Pancreatic Cancer Cell Migration and Metastasis Is Regulated by Chemokine-Biased Agonism and Bioenergetic Signaling

With an overall survival rate of only 6%, pancreatic cancer remains a poorly addressed health problem relative to other commonly diagnosed malignancies (1). The vast majority of patients with pancreatic ductal adenocarcinoma (PDAC) ultimately succumb to metastatic disease (2). Standard chemotherapies for PDAC provide a minimal improvement in survival for patients, particularly those with advanced stages of disease. The inability to chemotherapeutically target advanced PDAC reflects, in part, a current paucity of knowledge regarding the mechanisms driving metastasis. While data from transgenic mice suggested PDAC cell dissemination can occur early in disease progression (3), sequencing and modeling analysis after rapid autopsy of patients revealed that dissemination and metastasis of PDAC occurs subsequent to primary tumor development (4). Although no mechanism is known, human tissue analysis implicates the signaling protein Smad4 in multifocal metastasis of PDAC (5). However, the specific extracellular cues that regulate the invasion, migration, and metastatic homing of PDAC cells into distant tissues have yet to be identified.

As soluble/secreted proteins, chemokines, or chemotactic cytokines, direct cell movement in a concentration-dependent manner. In normal physiology, chemokines direct cell trafficking of a wide variety of cell types. Dysregulated expression of chemokines, or their cognate G-protein–coupled receptors (GPCR), is associated with pathologic cell migration in many diseases (6). Solid tumor progression has been linked with elevated expression of the chemokine receptor CXCR4 (7). Recently, we have uncovered epigenetic repression of the chemokine ligand CXCL12 in colorectal, mammary, and pancreas cancer, in parallel with increased expression of its cognate receptor CXCR4. Silencing of the ligand enables homing of CXCR4-expressing cancer cells to metastatic sites rich in CXCL12 (8–10). Subsequent histopathologic studies have also linked diminished CXCL12 expression with worse prognosis in gastric cancer, osteosarcoma, and breast cancer (11–13). Our more recent study in pancreatic cancer demonstrated that re-expression of CXCL12 inhibits PDAC cell growth as well as cell migration (10). Chemokines such as CXCL12 stimulate migration in a biphasic manner, with 10 nmol/L optimally stimulating epithelial cell movement and ≥100 nmol/L unable to induce migration (14–18). In contrast, chemokine signaling through the Gαi protein stimulates intracellular calcium flux over a broader concentration range (e.g., 1–1,000 nmol/L; refs. 16, 18). Importantly, the 1 to 1,000 nmol/L concentration range is comparable to concentrations of CXCL12 physiologically available in circulation and human tissue (15, 19–21). The disparity between these two downstream signaling effects represents a longstanding conundrum in GPCR function. Lefkowitz and colleagues first observed that distinct agonists of the same β-adrenergic receptor can elicit different signals and functions that are either G-protein–dependent or β-arrestin signaling–mediated and termed this functional selectivity as “biased agonism” (22). Thus, as defined by Reiter and colleagues, GPCR agonists can fall into three categories, G-protein–biased, β-arrestin–biased, or balanced, in which both G-proteins and arrestin are recruited and signaled. A preliminary report by Lefkowitz and colleagues also showed that the chemokines CCL19 and CCL21 could elicit differential β-arrestin–biased agonist signaling when binding to the chemokine receptor CCR7 (23).

Prior reports had shown that in physiologic solutions, chemokines were capable of reversibly forming multimeric structures, in a concentration-dependent manner (14, 24, 25). For CXCL12, at lower concentrations, the protein remains monomeric, but as the concentration increases, the presence of dimeric structures predominate (25). These data lead to the hypothesis that the dimeric and monomeric structures may possess functional selectivity for the biphasic migration response observed for CXCL12. Subsequently, in a series of prior reports using engineered variants of CXCL12, we discovered that the unique oligomeric forms of the ligand can function as biased agonists after binding to CXCR4 (26, 27), with monomeric CXCL12 eliciting “balanced” signaling and migration, whereas dimeric CXCL12 stimulated “G-protein–biased” signaling without eliciting migration. However, further investigation into a downstream signaling mechanism to link the lack of migration with the G-protein–biased agonism stimulated by dimeric CXCL12 is necessary.

Recent studies of PDAC illustrate that the there is a shift in metabolic flux toward anabolic pathways that facilitate nutrient synthesis (28) and that PDAC cells rely on glutamine metabolism to drive oncogene-dependent growth (29–31). However, little is known regarding the metabolic or bioenergetic profiles of migrating PDAC cells. Likewise, although hormone-dependent calcium signaling and cell polarity regulation have been associated with bioenergetic signaling through the homeostatic regulator AMP kinase (AMPK; refs. 32, 33), the potential relationship between cell metabolism and chemotaxis is unexplored. Herein, we establish that the decline in chemotactic migration at elevated CXCL12 concentrations reflects CXCR4-dependent AMPK and myosin light chain (MLC) phosphorylation. The lack of migration at high concentrations of CXCL12 was replicated using the engineered CXCL12 dimeric ligand. These are the first data demonstrating that receptor activation by a dimeric chemokine ligand can enhance a specific signaling pathway as a result of biased agonism. The direct stimulation of a nonmotile state by the dimeric ligand we herein have termed “ataxis.” Broadly, our work unearths an important energy signaling mechanism underlying chemotaxis and reveals a viable strategy for pharmaceutical targeting of cancer metastasis.

AICAR, LPA, AMD3100, metformin, and compound C were purchased from Sigma. BAPTA-AM, ionomycin, STO-609, and pertussis toxin were purchased from EMD Biosciences Calbiochem. The mitochondrial inhibitor Mito-CP was produced and used as previously published (34). Antibodies against total or phosphorylated AMPK, MLC, LKB1, and MYPT1 were purchased from Cell Signaling Technologies. The wild-type, locked monomer, and locked dimer variants of CXCL12 were expressed and purified as previously described (27, 35). The CXCL12 locked dimer (CXCL12-LD) mutations are L36C and A65C and create a dimeric protein. The CXCL12 locked monomer (CXCL12-LM) mutations are L55C and I58C. The engineered locked dimer or locked monomer proteins are structurally indistinguishable from native dimeric and monomeric protein, respectively, and retain full CXCR4 receptor–binding capability. Proteins were expressed in Escherichia coli and cells were lysed by French press. Fusion protein was purified through nickel chromatography and refolded by infinite dilution and ULP1 protease was used to cleave the 6XHIS-Sumo tag. Cation exchange and high-performance liquid chromatography was used for final purification.

The human pancreatic carcinoma cells Panc1 (CRL-1469) and MiaPaCa2 (CRL-1420) were purchased from the ATCC. Patient-derived PDAC MCW512, corresponding to MCW4 from our prior report, were obtained from the Medical College of Wisconsin Surgical Oncology Biobank using institutional review board–approved protocols and cultured as previously published (10). The cell lines K8282 and K8484 were derived from the original KRas^LSL.G12D/+-p53^R172H/+-Pdx^Cre (KPC) mice on the mixed 129/SvJae/C57BL/6 background and were the kind gift of Dr. Kenneth Olive (Columbia University, New York, NY). FC1199, FC1242, FC1245, and DT10022 cell lines were derived from KPC mice in which each of the founder mutant mice had been backcrossed to the C57BL/6 genetic background. KPC cells were maintained in high-glucose DMEM with 10% (v/v) FBS (Life Technologies Inc.). The Pan-02 cell lines were provided by the National Cancer Institute Cell Repository (Bethesda, MD) and maintained in RPMI-1640 with 10% (v/v) FBS.

SCID mice (cr-Prdkc^scid, Charles Rivers Laboratories) were anesthetized and orthotopically implanted with either 10⁶ Panc1 or MiaPaCa2 cells stably expressing firefly luciferase and tumor formation tracked in vivo by bioluminescent imaging (Lumina IVIS 100, Perkin Elmer) using our previously published technique (10). At 7 days postimplantation, mice were sorted into vehicle or treatment groups with equivalent average luminescence and treated twice weekly thereafter with 200 μL intraperitoneal injections of PBS or 5 μmol/L recombinant CXCL12 protein. Mice in the Panc1 model were allowed to survive until humanely euthanized when morbid, in accordance with an Institutional Animal Care and Use Committee-approved protocol, whereas MiaPaCa2 xenografted mice were sacrificed on day 70. Ex vivo analysis was performed with individual luminescence measurements of the liver, lung, and adjacent lymph nodes. The peritoneal cavity was visually inspected and imaged post-organ harvest to detect potential peritoneal movement of tumor cells.

Changes in bioenergetic flux in pancreatic cancer cells were measured using Seahorse Bioscience XF96 Extracellular Flux Analyzer (Seahorse Bioscience). MiaPaCa2 cells were first plated overnight in Seahorse plates and then equilibrated in unbuffered, serum-free medium containing only 5.5 mmol/L glucose and 4 mmol/L l-glutamine for 3 hours. Prior to the injection of chemokine into each well, 8 baseline measurements of oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were taken and averaged as a time zero energy measurement. Chemokines were then automatically dispensed into wells, with 8 measurements taken over 1 hour to determine basal energetic flux, after which glycolytic and oxidative stress tests were employed using the inhibitors, oligomycin, dinitrophenol, and antimycin A. Stress test inhibitors were injected sequentially, with three measurements taken after each individual treatment. Oligomycin was used to measure ATP-linked OCR and reserve ECAR, dinitrophenol was used to measure reserve OCR, and antimycin A was used to correct for nonspecific flux.

Cells were plated to 80% confluency in 60-mm dishes and then starved 24 hours for transfected cells or 5 hours for stimulated cells. Stimulations were performed in serum-free medium and inhibitors placed on cells 1 hour before stimulation. After stimulation, cells were washed twice in cold PBS and lysed using a modified RIPA buffer. Lysates were normalized for protein concentration, size separated using reducing SDS-PAGE, electrotransferred to polyvinylidene difluoride membranes (Millipore), and then probed using primary and horseradish peroxidase–conjugated secondary antibodies. Proteins were visualized by chemiluminescence with autoexposure and quantified by densitometric analysis using the FluorChem HD2 from Cell Biosciences. The optical densities of the proteins in immunoblots from independent experiments were obtained by densitometry.

Cells are grown to 80% confluency, then lifted using Enzyme-Free Dissociation Buffer (Invitrogen), and washed twice using cold PBS. Cells were incubated in a 1% (w/v) BSA blocking solution, incubated for 1 hour in primary antibody on ice, washed twice, and then incubated for 1 hour in fluorophore-conjugated secondary antibody. Cells were then fixed in 2% (w/v) paraformaldehyde and fluorescence intensity measured on a BD-LSR II flow cytometer and analyzed by FlowJo software (BD Biosciences).

Immunofluorescence analysis was performed as published previously (36). Briefly, cells were plated overnight at 50% confluency on glass coverslips, serum-starved 2 hours, and then treated. After treatment, cells were fixed using paraformaldehyde, blocked and permeabilized, and then incubated overnight at 4°C with a phosphor-specific antibody for MLC. Cells were then washed and incubated with a goat anti-rabbit AlexaFluor-488–conjugated antibody and phalloidin-AlexaFluor-594 for 2 hours at room temperature, counterstained with 4′,6-diamidino-2-phenylindole (DAPI) for 15 minutes, and coverslipped. Microscopy was performed and fluorescence intensity measured using the Nikon Eclipse Ti microscope and software at ×600 magnification under oil immersion. Percent of cells with stress fiber formation was quantified with images blinded for treatment and then scored by a separate investigator for presence of stress fiber or cortical actin.

RNA was isolated using the RNA Easy Kit from Qiagen and treated with DNAse I to remove gDNA contaminants. Conversion to cDNA was performed by priming with random hexamers and the SuperScript II cDNA Synthesis Kit (Life Technologies). PCR products were separated by agarose gel electrophoresis and visualized by ethidium bromide staining and ultraviolet imaging. Primers for the promoter region of the mannose-binding lectin gene were used to test for gDNA contaminant, while primers for the GAPDH transcript were used as a loading control. Amplification of chemokine receptors and chemokine was done using the following primers (5′-3′): mouse CXCL12 forward—ACCTCGGTGGTCCTCTTGCTGTCC, mouse CXCL12 reverse—GTTGGCTCTGGCGATGTGGCTCTC, mouse CXCR4 forward—TTGTCCACGCCACCAACAGTCA, mouse CXCR4 reverse—TGAAACACCACCATCCACAGGC, mouse CXCR7 forward—GGAGCCTGCAGCGCTCACCG, and mouse CXCR7 reverse—CTTAGCCTGGATATTCACCC.

shRNA sequences complementary to CXCR4 transcript, sequences 4052 (#1) and 4055 (#2), were purchased from the Open Biosystems dataset preinserted into the LL3.7-Puro vector. Mock, scramble, or CXCR4 shRNAs were cloned into competent E. coli. Lentiviral particles were constructed by transfecting each plasmid with viral accessory plasmids, pVSVG, pREV, and pRRE into HEK-293T cells using Mirus reagent (Mirus). After overnight transfection, medium was exchanged for harvest media and cells cultured for 48 hours. Harvested viral particles were then collected, filtered, and stored for later use. MiaPaCa2 cells were grown to about 50% confluency and then transduced with harvested mock, scramble, or CXCR4-encoding lentiviral particles using polybrene for 24 hours. Transduced cells were then selected using puromycin for 7 days, after which stable protein depletion was measured by both immunoblotting and flow cytometry.

Calcium mobilization was measured using the Fluo-4-AM cell permeable dye (Invitrogen) according to manufacturer's directions. Cells were plated to 90% confluency in 96-well plates and then serum-starved for 2 hours the next day. Cells were washed with calcium/magnesium-free Dulbecco's PBS, loaded with Fluo-4-AM for 30 minutes at 37°C, and then 10 minutes at room temperature. Stimulants, diluted in calcium/magnesium-free HBSS, were then loaded into auto-injectors of the Victor Wallac. Background readings were taken before stimulant injection (3 readings), and then 30 readings every 5 seconds were taken postinjection with ionomycin as a positive control.

Cells were grown to confluence in Transwell inserts (Corning Costar). The upper well of the Transwell membrane was coated with collagen I (15 μg/mL) and then incubated in 2% (w/v) BSA/PBS until cells were ready for plating. Cells were lifted using enzyme-free buffer, washed, and plated to the upper well, with chemoattractants added to the bottom well in serum-free media. MiaPaCa2 and MCW512 cell migration (10⁵ cells plated) was measured after 6 and 4 hours, respectively. Transwell migration was enumerated in a representative set of images taken using fluorescence microscopy after the upper side of each well was swabbed, fixed using 4% (w/v) paraformaldehyde, and stained with DAPI. Any functional inhibitors were mixed with cells plated in the top of Transwell chamber, with a pretreatment incubation time of 30 minutes.

All statistical analyses were performed using GraphPad Prism 4. Paired analyses were calculated using the Student t test. Multiple comparisons were analyzed using a one-way ANOVA and either a Dunnett post hoc analysis or a Tukey post hoc analysis to identify pairwise differences between distinct experimental groups. Statistical significance was defined as P≤0.05.

We recently demonstrated that stable re-expression of CXCL12 in human PDAC cells has a tumor-suppressive effect by slowing cancer cell proliferation and mobility (10). On the basis of those findings, we hypothesized that as a secreted protein, CXCL12 would function as a biologic response modifier of PDAC malignancy. Combining our previously established pancreatic orthotopic xenograft models with a therapeutic regimen and window defined in our previous studies using the protein in vivo for colorectal cancer and melanoma (10, 27, 37), we found that mice treated twice weekly with CXCL12 had less dissemination, relative to vehicle treatment (Fig. 1A and B). Ex vivo luminescence of metastatic destination organs, including liver, lung, and proximal lymph nodes, showed that CXCL12 treatment significantly lowered overall tumor dissemination, compared with controls (Fig. 1C and D). Notably, decreased metastasis was separately observed in both the MiaPaCa2 and Panc1 xenograft models, without any concomitant increase in peritoneal cavity dissemination. These data on the antimetastatic effects of CXCL12 mirror our prior report in colorectal cancer and melanoma (27).

One explanation for the antimetastatic effect of CXCL12 in vivo is that exogenous chemokine disrupts the concentration gradients necessary for chemotactic migration of cancer cells. On the basis of our observation that CXCL12 regulates in vitro pancreatic cell migration in a biphasic manner, wherein low nanomolar doses (≤10 nmol/L) stimulate migration whereas higher doses (≥100 nmol/L) are “ataxic” and unable to evoke migration, we hypothesized that the change from CXCL12 from a pro- to an antimetastatic ligand reflected functional selectivity or biased agonist signaling through the CXCR4 receptor (10). The mechanisms for functional selectivity of CXCL12 through CXCR4 are unknown. Using a Seahorse-XF analyzer, we first tested the notion that the biphasic migratory response to CXCL12 reflected changes in energy metabolism. A declining, dose-dependent trend in basal oxygen consumption and glycolysis rates in MiaPaCa2 cells stimulated with CXCL12 was measured (Supplementary Fig. S1; Fig. 2A and B). Significant decreases in ATP-linked oxidative phosphorylation, basal glycolysis, and reserve glycolytic capacity were observed in MiaPaCa2 cells stimulated with the ataxic 1,000 nmol/L dose of CXCL12, compared with vehicle control (Fig. 2A and B). When oxygen consumption and glycolytic rates were visualized together, an overall declining trend in cellular energy was observed as the concentration of CXCL12 increased (Fig. 2C).

We next predicted that CXCL12-dependent changes in cell energetics may be linked to altered activity of AMPK, a known regulator of metabolic stress. Immunoblot analysis demonstrated that CXCL12 dose dependently stimulated phosphorylation at the activating threonine-172 site of AMPK (Fig. 2D). Nonmotile 100 and 1,000 nmol/L doses of CXCL12 stimulated significant and sustained AMPK phosphorylation through 1 hour (Fig. 2D). CXCL12 regulation of migration and AMPK activity was next tested in a previously characterized patient-derived PDAC cell line (10). As predicted, patient-derived MCW512 cells migrated in response to exogenous CXCL12 stimulation in a biphasic fashion, with peak chemotaxis at 10 nmol/L CXCL12 (Supplementary Fig. S1C). As with MiaPaCa2 cells, 1,000 nmol/L CXCL12 stimulated significant increase in AMPK phosphorylation in patient-derived cells (Supplementary Fig. S1D). Together, these data demonstrate that ataxic doses of CXCL12 stimulate prolonged AMPK activity in human PDAC cells.

To address whether CXCL12-induced AMPK activity was species-dependent, we next acquired a battery of mouse-derived PDAC cell lines and first analyzed chemokine and receptor expression. RT-PCR analysis showed that 4 of 7 mouse lines lacked appreciable CXCL12 transcript, 4 of 7 mouse lines expressed CXCR4 transcript, and each line expressed CXCR7 mRNA (Supplementary Fig. S2A). Immunoblot and flow cytometric analyses confirmed protein expression of CXCR4 (Supplementary Fig. S2B–S2D). Of the 7 cell lines, FC1199 and Pan-02 most closely mimicked the expression pattern of CXCL12, CXCR4, and CXCR7 that we noted in human PDAC (10). Given its novel isolation from the primary tumor of C57BL/6 KRas^LSL.G12D/+-p53^R172H/+-Pdx^Cretg/+ (KPC) mouse model of PDAC, we identified the FC1199 cell line as a suitable model for further study. FC1199 cells had increased AMPK activity in a dose-dependent fashion after CXCL12 stimulation (Fig. 2E). Consistent with the human cell lines, CXCL12 stimulated AMPK phosphorylation was significantly increased 1 hour after stimulation with 100 or 1,000 nmol/L, but not with 1 or 10 nmol/L doses, in FC1199 cells (Fig. 2E). These data indicate that CXCL12-induced AMPK activity is a species-independent event.

In sum, our observations indicate that CXCL12 regulates bioenergy consumption rates as well as energy-sensing signaling in a dose-dependent manner. The optimal cell migratory dose of CXCL12 leads to transient AMPK activity and minimal change in glycolysis or oxidative phosphorylation. In contrast, doses exceeding 100 nmol/L CXCL12, which do not evoke cell migration, lead to sustained (≥1 hour) AMPK activity and decreased glycolysis and oxidative phosphorylation.

As our previous studies demonstrated that CXCL12-induced biphasic migration is mediated through CXCR4 binding (27), rather than CXCR7, we asked whether the AMPK phosphorylation stimulated by CXCL12 is dependent on CXCR4 signaling. As a first approach, immunoblot and flow cytometric analyses confirmed depletion of CXCR4 in 2 MiaPaCa2 clones (Supplementary Fig. S3A and S3B). Transwell migration assays then showed that the control and scramble sequence–expressing cells migrated toward 10 nmol/L CXCL12, whereas CXCR4-knockdown clones were unable to migrate in response to chemokine stimulation (Supplementary Fig. S3C and S3D). We subsequently assessed AMPK activity in those CXCR4-knockdown cells. As shown in Supplementary Fig. S3E and S3F, at the short time point of 15 minutes, CXCL12 stimulated little, if any, AMPK phosphorylation in CXCR4-depleted clones, whereas the scramble-transduced cells ably responded to nonmotility-inducing doses of chemokine. We next tested whether AMPK activation was due to Gαi protein–coupled signaling. Consistent with elevated CXCL12 signaling in a biased agonist fashion, immunoblot analysis showed that AMPK phosphorylation was decreased after pretreatment with either pertussis toxin (Supplementary Fig. S3G) or the small-molecule CXCR4 antagonist AMD3100 (Supplementary Fig. S3H). These data demonstrate that AMPK phosphorylation in response to the ataxic ≥100 nmol/L doses of CXCL12 reflects activation of the canonical CXCR4–Gαi signaling mechanism.

In previous studies, we and others established that, in contrast to the biphasic functional migration response, CXCL12 stimulates intracellular calcium flux in a saturable concentration-dependent fashion (25–27). In MiaPaCa2 cells, CXCL12 doses exceeding 10 nmol/L stimulated equivalent maximal levels of intracellular calcium flux with and without extracellular calcium ions (Fig. 3A–D). In patient-derived PDAC cells, 1,000 nmol/L CXCL12 stimulated a significant and sustained increase in calcium (Supplementary Fig. S4A and S4B). Given the similarity of concentration-dependent change in AMPK activity and calcium mobilization in response to CXCL12, we predicted that increased AMPK phosphorylation was dependent on calcium mobilization. Subsequently, the cell-permeable calcium ion chelator BAPTA-AM abrogated CXCL12 or ionomycin-induced AMPK phosphorylation in both MiaPaCa2 and patient-derived cells (Supplementary Fig. S4C and S4D).

Prior reports exploring the regulation of AMPK activity have shown that T172 can be phosphorylated by either the cell polarity regulator LKB1 or calcium-dependent calmodulin kinase kinase 2 (CamKK2). While LKB1 activity was not increased by stimulation with CXCL12 (Supplementary Fig. S4E), the CamKK2-specific inhibitor STO-609 reversed AMPK phosphorylation induced by 1,000 nmol/L CXCL12 (Fig. 3E and F). In total, our experiments examining CXCL12 downstream signaling in human and murine PDAC cells demonstrate that AMPK activity driven by CXCL12 is mediated through receptor-dependent intracellular calcium flux and CamKK2 activity.

RAPTOR, the established regulator of mTOR (mTORC1) signaling, is a key downstream target for AMPK-mediated phosphorylation. While AMPK is conventionally thought to regulate mTOR activity in the context of cell growth, we did not detect any altered RAPTOR signaling (data not shown), consistent with other studies demonstrating non–growth-related AMPK signaling (33). The repeated cycling of MLC phosphorylation and subsequent dephosphorylation is essential for the perpetual lever-like action that myosin exhibits on F-actin fibers, allowing cells to contract, relax, and then migrate (38, 39). Activation of AMPK has been proposed to affect MLC activity by preventing the dephosphorylation that occurs on the regulatory unit of MLC (40, 41). Mirroring our previous studies in intestinal epithelial cells (36), immunoblot analysis of MiaPaC2 cells showed that the chemotactic dose lead to peak phosphorylation of MLC at 15 minutes, which then returned to baseline levels by 60 minutes; a similar temporal pattern in MLC phosphorylation was observed with the migration-positive control (Fig. 4A). In contrast, both ataxic CXCL12 stimulation and the AICAR-positive control evoked a sustained increase in MLC phosphorylation over 60 minutes (Fig. 4B).

We hypothesized that differential MLC phosphorylation in response to CXCL12 was due to AMPK downstream signaling. The AMPK inhibitor compound C reversed MLC phosphorylation induced by 1,000 nmol/L CXCL12 (Fig. 4C). Previous studies indicated that AMPK regulates MLC by inhibitory phosphorylation of myosin phosphatase 1 (MYPT1; refs. 42, 43), an enzyme responsible for dephosphorylating MLC. We thus predicted that ataxic doses of CXCL12 would activate AMPK and in turn inhibit MYPT1 function. In a time course, CXCL12 lead to inhibitory phosphorylation of T853 of MYPT1 between 15 and 60 minutes (Supplementary Fig. S5). We then examined the AMPK dependence of CXCL12 regulation of MYPT1 and showed that compound C significantly reversed the ataxic CXCL12 or AICAR-induced inhibitory phosphorylation of MYPT1 (Fig. 4D), suggesting that elevated levels of CXCL12 disrupt MLC activity through AMPK.

We next questioned whether the CXCL12-stimulated change in MLC phosphorylation significantly altered the assembly of cell migration machinery through AMPK. Immunofluorescence microscopy revealed that after stimulation with ataxic doses of CXCL12, cells were rounded with few cytoplasmic projections, consistent with the nonmigratory phenotype (Fig. 5E). By comparison, the chemotactic dose of CXCL12 led to more cell spreading, with abundant filopodia and lamellopodia formation (Fig. 5E). Consistent with our immunoblot analysis, after 1-hour stimulation, immunofluorescence microscopy revealed that ataxic, and not the chemotactic, dose significantly increased levels of phosphorylated MLC in treated cells (Fig. 4F). In parallel, PDAC cells stimulated with 10 nmol/L CXCL12 had a higher percentage of actin stress fibers, whereas actin in 1,000 nmol/L CXCL12-treated cells was organized primarily in cortical bundles (Fig. 4G). Finally, compound C pretreatment of cells stimulated with 1,000 nmol/L CXCL12 reversed MLC phosphorylation and in turn increased formation of actin stress fibers and cell spreading (Fig. 4E–G).

We then sought to establish a link between metabolism and migration. We first showed in a Transwell migration assay, blocking glycolysis with 2-deoxyglucose (2-DG) and inhibiting mitochondria function with our previously established inhibitor Mito-CP (34), that abrogation of metabolic activity decreased both basal migration and chemokine–directed migration of MiaPaCa2 cells in a concentration-dependent manner (Fig. 5A). We then duplicated this result using dual inhibition of metabolic flux with 2-DG and metformin, an established inhibitor of pancreatic cancer cell metabolism (Fig. 5B; ref. 44). Last, we tested the notion that chemotaxis regulated by CXCL12 is dependent on AMPK activity in a MiaPaCa2 Transwell migration assay stimulated with a concentration curve of CXCL12 in the presence or absence of the AMPK inhibitor (Fig. 5C and D). In the absence of AMPK inhibition, CXCL12 stimulated a biphasic pattern of migration consistent with our previously published data (10). However, pretreatment with compound C completely reversed the biphasic pattern and lead to saturable CXCL12-induced migration. Importantly, the introduction of the AMPK inhibitor significantly increased migration at the previously ataxic doses of 100 and 1,000 nmol/L CXCL12 while continuing to permit chemotaxis at the optimal dose of 10 nmol/L CXCL12 (Fig. 5C and D).

Our prior work established not only that CXCL12 exists primarily as a monomer at low concentrations and as a dimer at elevated concentrations but also that an engineered locked dimer CXCL12 protein effectively interfered with colorectal cancer and melanoma metastasis, with increased stability in vivo compared with wild-type protein (27, 37). We tested whether the biophysical dimeric chemokine mechanism might explain CXCL12-induced ataxis through activation of the AMPK bioenergetic signaling mechanism. Using our previously established engineered mutants of CXCL12 locked in either the monomeric or dimeric state (27, 35), we first showed that 10 nmol/L CXCL12-LM stimulated migration of MiaPaCa2 cells equivalent to 10 nmol/L wild-type protein (Fig. 6A). As predicted, 10 nmol/L CXCL12-LD did not elicit significant Transwell migration of MiaPaCa2 cells but did stimulate equivalent calcium mobilization compared with monomer or wild-type protein in both MiaPaCa2 and FC1199 cells (Fig. 6B). Reflecting our observations using 1,000 nmol/L of the wild-type protein, CXCL12-LD stimulated prolonged AMPK activity compared with CXCL12-LM. Furthermore, 1- and 2-hour stimulation with locked dimer CXCL12 resulted in substantially more phosphorylation of AMPK compared with monomeric CXCL12 (Fig. 6C). Finally, we tested the notion that the lack of migration observed in CXCL12-LD–treated cells is dependent on AMPK activity. Using a Transwell assay of MiaPaCa2 cells, we found that pretreatment with the AMPK inhibitor compound C reversed the ataxis stimulated by 10 nmol/L CXCL12-LD. Instead, in the presence of compound C, CXCL12-LD stimulated a strong migration that was equivalent to that evoked by the same concentration of wild-type CXCL12 or CXLC12-LM (Fig. 6D).

Taken together, these data suggest that the concentration-dependent decline of chemotactic migration is mediated, in part, by AMPK. Ataxic doses of CXCL12 prevented dephosphorylation of MLC by inhibiting MYPT1 function. This loss in MLC dephosphorylation then prevented the uncoupling of MLC to F-actin, leading to the accumulation of cortical actin bundles. Thus, concentrations of CXCL12 exceeding 100 nmol/L require AMPK activity to actively prevent migration. These data indicate that the decline in migration is the result of biased agonism of CXCL12 binding to its cognate receptor CXCR4. Prolonged AMPK activity was similarly observed with stimulation by locked dimeric CXCL12, suggesting that this engineered protein has significant potential as a more stable in vivo regulator of PDAC metabolism and metastasis.

The CXCL12–CXCR4 chemokine axis has been implicated in numerous cancers, including highly metastatic PDAC, under the conventional model that elevated receptor expression correlates with malignancy (7). However, this model has not translated to successful clinical development of CXCR4-targeted anticancer therapies, suggesting a clear need for a novel pharmacologic approach. In particular, the mechanism underlying the ability of chemokines to control directional cellular migration at different doses is not known. The conventional model for chemokine involvement in cancer metastasis has been overexpression of CXCR4 leading to increased sensitization and passive migration toward distant tissue sources of CXCL12 such as the liver (45). This current dogma ignores the potential active role that the ligand CXCL12 plays in modulating cancer. Our work in highly malignant PDAC extends the current understanding of chemokines in cancer in 2 important ways. First, while other recent PDAC studies have focused on the role of chemokines in the primary tumor (46, 47), our data herein and in a previous study (10) identify an ataxic, protective role for CXCL12–CXCR4 in metastasis of pancreatic cancer. Second, although numerous studies have defined shifts in metabolism that occur during pancreatic cancer cell growth (29–31), our report is the first to observe the changes in bioenergetic metabolism that occur during pancreatic cancer cell migration. Our data indicate that CXCL12 can function as a balanced or biased agonist for its cognate receptor CXCR4, depending on the chemokine concentration. Elevated doses of CXCL12 effectively activate an AMPK-mediated bioenergetic brake that shifts pancreatic cancer cells into a nonmigratory response. We additionally demonstrate in a preclinical model that CXCL12 inhibits pancreatic cancer metastasis in vivo.

While canonical chemokine signaling occurs through G-protein–dependent calcium mobilization, the precise biochemical mechanisms downstream of chemokine ligand receptor binding are still emerging. In this study of PDAC, we show that 10 nmol/L CXCL12 stimulates maximal migration (chemotactic) whereas concentrations >100 nmol/L are unable to stimulate migration (ataxic). As we and others have shown for a number of chemokine axes (14, 16–18), over a concentration curve, CXCL12 stimulates migration in a biphasic fashion while eliciting intracellular calcium mobilization in a saturable manner. The decline in chemokine-induced migration at elevated doses has commonly been attributed to receptor desensitization but often without a mechanistic explanation (48). Our data reveal a novel mechanism that explains the lack of migration at elevated concentrations of chemokines through a biased agonist signaling cascade. Here, we show that CXCL12 dose dependently alters glycolysis, oxidative phosphorylation, and promotes activity of the homeostatic metabolism regulator AMPK. We confirmed that this dose-dependent effect on bioenergetic signaling was CXCR4–Gαi calcium signaling dependent. While others have demonstrated that small-molecule GPCR agonists are capable of inducing AMPK activity (33) in a calcium-dependent manner, our study is the first to link chemokines, as soluble proteins, with downstream AMPK signaling (Fig. 7). Furthermore, chemotactic CXCL12 concentrations induced cycling of phosphorylation of MLC, facilitating the perpetual motion of stress fiber actin (Fig. 7A) and in turn allowing directional movement (Fig. 7C). In contrast, cells stimulated with ataxic doses of CXCL12 stimulated the AMPK-dependent increase of MYPT1 inhibitory phosphorylation, which blocks MLC phosphorylation:dephosphorylation cycling (Fig. 7B), effectively locking the migration machinery and preventing cell movement (Fig. 7D). While the role of the alternative CXCL12 receptor CXCR7 was not explored, we predict that since CXCR7 signaling is G-protein–independent (27), it will not affect AMPK activity. In addition, as activity of AKT, a known downstream effector of chemokines, has recently been linked to inhibition of AMPK (33), further study into the interplay between AKT and AMPK in chemokine receptor signaling is necessary. Cumulatively, these data support a model wherein elevated concentrations of chemokine ligand, such as might be found once cells have arrived at their chemotactic destination, activate a bioenergetic molecular brake to stop moving.

Lefkowitz and colleagues first posited the concept of biased agonism wherein different ligands were capable of distinct signaling through a single GPCR (22). Importantly, their more recent study has shown that β-arrestin, a critical regulator of GPCR biased agonism, also modulates MLC phosphorylation downstream of angiotensin signaling (49). Our recent work identified chemokine oligomerization as a biophysical mechanism for changes in migration potential during chemotaxis and suggested that ligand concentration-dependent biased agonism could explain the associated changes in signaling (25–27). At low nanomolar concentrations, CXCL12 preferentially exists as a monomeric protein, whereas at high nanomolar and micromolar concentrations and in the presence of binding partners, the chemokine forms dimeric structures (25, 26, 50). Using engineered mutants locked into each form, we determined that monomeric CXCL12 was equipotent to the native protein and thus functions as a balanced agonist. In contrast, dimeric CXCL12 was incapable of stimulating migration but retained the equipotent G-protein–biased calcium signaling response as wild-type or monomeric forms of the ligand (27). As CXCL12 dimer-treated cells were able to migrate in response to CXCR4-independent stimulation, we first termed the dimer effect as “cellular idling” (27). We additionally showed that oligomers of CXCL12 differentially recruited β-arrestin and stimulated phosphorylation of ERK1/2 (27). In this study, we used engineered oligomeric mutants of CXCL12 to demonstrate that the ataxis of pancreatic cancer cells elicited by elevated concentrations of CXCL12 is mirrored with locked dimeric CXCL12. In addition, similar to ataxic, that is, >100 nmol/L, doses of wild-type CXCL12, dimeric CXCL12 elicited more prolonged AMPK activity compared with a monomeric variant of the ligand. Thus, these new data establish a mechanistic link between the observed biphasic regulation of migration by CXCL12 and biased agonism elicited by distinct concentrations or oligomeric states of the chemokine. These data are the first to identify a mechanism by which enhanced signaling drives the arrest of cellular migration, which we have termed “ataxis,” at elevated physiologic concentrations of the chemokine ligand.

Reexamining the role for CXCL12–CXCR4 in the context of biased signaling sheds important light on chemokine function in malignancy. Under the conventional model, some have suggested that CXCR4 is a prometastatic factor for pancreatic cancer (51), whereas others have shown that CXCL12 can stimulate increased activity of the cell survival factor AKT, implying that the chemokine is a progrowth factor (52). However, nearly all of those studies rely upon a single, typically migratory, dose of CXCL12 to examine signaling outputs and ignore potentially significant functional effects of higher concentrations of CXCL12 previously observed in vivo (20). Along with our previous study (10), these data suggest a model for CXCL12–CXCR4 in pancreatic cancer wherein low chemotactic concentrations elicit cancer cell movement to distant tissues but locally increased ataxic concentrations of the ligand CXCL12 abrogate cancer cell movement through bioenergetic signaling and cytoskeletal dysregulation. Translating this novel signaling mechanism, we found that systemic treatment with wild-type CXCL12 elicits a similar antimetastatic effect against malignant pancreatic cancer cells. Some have suggested that inhibition of CXCR4 is a strategy for blocking PDAC malignancy (46, 47). However, administration of small-molecule antagonists of CXCR4, such as AMD3100, passively blocks cell movement through competitive inhibition, primarily by blocking one of the two ligand-binding sites on the CXCR4 receptor. Given the CXCR4 antagonist binds only one of two necessary binding sites with micromolar binding affinity, it is perhaps not surprising that AMD3100 failed to provide clinical benefit in long-term treatment of patients with HIV (53). Our report suggests that the CXCL12 protein is an effective alternative to CXCR4 antagonists to inhibit PDAC metastasis. By specifically eliciting the antimigratory downstream signaling components, biased agonist variants of CXCL12 have potential to actively block pancreatic cancer cell movement. Further study is needed to fully uncover the pharmacokinetics/pharmacodynamics of this biologic response modifying drug, although extended biologic therapy for >75 days in our two models of pancreatic cancer herein showed few signs of toxicity. Future studies must also examine how CXCL12 biologic therapy will modulate the immunosurveillance in the pancreatic primary tumor environment. Finally, as pancreatic tumors are rich in GAGs (54, 55), which are known to alter chemokine binding, further investigation into the in vivo interaction between recombinant chemokine and extracellular components is necessary.

In sum, this work uncovers a biochemical mechanism for CXCL12 biphasic regulation of migration that links our previous reports in oligomeric chemokine biased agonism (27) and CXCL12 chemotactic signaling (36). We show that ataxic doses of CXCL12 lead to incomplete turnover of migratory signaling and subsequent locking of migration machinery that is bioenergetic signaling dependent. The cumulative findings of our recent reports underscore the antimetastatic potential of exploiting chemokine biased agonist signaling, address the dire need for identifying strategies to target metastasis, and delineate the importance of future comprehensive preclinical studies of chemokine ligand–specific therapies for pancreatic cancer.

B.F. Volkman and M.B. Dwinell have ownership interests (including patents) in Protein Foundry, LLC. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Roy, N.P. Zimmerman, B.F. Volkman, B. Kalyanaraman, M.B. Dwinell

Development of methodology: I. Roy, N.P. Zimmerman, B. Kalyanaraman, M.B. Dwinell

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Roy, D.M. McAllister, E. Gorse, K. Dixon, C.T. Piper, N.P. Zimmerman, A.E. Getschman, S. Tsai, D.D. Engle, B. Kalyanaraman, M.B. Dwinell

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Roy, E. Gorse, N.P. Zimmerman, B. Kalyanaraman, M.B. Dwinell

Writing, review, and/or revision of the manuscript: I. Roy, A.E. Getschman, D.D. Engle, D.B. Evans, B.F. Volkman, M.B. Dwinell

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.P. Zimmerman, M.B. Dwinell

Study supervision: B.F. Volkman, B. Kalyanaraman, M.B. Dwinell

The work was supported, in part, by CA178960 (M.B. Dwinell and B. Kalyanaraman), AI058072 (B.F. Volkman), as well as a grant from the MCW Cancer Center and philanthropic donations from the Bobbie Nick Voss Charitable Foundation (M.B. Dwinell). I. Roy is a member of the Medical Scientist Training Program at MCW, which is partially supported by an NIGMS training grant (T32GM080202).

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.