was kept for 4 minutes (40 minutes total). The electrospray and mass spec settings were as follows: spray voltage 3 kV (positive mode), capillary temperature 300°C, sheath gas flow 50 (arbitrary units), auxiliary gas flow 7 (arbitrary units), and sweep gas flow 5 (arbitrary units). The resolution was set to 70,000, automatic gain control to 1 × 10⁶ with the maximum injection time of 200 ms and the scan range of 300 to 900 m/z. The peak areas for the different lipid species were extracted and normalized to their respective internal standards.

Polar lipid and LPC (for accurate quantification) extracts reconstituted in methanol were analyzed using the same LC/MS system and settings as above, but with the following differences: Lipids were separated with the solvent gradient system containing (A) water and 0.05% NH₄OH and (B) methanol and 0.05% NH₄OH at the flow rate 300 μL/minute. The separation of lipids started with isocratic elution for 2 minutes and 50% B followed by a linear gradient from 50% to 70% B over the next 10 minutes, and 70% to 100% B over the next 8 minutes. Thereafter, the column was eluted with 100% B for 4 minutes and returned to initial conditions and equilibrated for 6 minutes. The peak area for the different lipid species was extracted and normalized to their respective internal standards. For both apolar and polar lipid analyses, data were analyzed with MAVEN and LipidSearch.

For targeted LPA analysis in both CM and tissues, a Vanquish UHPLC system coupled to TSQ Altis Triple Quadrupole mass spectrometer (Thermo) was used. Five microliters of the sample was loaded on a 1.7-μm particle 100 × 2.1 mm ID Waters Acquity CSH C18 column. Theoven temperature 50°C and the mobile phase consisted of (A) water and 0.05% NH₄OH and (B) methanol and 0.05% NH₄OH. The gradient elution program was as follows: 0–1 minute isocratic 55% B, 1–14 minutes 55%–85% B, 14–20 minutes 85%–100% B, 20–22 minutes 100% B, and 22–26 minutes 55% B. The total run time was 26 minutes, and the flow rate was 250 μL/minute. Heated electrospray ionization (H-ESI) parameters were set as follows: 2.5 kV negative mode of capillary voltage, sheath gas 50 arbitrary units, auxiliary gas 10 arbitrary units, ion transfer tube temperature 300°C, and vaporizer temperature 250°C. LPAs were analyzed in selected reaction monitoring mode and 30 V collision energy. Selected reaction monitoring transitions for LPA molecules were as follows: LPA(16:0) 409.236→152.995, 255.233; LPA(17:0) 423.252→152.995, 269.248; LPA(18:0) 437.267→152.995, 283.264; LPA(18:1) 435.251→152.995, 281.248; LPA(18:2) 433.236→152.995, 279.233; LPA(20:4) 457.236→152.995, 303.233. For each LPA molecule, two product ions were used: 152.995 m/z was used for quantification, and the second ion was used as fatty acid qualifier ion. The peak area for the different LPA species was extracted manually using Thermo Xcalibur 4.0 software and normalized to the peak area of the internal standard.

PSCs were cultured in DMEM containing 10% FBS until 80% confluency was reached. The medium was then changed to DMEM with 0.5% FFAF–BSA and 100 μmol/L of [U¹³C]-palmitate and 100 μmol/L of [U¹³C]-oleate (Sigma). After 24 hours, the medium was changed to DMEM containing 0.5% FFAF–BSA only, and cells were left to condition the medium for 48 hours. The medium was then harvested and concentrated as above, diluted in DMEM containing 0.5% FFAF–BSA to a final concentration of 4 mg/mL and provided to 60% confluent KPC cells. After 48 hours, lipids were extracted from KPC cells and analyzed as above. Data are presented as the percentage of the summed labeled isotopologs relative to all isotopologs in the respective lipids.

Male C57BL/6J mice at 8 weeks of age received orthotopic transplants of 5 × 10³ FC1199 cells. On day 24 after transplantation, tumors were removed and digested for 1 hour in DMEM containing 1 mg/mL collagenase IV, 0.1% soybean trypsin inhibitor, 50 U/mL DNase I, 0.125 mg/mL dispase at 37°C. Tumor digests were pelleted at 300 × g for 5 minutes, then further digested in 0.25% Trypsin-EDTA for 10 minutes at 37°C. Digests were washed with cold DMEM containing 10% FBS and filtered through a 100-μm strainer. The cell suspension was pelleted as above, washed with DMEM plus 10% FBS, then lysed in ACK lysis buffer (Thermo Fisher Scientific) at room temperature for 3 minutes. Cells were washed in FACS buffer (PBS plus 2% FBS), pelleted, and resuspended in FACS buffer at 10⁷ cells/mL. Anti-mouse CD16/CD32 (Fc block, BD #553142) was added at 1:20; after a 2-minute incubation, antibodies against CD45 (anti-CD45-FITC, BD #553080; 1:200) and Podoplanin (biotin anti-Podoplanin, BioLegend #127404; 1:200) were added and incubated on ice for 30 minutes. Cells were washed with FACS buffer, pelleted, and resuspended in FACS buffer containing APC–streptavidin (BD #554067, 1:1,000). After a 30-minute incubation on ice, cells were pelleted and resuspended in FACS buffer. FITC-positive, APC-positive, and double-negative cell populations were identified by FACS on a BD FACSAria Fusion and sorted into FACS buffer. After tumors had been entirely sorted, live cells were counted, and 2 × 10⁵ cells per population were seeded into 6-well plates in 2 mL DMEM containing 0.5% FFAF–BSA. After 12 hours, culture supernatants were harvested, spun down to pellet debris, and snap-frozen for LPC measurement by LC/MS as described above.

The pSpCas9(BB)-2A-Puro(PX459) v2.0 plasmid (Addgene #62988) was used to clone guide sequences targeting Enpp2 per supplier's protocol: sgRNA #1: CTTCCCTAATCTGTATACGC; sgRNA #2: TGGCCAGCGTATACAGATTA. The p53 2.1.1 PDAC cell line was transfected with control plasmid or plasmid containing either of the sgEnpp2 sequences and subject to selection with 2 μg/mL puromycin for 4 days. Single-cell clones were expanded and screened for autotaxin expression by Western blot. We noted that complete knockouts were not identified in any of four PDAC cell lines or with any of four sgRNAs tested. The degree of knockdown is demonstrated in Supplementary Fig. S5I.

The Institutional Animal Care and Use Committees (IACUC) of Oregon Health & Science University (OHSU) and the Beatson Institute approved all animal studies and procedures. For xenografts, 1 × 10⁵ KPC T832043 were injected alone or coinjected with 5 × 10⁵ ImPSC2 in the flanks of CD-1 nude female mice (Charles River). Tumor dimensions were manually taken triweekly, and tumor volume was calculated with the formula (width² × length)/2. For orthotopic transplantations, male C57BL/6J mice at 8–10 weeks of age were anesthetized with ketamine and xylazine, and pancreata were injected with 5 × 10³ FC1199 cells (provided by Dr. David Tuveson) or 1 × 10⁵ 4662 cells (provided by Dr. Robert Vonderheide), both derived from primary PDAC in KPC mice of pure C57BL/6J background. Beginning 14 days after transplantation, pancreata were imaged by high-resolution ultrasound using the Vevo 770 imaging system; mice were enrolled in the study when tumors reached 3–6 mm in diameter, and pretreatment tumor sizes were recorded. Mice were treated twice daily by oral gavage with 30 mg/kg ONO-8430506 dissolved in 7.5 mmol/L NaOH or vehicle control. After 10 days of treatment, posttreatment tumor measurements were performed on the Vevo 770, and mice were euthanized and tissues were harvested for analysis. The control and Enpp2 knockdown cell lines in the p53 2.1.1 background were orthotopically transplanted as above but into 8-week-old male FVB/n mice (5 × 10³ cells per mouse); tumors were imaged by high-resolution ultrasound on days 11 and 21 after transplantation. For the subcutaneous transplants in immune-competent hosts (Fig. 4D), 8-week-old male C57BL/6J mice were injected with 1 × 10⁵ FC1199 cells with or without 5 × 10⁵ ImPSC1 cells into the left flank. Recipients were treated with vehicle or ONO-8430506 twice daily as described above, starting on day 1 after injection. Tumor volumes were measured using digital calipers, and all mice were euthanized when the fasting-progressing group (PDAC + PSC, vehicle) had to be sacrificed due to tumor volume per OHSU IACUC standards (day 21).

All statistical analyses were performed using GraphPad PRISM software. Student t test was used to compare two groups to each other. One-way ANOVA was performed when multiple conditions were compared for one variable. Two-way ANOVA was used when multiple groups were compared for more than one variable. Tukey post hoc tests were used after both ANOVA analyses to perform multiple group comparison. Analysis with a P value < 0.05 was considered statistically significant.

For details of additional methods, please see Supplementary Methods.

J.J. Kamphorst is Director of Cellular Metabolism for Rheos Medicines and has ownership interest (including stock, patents, etc.) in the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F.R. Auciello, V. Bulusu, J.C. Norman, J.J. Kamphorst, M.H. Sherman

Development of methodology: F.R. Auciello, V. Bulusu, C. Oon, S. Bhattacharyya, J. Link, R.M. Evans, J.C. Norman, J.J. Kamphorst, M.H. Sherman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F.R. Auciello, V. Bulusu, J. Tait-Mulder, M. Berry, B.L. Allen-Petersen, J. Link, N.D. Kendsersky, E. Vringer, M. Schug, D. Novo, C. Dorrell, J.P. Morton, R.C. Sears, M.H. Sherman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F.R. Auciello, V. Bulusu, C. Oon, J. Tait-Mulder, S. Tumanov, M. Schug, D. Novo, J.C. Norman, J.J. Kamphorst, M.H. Sherman

Writing, review, and/or revision of the manuscript: F.R. Auciello, V. Bulusu, J. Link, R.F. Hwang, J.C. Norman, R.C. Sears, J.J. Kamphorst, M.H. Sherman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F.R. Auciello, C. Oon, J. Tait-Mulder, M. Berry, R.F. Hwang, C. Nixon

Study supervision: F.R. Auciello, J.C. Norman, J.J. Kamphorst, M.H. Sherman

We thank Sara Courtneidge, Dafna Bar-Sagi, Eyal Gottlieb, Joshua Rabinowitz, and Owen Sansom for providing feedback on the manuscript; Gillian Mackay and David Sumpton for support with mass spectrometry; the Cancer Research UK Glasgow Centre (C596/A18076) and the BSU facilities at the Cancer Research UK Beatson Institute (C596/A17196) and the Oregon Health & Science University Department of Comparative Medicine for support with in vivo work; and Ono Pharmaceutical Co. Ltd. for providing the autotaxin inhibitor. This work is supported by a Cancer Research UK Career Development Fellowship (C50242/A17728, to J.J. Kamphorst) and by a National Cancer Institute Pathway to Independence Award (CA188259-01) and grant R01CA229580, V Foundation V Scholar Award (V2017-009), American Cancer Society Research Scholar Grant (132898-RSG-18-142-01-CSM), and Hirshberg Foundation Seed Grant (all to M.H. Sherman).