Acinar-to-ductal metaplasia in the pancreas requires a glycolytic switch and functional mitochondria

Reprogramming of the cellular metabolism is a hallmark of pancreatic cancer, yet it remains unclear at what stage during carcinogenesis it occurs. Here, we investigated the metabolic requirements for acinar-to-ductal metaplasia (ADM), the first step in pancreatic carcinogenesis. We detected increased glycolytic marker expression in human ADM suggesting that a metabolic switch occurs during ADM formation. We report that this switch was similarly required for ADM formation in different oncogenic mouse models (KRAS, PI3K, and MEK1) and in ligand-induced ADM in mouse wild-type acini. In addition, we show that a functional electron transport chain (ETC), but not mitochondrial ATP production, was essential to ADM formation. We conclude that the ETC provides NAD+ for the de novo synthesis of serine from glycolysis intermediates. Our findings demonstrate that metabolic programming is essential for the initiation of pancreatic carcinogenesis and thus identifies potential targets for metabolic intervention.


Introduction
Metabolic requirements change throughout cancer progression 1 . Tumour development depends on nutrient uptake and biosynthesis. Compared with anaerobic glycolysis, oxidative phosphorylation (OxPhos) is more efficient in producing ATP. However, Otto Warburg described that cancer cells reprogram their metabolism to enhance glycolysis even under aerobic conditions, termed as aerobic glycolysis 2 . In pancreatic cancer oncogenic Kras favours glycolysis and pathways deriving from glycolysis. These changes provide cells with building blocks for anabolism and redox equivalents 3 . However, recent data suggest, that most cancer cells do not fully rely on glycolysis. Instead, they preserve metabolic plasticity to enable cancer cells to switch between aerobic glycolysis and OxPhos during tumourigenesis 4 . Cancer stem cells depend on OxPhos while the bulk of highly proliferative cancer favours glycolysis 5 . These concepts apply only to fully developed tumours. The metabolic state of early lesions remains unclear.
Pancreatic ductal adenocarcinoma (PDAC) is one of the most aggressive cancers and mostly incurable when diagnosed. The main reason for the dismal survival rate is that the patients become symptomatic at an advanced stage leaving only limited treatment options. Owing to the low incidence of pancreatic cancer, a population-based screening is not feasible. However, surveillance programs have been proposed for patients with hereditary tumour predisposition. Contrary to other tumour entities, prophylactic pancreatectomy is not indicated due to its high risk of morbidity and even mortality. Hence, novel strategies for PDAC prevention need to be developed, particularly for at-risk individuals.PDAC development and progression is a multi-step process characterized by activating KRAS mutations in more than 90% of ductal pancreatic adenocarcinoma, sequentially followed by the inactivation of tumour suppressor genes 6 . Previously, we have shown that TGFa can induce transdifferentiation of acinar cells into duct-like cells. This acinar-to-ductal metaplasia (ADM) was proposed as the first step in pancreatic carcinogenesis 7 . Indeed, lineage tracing experiments demonstrated a mainly acinar origin of PDAC 8 9,10 . As such, acinar cells represent the origin of pancreatic intraepithelial neoplasia (PanINs). However, PDAC can in principle also originate from duct cells.
Here, we investigated the metabolic requirements for ADM. We discovered that a metabolic switch with increased glycolysis and an ATP-independent ETC activity is required for acinarto-ductal metaplasia.

ADM in vivo is accompanied by metabolic changes
The reprogramming of the metabolism to enhance glycolysis including overexpression of glycolytic enzymes is a feature which has already been described for pancreatic cancer 11 . The onset of these changes during pancreatic cancer development was investigated by analysing the expression level of glycolytic markers in human tissue microarrays containing normal tissue and ADM ( Figures 1A-F). We detected significantly enhanced expression of glucose transporter type 1 (GLUT1), hexokinase 1 (HK1) and lactate dehydrogenase A (LDHA) in ADM lesions compared to normal tissue ( Figures 1G-I). This increased expression suggests that the glycolytic switch already occurs during ADM formation.
To investigate whether metabolic reprogramming was common among ADM lesions, we studied three preclinical mouse models of PDAC that recapitulate human pancreatic carcinogenesis by their driver mutations as well as by their morphology. Mice that endogenously express Kras G12D using Ptf1a Cre/+ develop ADM and present the full spectrum of PanIN lesions 12 . In addition, a gain of function mutation of PIK3CA (Pi3k CAH1047R ), which occurs in cases with wild-type KRAS, and a gain of function mutation of MEK1 (Mek1dd) results in the development of ADM and PanIN 13 . Mice with all three genotypes exhibited ADM formation at 10 weeks of age (Figures 2A-D). Quantification of these lesions revealed that compared with wild-type mice, Kras G12D mice developed only a few ADM lesions, whereas Pi3k CAH1047R or Mek1dd mice developed ADM at higher rates (3x higher in Pi3k CAH1047R and 5x higher in Mek1dd mice, compared to Kras G12D ) ( Figure 2E). These ADM lesions stained positive for the glycolytic markers GLUT1, HK1 and LDHA (Figures 2F-G and S1A-C, E-G, I-K, M-O), which were significantly upregulated compared to normal pancreatic acinar cells of the same animal ( Figures 2H-J). Notably, LDHA was already elevated in tissues from oncogene-harbouring mice that had a comparable appearance to those from wild-type animals. In contrast, the expression of ATP synthase F1 subunit delta (ATP5D), a marker of oxidative phosphorylation, did not change in ADM lesions ( Figures 2K and S1D, H, L, P). These data demonstrate that glycolytic markers are induced in ADM lesions of the three murine genetic pancreatic cancer models.
To analyse metabolic changes in detail, we performed nuclear magnetic resonance (NMR) metabolome analysis of tissue from 10-week-old mice of all three mouse models and wildtype animals. The principal component analysis revealed that samples clustered according to genotypes ( Figure 2L), with samples exhibiting higher ADM rates showing greater differences to wild-type samples. The glucose levels in the tissue increased as the number of ADM lesions increased ( Figure 2M), whereas, the level of the TCA metabolite succinate decreased ( Figure 2N). Collectively, these findings indicate that a metabolic switch towards increased glycolysis is associated with ADM formation in human and murine tissue.

Glycolysis is required for ADM formation
To further investigate these in vivo findings we used an in vitro ADM formation assay. In this well-established assay, normal acinar cells are isolated from the pancreas, embedded into a 3D collagen gel and transdifferentiated into ADM structures. In wild-type acini, ADM can be induced by the inflammatory ligands TGFa and interleukin 17a (IL17a) 7,14 , whereas, explanted acini from Kras G12D , Pi3k CAH1047R , or Mek1dd mice undergo ADM without further treatment ( Figure S2A-E). Low ADM rates were found in ligand induced ADM. The highest transdifferentiation rates were detected in Mek1dd, followed by Pi3k CAH1047R and Kras G12D acini ( Figure 3A).
The phenotypic switch of ADM is characterized by reduced expression of acinar markers (e.g. AMY2A) and enhanced expression of ductal markers such as KTR19 15 . Indeed, ductal-like structures were negative for AMY2A and stained positive for KRT19, as confirmed on RNA and protein level as well ( Figures S2F-M). Downregulation of acinar genes and induction of ductal genes was also confirmed when comparing transcriptomes of freshly isolated (basal) and transdifferentiated acini ( Figure S2N).
In addition to the changes in acinar and ductal genes following ADM formation, we observed increased Glut1 expression ( Figure S3A) and increased GLUT1 and LDHA protein levels in comparison to freshly isolated acini ("0 days") ( Figure 3B). Immunofluorescence staining showed that GLUT1 expression colocalised with KRT19, corroborating our in vivo findings of increased GLUT1 expression in ADM lesions ( Figures 3C-H). A further analysis of the transcriptional changes revealed that enrichment scores of gene sets related to glycolysis were upregulated during transdifferentiation in comparison to freshly isolated acini ("basal"); whereas OxPhos-related sets were predominantly downregulated ( Figure 3I). These results suggest that transdifferenting acini switch to glycolysis. Additionally, the metabolome was analysed by NMR: Here, we observed decreased glucose levels and increased lactate levels in transdifferentiated acinar cells and their supernatants compared to freshly isolated acini. These changes correlated with the different ADM rates (Figures 3J-K and S3B-C). This finding is in line with ADM-induced glucose consumption and lactate secretion. Isotopic tracing following the metabolism of 13 C-labeled glucose by LC-MS analysis revealed a decrease in intracellular hexose-P after 24 h ( Figure 3L). The labelling was comparable to the basal timepoint ("2 h"), indicating the consumption of glucose. In addition, TCA cycle intermediates citrate, succinate and fumarate decreased following ADM ( Figure 3M, S3D-E). Isotopic tracing analysis revealed that TCA cycle metabolites like alpha-ketoglutarate and fumarate decreased in both amount and labelling ( Figures 3N-O). Therefore, decreased levels TCA cycle intermediates are likely the result of reduced synthesis rather than increased consumption and turnover.
To directly assess glycolysis and mitochondrial respiration during ADM, the extracellular acidification rate (ECAR) and the oxygen consumption rate (OCR) were measured. Basal OCR was increased significantly and close to maximal rates in Mek1dd acini compared to all other genotypes ( Figure S4A, E). In addition, ECAR was elevated at basal levels in Mek1dd and Pi3k CAH1047R acini, suggesting an increased energy demand. Upon addition of oligomycin to block mitochondrial ATP production, ECAR increased significantly, demonstrating a pronounced spare glycolytic capacity in Pi3k CAH1047R and Mek1dd acini ( Figure 4A, S4F). ATP production rates calculated from flux measurements revealed that Mek1dd acini displayed highest levels of ATP production compared to all other genotypes Figure 4B). In all genotypes, glycolysis-derived ATP increased compared to medium-treated wild-type acini. The glycolysis-derived ATP followed the same trend as the transdifferentiation rates measured 24 h later supporting the association between glycolysis and ADM formation.
To investigate the functional need for glycolysis during ADM, we blocked glycolysis at several levels. The GLUT1 inhibitor WZB117 effectively inhibited transdifferentiation ( Figure 4C). Low transdifferentiation rates were already effectively inhibited at low WZB117 concentrations. However, higher doses of WZB117 were required to block ADM in Mek1dd or Pi3k CAH1047R acini. Similarly, HK1 inhibition by 2-deoxy-d-glucose (2-DG) led to a dosedependent reduction of ADM ( Figure 4D) and treatment with galloflavin, an LDHA inhibitor, reduced ADM as well ( Figure 4E). We observed no treatment-induced cytotoxicity in acini treated with glycolysis inhibitors ( Figure S4B-D). To validate the murine findings, human acini were isolated from healthy pancreata and exposed to TGFα ( Figure 4F). Similar to murine acini, WZB117, 2-DG or galloflavin inhibited ADM formation in human acini. Taken together, acinar cells undergoing ADM increase their glycolytic capacity in vitro and in vivo. This metabolic switch is a functional requirement during murine and human ADM formation.

c-MYC reduces glucose flux to the mitochondria
In addition to the metabolic changes, we analysed expressional changes during ADM. Specifically, we looked for transcription factors which are involved in metabolic reprogramming. Strikingly, c-Myc (MYC) target genes were upregulated following ADM ( Figure 5A), especially in ADM from acini harbouring oncogenes. Notably, increased MYC protein levels were detected as early as 3 h after isolation ( Figure 5B). The MYC inhibitor 10058-F4 attenuated ADM formation independently of the stimulus; however, at low 10058-F4 concentrations, ADM formation in Mek1dd of Pi3k CAH1047R acini was only slightly inhibited ( Figure 5C). No changes in cell death were observed following MYC inhibition ( Figure S5A). Likewise, acinar cells with a heterozygous deletion of MYC showed attenuation of ADM when exposed to TGFα or IL17a ( Figure 5D). These data demonstrate that MYC is essential for ADM development.
To determine the effect of MYC on the glycolytic phenotype, we combined MYC and GLUT-1 inhibition during ADM formation. Whereas treatment of Mek1dd or Pi3k CAH1047R acini with low 10058-F4 or WZB117 concentrations resulted only in minor effects on transdifferentiation (25% or 5% reduction for 10058-F4 and 35% or 15% reduction for WZB117 in PI3K CAH1047R or Mek1dd, respectively), the combination of both inhibitors caused a significant synergistic attenuation of ADM ( Figure 5E). The combination of both inhibitors caused no increased cell death ( Figure S5B). Additionally, treatment of acini with 10058-F4 displayed a decreased in hexose-P levels as detected by 13 C-glucose isotopic tracing ( Figure  S5C). Investigating the fate of the metabolised hexose-P, we found decreased labelled aspartate and glutamate after 24 h, which are both side branches of the TCA cycle (Figures 5F, G). Importantly, 10058-F4 treatment reversed this decrease indicating that MYC was involved in directing the metabolic flux away from the mitochondria during ADM. 16 including not only glycolysis but also glutamine metabolism. Enzymes of the glutaminolysis like glutaminase (Gls), Glud and Gpt2 as well as transporters for glutamine and glutamate were either unchanged or downregulated during ADM and only Got2 was upregulated ( Figure  S5D). Additionally, intracellular amounts of glutamine and glutamate decreased during transdifferentiation independently of the transdifferentiation rate ( Figure S5E-F). Also, extracellular levels of glutamine were independent of the transdifferentiation rate ( Figure  S5G-H). However, the extracellular glutamate levels increased during ADM, which was less pronounced at higher transdifferentiation rates ( Figure S5H). Blocking GLS using BPTES which inhibits glutamine to glutamate conversion did not affect ADM ( Figure S5I). Taken together, the metabolic role of MYC during ADM rather appears to be the strengthening of glycolysis than induction of glutaminolysis.

Mitochondrial activity is needed for ADM formation
So far the preserved respiration seen in the OCR measurements seems unclear considering increased glycolysis, the reduction of TCA metabolites and independence of glutaminolysis. The first step of oxidative phosphorylation (OxPhos) is performed by the pyruvate dehydrogenase (PDH). The amount of phosphorylated (inactive) PDH decreased during transdifferentiation ( Figure S6A) and PDH inhibition reduced ADM formation in all stimulating conditions ( Figure S6B). Additionally, inhibition of complexes I, III, or V of the electron transport chain (ETC) using rotenone, antimycin A, or oligomycin, respectively, attenuated transdifferentiation ( Figure  6A). This data clearly demonstrate that, besides glycolysis, mitochondrial function is essential for ADM formation. Surprisingly, mitochondrial uncoupling with carbonyl cyanide m-chlorophenylhydrazone (CCCP) did not affect the transdifferentiation rates ( Figure 6B). Importantly, addition of CCCP to oligomycin-treated acini partially rescued the transdifferentiation of TGFa-stimulated wild-type or Kras G12D acini and fully rescued ADM formation in the other inducers ( Figure 6B). These data demonstrate that the ETC is required whereas mitochondrial ATP production is dispensable during ADM. To understand the requirement of a functional ETC, we performed isotope tracing on acini treated with oligomycin, CCCP or both using fully labelled 13 C-glucose. Afterwards, metabolites were screened for a pattern correlating with ADM formation: Decrease with oligomycin and rescue by CCCP. As a result, we found that intracellular serine was strongly increased during ADM ( Figure 6C). This increase was partially blocked by oligomycin and rescued by CCCP paralleling the transdifferentiation rates. Isotopic tracing revealed that parts of the serine were synthesised from glucose within 2 h of labelling ( Figure S6C). As a functional test, we supplemented serine in a concentration comparable to Dulbecco's modified Eagle's medium (DMEM) to rescue the oligomycin-induced block in ADM formation. Whereas Kras G12D acini only showed a weak response, all other conditions demonstrated a partial to full rescue ( Figure 6D). It has been reported already that an impairment of the ETC results in a decrease of NAD + which is required for serine synthesis. To this end, we treated acini with nicotinamide (NAM), a membrane-penetrating precursor of NAD + . The treatment with NAM rescued the oligomycin-induced block in ADM formation, paralleling the CCCP and serine treatment. Additionally, inhibition of serine biosynthesis by NTC-502 attenuated ADM formation comparable to oligomycin ( Figure 6E). These results indicate that a functional ETC is mandatory for ADM formation by providing NAD + for serine synthesis.
Serine is involved in multiple pathways like synthesis of DNA, lipids and membranes, methylation processes and ROS defence. Comparing transcriptional levels of freshly isolated and transdifferentiated acini revealed no significant upregulation of gene sets covering serine-glycine-one-carbon-metabolism (SGOC) 17 , DNA or lipid biosynthesis ( Figure S6D-S6G). In contrast, genes associated with glutathione (GSH) regeneration were significantly upregulated ( Figure S6H) and the precursor of GSH, cystathionine, increased in abundance as well ( Figure 7B). Strikingly, both serine and cystathionine levels decreased after MYCinhibition ( Figure 7A-B), linking the sensitivity of ADM towards MYC-inhibition and serine depletion. At the same time, MYC-inhibition led to an increase in the mRNA for Gclc, the rate-limiting enzyme of GSH synthesis ( Figure 7C). Compensatory increases of Gclc have been described in situations of GSH-depletion.
Additionally, we inhibited GCLC by BSO (Buthioninsulfoximin) to test for the functional relevance of GSH synthesis during ADM formation. In line with our previous data, inhibition of GSH synthesis led to a significant impairment of ADM formation in all tested conditions ( Figure 7D). These results suggest that during ADM, MYC-mediated serine synthesis is needed for GSH production.

Discussion
One of the critical challenges of cancer prevention is the lack of understanding of key molecular changes during cancer initiation. Reprogramming of cellular metabolism is a recognized hallmark of cancer 1 . Cancer cells are dependent on these metabolic changes to maintain energy balance and support biosynthesis 18 . Unlike the abundant work on genomic and transcriptional alterations in precancerous tissues, a comprehensive analysis of metabolic reprogramming has not been performed yet.
Here, we show that murine and human ADM lesions display increased expression of glycolytic markers GLUT1, HK1, and LDHA. Increased expression of GLUT1 is invariably detected in KRAS mutant pancreatic lesions 19 . HK1 and 2, rate limiting enzymes in the glycolytic pathway, are upregulated in primary pancreatic cancer specimens and lactate dehydrogenase A (LDHA), executing the final step of aerobic glycolysis, is elevated in pancreatic cancer patients 20 . The increased expression of these glycolytic markers in precursor lesions of pancreatic cancer suggests that a metabolic switch, increased aerobic glycolysis and attenuated OxPhos (Warburg effect), might already occur during ADM. In addition, the amount of glucose in the tissue increased with the number of ADM lesions. At the same time, the amount of succinate, a metabolite of the TCA cycle decreased. In vitro studies revealed that glucose consumption and lactate secretion correlated with the transdifferentiation rate. Meanwhile, TCA metabolites decreased during ADM. Analysis of the oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) revealed a greater increase in glycolysis-derived ATP with increasing ADM rates compared to OxPhosderived ATP. Finally, glycolysis inhibition blocked ADM formation of isolated murine and human acini. Our findings are consistent with previous observations that freshly isolated wild-type acinar cells rely on both glycolysis and OxPhos 21 , with glycolysis inhibition having a major impact on ATP homeostasis. Also, hypoxia increased glycolysis in acinar cells 22 . Moreover, ceruleintreatment, which causes hyperstimulation of acinar cell and ADM, reduced OxPhos in wildtype acini 23 . Only a few descriptive studies indicate that cell metabolism may be reprogrammed at early stages of tumourigenesis. For example, hexokinase expression was described in liver neoplasia 24 , whereas preneoplastic rat renal basophilic cell lesions display increased Glut1 expression 25 . GLUT1 expression increases from low-to high-grade dysplasia in colon adenomas 26 . The comparison of precursor lesions and lung cancer samples indicated an early metabolic change as well 27 . A similar study was performed in hepatocellular cancer patients 28 . An increase in metabolic alterations was observed during tumour progression in these two studies. However, in both studies the "normal tissue" which was used as a control tissue was taken from resections of cancer patients. Therefore, paracrine metabolic alterations induced by neoplastic lesions in close proximity cannot be excluded. This limitation also applies to our analysis of human precursor lesions. In contrast, our murine in vitro data from stimulated wild-type acinar cells or active oncogenes as sole drivers exclude the influence of neighbouring tumour cells. From these complementary results in human and mouse samples, we conclude that a metabolic switch occurs during ADM and is required for ADM formation. We propose that metabolic reprogramming is essential for the initiation of pancreatic cancer and a major driver in its progression. Glutamine is a major source of energy entering the TCA cycle via a-ketoglutarate (a-KG). It is known that pancreatic cancer cells heavily rely on glutamine 29 . In contrast, our data demonstrate that acinar cells undergoing ADM do not depend on glutamine. Several glutamine transporters including SLC1A5 (ASCT2), SLC38A5 and the mitochondrial glutamate transporter SLC25A1 as well as enzymes of the glutamine metabolism like GLUD1, GLS and GOT are upregulated in many types of cancer 30 . Following ADM development, these transporters and enzymes were either unchanged or downregulated, with only GOT2 being slightly increased. Moreover, intracellular and extracellular amounts of glutamine did not correlate with transdifferentiation rates. Blocking the enzymatic conversion of glutamine to glutamate using the GLS inhibitor BPTES did not inhibit ADM formation. Thus, our data suggest that glutamine metabolism is not essential during ADM. It might be hypothesized that enhanced glutamine metabolism might be beneficial for cancer cells by promoting proliferation but is not required for transdifferentiation. The transcription factor MYC is an essential downstream effector of oncogenic KRAS in the pancreas 31 . Oncogenic levels of MYC reprogram cellular metabolism, and are an integral part of the adaptation of cells to nutrient deprivation, a hallmark of cancer development. In vivo studies confirmed that MYC impacts the metabolism of glutamine mainly by enhancing the expression of GLS 32 . However, our data suggest that glutamine metabolism is not essential during ADM. A reason for this discrepancy might be the only transient MYC-activation observed in our system. Recent data from PDAC cells demonstrated a MYC-dependent regulation of TCA cycle activity 33 . Additionally, upregulation of MYC favours glycolysis over OxPhos in cancer stem cells 34 . These results are in line with our finding that MYC supports glycolysis during ADM. Apart from glycolysis, serine metabolism is required for ADM formation. Serine contributes to a number of processes including protein, lipid and nucleotide synthesis, glutathione and NADPH generation and the provision of one carbon units for the folate cycle and methylation 35 . De novo synthesis of serine from glucose starts by the oxidation of 3-phosphoglycerate (3-PG) which requires NAD + . Lactate production alone only restores as much NAD + as needed for the glycolysis. Hence, additional regeneration of the NAD + pool requires a functional ETC ( Figure 7E). Inhibition of the ETC either pharmacologically or by hypoxia has been shown to disturb the NADH/NAD + balance 36 . Therefore, de novo serine synthesis is dependent on a functional ETC 37 . We show that oligomycin blocks ADM formation, however, uncoupling with CCCP rescued oligomycin-inhibited ADM formation. These data indicate that ATP production is unlikely to be the main function of mitochondria during ADM. In addition, supplementing serine as well as addition of NAM was also able to rescue the inhibitory effect of oligomycin, while the effect of serine synthesis inhibition was comparable to oligomycin. Therefore, our data suggest that mitochondria are needed to provide NAD + for de novo serine synthesis. To our knowledge, a requirement for serine synthesis in early lesions has so far not been demonstrated. In fully developed PDAC, serine was shown to be essential for tumour survival 38 . Tumours which are unable to synthesise serine are dependent on extracellular supplementation. Our culture medium did not contain serine. Therefore, our in vitro experiments were performed under serine-depleted conditions. As for the fate of serine, gene sets covering DNA, lipid or membrane synthesis or one-carbon metabolism were either unchanged or downregulated. On the other hand, we demonstrate that increased intermediates of GSH during ADM formation were MYC-dependent. In addition, we could show that inhibiting GSH synthesis blocked ADM formation. The requirement for serine synthesis in the initiation of early lesions might open strategies to prevent early pancreatic carcinogenesis with dietary of pharmacologic interventions. A serine-free diet in combination with a serine synthesis inhibitor resulted in reduced tumour growth in a Kras G12D -driven mouse model of colorectal cancer 39 . Thus, dietary serine restriction might inhibit pancreatic carcinogenesis as well. In summary, we demonstrate for the first time that metabolic reprogramming is an essential feature of acinar-to-ductal metaplasia the first step in pancreatic carcinogenesis. This reprogramming comprises increased glycolysis and a functional electron transport chain (ETC) to ensure serine synthesis and GSH synthesis. The observed changes are conserved among oncogene-and ligand-induced ADM formation. While the metabolic switch already takes place during the early stages of PDAC development, the diversification into metabolic subtypes must occur at later stages.

Declaration of interests
None.

Human acinar cell isolation
Healthy human pancreata were obtained from Surgery Department of the MRI. After soaking in SOL1 (McCoy'2 5A Medium, 0.02% Trypsin inhibitor from soy bean, 0.1% BSA) for 10 min, RT, specimens were cut into pieces and digested twice with 10 ml/g tissue of 50 U/ml collagenase (CLSPA Worthington Biochemical Corporation) in SOL1 for 10 min, stirred at 37°C. Cells were passed through a 200 µm cell strainer, washed with SOL1 and centrifuged at 157 xg. Cells were recovered in culture medium (see murine acinar cells) + 30% FCS and incubated for 30-60 min at 37°C. Afterwards, cells were collected at 14 xg for collagen embedding.

In vitro transdifferentiation assay
Isolated acinar cells were recovered in culture medium, mixed at a ratio of 1:1 with neutralized collagen I from rat tail (Corning) and plated on a solid layer of 2.5 mg/ml collagen containing 10% 10x PBS. After solidification, an additional layer of collagen was added. Culture medium was added on top of the three-layer culture system. On day 1, the medium was changed. Transdifferentiation rates were determined at d2 (Mek1dd, Pi3k CAH1047R ) or d3 (Kras G12D , stimulated wild-type

Cytotoxicity assay
To determine treatment induced cytotoxicity, LDH release into the medium was measured. All supernatants from the transdifferentiation assay were collected. Acinar cells lysed with 2% TritonX-100 (Sigma) were used as positive control; supernatant from medium-treated acinar cells as negative control. LDH activity was assayed using the Cytotoxicity Detection Kit (Roche) according to the manufacturer's protocol.

OCR and ECAR measurement
Seahorse measurements were performed as described before 43 . Briefly, acinar explants were seeded in collagen one day before measurement. 1 h before the measurement, the medium was changed to XF assay medium (Agilent) containing 5 g/l glucose. One assay cycle consisted of 1 min mixing, 2 min waiting and 4 min measuring. After basal respiration measurements (5 cycles), oligomycin was added (2 µg/ml) to inhibit ATP production and proton leak respiration measured (4 assay cycles). Afterwards, carbonyl cyanide mchlorophenylhydrazone (CCCP) (1 µM) and pyruvate (5 mM) were added to determine the maximal respiratiory rate (4 cycles). Addition of rotenone (2.5 µM) and antimycin A (2.5 µM) served to fully inhibit the respiratory chain (3 cycles). 2-deoxyglucose (2DG) (100 mM) was added to inhibit glycolysis and correct for non-glycolytic acidification (6 cycles). ATP production rates were calculated as published previously 44 .

Immunohistochemistry
Evaluation of expression of Hexokinase1, GLUT1 and LDHA was evaluated on human tissues using tissue microarrays (TMAs). For TMA construction, tissue blocks from in total 63 patients from the archive of the Institute of Pathology of TUM were selected, containing either PanIN and/or acinar-to-ductal metaplasia. The slides were annotated and TMAs were constructed using a tissue microarrayer (Beecher Instruments). For murine tissues, additionally expression of ATP5D was evaluated immunohistochemically. 2 µm thin slides were then prepared and immunohistochemistry was performed on a Bond Rxm system (Leica) with primary antibodies against Hexokinase 1 (1:500, Cell signaling, 2024), LDHA (1:10.000, Novus Biologicals, NBP1-48336), GLUT1 (1:750, abcam, ab115730) or ATP5D (1:100, Sigma Aldrich, HPA002865). Briefly, after deparaffinization, slides were pretreated with antigen retrieval solution 1 (ER1, Leica) for 30 or 40 minutes and primary antibody binding was visualized using a Polymer Refine Detection Kit (Leica). Slides were scanned with an AT2 (Leica) scanning system and evaluated semiquantitatively by a board-certified pathologist (KS). Marker expression in human tissue was scored with respect to the specific pathology using a 4-tier scoring system (0-no expression, 1-slight expression, 2-moderate expression, 3-strong expression). For the murine specimen, expression for all marker was evaluated with respect to normal acinar tissue in comparison to acinar-to-ductal metaplasia (ADM) using a 6-tier scoring system reflecting expression frequency and intensity (from 0-no expression to 5-strong expression in all cells).

Isolation of RNA
Cultivated acinar cells were explanted from the collagen gel using SOL2 (see murine acinar isolation). Afterwards, cells were pelleted at 300 g and lysed in 200 µL homogenisation buffer including 2% 1-thioglycerol. The isolation was done using the Maxwell Kit (all Promega) according to the manufacture's protocol.

RNASeq
Library preparation for bulk-sequencing of poly(A)-RNA was done as described previously 45 . Briefly, barcoded cDNA was generated with a Maxima RT polymerase (ThermoFisher) using oligo-dT primer containing barcodes, unique molecular identifiers (UMIs) and an adaptor. Ends of the cDNAs were extended by a template switch oligo (TSO) and full-length cDNA was amplified with primers binding to the TSO-site and the adaptor. NEB UltraII FS kit was used to fragment cDNA. After end repair and A-tailing, a TruSeq adapter was ligated and 3'-endfragments were amplified using primers with Illumina P5 and P7 overhangs. In comparison to Parekh et al. (2016), the P5 and P7 sites were exchanged to allow sequencing of the cDNA in read1 and barcodes and UMIs in read2. The library was sequenced on a NextSeq 500 (Illumina) with 67 cycles for the cDNA in read1 and 16 cycles for the barcodes and UMIs in read2. Data was processed using the published Drop-seq pipeline (v1.0) to generate sampleand gene-wise UMI tables 46 . Reference genome (GRCm38) was used for alignment. Transcript and gene definitions were used according to the GENCODE Version M25. RNASeq data were normalized with the Deseq2 package in R.

Western Blot
Acinar cells were lysed with RIPA buffer containing protease and phosphatase inhibitors. Protein extracts were separate using SDS-PAGE, transferred to a nitrocellulose membrane and tested with the following antibodies:  All MS-analyses were performed on a high-resolution QExactive mass spectrometer (ThermoScientific) in alternating positive-and negative full MS mode with a Scan Range of 69.0 -1000 m/z at 70K Resolution and the following ESI source parameters: Sheath gas: 30, auxiliary gas: 10, sweep gas: 3, Aux Gas Heater temperature: 120 °C. Spray voltage: 2.5 kV in positive ion mode and 3.6 kV in negative ion mode, Capillary temperature: 320 °C, S-lens RF level: 55.0. Signal determination and quantitation was performed using TraceFinder™ Software Version 3.3 (ThermoFisher). All analyses were performed with five independent biological replicates.

Statistical analysis
The graphical arrangement of data and statistically testing for significance was done using GraphPad Prism 8 (GraphPad Software). For parametric data a Student's t-test or an ANOVA was performed. For non-parametric data (histological scores) a Kruskal-Wallis test was used.            Sup. Fig. 4 Seahorse measurement and LDHA release assay   Sup. Fig. 6 Mitochondrial regulator PDH is activated and required for ADM and serine levels correlate with mitochondrial activity