Enhanced Glutaminolysis Drives Hypoxia-Induced Chemoresistance in Pancreatic Cancer

Pancreatic cancer is a fatal tumor in humans; approximately 90% of cases are diagnosed as pancreatic ductal adenocarcinoma (PDAC). Despite ongoing advances in PDAC therapies, PDAC displays high resistance against chemotherapies (1, 2). FOLFIRINOX (5-fluorouracil, leucovorin, irinotecan, oxaliplatin) is the current standard of care for PDAC chemotherapy (3), but the overall response rate of patients with metastatic pancreatic cancer to this regimen is less than 32% (4). Therefore, PDAC is still a lethal disease with a 5-year survival rate of less than 10% (5).

PDAC chemoresistance can be induced by cellular and metabolic changes such as altered expression of metabolic enzymes (6, 7), increased drug efflux transporter (8), inhibition of the apoptosis pathway (9), presence of cancer stem cells, and enhanced epithelial-to-mesenchymal transition (EMT; ref. 10). Furthermore, the underlying molecular events of these chemoresistant mechanisms are associated with hypoxia (11–14).

PDAC is characterized by hypoxia (15). Tumor hypoxia is determined by the low oxygenation status of the tumor, derived from an insufficient supply and surplus oxygen consumption (16–18). However, due to the lack of an experimental model that can appropriately replicate the spontaneous hypoxia that occurred in tumor tissue, the possible contributions to hypoxia by enhanced oxygen consumption by cancer cells have not yet been evaluated. To unravel this problem, we invented extracellular matrix (ECM)–based in vitro 3D printing models that efficiently imitate the spontaneous hypoxia of PDAC tumor tissue that is hardly observed in in vitro 2D culture. We conducted a systemic approach using an ECM-based 3D-printed PDAC model to determine whether the enhanced cellular metabolism could be the cause of hypoxia, and whether targeting the altered metabolism in cancer cells could be an effective strategy to overcome chemoresistance.

In response to low-oxygen conditions, cancer cells direct glucose-derived pyruvate away from the TCA cycle (19) and promote glutamine uptake by increasing the levels of glutamine transporters for mitochondrial anaplerosis (20, 21). Glutamine-derived TCA cycle metabolites support reductive carboxylation for lipogenesis (22), and glutamine-driven OXPHOS (oxidative phosphorylation) accounts for the majority of ATP generation even in hypoxic cancer cells (23, 24). Though hypoxia-induced glutamine metabolism is well known, how glutamine metabolism can alter hypoxic responses in cancer cells has not yet been investigated.

In this study, we found that enhanced glutaminolysis within PDACs increases oxygen consumption, resulting in intracellular hypoxia and chemoresistance using in vitro 3D printing models and in vivo mouse models. Consequently, we suggest that targeting the glutaminolysis–oxygen consumption–hypoxic axis has potential as a strategy for overcoming the therapeutic resistance of PDACs.

Capan-1 (RRID:CVCL_0237), HPAF-II (RRID:CVCL_0313), CFPAC-1 (RRID:CVCL_1119), Panc10.05 (RRID:CVCL_1639), Capan-2 (RRID:CVCL_0026), Panc-1 (RRID:CVCL_0480), MiaPaCa-2 (RRID:CVCL_0428), SW1990 (RRID:CVCL_1723), BxPC-3 (RRID:CVCL_0186), and AsPC-1 (RRID:CVCL_0152) cells were obtained from the ATCC. SNU-213 (RRID:CVCL_5034) and SNU-324 (RRID:CVCL_5051) cells were obtained from the Korean Cell Line Bank. Patient-derived PDAC cells (YPAC-02, YPAC-05, YPAC-16, YPAC-25, YPAC-26 and YPAC-30 cells) were provided by Dr. Seungmin Bang (Yonsei University, College of Medicine, Seoul, Korea). All cells were routinely checked for Mycoplasma negativity using MycoAlert Mycoplasma Detection Kit (Lonza) and were used for less than six months within 10 passages. Panc-1, MiaPaCa-2, Capan-1, HPAF-II, SNU-213 and SNU-324 cells were cultured in DMEM with 10% FBS. AsPC-1, BxPC3, Panc10.05 were cultured in RPMI medium with 10% FBS. CFPAC-1 cells were cultured in IMDM. Capan-2 and SW1990 cells were cultured in McCoy's 5A medium and Leibovitz's L-15 medium with 10% FBS, respectively. YPAC-02, YPAC-05, YPAC-16, YPAC-25, YPAC-26, and YPAC-30 cells were cultured in F-medium (25). All cells were grown in a 37°C incubator in a 5% CO₂.

Western blot antibodies were obtained from the following sources: Rabbit polyclonal anti-ABCB1 (NB100–80870, RRID:AB_2174428, 1:1,000), rabbit polyclonal anti-DCK (NBP2–32179, RRID:AB_2927426, 1:500), rabbit polyclonal anti-SLC31A1 (NBP2–93123, RRID:AB_2927427, 1:1,000), rabbit polyclonal anti-GOT2 (NBP2–32241, RRID:AB_2927428, 1:1,000), rabbit monoclonal anti-HIF1A (NB100–105, RRID:AB_10001154, 1:1,000), and rabbit polyclonal anti-EPAS1 (NB100–122, RRID:AB_10002593, 1:1,000) from Novus Biologicals; rabbit monoclonal anti-ABCG2 (sc-58224, RRID:AB_771493, 1:500), rabbit polyclonal anti-SLC1A5 (sc-99002, RRID:AB_2239446, 1:1000), rabbit monoclonal anti-GPT2 (sc-398383, RRID:AB_2927429, 1:1,000), mouse monoclonal anti-parp (sc-8007, RRID:AB_628105, 1:1000), mouse monoclonal anti-PDK1 (sc-515944, RRID:AB_2927430, 1:1,000), mouse monoclonal anti-MYCBP (sc-398624, RRID:AB_2927431, 1:1,000), and mouse monoclonal anti-ACTB (sc-8432, RRID:AB_626630, 1:1,000) from Santa Cruz Biotechnology; rabbit monoclonal anti-SLC38A1 (36057, RRID:AB_2799092, 1:1,000), rabbit monoclonal anti-cleaved caspase-3 (9664, RRID:AB_2070042, 1:1,000), rabbit polyclonal anti–caspase-3 (9662, RRID:AB_331439, 1:1,000), and rabbit polyclonal anti-cleaved parp (9541, RRID:AB_331426, 1:1,000) from Cell Signaling Technology; rabbit polyclonal anti-GLS (MBS9408666, RRID:AB_2927432, 1:1,000) and rabbit monoclonal anti-PPAT (CF504457, RRID:AB_2927433, 1:1,000) from Thermo Fisher Scientific; rabbit polyclonal anti-VEGFA (ab267566, RRID:AB_2927434, 1:500) from Abcam; and goat anti-mouse IgG, HRP (31430, RRID: AB_228307, 1:10,000) and goat anti-rabbit IgG, HRP (31460, RRID:AB_228341, 1:10,000) secondary antibodies from Invitrogen. IHC antibodies were obtained from the following sources: Rabbit polyclonal anti-SLC1A5 (A304–353A, RRID:AB_2620548, 1:1,000) from Bethyl Laboratories; rabbit polyclonal anti-ABCB1 (NBP1–90291, RRID:AB_11026685, 1:500) and rabbit monoclonal anti-CA9 (NBP2–48405, RRID:AB_1959825, 1:1,000) from Novus Biologicals; and rabbit monoclonal anti-cleaved caspase-3 (9664, RRID:AB_2070042, 1:500) from Cell Signaling Technology. The reagents gemcitabine (PHR2582), 5-FU (G6423), oxaliplatin (O9512), cisplatin (PHR1624), BPTES (SML0601), 2-DG (D6134), dimethyl succinate (S0755), metformin (PHR1084), and aciflavine (A8126) were obtained from Sigma-Aldrich.

The pLKO.1 puro vector (RRID:Addgene_8453) was provided by Dr. Jinu Lee (Yonsei University, College of Pharmacy). The pIRESpuro3 vector (RRID:Addgene_64358) was purchased from Clontech. pLKO.1-puro-shSLC1A5_var and pIRESpuro3-SLC1A5_var were established as described previously (20). The pGL4.22-VEGF-HRE:dLUC (5xHRE-reporter construct, RRID:Addgene_128096) was purchased from Addgene. The following primers were used for quantitative PCR: SLC1A5 for, 5′-CTCGATTCGTTCCTGGATCTT-3′, rev, 5′-GTTCCGGTGATATTCCTCTCTTC-3′; SLC1A5_var for, 5′-GCCCTCCCACTATGTACTCTA-3′, rev, 5′-CTACCAAGCCCAGGATGTTC-3′; SLC38A1 for, 5′-CACAGACCAGGATGGAGATAAAG-3′, rev, 5′- GGAATGCTGACCAAGGAGAA-3′; GLS for, 5′-AGGAATGACACCAGGGTTTG-3′, rev, 5′-TCAGACTCACCAACAGCAATAC-3′; GOT2 for, 5′-GTTTGCCTCTGCCAATCATATG-3′, rev, 5′-GAGGGTTGGAATACATGGGAC-3′; GPT2 for, 5′-CATGGACATTGTCTGAACC-3′, rev, 5′-TTACCCAGGACCGACTCCTT-3′; PPAT for, 5′-ACAGTGGCAGAGCAAGATGA-3′, rev, 5′-ACTTCTCCAGTTCAGAGGCT-3′; PDK1 for, 5′-GCTAGGCGTCTGTGTGATTT-3′, rev, 5′-GTATTGGCTGTCCTGGTGATT-3′; VEGFA for, 5′-TTGCCTTGCTGCTCTACCTCCA-3′, rev, 5′-GTATTGGCTGTCCTGGTGATT-3′; EPAS1 for, 5′-GACTTACACAGGTGGAGCTAAC-3′, rev, 5′-GAGACTCAGGTTCTCACGAATC-3′; MYCBP for, 5′-CTCCACACATCAGCACAACTA-3′, rev, 5′-TGTCCAACTTGACCCTCTTG-3′; SLC38A1 for, 5′-CACAGACCAGGATGGAGATAAAG-3′, rev, 5′- GGAATGCTGACCAAGGAGAA- 3′; ABCB1 for, 5′-GCTGTCAAGGAAGCCAATGCCT-3′, rev, 5′-TGCAATGGCGATCCTCTGCTTC-3′; ABCG2 for, 5′-GTTCTCAGCAGCTCTTCGGCTT-3′, rev, 5′-TCCTCCAGACACACCACGGATA-3′; DCK for, 5′-AGTGGTTCCTGAACCTGTTGCC-3′, rev, 5′-GACCATCGTTCAGGTTTCTCATAC-3′; SLC31A1 for, 5′-CGCTACAATTCCATGCCTGTCC-3′, rev, 5′-GACTACCTGGATGATGTGCAGC-3′; and ACTB for, 5′-CTACGTCGCCCTGGACTTCGAGC-3′, rev, 5′- GATGGAGCCGCCGATCCACACGG-3′.

All animal work was performed using protocols approved by the Institutional Animal Care and Use Committee at the Korea Research Institute of Bioscience and Biotechnology (IACUC-A-202103–1230–01). For the in vivo orthotopic tumor assay, 4-week-old male athymic nu/nu mice (RRID:MGI:5652489, 20–22g, Koatech Co., Pyeongtaek, Republic of Korea) were used. The mice were housed at 4 per cage at 24°C under a 12 hour light/dark cycle and were allowed access to food (38057, Purina) and water ad libitum in a specific pathogen-free facility, and their health status was monitored by culture, serum, and microscopic examination.

Each patient was informed about this research and written informed consent was obtained from them. All experiments were performed in accordance with the ethical guidelines of the 1975 Declaration of Helsinki and approved by the Institutional Review Board of Severance Hospital, Seoul, Korea (IRB number 4–2015–0297). Tumor specimens (≤1 cm) were obtained from surgically resected tissues from patients diagnosed with PDAC. Patients with unresectable PDAC underwent endoscopic ultrasonography–guided biopsy or percutaneous biopsy to collect tumor specimens. All tissues were placed into medium containing antibiotics. Adipose tissue was removed using forceps and a scalpel. Tumor tissues were finely cut into small pieces of 1–2 mm with sterile scissors. The dissected specimens were placed in medium. Primary cell lines were isolated within 1–2 hours of tumor resection. If the specimens could not be treated immediately to prepare conditionally reprogrammed cells, the tumor cells were frozen in liquid nitrogen for long-term storage. Tissues were resuspended in collagenase (1 mg/mL, Sigma) in medium and incubated at 37°C for 30 minutes with stirring to dissociate tumor tissue from the collagen matrix. Medium (5x F) was added for neutralization and centrifuged at 1,500 rpm for 3 minutes at 4°C. The supernatant was filtered through a cell strainer (70 μmol/L nylon, Falcon). The filtered tumor cells were resuspended in F-medium consisting of Keratinocyte-SFM (Life Technologies) supplemented with prequalified recombinant epidermal growth factor and bovine pituitary extract (Life Technologies), 2% FBS (Sigma), and 1% antibiotic–antimycotic (Life Technologies). The filtered tumor cells were seeded into 6-well plates with lethally irradiated (30 Gy) J2 mouse fibroblasts. Cells were cultured at 37°C with 5% CO₂. Tumor cells on the plates were readily apparent by morphology relative to stromal elements (e.g., fibroblasts). Contaminated stromal elements were removed by differential trypsin treatment or by selectively scraping the plate as needed. In every passage, we scrutinized using a phase-contrast microscope and tapped the cultures gently to detach the cells without contaminated stromal elements during trypsinization. Cell lines were pretreated with 500 ng/mL Mycoplasma remover (MP Biomedicals). The resulting cell lines were regularly checked for Mycoplasma infection. After the establishment of patient-derived PDAC cell lines, we did not use an irradiated J2 mouse fibroblasts feeder cell for supporting patient-derived PDAC cells. We established all patient-derived PDAC cells in 2D culture plates and then used them for 3D printing.

Gelatin powder (Sigma) was dissolved in a 0.9% NaCl solution (w/v) at 20% (w/v). Sodium alginate powder (Sigma) was dissolved in a 0.9% NaCl solution (w/v) at 4% (w/v). To optimize the gelatin: Alginate ratio, a hydrogel preliminary test was carried out by preparing ratios of 12:5, 10:5, 8:5, 6:5, and 4:5. In the main experiment, the optimal ratio (10:5) was used. Both the gelatin solution and the sodium alginate solution were sterilized by heating three times in an oven (70°C) for 30 minutes. Fibrinogen powder (Sigma), hyaluronic acid (Sigma), and collagen powder (Sigma) were dissolved in the gelatin/alginate solution at a concentration of 0.3∼2.0 mg/mL. The solutions were heated three times at 70°C in an oven for 1 hour to complete sterilization before being stored at 4°C.

For 3D printing, patient-derived PDAC cells and pancreatic cancer cell lines were collected by centrifugation at 900 rpm for 3 minutes and suspended in an ECM-based hydrogel solution at 37°C. The hydrogel/cell solution was loaded into a metal syringe, which was heated at 37°C to melt the hydrogel solution. In the printing process, the hydrogel was printed through a 24-gauge needle at a pressure of 50∼80 kPa. The line pattern was extracted at a dispensing speed of 0.1182 mm/s, and the velocity of x–y axis movement was a feed rate of 100 mm/min. The dotting pattern was extruded at a dispensing speed of 0.0534 mm/s. Cancer cells were plated in 96-well plates at a density of 3×10³ cells per well per one dotting. The temperature in the chamber was maintained at 10°C during the printing process. After printing, the construct was then immersed in thrombin (Sigma, 20 U/mL) and CaCl₂ (3%, w/v) for 15 minutes to crosslink fibrinogen and alginate. The 3D-printed constructs were gently washed in cold PBS, and growth medium was added. 3D round bottom cultures were generated in ultra-low attachment (ULA) 96-well round-bottomed plates (Corning) at a density of 3 × 10³ and cultured under standard culture conditions.

YPAC-25, YPAC-26, YPAC-30, YPAC-02, YPAC-05, YPAC-16, Capan-1, and Panc-1 cells were transfected with cDNA constructs by using TurboFect transfection reagent (Thermo Fisher Scientific, R0531). The experiments performed in this study used cells 24 hours after transfection. For siRNA transfection, cells were transfected with siRNA by using Lipofectamine 2000 reagent (Thermo Fisher Scientific, 11668500). Forty-eight hours after transfection, the cells were used for the experiments. The following siRNA sequences were used in the experiment. siNDUFS2: 1: CACAGAGAAGUCUGCUACA, 2: UGUAGCAGACUUCUCUGUG, 5′UTR siHIF1A: 1: GUGGUUGGAUCUAACACUA, 2: UAGUGUUAGAUCCAACCAC.

YPAC-25, YPAC-26, YPAC-30, YPAC-02, YPAC-05, YPAC-16, Capan-1, and Panc-1 cells were transfected with pIRESpuro3, including the 5xHRE-reporter construct (pGL4.22-VEGF-HRE::dLUC) cDNA using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's protocol. Forty-eight hours after transfection, the cells were subcultured in medium containing puromycin (5 μg/mL). After two weeks, the surviving cells were observed; large, healthy colonies were isolated using a cloning cylinder (Sigma) and continually maintained in medium containing puromycin (Sigma). Single cells from resistant colonies were transfected in 96-well plates to confirm that they could grow as puromycin-resistant colonies. YPAC-25, YPAC-26, YPAC-30, and Capan-1 cells were transfected with pIRESpuro3 that included SLC1A5_var cDNA using Lipofectamine 2000 reagent (Invitrogen) as described above. To generate stable shSLC1A5_var-knockdown cells, YPAC-02, YPAC-05, YPAC-16 and Panc-1 cells were transfected with shSLC1A5_var with psPAX2 (RRID:Addgene_12260) 2nd packaging and pMD2.G envelope plasmids (RRID:Addgene_12259) using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's protocol. Virus-containing supernatants were collected 48 hours after transfection. Cells were infected with 0.45-μm–filtered viral supernatant in the presence of 10 μg/mL polybrene (Sigma) for 24 hours. Infected cells were selected with 5 μg/mL puromycin.

Two-dimensional–cultured and 3D-printed pancreatic cancer cells were plated in 96-well plates at a density of 3 × 10³ cells per well and initially cultured in the appropriate medium for 6 days. For viable cell counting, the cells were collected by centrifugation at 900 rpm for 3 minutes and suspended in an ECM composition solution. Then, we spun the culture down to separate the cancer cells from the hydrogel. Viable cell counting was performed in triplicate using the Accu-chip and a digital cell counter (Digital Bio.). The viability was automatically calculated in ADAM software after the total cell count and the nonviable cell count were measured separately. To measure growth inhibition, cells were cultured for 5 days under 2D- and 3D printed conditions and treated with a combination of gemcitabine, 5-FU, oxaliplatin or cisplatin (10 and 100 nmol/L, 1, 10, 50, 100, 250, 500, and 750 μmol/L, or 1 mmol/L) and BPTES (500 μmol/L), metformin (1 mmol/L) or acriflavine (1 μmol/L) for 24 hours. Cell proliferation was measured using the WST-1 assay, in which the highly sensitive water-soluble tetrazolium salt is used to produce a water-soluble formazan dye upon reduction in the presence of mitochondrial dehydrogenases. The absorbance at 450 nm was measured using a microplate reader (Infinite 200 PRO, Tecan) and normalized to the control.

Two-dimensional–cultured and 3D-printed pancreatic cancer cells were plated in 96-well plates at a density of 3 × 10³ cells per well and kept initially in the appropriate medium. After 5 days, cancer cells treated with a combination of gemcitabine (50 μmol/L), 5-FU (50 μmol/L), oxaliplatin (100 μmol/L) or cisplatin (100 μmol/L) and BPTES (500 μmol/L), metformin (1 mmol/L) or acriflavine (1 μmol/L) and verapamil (50 μmol/L) for 24 hours. The Annexin V assay (Thermo Fisher Scientific) was performed according to the manufacturer's protocol. To measure apoptosis, cells were harvested, stained with Annexin V-red (Sartorius) and incubated for 30 minutes at room temperature in the dark, followed by analysis using flow cytometry (BD FACSAria III). The effects of glutamine and serine starvation or metabolic inhibitors BPTES (500 μmol/L), 2-DG (5 mmol/L), or etomoxir (50 μmol/L) and verapamil (50 μmol/L) on calcein AM retention were determined using the Vybrant Multidrug Resistance Assay Kit (Thermo Fisher Scientific). Cells were pretreated with glutamine and serine starvation or metabolic inhibitors as indicated for 24 hours, followed by the addition of calcein AM (0.25 μmol/L) for 30 minutes. Cells were then washed and dissolved in DMSO, and the absorbance of each sample at 515 nm was then measured using a microplate reader (Infinite 200 PRO, Tecan). Calcein retention was normalized by viable cells counting using the Accu-chip and an ADAM digital cell counter.

For real-time PCR analysis, RNA was isolated using the MiniBEST Universal RNA Extraction Kit (TAKARA) according to the manufacturer's instructions. Reverse transcription of 1,000 ng of RNA was performed using the PrimeScript 1st strand cDNA Synthesis Kit (TAKARA). For qRT-PCR analysis, cDNA was diluted in nuclease-free water (1:5), and gene expression levels were analyzed using Step One Plus (Applied Biosystems). Expression levels were normalized to ACTB.

After the cells were lysed with buffer containing 40 mmol/L HEPES (pH 7.4), 0.5% Triton X-100, 10 mmol/L β-glycerol phosphate, 10 mmol/L pyrophosphate, and 2.5 mmol/L MgCl₂ supplemented with 5 μg/mL protease inhibitor, the lysates were incubated on ice for 30 minutes and centrifuged at 12,500 × g for 10 minutes at 4°C. After lysis, cell lysates (20 μg of total protein) were diluted in SDS-sample buffer and heated at 95°C for 5 minutes before being loaded on a 10% SDS-polyacrylamide gel and electrophoresed. Then, the proteins were transferred to polyvinylidene difluoride membranes (Merck) using a Trans-Blot Turbo Blotting System (Bio-Rad). After blocking in TBST buffer containing 5% skim milk or 5% BSA, the membranes were incubated with individual primary antibodies overnight. The next day, the membranes were incubated with either anti-mouse or anti-rabbit IgG conjugated with horseradish peroxidase (Invitrogen). Immunoblot signals were detected by a MicroChemi (DNR Bio-Imaging Systems) with enhanced chemiluminescence, EzWestLumi (ATTO).

Glucose levels were determined using a Glucose Colorimetric Assay Kit II (BioVision). Glutamine levels were determined using a Glutamine Detection Assay Kit (BioVision). Fatty acid levels were determined using a Free Fatty Acid Quantification Colorimetric Kit (BioVision) in accordance with the manufacturer's instructions. Reactions with the compound of interest produced stable colorimetric signals that were measured as the absorbance at a specific wavelength (Glucose Colorimetric Assay Kit II and Glutamine Detection Assay Kit: 450nm, Free Fatty Acid Quantification Colorimetric Kit: 570 nm) by a microplate reader (Infinite 200 PRO, Tecan). Plates included vehicle wells with no sample for background subtraction. The amounts of each nutrient were normalized by viable cells counting using the Accu-chip and an ADAM digital cell counter.

Specifically, a total of 1 × 10⁷ cells were collected from 2D-cultured and 3D-printed pancreatic cancer cells. Cells were collected by centrifugation at 900 rpm for 3 minutes and suspended in an ECM composition solution. The cell pellet was then quickly washed twice with DPBS, followed by a brief wash with Optima LC/MS water and subsequent dilution with 1 mL of 80% methanol in water. Samples were vortexed vigorously for 1 minute and spun down at 17,500 × g for 15 minutes, and the supernatants were analyzed with LC/MS to determine the total moles of metabolite in each sample. Whole-cell metabolites were calculated as relative measures of “scaled intensity” after normalization to protein. For ultraperformance liquid chromatography-triple-quadrupole mass spectrometry (UPLC-TQ-MS) analysis, the metabolite extract was dried and redissolved with 20% (v/v) acetonitrile in water. Agilent 1290 Infinity II LC and Agilent 6495 Triple Quadrupole MS systems equipped with an Agilent Jet Stream ESI source (Agilent Technologies) were used for analysis. MassHunter (v B.06.00, Agilent Technologies) software was used for data acquisition and analysis. LC separations were carried out on a Kinetex C18 column (100 × 2.1 mm, particle size of 1.7 μmol/L, Phemomenex) and a Kinetex Polar C18 column (100 × 2.1 mm, particle size of 2.6 μmol/L, Phemomenex). MS was performed with the following settings: Gas temperature of 220°C, nebulizer gas of nitrogen at 40 psi, sheath gas temperature of 300°C, and sheath gas flow rate of 12 L/min. Quantification was performed in multiple reaction monitoring mode, and the optimized conditions for each metabolite were achieved by using flow injection of individual standard compound solutions (100 ng/mL) into the mass spectrometer.

The oxygen consumption rate (OCR) was determined with an XFe24 extracellular flux analyzer (Agilent Technologies) as described in the manufacturer's protocol. A total of 3 × 10³ cells cultured in 2D and 3D were seeded per well in 24-well microcell culture plates (Agilent Technologies) in the appropriate medium and incubated at 37°C overnight in a 5% CO₂ incubator for 5 days. Then, the growth medium was replaced with phenol red- and bicarbonate-free DMEM at a pH of 7.4, and the cells were incubated at 37°C in a non-CO₂ incubator to equilibrate the CO₂ level to that in the atmosphere. Using the XFe24 analyzer, OCRs were measured under baseline conditions and under treatment with glutamine or serine starvation, with BPTES (500 μmol/L), etomoxir (50 μmol/L), metformin (1 mmol/L), and 2-DG (2-deoxyglucose, 5 mmol/L), with dimethyl succinate (4 mmol/L) for 24 hours. The basal OCR values were normalized to the viable cell number and analyzed using WAVE software (Agilent Technologies).

Staining with the hypoxia-sensitive probe MAR (Goryo Chemical) was used to detect intracellular oxygen tension. 2D-cultured and 3D-printed pancreatic cancer cells were seeded in 96-well plates at a density of 3×10³ cells per well. Then, the 2D-cultured and 3D-printed cells were incubated for 6 days, stained with 500 nmol/L MAR dye for 24 hours and analyzed with the IncuCyte ZOOM system (Sartorius), and quantitation was performed with the accompanying commercial software.

Pancreatic cancer cells stably expressing a hypoxia response element (HRE)–dependent luciferase reporter construct (pGL4.22-VEGF-HRE:dLUC) were established. Cells were plated in replicate sets at 3×10³ cells/well in 96-well plates for 2D and 3D culture. The cells were exposed to BPTES (500 μmol/L), metformin (1 mmol/L), acriflavine (1 μmol/L), N-acetyl-L-cysteine (NAC, 10 mmol/L), or H₂O₂ (50 μmol/L) treatment for 24 hours. Medium containing D-luciferin potassium salt (Sigma) at a final concentration of 150 μg/L was added. After 5 minutes of incubation, the cells were imaged for 1–5 seconds in an in vivo imaging (IVIS) 100 system (Xenogen). Luciferase activity was determined using an IVIS Lumina (PerkinElmer) with a 1-minute exposure time and medium binning.

Two-dimensional–cultured and 3D-printed pancreatic cancer cells were plated in 96-well plates at a density of 3 × 10³ cells per well and kept initially in the appropriate medium. After 5 days, the cells were exposed to NAC (10 mmol/L), BPTES (500 μmol/L) or H₂O₂ (50 μmol/L) treatment for 24 hours. Intracellular reactive oxygen species (ROS) were determined by staining cells with 1 μmol/L of 5-(and-6)-carboxy-2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA, Invitrogen) for 15 minutes according to the manufacturer's protocol (492/517 nm). Stained cells were analyzed using a flow cytometer (BD FACS ARIA III).

RNA-seq library processing was conducted as described below. For the RNA-seq data from each sample, the quality of the raw data was controlled through FastQC (RRID:SCR_014583). TopHat (RRID:SCR_013035) was used to map reads from the RNA-seq experiment to a reference genome (Homo sapiens GRCh38.94). Then, the mapped reads were counted using HTSeq (RRID:SCR_005514) with the following parameters: htseq-count -s no -m intersection-nonempty -f bam. We conducted differential gene expression analysis using the DESeq2 package (v. 1.24.0, RRID:SCR_000154) and manipulated the data in R (v3.6.1). We filtered out the genes with low counts (<10) and applied a regularized log transformation to minimize differences between samples with small counts. Then, the fold change (FC) value was calculated on the basis of the housekeeping gene ACTB. The regularized log-transformed FPKM value of ACTB was subtracted from the regularized log-transformed FPKM values. Finally, the regularized log-transformed FPKM and rlog FPKM FC values were visualized using various software or libraries in R.

We validated the genetic similarity of the established 2D and 3D-printed cells with the original PDAC tissue by comparing gene DNA sequences to develop representative genomic data of PDAC. Targeted sequencing was performed using the Cancer-SCAN panel (83-gene panel at ∼900×), and whole-exome sequencing (WES) was performed using the Novaseq6000 system. Initially, DNA obtained by the microdissection of formalin-fixed paraffin-embedded PDAC tissue and DNA from pancreatic 2D and 3D-printed cells of the same patient were sequenced. DNA was extracted using the QIAamp DNA Mini Kit (Illumina). The quality of the DNA was evaluated using a Nanodrop spectrophotometer (Thermo Fisher Scientific).

DNA preprocessing of raw sequencing data was conducted by quality filtering using FastQC and trimming adapters using trim_galore (v0.6.7, RRID:SCR_011847). Each trimmed sequencing read was aligned to the human reference genome (UCSC hg38, RRID:SCR_005780) using BWA-MEM (v0.7.17, RRID:SCR_017619). After converting bam files using Samtools (v1.11, RRID:SCR_002105), we regulated duplicate bias by Picard tools (v2.27, http://broadinstitute.github.io/picard/, RRID:SCR_006525). According to the Genome Analysis Toolkit (GATK, RRID:SCR_001876) best practice workflow, the location of insertions and deletions was recalibrated on the basis of the dbSNP database (v150 RRID:SCR_002338), of known variants. SNPs and indels were identified in each sample by GATK- Mutect2 (v4.2.6.1). Finally, all somatic mutations were annotated with genetic features using the ANNOVAR tool to obtain the common somatic mutation list based on the Catalog of Somatic Mutation in Cancer database (COSMIC, RRID:SCR_002260). The mutations detected in both the primary tumor and paired 2D or the primary tumor and paired 3D printing of the same patient were defined as concordance mutations. The SNP similarity (concordance rate) was calculated using the formula below for primary-2D or -3D printing pairs. SNP similarity (concordance rate) = count of shared SNPs/count of total SNPs.

Gene set enrichment analysis (GSEA; RRID:SCR_003199) measured the differences in the non-R (nonresponder) group compared with the R (responder) group in 2D- and 3D-printed culture conditions. According to the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway metabolic gene set, we divided subsets and calculated rlog FPKM FC values. Then, for each gene set, the mean of the rlog FPKM FC values of each sample was calculated. This enabled an overall comparison between responder samples and nonresponder samples in the 2D and 3D groups. We subtracted the mean value of the 2D-R samples from the mean value of the 3D-NR samples. Likewise, the mean value of the 2D-NR samples was subtracted from the mean value of the 2D-R samples. Then, a plot was created using the graphics package (v3.6.1) in R. We visualized rlog FPKM FC values with a Heatmap to see the overall changes in gene expression. The Heatmap (v1.0.12) package in R was used without clustering and with row scaling.

To prepare the scatter plot, we calculated the mean rlog FPKM FC value for genes related to glutamine metabolism, multidrug resistance, and the hypoxia pathway in each sample. By using a graphics package (v3.6.1) in R, we ensured that the points represent the mean value of each sample.

The correlation between metabolic pathways and genes was automatically calculated by Cytoscape software (v3.8.2, RRID:SCR_ 003032; ref. 26). The correlation coefficients compared with the rlog FPKM FC values were calculated by using the stat package (v4.0.3) with the “Pearson” method in R. Then, the genes that were highly correlated with another gene (> 0.9) were selected as input and used to create the Cytoscape network. The input file consisted of 5 metabolic pathway columns; correlation values between gene groups are shown.

The Cancer Genome Atlas (TCGA) RNA-seq and clinical datasets (i.e., level 3 data) from patients with pancreatic cancer were obtained from the Broad Institute GDAC Firehose (v2016_01_28). According to the Ragnum signatures and Creighton signatures, hypoxia scores were quantified on the basis of TCGA PAAD gene expression values. These scores were combined with the corresponding clinical data, and comparison analysis between the gemcitabine responder and nonresponder groups was performed using the ggplot2 (RRID:SCR_014601) package in R (v4.0.3). The raw sequencing files obtained from pancreatic cancer cells that were cultured under different conditions (2D and 3D) were trimmed and aligned with a reference Homo sapiens genome (GRCh38.p13, available at http://ftp.ensembl.org/pub/release-102/fasta/homo_sapiens/dna/) using Trim Galore (v0.6.5, RRID:SCR_011847) and HISAT (v2.2.0). We used featureCounts (v2.0.0) software to generate the count matrix and the TPM normalization method to calculate the log FC and P value. Hierarchical clustering is presented in the form of a heatmap prepared with the Pheatmap (RRID:SCR_016418) R package using the Euclidean distance. For GSEA, we organized our genes by their log fold-change values. Different collections of molecular signatures database (MSigDB) gene sets were used to detect the pathway enrichment scores.

Four-week-old male athymic nu/nu mice were used for the stereotactic injection of pancreatic cancer cells. Pancreatic cancer cells were treated with trypsin/EDTA (Sigma) and washed twice with FBS-supplemented serum-free medium. The mice were anesthetized using isoflurane, and a small incision was made in the left abdomen. A cell (5 × 10⁵ cells) suspension was injected into the tail of the mouse pancreas under isoflurane (Pfizer). A swab was held at the injection site for 1 minute to prevent leakage of the pancreatic tumor cells. The peritoneum and skin incisions were sequentially sutured with absorbable sutures. Buprenorphine was administered to the mouse as an analgesic every 8 hours for 2 days. Tumor volume (n = 6) and survival time to death (n = 8) were measured. Two weeks after inoculation, when the tumor size reached approximately 30 mm³, mice were treated with control (10% DMSO in PBS, every 3 days), gemcitabine (100 mg/kg, i.p., every 3 days) or gemcitabine (100 mg/kg, i.p., every 3 days) with BPTES (12.5 mg/kg, i.p., every 3 days), metformin (250 mg/kg, i.p., every 3 days), or acriflavine (1 mg/kg, i.p., every 3 days) for 3 weeks. At the end of the study, the mice were sacrificed by CO₂ asphyxiation and the tumor volumes were measured. Survival curves were evaluated using the Kaplan–Meier method by counting dead mice; drugs were administered up to 80 days after orthotopic transplantation in the same manner as above.

Stably transfected 5xHRE luciferase-expressing pancreatic cells were orthotopically transplanted into the pancreas by the above method. The growth of the orthotopically implanted tumors was monitored by measuring luminescence via an IVIS (Caliper Life Sciences). Two weeks after inoculation, treatment with control (10% DMSO in PBS, every 3 days), BPTES (12.5 mg/kg, i.p., every 3 days), metformin (250 mg/kg, i.p., every 3 days), or acriflavine (1 mg/kg, i.p., every 3 days) was performed for 3 weeks. To determine the basal luminescence value, mice were imaged on the same day after the injection of D-luciferin (3 mg/mouse, i.p.) for 10 minutes, and luminescence was measured. The experiment was terminated on day 30 by humanely euthanizing the mice using a CO₂ overdose, and tumor volume was measured for normalization of luciferase activity. Representative mice from the control and treatment groups were imaged for luminescence.

Cells were placed in a hypoxia chamber (STEMCELL Technologies) flushed with 100 L of a gaseous mixture containing 1% O₂ and 5% CO₂ balanced with N₂ and then incubated at 37°C. The hypoxia chambers were used to measure growth inhibition and cell death.

Patient pancreatic cancer tissue was provided by Dr. Seungmin Bang (Yonsei University, College of Medicine). A pancreas cancer tissue array with 96 cases/192 cores was purchased from US Biomax (PA1921a). IHC was carried out on a fully automated Ventana Discovery Ultra instrument (Roche Diagnostics International AG). Formalin-fixed pancreatic cancer tissue sections were deparaffinized and antigen retrieval was performed with CC1 buffer (Roche Diagnostics). Each sample was treated with anti-CA9 antibody (1:1,000 dilution), ABCB1 antibody (1:500 dilution) and SLC1A5 (1:1,000 dilution) and incubated at 37°C for 32 minutes. The slides were developed using the Ultramap DAB Staining Kit (Roche Diagnostics) according to the manufacturer's instructions. The percentage of tumor cells positive for antibody binding was determined in at least three areas at ×100 magnifications and the values were averaged. On the basis of the mean percentage, the tissue sections were then assigned to one of five categories: 0, no cancer cells stained; 1, 0%–10% of cancer cells stained; 2, 11%–50% of cancer cells stained; 3, 51%–75% of cancer cells stained; 4, more than 75% of cancer cells stained. All IHC results were obtained by multiplying the positive cell percentage and staining intensity for each case.

All fluorescence images were analyzed with Zen imaging software (Zeiss), and the background was quantified by subtraction with ImageJ software (RRID:SCR_003070). All statistical analyses were performed using GraphPad Prism 8 software (RRID:SCR_002798). Statistical data were analyzed using one-way ANOVA and Tukey's multiple comparison tests; the quantitative data are shown as the mean ± SD. Associated P values, indicating statistical significance are indicated as follows: *, P < 0.05; **, P < 0.001; ***, P < 0.001; ns, not significant P > 0.05. No additional statistical tests for data distributions were performed.

All sequencing data created within this study were uploaded to the NCBI Gene Expression Omnibus (GEO, RRID:SCR_005012) and are available under the accession codes GSE198192 and GSE214894 for RNA-seq (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE198192, www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE214894). Source data are provided with this paper. Other information needed is available from the corresponding authors upon reasonable request.

To obtain an in vitro experimental model that replicates the clinical properties of in vivo pancreatic cancer tissues, we developed a method for creating 3D cell constructs that mimic the ECM of PDAC tissues by using a gelatin–alginate hydrogel for 3D-printing and ECM components, including fibrinogen, hyaluronic acid, and collagen, which are known components of the PDAC microenvironment (Fig. 1A; Supplementary Fig. S1A; refs. 27–31). The optimal condition was constructed by combining the gelatin-alginate and ECM components in a ratio suitable for Panc-1 cell growth (Supplementary Fig. S1B–S1D). The 3D-printed lines with cells were smooth with a consistent thickness, and the dots were finely printed with a consistent diameter and volume (Supplementary Fig. S1E–S1I).

On the basis of responsiveness to chemotherapy, we divided the patients with PDAC into responder (YPAC-25, 26, 30) and nonresponder groups (YPAC-02, 05, 16) and established the corresponding patient-derived PDAC cells using the “conditional reprogramming” technique (32), enabling indefinite growth under defined conditions (ref. 25; Supplementary Fig. S1J). In an orthotopic mouse model using these patient-derived PDAC cells, we confirmed a significant inhibition of tumor growth in the responders compared with the nonresponders in the gemcitabine-treated group, indicating that the chemoresistance of the patient-derived PDAC cells was preserved in an orthotopic mouse model (Supplementary Fig. S1K and S1L).

Next, we compared the growth of the patient-derived PDAC cells and conventional PDAC cell lines cultured in a 2D plane, the 3D-printed, and a 3D round-bottom culture system and observed that 3D-printed cells showed significantly higher cell growth compared with those cultured in a 2D plane or the 3D round-bottom culture system (Supplementary Fig. S1M and S1N) and decreased cell death compared with those cultured in the 3D round-bottom culture system (Fig. 1B and C). We confirmed whether these culture conditions replicate responsiveness to chemotherapy by monitoring the intracellular uptake of calcein AM, which is a marker of chemoresistance (33). Interestingly, in the 3D-printed system, the patient-derived PDAC cells from the nonresponder group showed significantly reduced calcein AM uptake, reflecting the clinical outcome of drug treatment (Fig. 1B and D), and significantly elevated mRNA and protein expression of ABCB1, ABCG2, DCK, and SLC31A1, which are genes related to drug efflux (ref. 34; Fig. 1E and F). Furthermore, when the GI₅₀ values of the anticancer drugs gemcitabine, 5-fluorouracil, oxaliplatin, cisplatin, and FOLFIRINOX were measured under 2D or 3D-printed conditions of patient-derived PDAC cells (Supplementary Fig. S2A–2SE), the 3D-printed environment considerably increased the GI₅₀ values for the anticancer agents in cells from the nonresponder group compared with cells from the responder group (Fig. 1G–K). Similarly, among conventional PDAC cells, we found that Panc-1, MiaPaCa-2, SNU-213, SW1990, BxPC-3, and AsPC-1 cells showed significantly increased GI₅₀ values for the anticancer agents (Supplementary Fig. S3A–S3E) and increased protein expression of drug-efflux genes in the 3D-printed environment compared with 2D planar culture (Supplementary Fig. S3F). Consistent with their increased chemoresistance, 3D-printed nonresponder group displayed decreased anticancer drug-induced cell death compared with that of the 3D-printed responder group (Fig. 1L–O; Supplementary Fig. S3G–S3R). Three-dimensional–printed Panc-1 and MiaPaCa-2 (nonresponder-like) cells showed a level of chemoresistance similar to that of 3D-printed cells from the nonresponder group and showed decreased anticancer drug-induced cell death compared with that of 3D-printed Capan-1 or HPAF-II (responder-like) cells (Supplementary Fig. S3S). To strengthen the relationship between multidrug resistance (MDR) and chemoresistance in the 3D-printed nonresponder group, we examined the effect of MDR pump inhibitor verapamil on anticancer drug responses in PDAC cells. Verapamil treatment significantly rescued anticancer drug-induced cell death in the 3D-printed nonresponder group, indicating the essential role of drug-efflux activity in chemoresistance (Fig. 1L–O; Supplementary Fig. S3S). These data suggest that the ECM-based 3D printing model could better replicate anticancer drug responses in PDAC cells than in a 2D planar and might be beneficial to reflect drug responses from individual patients with PDAC.

To further investigate the clinical relevance of the 3D-printed system, we analyzed and compared the global transcriptome through RNA-seq of patient-derived PDAC cells from 2D planar, 3D-printed, orthotopic mouse xenograft, and human primary PDAC tissues. Because primary tumor tissues from YPAC-05 (nonresponder) and YPAC-25 (responder) patients are only available for RNA-seq, we used only these two patient primary tissue samples for transcriptome analysis. On the basis of the heatmap data from global transcriptome analysis (7,595 genes total), we found similar gene expression patterns in the responder and nonresponder groups under 2D culture conditions (Supplementary Fig. S4A). In contrast, under 3D-printed conditions, the nonresponder group displayed different gene expression patterns from the responder group (Supplementary Fig. S4A). In addition, the gene expression signatures of orthotopic and primary tumor tissues of the nonresponder group resembled those of the 3D-printed nonresponder group more than those of the 2D planar group (Supplementary Fig. S4A). For quantitative comparison, we screened differentially expressed genes (DEG) that increased or decreased concurrently with orthotopic or primary tumor tissues in 2D planar or 3D-printed systems with patient-derived PDAC cells; we identified 493/7,595 (6.5%) shared DEGs among PDAC cells in 3D printed, orthotopic, and primary tissues (Supplementary Fig. S4B), whereas 133/7595 (1.8%) DEGs were shared by 2D culture, orthotopic, and primary tissues (Supplementary Fig. S4C). These data suggest that the 3D-printed model could recapitulate the transcriptome of the primary PDAC tissues.

To determine whether the 3D-printed system maintains the genetic information of the patient tissues, we performed WES of patient-derived PDAC cells in a 2D planar and a 3D-printed system. Because we had surgically resected YPAC-05 and -25 tissues available and genetic information from YPAC-02 and -26 obtained through WES immediately after biopsy, we could compare this genetic information with that of 2D-cultured or 3D-printed patient-derived PDAC cells. The genetic information of the PDAC cells was not changed, and similarity analysis displayed consistency of over 97% (Supplementary Fig. S4D) and the 2D-cultured or 3D-printed PDAC cells and primary PDAC tissues had a shared identity of over 98% (Supplementary Fig. S4D). Furthermore, we observed 100% shared identity of SNP, deletion, and frameshift mutations of 32 key genes, including KRAS, TP53, and PTEN, between 2D-cultured, 3D-printed, and primary PDAC tissues, indicating that the genetic alterations of the patient-derived PDAC cells are consistently maintained without being degraded in the process of 2D culture or 3D-printing experiments (Supplementary Fig. S4E).

To understand and identify the cause of chemoresistance of the 3D-printed nonresponder group, we performed transcriptional profiling of cells cultured in either a 2D plane or the 3D-printed system. The 3D-printed nonresponder group had highly enriched gene signatures, including metabolic pathways, growth factor binding, ECM organization, and hypoxic signaling (Fig. 2A). To further analyze the metabolic pathway-related genes most changed in the 3D-printed nonresponder group, we investigated the KEGG pathways enriched in the nonresponder group compared with the responder group. Interestingly, amino acid metabolism pathways, including the glutamine and glutamate metabolism pathways, carbohydrate metabolism pathways, and lipid metabolism pathways, were highly enriched in cells from the nonresponder group cultured in the 3D-printed system compared with the responder group (Fig. 2B). Among genes in glutamine metabolism pathways, GLS, GOT, GPT, PPAT, SLC1A5, and SLC38A1 were specifically enriched in cells from the nonresponder group in the 3D-printed system (Fig. 2C). Interestingly, we observed a strong correlation between MDR gene expression and glutamine metabolism gene expression in the 3D-printed nonresponder group but not the 3D-printed responder group or cells from either group cultured in a 2D plane (Fig. 2D). Increased mRNA and protein levels of glutamine metabolism-related factors were observed in the 3D-printed nonresponder group compared with the responder group (Fig. 2E; Supplementary Fig. S5A–S5C). These results indicated that the 3D-printing conditions allowed for the discrimination of specific differences in glutamine metabolism-related gene and protein expression between the responder and nonresponder groups.

Consistent with the increased glutamine metabolism-related gene and protein expression in 3D-printed cells from the nonresponder group compared with those from the responder group, glutamine uptake, but not glucose and fatty acid uptake (Fig. 2F–H), and glutamine-derived TCA metabolites generated through mitochondrial glutaminolysis (Fig. 2I–L) were specifically increased in the 3D-printed nonresponder group. Similarly, 3D-printed Panc-1 and MiaPaCa-2 (nonresponder-like) cells, but not Capan-1 or HPAF-II (responder-like) cells, showed increased glutamine uptake (Supplementary Fig. S5D–S5F) and glutamine-derived TCA metabolites (Supplementary Fig. S5G–S5J). Deprivation of glutamine, but not serine, selectively induced cell death (Fig. 2M; Supplementary Fig. S5K) and reduced MDR as based on calcein AM retention (Fig. 2N; Supplementary Fig. S5L) in the 3D-printed nonresponder group and in Panc-1 and MiaPaCa-2 (nonresponder-like) cells, indicating the glutaminolysis dependency of cell viability and chemoresistance.

Because amino acid metabolism, carbohydrate metabolism, and lipid metabolism pathways were highly enriched in cells from the nonresponder group cultured in the 3D-printed system (Fig. 2B), we examined which metabolic pathway is primarily involved in cell viability and chemoresistance. The glutaminase inhibitor BPTES, but not the glucose analog 2-DG or the fatty acid oxidation inhibitor etomoxir, selectively induced cell death (Fig. 2O; Supplementary Fig. S5M) and reduced MDR (Fig. 2P; Supplementary Fig. S5N) in the 3D-printed nonresponder group and in Panc-1 and MiaPaCa-2 (nonresponder-like) cells. Together, these results suggest that increased glutamine metabolism is related to chemoresistance in PDAC.

Because chemoresistant PDAC cells showed glutamine dependency and the chemoresistance of pancreatic tumors is related to hypoxia (35–37), we hypothesized that increased glutaminolysis would be related to chemoresistance by contributing to tumor hypoxia. Therefore, we first examined whether glutamine metabolism and hypoxia are correlated in chemoresistant PDAC. Cytoscape network analysis of transcriptional profiling between the responder and nonresponder groups showed that glutamine metabolism-related genes were closely related to hypoxia-related genes in the 3D-printed nonresponder group (Fig. 3A). In addition, the expression of glutamine metabolism-related genes was positively correlated with that of hypoxia-related genes in the 3D-printed nonresponder group, but not in the 3D-printed responder group or 2D-cultured cells (Fig. 3B). Furthermore, we observed a strong correlation between MDR gene expression and hypoxia-related gene expression in the 3D-printed nonresponder group, but not the 3D-printed responder group or 2D-cultured cells (Fig. 3C). Consistently, an analysis of TCGA pancreatic cancer data (PAAD) showed a positive correlation between the expression of glutamine metabolism-related genes such as SLC1A5, SLC38A1, SLD38A5, ASL, CAD, PPAT, CIPS2, GOT2, GLUD1, GMPS, and ALDH18A1 and that of hypoxia-related genes in the nonresponder group but not the responder group (Supplementary Fig. S6A). In addition, single-sample GSEA of responses to gemcitabine showed that the enrichment scores of genes related to glutamine metabolism and hypoxia were higher in samples with a poor response than those in samples with a good response (Fig. 3D; Supplementary Fig. S6B).

Furthermore, immunohistochemical analysis of the expression of glutamine metabolism, hypoxia, and MDR markers in human PDAC tissue samples from patient-derived PDAC cells and additional independent patients with PDAC (Fig. 3E and F) and a PDAC tissue microarray consisting of 96 human pancreatic cancer cores that underwent pathological assessment (Fig. 3G and H) indicated that expression of the MDR marker ABCB1 H-score was significantly associated with expression of the glutamine metabolism marker SLC1A5 H-score (Fig. 3I); that of the hypoxia marker CA-9 H-score (Fig. 3J) and IHC quantification in the form of H-scores showed a positive correlation between the expression of these genes (Fig. 3K–M). Indeed, PDAC cells from the nonresponder group and the nonresponder-like cell lines (Panc-1, MiaPaCa2, SNU-213, SW1990, BxPC-3, and ASPC-1) cultured in the 3D-printed system displayed significantly increased MAR staining, indicating hypoxia (Supplementary Fig. S7A and S7B). These results indicated a strong positive correlation between glutamine metabolism and hypoxia in chemoresistant PDAC cells.

We then aimed to study how glutamine metabolism contributed to tumor hypoxia. We hypothesized that enhanced glutaminolysis resulted in increased oxygen consumption that would disrupt the intracellular oxygen homeostasis, which would lead to enhanced hypoxia in chemoresistant PDAC cells. When we analyzed staining for the hypoxia probe MAR and the basal mitochondrial OCR, we found that the nonresponder group showed higher hypoxic marker staining (Supplementary Fig. S7A and S7B) and a significantly higher OCR than the responder group (Supplementary Fig. S8A), with a strong positive correlation between these two parameters (Fig. 4A), indicating a positive association between mitochondrial oxygen consumption and intracellular hypoxia in the 3D-printed nonresponder group. Importantly, the increased basal OCR in the 3D-printed nonresponder group was suppressed by glutamine deprivation, but not serine deprivation (Fig. 4B; Supplementary Fig. S8B), indicating that enhanced glutaminolysis-induced oxygen consumption in chemoresistant PDAC cells.

As extracellular glutamate can activate HIF1A in breast cancer (38), we investigated the role of extracellular glutamate in regulating hypoxic response in PDAC cells using MAR and HRE reporters. The MAR signal of PDAC cells in the 3D-printed nonresponder group was suppressed by glutamine deprivation, but this was not rescued by glutamate supplementation (Supplementary Fig. S8C and S8D), suggesting that extramitochondrial glutamate produced from glutamine did not impact the intracellular hypoxia status of PDAC cells in the nonresponder group.

Next, we investigated whether the glutaminolysis-induced increase in oxygen consumption could be the cause of the intracellular hypoxia that appeared specifically in the 3D-printed nonresponder group. To this end, we generated cells that stably expressed the luciferase-based reporter of HRE. Interestingly, we found spontaneous hypoxia reporter activity in the 3D-printed nonresponder group; this increased hypoxia reporter activity was dependent on glutamine availability, but not serine availability (Fig. 4C and D). Similarly, nonresponder-like Panc-1 cells also showed enhanced hypoxia reporter activity based on glutamine availability whereas responder-like Capan-1 cells did not (Supplementary Fig. S8E and S8F).

To further verify that glutaminolysis induces intracellular hypoxia via enhanced oxygen consumption, we blocked glutaminolysis, glycolysis, and fatty acid oxidation using the glutaminase inhibitor BPTES, the glucose analog 2-DG, the fatty acid oxidation inhibitor etomoxir, and the OXPHOS inhibitor metformin (39), and we measured OCR and hypoxia induction. BPTES and metformin, but not 2-DG or etomoxir, significantly suppressed the glutaminolysis-induced increases in OCR and hypoxia reporter activity in the 3D-printed nonresponder group (Fig. 4E–G; Supplementary Fig. S8G–S8I). Next, we confirmed whether the reduced oxygen consumption under metformin treatment was a consequence of inhibition of mitochondrial complex I. In NDUFS2-downregulated PDAC cells in the 3D-printed nonresponder group, basal OCR (Fig. 4H; Supplementary Fig. S8J) and HRE reporter activity (Fig. 4I and J; Supplementary Fig. S8K and S8L) was suppressed. However, exogenous expression of NDI1, a yeast NADH dehydrogenase as an orthologous human complex I, rescued both basal OCR and HRE reporter activity under NDUFS2 downregulation (Fig. 4H–J; Supplementary Fig. S8J–S8L). These results indicate that metformin's ability to inhibit mitochondrial complex I is directly related to basal OCR and intracellular hypoxia. In addition, BPTES and metformin selectively suppressed the mRNA and protein expression of the hypoxia-related genes PDK1, VEGFA, EPAS1, and MYCBP and proteins PDK1, VEGFA, HIF1A, EPAS1, and MYCBP, which were overexpressed in the 3D-printed nonresponder group and nonresponder-like cells (Fig. 4K and L; Supplementary Fig. S8M and S8N).

Hypoxia can be triggered by the generation of cellular ROS (40). The 3D-printed nonresponder group and nonresponder-like Panc-1 cells showed upregulated ROS generation suppressed by NAC and BPTES (Supplementary Fig. S9A and S9B). However, ROS generation did not correlate with hypoxia induction (Supplementary Fig. S9C–S9F), indicating that glutaminolysis-induced hypoxia induction is independent of ROS generation. BPTES can increase ROS by reducing cellular NADPH and GSH, which can be synthesized from glutamine utilization (41). Simultaneously, BPTES can decrease cellular ROS because glutaminolysis inhibition can reduce mitochondrial ROS by impairing glutamine-derived supplementation for OXPHOS (41). Consistent with previous reports, in 2D-cultured conditions, both responder and nonresponder PDAC cells showed elevated cellular ROS levels under BPTES treatment; these elevated ROS levels were not suppressed by metformin treatment, suggesting that OXPHOS is not involved in BPTES-induced ROS elevation in 2D-cultured conditions (Supplementary Fig. S9G and S9H). However, in 3D-printed conditions, cellular ROS in both responder and nonresponder groups was suppressed by BPTES and metformin treatment, suggesting that OXPHOS is involved in ROS levels in 3D-printed conditions (Supplementary Fig. S9G and S9H). These results suggest that nonresponder PDAC cells in 3D-printed conditions use more glutamine for OXPHOS than those in 2D culture conditions and that enhanced glutaminolysis drives intracellular oxygen consumption and disrupts the homeostasis of oxygenation, leading to hypoxia in chemoresistant PDAC.

To clarify the mechanism of glutaminolysis-induced intracellular hypoxia, we knocked down the mitochondrial glutamine transporter SLC1A5_var, which is critical for mitochondrial glutamine transport and glutaminolysis (20); added cell-permeable dimethyl succinate, a downstream metabolite of glutaminolysis, in a rescue experiment; and measured OCR and hypoxia reporter activity. SLC1A5_var knockdown decreased both the OCR and hypoxia reporter activity in the 3D-printed nonresponder group and in nonresponder-like Panc-1 cells, whereas cell-permeable dimethyl succinate restored these values (Fig. 5A–F). Conversely, SLC1A5_var overexpression in the 3D-printed responder group and responder-like Capan-1 cells resulted in higher basal OCR and higher hypoxia reporter activity compared with those in control cells (Fig. 5G–L). In addition, glutamine deprivation or metformin treatment ablated SLC1A5_var overexpression-induced OCR and intracellular hypoxia (Fig. 5G–L). Overall, these results indicate that mitochondrial glutaminolysis in cells increases oxygen consumption via OXPHOS and therefore promotes intracellular hypoxia.

Next, to examine whether glutaminolysis-induced hypoxia in the 3D-printed model could be recapitulated in vivo, we orthotopically transplanted PDAC cells expressing the 5xHRE-reporter and the mitochondrial glutamine transporter SLC1A5_var into the pancreas of immunodeficient mice and monitored the tumor hypoxia luciferase activity before and after targeting glutaminolysis or OXPHOS. Tumors derived from nonresponder YPAC-05 cells showed a significantly higher hypoxia reporter activity than tumors derived from responder YPAC-25 cells (Fig. 5M and N). However, tumors derived from YPAC-25 cells overexpressing SLC1A5_var showed a hypoxia reporter activity comparable with that of tumors derived from YPAC-05 cells (Fig. 5M and N). The administration of BPTES or metformin suppressed hypoxia reporter activity in both YPAC-05 and YPAC-26 cells overexpressing SLC1A5_var (Fig. 5M and N). These in vivo results support that mitochondrial glutaminolysis and subsequently activated OXPHOS regulate intracellular hypoxia in PDAC cells.

Because glutaminolysis-induced intracellular hypoxia was observed in the 3D-printed nonresponder group, we conducted a correlation analysis to investigate the relationship between the sensitivity to anticancer agents and the OCR in PDAC cells. We found a significant positive correlation between the OCR and GI₅₀ values of the anticancer drugs gemcitabine (r = 0.8588), 5-FU (r = 0.9216), oxaliplatin (r = 0.8462), and cisplatin (r = 0.9360) in 3D-printed PDAC cells (Supplementary Fig. S10A–S10D), whereas there was no apparent relationship between the OCR and GI₅₀ values of these anticancer drugs in 2D-cultured PDAC cells (Supplementary Fig. S10E–S10H). Notably, the 3D-printed nonresponder group clearly showed a higher basal OCR and higher GI₅₀ values of the anticancer drugs than the 3D-printed responder group (Supplementary Fig. S10A–S10D). We also observed a strong positive correlation between the level of hypoxia marker staining and GI₅₀ values of the anticancer drugs gemcitabine (r = 0.7913), 5-FU (r = 0.8850), oxaliplatin (r = 0.8391), and cisplatin (r = 0.8537) in the 3D-printed PDAC cells (Supplementary Fig. S10I–S10L). Overall, these results show a strong positive correlation among oxygen consumption, hypoxia, and GI₅₀ values of the anticancer drugs in 3D-printed PDAC cells, and suggest that the OCR values can be a marker for anticancer drug responsiveness in PDAC.

Next, we investigated whether the inhibition of glutaminolysis or OXPHOS relieves hypoxia and sensitizes chemoresistant PDAC cells to anticancer drugs. Strikingly, treatment with BPTES, metformin, or the hypoxia inhibitor acriflavine (42) significantly decreased the GI₅₀ values of the anticancer drugs gemcitabine, 5-FU, oxaliplatin, and cisplatin (Fig. 6A; Supplementary Fig. S11A–S11G) and increased cell death (Fig. 6B; Supplementary Fig. S11H–S11N) in the 3D-printed nonresponder group and nonresponder-like Panc-1 cells. On the other hand, incubation in the hypoxic chamber rescued the resistance of the cells to anticancer drugs. Even under treatment with BPTES or metformin, the hypoxic environment partially restored the resistance to anticancer drugs (Fig. 6A; Supplementary Fig. S11A–S11G) and suppressed anticancer-induced cell death (Fig. 6B; Supplementary Fig. S11H–S11N) in cells from the nonresponder group. Related to resistance to anticancer drugs, we measured MDR gene expression under BPTES, metformin, or acriflavine treatment. Indeed, BPTES, metformin, and acriflavine attenuated the expression of MDR genes (ABCB1, ABCG2, DCK, and SLC31A1) in the 3D-printed nonresponder group (Fig. 6C and D; Supplementary Fig. S11O and S11P); calcein AM retention was also increased (Fig. 6E; Supplementary Fig. S11Q).

To verify that glutaminolysis-induced intracellular hypoxia is a valuable therapeutic target for pancreatic cancer treatment, we assessed the effect of targeting glutaminolysis, OXPHOS, or hypoxia on tumor growth and chemoresistance using orthotopic xenografts in mice. Tumors derived from YPAC-05 cells and SLC1A5_var-overexpressing YPAC-25 cells, but not tumors derived from YPAC-25 cells, were resistant to gemcitabine treatment (Fig. 6F–G). However, BPTES, metformin, or acriflavine treatment-sensitized tumors from YPAC-05 cells and SLC1A5_var-overexpressing YPAC-25 cells to gemcitabine treatment (Fig. 6F–G). Increased protein expression of the hypoxia markers HIF1A and EPAS1 and the MDR markers ABCB1 and ABCG2 was suppressed by the combination of gemcitabine with BPTES, metformin, or acriflavine in tumors derived from YPAC-05 cells or SLC1A5_var-overexpressing YPAC-25 cells (Fig. 6H). In addition, increased IHC staining scores of the hypoxia marker CA9 and MDR marker ABCB1 were suppressed by the combination of gemcitabine with BPTES, metformin, or acriflavine in tumors derived from YPAC-05 cells or SLC1A5_var-overexpressing YPAC-25 cells (Fig. 6I–K). The IHC staining score of cleaved caspase-3 was increased by gemcitabine treatment in tumors derived from YPAC-25, but not in tumors derived from YPAC-05 cells or SLC1A5_var-overexpressing YPAC-25 cells (Fig. 6I and L). However, the combination of gemcitabine with BPTES, metformin, or acriflavine increased the IHC staining score of cleaved caspase-3 in tumors derived from YPAC-05 cells or SLC1A5_var-overexpressing YPAC-25 cells (Fig. 6I and L). Finally, targeting glutaminolysis, OXPHOS, or hypoxia through the application of BPTES, metformin, or acriflavine significantly extended the survival of mice with tumors derived from YPAC-05 cells or SLC1A5_var-overexpressing YPAC-25 cells (Fig. 6M–Q). Taken together, these results suggest that augmented glutaminolysis increases oxygen consumption, resulting in intracellular hypoxia and chemoresistance in PDAC, and that targeting the glutaminolysis–oxygen consumption–hypoxia axis holds promise as a strategy for overcoming therapeutic resistance in PDACs.

It is very important to determine the cause of hypoxia in PDAC, which is related to anticancer chemoresistance. However, until now, the cause of hypoxia has been unclear. Therefore, we investigated the cause of hypoxia in PDAC through this study. We established an ECM component-based 3D cell printing model that replicates the anticancer chemoresistance observed in human patients with PDAC. Our study shows that patient-derived PDAC cells and pancreatic cancer cell lines with low chemotherapeutic drug sensitivities exhibit enhanced glutaminolysis under 3D-printed conditions. We provide direct evidence that augmented glutaminolysis in cells elicits increased oxygen consumption via OXPHOS, resulting in glutaminolysis-induced intracellular hypoxia.

Glutaminase inhibitor CB-839 sensitized gemcitabine-resistant cancer cells, thereby improving the therapeutic effect (43). Moreover, in KRAS ablation through genetic deletion or pharmacologically inhibition, survived pancreatic cancer cells induce tumor recurrence and display OXPHOS dependency, and targeting OXPHOS impairs the tumor-initiating ability of cancer cells (44), highlighting the combinatorial usage of OXPHOS inhibition. In this study, OXPHOS suppression using the GLS inhibitor BPTES or the complex I inhibitor metformin effectively abolished glutaminolysis-induced intracellular hypoxia in vitro and in vivo. We further show that targeting glutaminolysis or OXPHOS in combination with conventional anticancer agents for PDAC therapy exhibits in vivo efficacy against the drug-resistant characteristics of PDACs. Therefore, alleviating intracellular hypoxia by suppressing glutaminolysis and oxygen consumption may be another strategy to modulate chemoresistance in PDAC.

Glutamine is a conditionally essential amino acid that is highly abundant in human plasma (41) and plays diverse roles in cellular metabolism (45, 46). The carbon scaffold of glutamine is essential to the construction of TCA cycle intermediates and supports OXPHOS (41). PDAC cells are highly dependent on glutamine (46–49) and maintain proliferation and oxidative metabolism under severe hypoxia (50). This characteristic of PDAC cells may underlie the dismal drug response and severe hypoxia observed in PDAC (51, 52), which result in exceptionally lower treatment success compared with that of other cancer types. Considering the lack of methods for the early detection of PDAC and rapid disease progression (5), the ECM-component–based 3D-printing system combined with patient-derived cells is useful for investigating the dependence of PDAC cells on glutamine metabolism associated with chemoresistance. Furthermore, given that the chemoresistant PDAC group showed increased glutamine uptake and glutamine catabolism compared with the chemoresponsive PDAC group, glutamine-based PET imaging (53) could be useful to predict the responsiveness of patients with pancreatic cancer to anticancer agents. These efforts will further increase the possibility of personalized medicine for each patient with pancreatic cancer.

Although many studies have been reported that oncogenic KRAS promotes metabolic reprogramming in cancer cells, shifting them toward an anabolic metabolism, including glycolysis (54), glutaminolysis (47), lipid biosynthesis (55), nutrient scavenging (56), and biomass production for rapid proliferation (57), we did not observe distinguished characteristics originating from the KRAS status of patient-derived PDAC cells and PDAC cell lines in our study because the KRAS mutation frequency was similar between the responder and nonresponder groups. Therefore, the presence or absence of KRAS mutation does not seem to be an important factor in changes in therapy response and glutamine metabolism in the 3D-printed nonresponder group. It would be interesting to determine the cause of metabolic reprogramming in the nonresponder group.

The advantage of the 3D printing culture system is that it shows clinical relevance compared with the 2D monolayer culture model. First, 3D-printed PDAC cells are more similar to gene expression patterns of orthotopic PDAC tumors and primary tumors than 2D-cultured PDAC cells (Supplementary Fig. S4B and S4C). Second, compared with 2D-cultured cells, 3D-printed cells respond to various types of anticancer drugs similarly to patients’ responses to anticancer drugs (Supplementary Fig. S2A–S2E). Third, although the gene expression is significantly different between 2D-cultured cells and their corresponding orthotopic or primary tumors, when the 2D-cultured cells are implemented as a 3D-printed model, the gene expression becomes similar to those of the orthotopic or primary tumors (Supplementary Fig. S4A). Besides these advantages, we can observe spontaneous intracellular hypoxic responses derived from metabolic reprogramming in 3D-printed model.

Because the 3D-printed model is constructed using cells established through 2D monolayer culture, minimizing genomic instability is key. In a previous report, we showed that the 2D cultured patient-derived PDAC cells faithfully preserved genetic profiles of primary tumors, including mutation patterns and copy-number variations (25). In this study, we demonstrate that the somatic mutations are shared in the primary tumors and patient-derived cells and do not alter even after 3D printing. These features could keep clinical relevance and support the value of the 3D-printing model.

In summary, we demonstrate that increased glutaminolysis is the cause of hypoxia, via imbalanced oxygenation status due to an increase in oxygen consumption, and thus drives anticancer drug resistance in patient-derived models of PDAC. These studies support the therapeutic potential of combining glutaminolysis inhibition and conventional anticancer drugs to overcome therapeutic resistance in PDAC.

No disclosures were reported.

S.J. Park: Resources, data curation, formal analysis, validation, methodology. H.C. Yoo: Conceptualization, formal analysis, validation, writing–original draft. E. Ahn: Conceptualization, resources, data curation. E. Luo: Resources, data curation, software. Y. Kim: Resources, data curation, software. Y. Sung: Resources, software. Y.C. Yu: Resources, software. K. Kim: Resources, software. D.S. Min: Supervision. H.S. Lee: Resources, supervision. G.-S. Hwang: Formal analysis, supervision. T. Ahn: Software, supervision. J. Choi: Formal analysis, supervision. S. Bang: Resources, supervision. J.M. Han: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported by the Basic Science Research Program through the National Research Foundation (NRF) of Korea funded by the Ministry of Education (2018R1A6A1A03023718 and 2020R1I1A1A01067423), the Bio and Medical Technology Development Program of the National Research Foundation (NRF) funded by the Ministry of Science and ICT (2020M3E5E2040282), and a National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (MSIT; 2020R1A2C2099586, 2021R1C1C2006283, and 2019R1C1C1008185). Y. Sung was supported by the Health Fellowship Foundation.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.