Mutant IDH1 confers resistance to energy stress in normal biliary cells through PFKP-induced aerobic glycolysis and AMPK activation

Metabolism is a critical regulator of cell fate determination. Recently, the significance of metabolic reprogramming in environmental adaptation during tumorigenesis has attracted much attention in cancer research. Recurrent mutations in the isocitrate dehydrogenase (IDH) 1 or 2 genes have been identified in several cancers, including intrahepatic cholangiocarcinoma (ICC). Mutant IDHs convert α-ketoglutarate (α-KG) to 2-hydroxyglutarate (2-HG), which affects the activity of multiple α-KG-dependent dioxygenases including histone lysine demethylases. Although mutant IDH can be detected even in the early stages of neoplasia, how IDH mutations function as oncogenic drivers remains unclear. In this study, we aimed to address the biological effects of IDH1 mutation using intrahepatic biliary organoids (IBOs). We demonstrated that mutant IDH1 increased the formation of IBOs as well as accelerated glucose metabolism. Gene expression analysis and ChIP results revealed the upregulation of platelet isoform of phosphofructokinase-1 (PFKP), which is a rate-limiting glycolytic enzyme, through the alteration of histone modification. Knockdown of the Pfkp gene alleviated the mutant IDH1-induced increase in IBO formation. Notably, the high expression of PFKP was observed more frequently in patients with IDH-mutant ICC compared to in those with wild-type IDH (p < 0.01, 80.9% vs. 42.5%, respectively). Furthermore, IBOs expressing mutant IDH1 survived the suppression of ATP production caused by growth factor depletion and matrix detachment by retaining high ATP levels through 5ʹ adenosine monophosphate-activated protein kinase (AMPK) activation. Our findings provide a systematic understanding as to how mutant IDH induces tumorigenic preconditioning by metabolic rewiring in intrahepatic cholangiocytes.


Results
IDH1 mutation enhances the formation of biliary organoids established from murine liver. To elucidate how mutant IDH affects the molecular or biological features of normal intrahepatic cholangiocytes, we established IBOs from murine normal liver cells ( Fig. 1A and Supplementary Fig. 1A) 24 . Given that organoids maintained the characteristics of primary cells 24 , we applied this system to estimate the biological and metabolic traits triggered by IDH1 mutation in normal biliary epithelial cells. IBOs were composed of a monolayer of biliary lineage cells expressing biliary marker genes, including cytokeratin 7 (Ck7), cytokeratin 19 (Ck19), and hepatocyte nuclear factor 1-beta (Hnf1β), but not hepatocyte marker genes such as transthyretin (Ttr) and aldolase B (AldoB) ( Fig. 1A and Supplementary Fig. 1A). Immunohistochemistry confirmed the expression of CK19 and SRY (sex determining region Y)-box 9 (SOX9) as biliary marker proteins in IBOs (Fig. 1B). The established IBOs were transduced with lentiviral vectors expressing human wild-type or mutant IDH1 (R132C) ( Supplementary  Fig. 1B). The activity of transduced mutated IDH1 was confirmed by production of 2-HG (Fig. 1C). We found that IBOs expressing mutant IDH1 (mut-IBOs) exhibited enhanced organoid-forming efficiency compared to IBOs expressing wild-type IDH1 (wt-IBOs), constitutively through serial passages (Fig. 1D, E). The 2-HG addition also enhanced the formation of IBOs after the third passage (Fig. 1F). In addition, a specific inhibitor of mutant IDH1 25 , AGI-5198, significantly reduced the 2-HG production and hampered the formation of mut-IBOs. The effects were specific for mut-IBOs because AGI-5198 did not affect the growth of wt-IBOs (Fig. 1G,I). These results indicated that mutant IDH1 has the biological potential to enhance the IBO forming ability.

Mutant IDH1 upregulates glucose metabolism in IBOs.
To uncover metabolic traits underlying the increased forming ability in mut-IBOs, metabolic profiles of the IBOs were analyzed by capillary electrophoresis-mass spectrometry (CE-MS) ( Fig. 2A and Supplementary Fig. 2). Notably, mut-IBOs displayed enrichment of glycolytic intermediates including fructose-1,6-bisphosphate (F1,6 P), 3-phosphoglycerate (3PG), 2-phosphoglycerate (2PG), phosphoenolpyruvate (PEP), and lactate compared with wt-IBO ( Fig. 2A). In addition, the metabolites of other glucose-utilizing pathways sedoheptulose-7-phosphate (S7P), adenosine monophosphate (AMP), uridine monophosphate (UMP) in the pentose phosphate pathway (PPP), and UDP-N-acetylglucosamine in the hexosamine biosynthesis pathway (HBP), were significantly accumulated in mut-IBOs ( Fig. 2A). Consistent with an increase in glycolysis, glucose uptake was significantly enhanced in mut-IBOs compared to wt-IBOs (Fig. 2B). Given that all of these metabolic profiles were identified under aerobic culture conditions with sufficient glucose and glutamine, such preferred aerobic glycolysis seemed to be similar to the 'Warburg effect' -a well-known hallmark of rapidly proliferating mammalian cells 26 . However, some metabolites in the TCA cycle, including citrate, fumarate, and malate, were also elevated in mut-IBOs ( Supplementary  Fig. 2), indicating that mutations in IDH1 lead to higher glucose metabolism. Since reprogramming from differentiated cells to cancer cells is often accompanied by aerobic glycolytic phenotypes, we aimed to investigate if activated glycolysis correlates with the enhanced formation of mut-IBOs by treatment with a glycolytic inhibitor, 2-deoxy-d-glucose (2DG) (Fig. 3A). Although 2DG reduced the organoid-forming efficiency of both wt-and mut-IBOs in a dose-dependent manner, the formation of mut-IBOs was relatively resistant to the inhibitory effects of 2DG compared to wt-IBOs (Fig. 3B,C). The forming capacity of mut-IBOs compared to wt-IBOs was relatively conserved under a low glucose condition (Fig. 3D,E). These results suggest that the formation of IBOs depends on glucose as a critical energy source, and IDH1 mutation allows for IBOs to become resistant to glucose restriction.
Mutant IDH1 activates glycolysis through the upregulation of Pfkp gene. To address the molecular mechanisms underlying increased glycolysis in mut-IBO, we performed microarray gene expression analysis involving >40,000 oligonucleotide probes. Regardless of the drastic changes in organoid-forming efficiency and metabolism by IDH1 mutation, the analysis showed that only 19 genes were upregulated >2-fold, and 27 genes were downregulated <2-fold in mut-IBOs compared to wt-IBOs (P < 0.05, Supplementary Fig. 3A and Supplementary Table 1). Consistent with the effects of IDH1 mutation in terms of catalytic activities of αKG-dependent histone lysine demethylases, the levels of global histone H3 lysine 4 trimethylation (H3K4me3) and lysine 27 trimethylation (H3K27me3) were elevated in mut-IBOs, compared to wt-IBOs ( Supplementary  Fig. 3B) as previously reported in cancer cells 15,16 . Expression of genes encoding enzymes such as Jarid1a, Jarid1b, or Jmjd3 were not affected ( Supplementary Fig. 3C). In contrast, there were no significant differences in the levels     Fig. 3C-E). In addition, we analyzed the effects on chromatin structure caused by mutant IDH1 using an assay for transposase-accessible chromatin using sequencing (ATAC-seq) 27 . Although different peak patterns were detected in certain locations, there were no significant differences in specific motif sequences binding transcription factors (TFs) between wt-and mut-IBOs (Supplementary Fig. 4A-C). These results appear to be compatible with the lack of differences in DNA methylation, and the limited number of differentially expressed genes between wt-and mut-IBOs. Notably, however, we found a significant upregulation in the expression of the platelet isoform of phosphofructokinase-1 (Pfkp) gene in mut-IBOs (Fig. 4A). Mammalian phosphofructokinase-1 (PFK-1) is a tetramer consisting of three isoforms; Pfkl, Pfkm, and Pfkp 28 and catalyzes the conversion of fructose 6-phosphate (F6P) to F1,6 P, which is a rate-limiting step in the glycolytic pathway. Intriguingly, the promotion of glycolysis in mut-IBOs occurred immediately downstream the transition of F6P to F1,6 P ( Fig. 2A). This led us to hypothesize that IDH1 mutation drives glycolysis through the upregulation of Pfk-1 activity. Supporting our hypothesis, we confirmed not only increases in Pfkp mRNA and protein levels but also identified elevated Pfk-1 activity in mut-IBOs ( Fig. 4A-D). To analyze if PFKP contributes to IBO formation, we established two distinct mut-IBO lines in which Pfkp was stably knocked-down (Fig. 4E). Both mut-IBO lines decreased the efficiency of IBO formation by mutant IDH1 (Fig. 4F), emphasizing the notion The number of organoids (>100 μm) at 5 d after plating in medium with or without 2DG (normalized to that in the 2DG-free medium, n = 4, 5,000 cells per group *P < 0.05). (C) Representative images. Scale bars, 250 μm. (D) The number of IBOs (>100 μm) was counted at 7 d after plating under the normal culture condition (Normal) and the low glucose condition (Low glucose), replacing advanced-DMEM/F12 (3 g/L glucose) to L0091, glucose-free DMEM/F12 instead (n = 4, 5,000 cells per group, **P < 0.01). Low glucose medium contains a little glucose derived from Wnt3a-and RSPO1-conditioned medium; the final concentration of glucose is 0.2 g/L. (E) Representative images. Scale bars, 500 μm. (2019) 9:18859 | https://doi.org/10.1038/s41598-019-55211-w www.nature.com/scientificreports www.nature.com/scientificreports/ that IDH1 mutation promotes IBO formation by enhancing glycolysis via upregulation of PFKP. To examine the dysregulation of Pfkp directly caused by IDH1 mutation, the effect of the inhibitor AGI-5198 on the expression of the Pfkp gene was analyzed 25 . AGI-5198 decreased the expression of Pfkp in mut-IBOs (Fig. 4G), and, in turn, 2-HG treatment increased the expression of Pfkp in IBOs from normal mouse liver (Fig. 4H), suggesting that the upregulation of Pfkp was attributed to IDH1 mutation. In addition, although ATAC-seq data showed comparable accessibility surrounding the Pfkp locus, the level of H3K4me3 as an active transcriptional mark was significantly increased on the promoter region of Pfkp in mut-IBOs compared to wt-IBOs (Supplementary Fig. 4D and Fig. 4I), indicating that the upregulation of Pfkp was mediated by aberrant histone modification. These data indicated that the epigenetic disturbance induced by IDH1 mutation was focally limited to specific targets, such as Pfkp in normal biliary cells. Furthermore, we investigated the expression status of PFKP protein in 101 surgically resected ICC specimens from our hospital. PFKP expression was detected in 69 cases (68%) (Fig. 5A,B), and the expression levels were significantly higher in ICCs harboring IDH mutations (P < 0.01, Fischer's exact test) www.nature.com/scientificreports www.nature.com/scientificreports/ (Fig. 5B). Considering the significance of PFKP for growth advantage in many human cancers 29,30 , PFKP could serve as a novel metabolic target for IDH1-mutant ICCs.

IDH1 mutation provides EGF-and HGF-independent survival in IBOs.
We sought to assess the biological consequences of mutant IDH1-induced glycolytic activation. Consistent with previous reports showing that mutant IDH1 alone is insufficient for transformation in several cell linages 16,20 , subcutaneous inoculation of either wt-IBO or mut-IBO into nude mice failed to generate tumors (2,000,000 cells per group, n = 6, data not shown). The organoid culture systems are applied to the assessment of extracellular factor-dependency 31 . Based on the findings that cultured organoids gradually become independent of growth factors followed by oncogenic mutations 31 , we evaluated the effect of IDH1 mutation for growth factor-dependency in the survival of IBOs. Depletion of EGF, FGF10, HGF, or R-spondin-1 significantly reduced the organoid-forming efficiency of wt-IBO ( Supplementary Fig. 5A,B). Although the efficiency of mut-IBO formation decreased in the absence of EGF, FGF10, HGF, or R-spondin-1, the higher organoid-forming efficiency compared to wt-IBO was still observed www.nature.com/scientificreports www.nature.com/scientificreports/ ( Supplementary Fig. 5A,B). Remarkably, double depletion of EGF and HGF (EH-depletion), completely blunted the formation of wt-IBOs, whereas mut-IBOs persisted (Fig. 6A,B).

Mutant IDH1-induced AMPK activation makes IBOs override the EGF and HGF dependency.
To address the molecular mechanisms underlying the sustained survival after EH-depletion, we examined the status of signal transduction pathways related to cell survival ( Fig. 7A and Supplementary Fig. 6). The levels of phosphorylation of Erk, Akt, or Stat3 were comparable between wt-IBOs and mut-IBOs (Fig. 7A). In addition, these phosphorylation levels were not affected by EH-depletion (Fig. 7A). In contrast, we identified enhanced phosphorylation of AMP-activated protein kinase (AMPK) in mut-IBOs compared to wt-IBOs, regardless of EH-depletion (Fig. 7A). The two primary upstream regulators of AMPK, Lkb1, and Camkk2 were not differently activated between mut-IBOs and wt-IBOs ( Supplementary Fig. 7). Intriguingly, among the multiple downstream targets of the AMPK signaling pathway 32 , Unc-51 like kinase-1 (Ulk1), which is a crucial initiator of autophagy 33 , was activated by EH-depletion only in mut-IBOs ( Fig. 7A and Supplementary Fig. 7). To address whether AMPK activation contributed to the sustained formation of IBOs under EH-depletion, mut-IBOs were treated with the AMPK inhibitor, Compound C 34 . The organoid-forming capacity after EH-depletion in mut-IBOs was entirely inhibited by Compound C (Fig. 7B,C), suggesting that AMPK contributes to the growth-factor independence in cooperation with IDH1 mutation. Given that 5-aminoimidazole-4-carboxamide-1-β-D-ribofuranoside (AICAR), which is a well-known AMP analog that activates AMPK, inhibited the formation of wt-IBOs ( Supplementary  Fig. 8). Thus, the role of AMPK seems to be context-dependent, as previously reported 35 .
It is well-known that growth factor deprivation, as well as detachment from the extracellular matrix (ECM), induces ATP loss and subsequent cell death in normal epithelial cells 36 . When organoids are passaged as dissociated single cells, they must be exposed to energy stress without growth factors or ECM. Supporting this notion, wt-IBOs were not formed in the settings without matrigel after passage procedures. Thus, we compared the levels of cellular ATP in dissociated IBO cells just after the passage procedure (Fig. 7D,E). Mounting of the dissociated wt-IBO cells on matrigel with complete medium after passage recovered ATP production in a time-dependent manner, but EH-depletion from medium inhibited the recovery (Fig. 7E). In contrast, the dissociated mut-IBO cells restored the high levels of ATP after passage, even under EH-depletion (Fig. 7E). Given the function of AMPK in replenishing ATP after the loss of ECM attachment 36 , we considered that AMPK activation contributed to ATP production in dissociated mut-IBO cells. Consistently, AMPK inhibition significantly decreased ATP levels in the dissociated mut-IBO cells (Fig. 7F). These data supported the notion that AMPK activity contributes to ATP production under EH-depletion in mut-IBOs.

Discussion
In this study, we found a unique metabolic status in the IDH-mutant biliary cells. Several previous reports have described the phenotype of IDH-mutant cancer cells related to hypoxia-inducible factor 1 (HIF-1) activity. For example, the survival and proliferation of IDH-mutant glioma cells were promoted by the Warburg effect due to the stabilization of HIF-1α 37 . In contrast, other studies have shown that IDH mutations downregulate the HIF-1 pathway and induce limited glycolysis, contributing to the slower growth and better prognosis in glioma cells 38 . Although we do not entirely exclude the possible involvement of Hif-1α in the IDH-mutant biliary cells, the expression of Hif-1α protein and the majority of its downstream glycolytic genes were not markedly upregulated in mut-IBOs (Supplementary Fig. 9). Upregulated HIF-1α expression by IDH1 mutation reduced the energy flux through the TCA cycle and increased the clonogenicity of normal human astrocytes due to the activation of pyruvate dehydrogenase kinases (PDKs) 39 . In contrast, levels of some metabolites in the TCA cycle, including citrate, fumarate, and malate were elevated in mut-IBOs ( Supplementary Fig. 2). Additionally, decreased levels of glutamate identified in glioma or astrocytes harboring IDH1 mutation 40 were not observed in the case of mut-IBOs (Supplementary Fig. 2). These findings propose a possibility that metabolic rewiring traits triggered by IDH mutation are different among distinct cell lineages. Furthermore, while mutant IDH in patient material www.nature.com/scientificreports www.nature.com/scientificreports/ of glioma and glioblastoma was associated with a non-Warburg phenotype in comparison with wild-type IDH tumors, the artificial introduction of mutant IDH into glioblastoma cell lines resulted in aerobic glycolysis [41][42][43] . Thus, the metabolic effects of mutant IDH can also vary between patient-derived materials and those established experimentally. To assess the possibility of clinical trials focusing on the metabolic traits of mutant IDH tumors as in gliomas 44 , the metabolome analysis of clinical samples of ICC would be needed in the near future.
It was noteworthy that there was a difference in the organoid-forming efficiency between mut-IBOs and wt-IBOs treated with 2-HG ( Fig. 1D-F). We speculate two possibilities for the different extent of growth promotion. The first is a possibility that the intracellular concentration of 2-HG in wt-IBO cells treated with 2-HG was lower than in mut-IBOs due to the limited cell permeability of R-2-HG or the run-in effect 17 . The second is a possible 2-HG independent function of mutant IDH1, which might be supported by a previous paper showing that DNA methylation profiles of 2-HG treated cells could not entirely recapitulate those of mutant IDH1 cells 45 .
In general, as the increasing level of AMP mirrors the decrease in ATP under metabolic stress conditions in normal cells 32 , the metabolic situation caused by IDH1 mutation seems to be extraordinary, where both AMP and ATP accumulated. Similarly, the activation of AMPK in mut-IBOs was rather paradoxical in terms of its energy sensor function 32 because ATP levels remained unperturbed in mut-IBOs (Fig. 2). As the AMP-induced activation of AMPK is allosterically much more potent than the inhibitory action of ATP based on structural research 46 , AMPK activation in mut-IBOs was likely to be induced by AMP accumulation (Fig. 2).
Despite the tumor-suppressive roles of AMPK in many cancers, recent studies have shown the pleiotropic properties of AMPK for tumor cell survival under metabolic stress 35 . In terms of the relationships of AMPK and www.nature.com/scientificreports www.nature.com/scientificreports/ IDH mutation, previous reports demonstrated the metabolic vulnerability of mutant IDH gliomas or glioblastomas to biguanides, such as metformin and phenformin, which activate AMPK by inhibiting mitochondrial respiratory chain complex 1 [47][48][49] . Therefore, our findings emphasize the notion that the role of AMPK seems to be pleiotropic and context-dependent 35 .
Finally, we identified a two-way metabolic rewiring status; PFKP upregulation and AMPK activation induced by IDH1 mutation in normal biliary cells (Fig. 8). Nonetheless, it is still necessary to elucidate the whole mechanism acquired by IDH1 mutation that finally accomplishes biliary tumor development.

Construction of IDH1 mutation and lentiviral infection.
Human wild-type IDH1 expression vector (plasmid pFN21AE0433) was purchased from Kazusa DNA Research Institute (Chiba, Japan). The IDH1 R132C mutant was generated from the wild-type IDH1 open reading frame of this plasmid using the QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Rosa, CA) and primers 5′-GATCCCCA TAAGCATGACAACCTATGATGATAGGTTC-3′ for sense and 5′-GAACCTATCATCATAGGTTG TCATGCTTATGGGGATC-3′ for antisense. The wild-type and mutant IDH1 were ligated into the multiple cloning sites of pLVSIN-EF1α Vector (TAKARA, Shiga, Japan) and these lentiviral vectors were transfected into 293 T/17 cells. Organoids were infected with concentrated lentivirus supernatant for 9-12 h and selected with puromycin (2 μg/mL).
shRNA. Lentiviral (pLKO.1) Pfkp shRNA vectors TRCN0000025916 and TRCN0000025962 were purchased from Dharmacon Inc. (Lafayette, CO). and used as shPfkp-2 and shPfkp-5, respectively. The puromycin-resistance gene was excised with restriction enzymes, and the hygromycin-resistance gene was ligated into the same region of each vector. Clinical samples, molecular analysis, and PFKP scoring system. Formalin-fixed, paraffin-embedded (FFPE) tissue blocks from 101 patients with ICC who underwent surgical treatment at the University of Tokyo Hospital were used for the extraction of tumor DNA. Mutational analysis of IDH was performed by direct sequencing of PCR products. FFPE tissue sections were immunohistochemically stained using the primary antibody against PFKP (ab119796, Abcam). Staining results were scored by two pathologists (K.M. and A.H.) blinded to the mutational profile of IDH. The scoring system was validated by the percentages of positive cells as follows: score 0, <1%; score 1, 1-5%; score 2, 6-25%; score 3, 26-50%; and score 4, >51%. Simultaneously, scores 0 and 1 were defined as negative expression, score 2 as low, score 3 as moderate, and score 4 was defined as high expression, respectively. Detailed procedures for immunohistochemistry and DNA sequencing were previously described 51 . This study was approved by the Ethics Committee of the University of Tokyo, and carried out in accordance with the Declaration of Helsinki. Informed consent in writing was obtained from all patients. PFK-1 activity assay. 6-Phosphofructokinase Activity Assay Kit (ab155898, Abcam) was used in accordance with the manufacturer's protocol. Briefly, for each assay, 900,000 cells were harvested and washed in cold PBS. After suspension and appropriate dilution in the assay buffer, cellular extracts were mixed with the reaction solution containing ATP and F6P. NADH produced by the reaction was monitored spectrophotometrically at 450 nm for 100 min. PFK-1 activity was calculated from an NADH standard curve.
Glucose uptake assay. Glucose Uptake Assay Kit (ab136955, Abcam) was used in accordance with the manufacturer's protocol. Briefly, organoids were treated in Krebs-Ringer-Phosphate-Hepes (KRPH) buffer: 20 mM HEPES, 5 mM KH2PO4, 1 mM MgSO4, 1 mM CaCl2, 136 mM NaCl, and 4.7 mM KCl, at a pH 7.4 containing 2% bovine serum albumin (BSA), with or without 2-DG (1 mM), for 20 min. Organoids were harvested and washed in cold PBS. For each assay, we used 70,000 cells. Harvested cells were lysed with extraction buffer, frozen at −80 °C for 15 min, and heated at 85 °C for 40 min. After cooling on ice for 5 min, the lysates were neutralized by adding neutralization buffer and diluted appropriately with assay buffer. Lastly, by two amplification steps, according to the manufacturer's instructions, the colorimetric end product was detected at 412 nm. From the 2-DG-6-phosphate (2-DG6P) standard curve, 2-DG (or glucose) uptake was calculated.