Oncogenic KRAS Induces NIX-Mediated Mitophagy to Promote Pancreatic Cancer

Pancreatic cancer is a deadly malignancy with a dismal 5-year survival rate of 8% (1). The most prevalent type, pancreatic ductal adenocarcinoma (PDAC), is characterized by near-universal mutational activation of KRAS (2). Oncogenic KRAS can promote proliferation under the nutrient-limiting conditions found in tumors by altering both the uptake of nutrients from the environment and the expression and activity of metabolic enzymes (3–6). KRAS-mutant cancer cells have also been shown to use autophagy to meet the metabolic demands of proliferation and tumorigenesis in vitro and in vivo (7–10). In these studies, pharmacologic inhibition and/or genetic ablation of general autophagy and lysosomal programs led to the accumulation of defective mitochondria and consequent metabolic insufficiencies. In addition, cells surviving loss of oncogenic Kras expression in an inducible model of pancreatic cancer had increased mitochondrial content and exhibited increased sensitivity to mitochondrial inhibitors (11). These findings suggest a possible connection between the mitochondrial content of Kras-mutant cells and their resultant metabolic alterations, which we hypothesized may affect tumor development in vivo.

We previously reported that Kras^G12D activated the Nfe2l2/ NRF2 antioxidant pathway to lower cytoplasmic reactive oxygen species (ROS), promoting cell proliferation and the initiation of early lung and pancreatic cancers (12). As part of these findings, we observed a Kras^G12D-dependent reduction in mitochondrial ROS levels but did not further interrogate the mitochondrial response to oncogenic KRAS. Accordingly, we decided to further evaluate the acute effects of endogenous Kras^G12D expression on the mitochondrial network using mouse embryonic fibroblasts (MEF) and pancreatic ductal organoids generated from LSL-Kras^G12D/+ mice (13, 14). Cells were cultured in low glucose medium [0.5 mmol/L and 2 mmol/L glucose for two-dimensional (2-D) cells and organoids, respectively], because these limited conditions select for the emergence and outgrowth of cells harboring Kras mutations in vitro and are a closer approximation than normal commercial media for the nutrient environment of PDAC in vivo (15, 16). Consistent with our published work, we found that Kras^G12D decreased both cytoplasmic and mitochondrial ROS levels, indicated by diminished DCF-DA and MitoSox Red fluorescence intensity, respectively, in MEFs (Fig. 1A). Although levels of cytoplasmic ROS in Kras-mutant MEFs remained at Kras-wild type (WT) levels in the absence of NRF2, the reduction in mitochondrial ROS persisted, suggesting that the effects of mutant Kras on mitochondrial ROS are independent of the NRF2 program (Supplementary Fig. S1A). Alongside diminished mitochondrial ROS, the expression of mutant Kras in MEFs and organoids also led to decreased mitochondrial mass and total mitochondrial membrane potential per cell (Fig. 1A and B), suggesting a reduced mitochondrial network. Quantification of transmission electron microscopy (TEM) images confirmed a reduction in the mitochondrial fraction of Kras-mutant MEFs and organoids (Fig. 1C and D; Supplementary Fig. S1B and S1C, arrows mark mitochondria). Membrane-bound autophagosomes containing engulfed mitochondria could also be found in these micrographs (Supplementary Fig. S1B). Determination of the ratio of mitochondrial DNA (mtDNA) to nuclear DNA (nDNA), an alternative method of evaluating mitochondrial content, also showed a decreased ratio in Kras^G12D MEFs compared with Kras-WT MEFs, further indicating a suppression of mitochondrial content (Supplementary Fig. S1D; ref. 17).

To study the effects of loss of mutant Kras on this mitochondrial phenotype, we crossed mice harboring an excisable endogenous mutant Kras allele (FRT-LSL-Kras^G12V-FRT; Supplementary Fig. S1E) with mice harboring LSL-Trp53^R172H/+, Pdx1-Cre and Rosa26-FlpOERT2 alleles to generate an autochthonous “FPC” mouse model of PDAC in which we could delete mutant Kras using tamoxifen. Cell lines derived from tumors arising in FPC mice were cultured with 4-OH-tamoxifen (4OHT) for 96 hours, which led to excision of the mutant Kras allele (Supplementary Fig. S1F). After 4OHT treatment, FPC cells had significantly increased mitochondrial mass measured by MitoTracker Green staining (Fig. 1E). Moreover, in the setting of KRAS-mutant human PDAC cell lines (Suit2 and FA6), siRNA-mediated knockdown of KRAS similarly led to an increase in mitochondrial mass as measured by flow cytometry, electron microscopy, and mtDNA/nDNA ratio (Fig. 1F; Supplementary Fig. S2A–S2G).

Mitochondrial mass is determined by the balance between the biogenesis of new mitochondria and the selective degradation of existing mitochondria via mitophagy (18). Given the increased occurrence of engulfed mitochondria that we observed in Kras-mutant organoids—and to a lesser extent in Kras-mutant MEFs—we hypothesized that mitochondrial degradation was increased in these cells. Consistent with this prediction, we did not observe significant changes in the expression of key transcription factors involved in mitochondrial biogenesis, including Pgc1a, Nrf1, and Tfam, following activation of mutant Kras, suggesting that a decrease in biogenesis was not the primary explanation for decreased mitochondrial mass (Supplementary Fig. S3A and S3B). To assess mitochondrial degradation, we first examined the expression of several mitophagy genes in Kras^G12D-expressing MEFs. In this setting, expression of Kras^G12D resulted in increased mRNA expression of the mitophagy mediator BCL2/adenovirus E1B 19-kDa–interacting protein 3-like (Bnip3l/Nix) but not of other mediators of mitophagy such as p62/SQSTM1, PINK1, or PARK2 (Supplementary Fig. S3C). Consistent with this observation, the expression of oncogenic Kras led to increased NIX protein in the mitochondrial fraction of MEFs without any change in the amount of p62 (Fig. 2A). In organoids, we found that activation of mutant Kras also led to an increase in NIX protein and a trend toward increased Nix mRNA level, with a concurrent increase in levels of lipidated LC3 (LC3-II), a marker of active autophagy (Fig. 2B; Supplementary Fig. S3D). In parallel, the mRNA levels of Acsm3, a gene we previously identified as Kras-dependent in murine pancreatic organoids, increased notably (Supplementary Fig. S3D; ref. 14). We also observed increases in both Nix and p62 mRNA levels in tumor tissue isolated from the LSL-Kras^G12D/+; LSL-Trp53^R172H/+; Pdx1-Cre (KPC) murine model of PDAC compared with control pancreas tissue from LSL-Trp53^R172H/+; Pdx1-Cre (PC) mice, suggesting that as in MEFs, oncogenic KRAS engages Nix in vivo (Supplementary Fig. S3E). Consistent with these findings, the mitochondrial fractions from a series of organoids derived from murine pancreatic intraepithelial neoplasia (PanIN; KC) and PDAC (KPC) tumors also exhibited higher levels of NIX protein compared with normal pancreas organoids (Fig. 2C).

To uncover how NIX expression varies during the progression of PDAC in vivo, we examined NIX levels by IHC in murine normal pancreata, PanIN, PDAC, and lung metastatic lesions (Fig. 2D). In this setting, we observed increased NIX staining intensity in PanIN lesions relative to normal ducts. NIX staining further intensified in PDAC and lung metastatic lesions. In contrast with the early induction of NIX, levels of p62 were not increased until later in the progression of murine PDAC, suggesting that NIX-mediated mitophagy represents a distinct pathway from p62-mediated mitophagy (Fig. 2D). In agreement with our mouse IHC data, we also observed higher levels of NIX staining in human PanIN lesions compared with adjacent normal ducts as well as intense staining for NIX in human PDAC sections, suggesting that NIX induction is an early and sustained event during pancreatic tumorigenesis (Fig. 2E).

To confirm that oncogenic KRAS maintains NIX expression, we assessed FPC 2D cells following the deletion of Kras^G12V. Excision of mutant Kras led to a reduction in Nix mRNA levels and NIX protein levels in FPC organoids and 2-D cells, respectively (Fig. 2F; Supplementary Fig. S3F), demonstrating that sustained KRAS signaling is required to maintain NIX levels in Kras-mutant cells. In support of this finding, we also observed decreased levels of NIX in the mitochondrial fraction of Suit2 and FA6 human pancreatic cancer cells that were treated with KRAS siRNA (Supplementary Figs. S2E–S2G, S3G, and S3H). Moreover, decreased mitochondrial NIX was also observed in FA6 cells treated with either an siRNA directed against MAP2K2/MEK2 or the MEK inhibitor AZD6244, demonstrating a requirement for MAPK signaling through MEK downstream of oncogenic KRAS to maintain NIX levels (Supplementary Fig. S3G–S3J).

To assess the relationship between NIX expression and mitochondrial content in vivo, we compared NIX protein levels in the mitochondrial fraction derived from the pancreata of WT mice and PanIN-bearing KC mice and observed higher levels of NIX protein in KC pancreata compared with WT controls (Fig. 3A; Supplementary Fig. S4A). To gauge mitochondrial content, we isolated DNA from these pancreata and determined the ratio of mtDNA to nDNA (Fig. 3B). As in our in vitro experiments, the mtDNA/nDNA ratio for KC pancreata was significantly lower than that of WT pancreata and showed an inverse relationship with the amount of NIX protein in the mitochondrial fraction of the tissue, confirming that increased NIX levels are associated with less mitochondrial content in vivo (Supplementary Fig. S4B).

Building upon these observations, we sought to confirm whether NIX-mediated mitophagy is responsible for the suppression of mitochondrial content observed after mutant Kras expression in vitro and in vivo. To this end, we assessed mitochondrial mass in Kras-mutant MEFs and in two KRAS-mutant human PDAC cell lines following siRNA-mediated depletion of NIX. In Kras-mutant MEFs, Nix knockdown led to increased mitochondrial mass, mitochondrial ROS, and mitochondrial membrane potential compared with Kras-mutant cells treated with control siRNA (Fig. 3C). Similar results were observed following NIX knockdown in FA6 and Suit2 cells (Fig. 3D). These changes were also recapitulated in both MEFs and FA6 cells treated with the MEK inhibitor AZD6244 (Supplementary Fig. S4C and S4D) as well as in FA6 cells treated with siRNA against MAP2K2/MEK2 (Fig. 3E), confirming that MAPK signaling is important for maintaining both NIX and the mitophagy program. In the human cancer cell lines, these findings were further validated by confirming that siRNA depletion of NIX increased both the mitochondrial fraction in TEM images and the mtDNA/nDNA ratio of the siNIX cells compared with controls (Supplementary Fig. S2A–S2C), similar to the changes observed following the knockdown of KRAS in these cells.

Because NIX is known to be a facilitator of mitophagy and higher NIX levels correlated with reduced mitochondrial mass across our experimental systems, we sought to confirm whether perturbing NIX would alter mitochondrial flux in Kras-mutant cells (19, 20). We generated FA6 and Suit2 cells that stably express the mitoQC mitophagy probe (FA6-mtQC and Suit2-mtQC), a mitochondrially targeted tandem GFP-mCherry probe that allows for analysis of relative mitophagic flux (21). In cells expressing the probe, mitochondria will exhibit dual mCherry and GFP fluorescence at steady state whereas those mitochondria found within lysosomes, that is, mitochondria undergoing mitophagy, will lose GFP fluorescence, which is quenched in an acidic environment. In an analogous way to the tandem LC3 general autophagy probe and the mt-Keima mitophagy probe, the ratio of mCherry to GFP mitoQC fluorescence as determined by flow cytometry can reveal differences in mitophagic flux (22, 23). NIX-depleted FA6-mtQC and Suit2-mtQC cells showed a modest reduction in their mCherry/GFP ratios compared with control cells, consistent with decreased mitophagic flux in these cells (Supplementary Fig. S4E and S4F). In comparison, knockdown of p62 led to greater suppression of mitophagy in both FA6 and Suit2 cells, likely reflecting p62′s more expansive role as a mediator of various selective autophagy programs, including the PINK-PARKIN-p62 mitophagy program, which can clear damaged mitochondria independently of NIX (ref. 18; Supplementary Fig. S4E and S4F).

In glucose-limited environments, suppression of mitochondrial content and mitochondrial network activity could act to directly limit the amount of glucose available for mitochondrial oxidative phosphorylation. In such a model, mitophagy would divert glucose away from the tricarboxylic acid (TCA) cycle and allow for its greater utilization in aerobic glycolysis and other anabolic pathways. Such channeling of glucose to lactate could also support the high level of NAD+ generation required to maintain the elevated glycolytic rate of cancer cells (24). Because it has been shown that KRAS mutations provide a growth advantage to cells cultured in glucose-limited medium (15), we investigated whether the KRAS–NIX mitophagy program might support increased proliferation in these conditions. In agreement with this hypothesis, we found that although siRNA-mediated depletion of NIX in both FA6 and Suit2 cells did not alter proliferation in high glucose (25 mmol/L glucose) media, proliferation was reduced in low-glucose (0.5 mmol/L glucose) conditions (Fig. 3F). This proliferative defect following Nix loss was even more profound in Kras^G12D-expressing MEFs and occurred even in high-glucose conditions (Supplementary Fig. S4G).

Oncogenic RAS has been previously reported to engage a p62-dependent autophagy program that is required for pancreatic cancer progression (7, 25). This requirement for p62 in RAS-mutant cells may reflect both its central function in selective autophagy/mitophagy programs and subsequent maintenance of a pool of functional mitochondria (6) as well as its function in promoting Nrf2 antioxidant signaling through interactions with KEAP1 (26). As such, we considered whether the NIX program might similarly prevent the accumulation of damaged mitochondria or instead reduce pools of functional mitochondria—a phenomenon reported during red blood cell development (19, 20). To assess whether mitochondria restored by the loss of NIX in Kras-mutant cells were functional, we measured the oxygen consumption rate (OCR) following constitutive knockdown of Nix in mouse tumor (mT) organoids, Kras^G12D-expressing MEFs, and human PDAC cells. In organoids, short hairpin RNA (shRNA)–mediated knockdown of Nix led to increased basal respiration and a greater maximal respiratory rate in response to the mitochondrial uncoupler FCCP, suggesting that restored mitochondria in organoids are indeed functional (Fig. 4A). Similarly, in Kras^G12D-expressing MEFs and in Suit2 and FA6 cells, NIX knockdown increased basal and maximal respiratory capacity (Supplementary Figs. S4B and S5A, respectively). To further interrogate the functional state of mitochondria restored to NIX-depleted cells, we quantified the number of cristae per mitochondrion in our TEM images of Suit2 and FA6 cells following NIX depletion. A decrease in this metric has been shown to correspond with faulty mitochondrial function (27, 28). Consistent with the mitochondria in siNIX cells being functional, we did not detect significant alterations in cristae per mitochondrion in siNIX cells (Fig. 4C and D). Similarly, we found that although mitochondrial ROS per cell increased after NIX knockdown in vitro (Figs. 1A and 3D), the mitochondrial ROS per mitochondrial mass ratio did not increase as would be expected if the mitochondria were damaged (Fig. 4E). We then challenged cells with the mitochondrial-specific ROS inducer mitoParaquat (mitoPQ; ref. 29) to see whether mitochondria in NIX-depleted cells were more sensitive. As expected, acute treatment with mitoPQ increased the mitochondrial ROS/mass ratio compared with control-treated cells (Supplementary Fig. S5B). However, this occurred to a similar extent independently of the NIX status of the treated cells, suggesting that the mitochondria in NIX-depleted cells were not more sensitive to the treatment. Finally, we cultured FA6 and Suit2 cells in galactose media, an environment where mitochondrial function is required to generate ATP from galactose (30). Intriguingly, we observed increased proliferation in the NIX-depleted cells compared with controls (Fig. 4F), supporting our conclusion that the KRAS–NIX mitophagy program clears otherwise functional mitochondria.

Given that an increased mitochondrial network might consume greater amounts of available cellular glucose, we evaluated whether Kras^G12D-driven mitophagy reduces mitochondrial glucose flux. Consistent with this hypothesis, we observed smaller steady-state pools of the TCA cycle intermediates succinate and malate in Kras^G12D MEFs compared with WT MEFs, with a restoration of these pools to WT levels in Kras^G12D MEFs treated with Nix siRNA (Fig. 4G). In agreement with this observation, siRNA-mediated NIX depletion in both FA6 and Suit2 cells led to greater accumulation of ¹³C₆-glucose–derived label in TCA cycle intermediates, especially citrate and malate, following a 45-minute incubation with 2 mmol/L ¹³C₆-glucose, with comparable labeling of glycolytic pools of glucose 6 phosphate (G6P), dihydroxyacetone phosphate/glyceraldehyde-3-phosphate (DHAP/G3P), and pyruvate (Fig. 4H; Supplementary Fig. S5C and S5E). Although the intracellular lactate pool modestly increased in siNIX cells, relative partitioning of glucose label between citrate in the TCA cycle and lactate, as judged by the citrate m+2/lactate m+3 ratio, was increased (Fig. 4I; Supplementary Fig. S5F). To further interrogate glucose flux in response to perturbing Nix in a more physiologically relevant three-dimensional setting, mT organoids were cultured with 2 mmol/L uniformly labeled ¹³C₆-glucose for 45 minutes, and glucose metabolism via glycolysis and the TCA cycle was analyzed. Both constitutive knockdown of Nix in mT8 organoids and its inducible knockdown in mT5 organoids led to a significant increase in the accumulation of labeled isotopologs of TCA cycle intermediates in shNix organoids compared with controls (Supplementary Fig. S5G–S5I). As in the 2-D cells, increased glucose flux into the TCA cycle correlated with reduced relative glucose flux to lactate (Supplementary Fig. S5J and S5K). These findings demonstrate that suppression of mitochondrial content in Kras^G12D-expressing cells facilitates reduced cycling of glucose-derived carbon within mitochondria and its greater relative conversion to lactate.

Previous studies have demonstrated that pancreatic cancer cells upregulate the nonoxidative arm of the pentose phosphate pathway (PPP) to generate nucleic acids while relying on the NRF2 antioxidant program to limit intracellular ROS (4, 12). Tracing experiments with 2 mmol/L ¹³C₆-glucose demonstrated that NIX depletion led to greater flux from glucose to ribose phosphate (R5P) through the PPP in our organoid and 2-D cultured cells during the 45-minute labeling period (Fig. 4H; Supplementary Fig. S5D, S5H, and S5I). Given that siNIX cells have higher levels of mitochondrial ROS, we predicted that the observed increased glucose flux to R5P might be through the oxidative arm of the PPP to generate NADPH. To test this idea, we incubated FA6 cells for 3 hours with 2 mmol/L 1-¹³C₁-glucose, a glucose isotopolog that can be used to distinguish between the oxidative and nonoxidative routes through the PPP to form R5P (Supplementary Fig. S6A). In this experiment, the ¹³C label is retained on R5P if glucose is metabolized through the nonoxidative branch of the PPP but is lost if it instead traverses the oxidative arm of the PPP. Thus, the relative m+0/m+1 ratio of labels in ribose-5-phosphate reflects the contribution of oxidative versus nonoxidative PPP activity in the cells (31). In unstressed conditions, FA6 siNIX cells exhibited a larger m+0/m+1 ratio than nontargeting (NT) controls, suggesting greater basal requirements for oxidative PPP flux (Fig. 4J)—albeit in maintenance of a more oxidized baseline NADPH/NADP⁺ redox ratio (Fig. 4K). This finding generalized to both Suit2 cells and organoids wherein NIX depletion resulted in reduced steady-state NADPH/NADP⁺ ratios in each system (Fig. 4L and M). In the FA6 cells, applying redox stress using menadione further increased flux through the oxidative branch of the PPP independently of NIX status, as expected, but to a greater extent in siNIX cells, underscoring their increased redox burden (Fig. 4J and K).

PDAC cells have been shown to largely use glutamine to fuel their TCA cycle in vitro (3). As such, we investigated whether NIX depletion also altered glutamine flux by incubating cells for 45 minutes with 2 mmol/L uniformly carbon- and nitrogen-labeled (¹³C₆,¹⁵N₂) glutamine (Supplementary Fig. S6B). Consistent with our glucose tracing experiments, upon knockdown of NIX in FA6 and Suit2 cells and organoids, we observed increased TCA accumulation of glutamine label (Supplementary Fig. S6C–S6F). Malic enzyme activity, in addition to oxidative PPP activity, can generate NADPH and has been shown to be important in PDAC (3, 32). Malic enzyme activity can be inferred by examining the generation of m+3 pyruvate from glutamine-labeled malate (Supplementary Fig. S6G). In glutamine-labeled PDAC cells and organoids depleted of NIX, however, we did not observe significantly different accumulation of m+3 pyruvate between conditions—although labeling of m+3 pyruvate in each case was low, as also reported by Son and colleagues (3)—suggesting that malic enzyme activity was similar (Supplementary Fig. S6C-F). Increased reductive carboxylation has been revealed as a strategy used by nonadherent cells to cope with redox stress (33). In nonadherent cells, reductive carboxylation of glutamine to citrate followed by its subsequent oxidation in the mitochondria acts to shuttle reducing equivalents from the cytosol to the mitochondria. In the FA6 cells and both organoid systems, we found that reductive carboxylation of glutamine to citrate, as judged by the m+5/m+4 citrate ratio in glutamine-labeled cells (34), increased upon NIX knockdown (Supplementary Fig. S6B and S6H–S6K). In Suit2 cells, this was not the case, although these cells had by far the highest levels of m+5 citrate label accumulation independent of NIX status (Supplementary Fig. S6D), suggesting that reductive carboxylation of glutamine to citrate is already occurring at high levels in these cells. The reductive carboxylation of glutamine to citrate can act to shuttle NADPH-reducing equivalents to the mitochondria as follows: NADPH is first consumed by cytoplasmic IDH1 to make citrate from aKG. As long as mitochondria are functional, the subsequent oxidation of this citrate then regenerates NADPH in the mitochondria (33). Redox cycling through the pathway can be inferred by determining whether citrate continues oxidatively to malate, yielding m+4 malate (and mitochondrial NADPH) or proceeds reductively, yielding m+3 malate (but no mitochondrial NADPH; Supplementary Fig. S6G). We did not observe a decrease in the malate m+4/m+3 ratio in any of the cell types tested, as occurs in hypoxia when cells suppress mitochondrial oxidation activity, and even observed an increase in the ratio in Suit2 cells (Supplementary Fig. S6L–S6O; refs. 35, 36). This shows that NIX-depleted cells are capable of completing the NADPH shuttle. Considered alongside our other data, we believe this finding suggests that NADPH shuttling may help to augment mitochondrial redox control in NIX-depleted cells. Taken together, our results suggest that the greater mitochondrial ROS production associated with the increased functional mitochondrial pool in NIX-depleted cells increases their requirement for oxidative PPP flux to generate NADPH. Some of these reducing equivalents may then be shuttled to mitochondria via increased reductive carboxylation of glutamine to citrate.

To determine whether the NIX-mediated mitochondrial and metabolic alterations observed in vitro influence the development of pancreatic cancer, we conditionally deleted NIX in the pancreata of KC and KPC mice using a floxed Nix allele (37). At 3 months of age, KCNix^FL/FL (KCNix^Δ/Δ) mice formed low-grade PanIN-1A lesions similar to KC mice (Supplementary Fig. S7A). We confirmed that NIX was deleted throughout the exocrine pancreas compartment in these mice (i.e., acinar cells and ductal cells) with expression retained in cells of other lineages (i.e., fibroblasts and immune cells; Supplementary Fig. S7B). Despite NIX deletion, PanIN-1 lesions in KCNix^Δ/Δ mice had similar levels of Ki67 and phospo-ERK expression compared with controls (Supplementary Fig. S7B), suggesting that loss of NIX did not initially impair the proliferative effects of oncogenic KRAS signaling. Nonetheless, by later time points (9–12 months) when all KC mice harbored extensive PanIN lesions and possessed little normal tissue, KCNix^Δ/Δ mice retained a significantly larger fraction of disease-free normal pancreas tissue, demonstrating a delay in PanIN progression (Fig. 5A and B). Indeed, although all KC mice examined harbored PanIN-2 and PanIN-3 lesions, most KCNix^Δ/Δ mice still exhibited only low-grade PanIN-1A and PanIN-1B lesions (Fig. 5C and D). Extending these observations to the aggressive KPC genetically engineered mouse model of PDAC, the pancreata of KPCNix^FL/FL (KPCNix^Δ/Δ) mice sacrificed at 4 months of age had mostly low-grade PanIN-1A and PanIN-1B lesions, with only 1 mouse (out of 5 mice analyzed) harboring a solitary cystic papillary neoplasm (CPN), and no mice harboring PDAC within the cohort (Fig. 5E and F). In age-matched KPC mice, PDAC was observed in 60% of KPC mice, and all mice showed at least PanIN-2 lesions (Fig. 5E and F).

Given the differences that we observed in PanIN progression to PDAC in both KC and KPC pancreatic cancer models with Nix deletion, we decided to examine whether this delay also correlated with increased survival within the KPC model. We first confirmed that mitochondrial content was disrupted in KPCNix^Δ/Δ mice. As expected, 16-week-old KPCNix^Δ/Δ mice exhibited increased mitochondrial content in PanIN lesions as judged by the immunofluorescence detection of Tom20, a mitochondrial marker (Fig. 6A and B). We followed a cohort of mice to humane endpoint and found that Nix ablation in the KPC model led to a significant extension of median survival from 21.86 weeks to 34.86 weeks in KPCNix^Δ/Δmice, although KPCNix^Δ/Δ mice eventually all succumbed to malignant disease (Fig. 6C). Interestingly, analysis of The Cancer Genome Atlas (TCGA) data from patients with resected PDAC showed that elevated NIX mRNA expression was associated with significantly shorter survival in humans as well (Fig. 6D). Even though NIX loss was maintained in PDAC tumors from KPCNix^Δ/Δ mice (Supplementary Fig. S8A), the mitochondrial content in PDAC lesions of KPC and KPCNix^Δ/Δ mice had normalized by survival endpoint (Fig. 6E and F), suggesting that PDAC tumors in KPCNix^Δ/Δ mice had acquired the ability to overcome deficits in the NIX mitophagy program. This observation was recapitulated in organoids derived from KPC and KPCNix^Δ/Δ mice grown in low glucose medium, wherein mitochondrial content in the groups was not significantly different—confirming a selective pressure to maintain mitochondrial content suppression in KRAS-mutant cells (Supplementary Fig. S8B and S8C). We hypothesized that the normalization of mitochondrial content by survival endpoint might reflect compensatory upregulation of autophagy or mitophagy pathways. Indeed, in two of three organoid lines derived from KPCNix^Δ/Δ mice at survival endpoint, we found elevated levels of BNIP3, a hypoxia-inducible mitophagy adaptor protein (Fig. 6G). The third KPCNix^Δ/Δ organoid line expressed lower levels of BNIP3; however, upon addition of chloroquine, this line demonstrated increased accumulation of LC3-II, indicating increased autophagic flux (Supplementary Fig. S6G). In addition, when Suit2 and FA6 cells were grown in low glucose media in hypoxic (1% O₂) culture conditions, knockdown of NIX led to induction of BNIP3 and loss of the proliferation defect seen in normoxic low glucose conditions (Supplementary Fig. S8D and S8E). Similarly, in a panel of organoids derived from 16-week-old KPC and KPCNix^Δ/Δ mice, hypoxic conditions led to an induction of BNIP3 in NIX-null organoids, possibly as a compensatory mechanism for NIX-null cells to adapt to hypoxic conditions like that characteristic of the PDAC microenvironment (Supplementary Fig. S8F). Staining of tumor sections revealed that a greater fraction of KPCNix^Δ/Δ tumors had high positivity of BNIP3 staining compared with KPC controls (Fig. 6H and I), supporting our in vitro findings that NIX ablation was accompanied by upregulation of BNIP3. Collectively, our results demonstrate that NIX promotes the progression of KRAS^G12D-driven PanIN to PDAC and nominate the NIX pathway as a new dependency in PDAC.

General autophagy has previously been shown to support a glycolytic metabolic shift during RAS-driven cellular transformation (38). More recently, selective mitophagy has been implicated as a driver of glycolysis in the setting of normal physiologic development (39). Here, we present evidence that KRAS-mutant cancer cells can channel their glucose metabolism away from the mitochondria by a previously unappreciated means: programmed mitophagy via the mediator BNIP3L/NIX. We find that oncogenic KRAS induces NIX expression and suppression of mitochondrial content in a MAPK pathway–dependent fashion. Loss of NIX led to a shift toward oxidative glucose metabolism, with a compensatory increase in flux through the oxidative arm of the PPP and reductive carboxylation of glutamine—pathways that regenerate NADPH and can facilitate its shuttling into mitochondria. Because we did not analyze mitochondrial NADPH levels directly, more work will be required to confirm the importance of redox shuttling for mitochondrial redox homeostasis in NIX-depleted cells. In vivo, NIX ablation led to an increase in mitochondrial content within PanIN lesions and a delay in progression of these premalignant lesions to higher-grade PanIN and to bona fide PDAC.

The role of selective autophagy has been less well studied than that of general autophagy in PDAC. Studies of the Pink/Parkin pathway and BNIP3 suggest that these pathways can have tumor-suppressive functions as well as mitophagy-independent activities in various cancer types (18, 40). In the KC mouse model, for example, conditional deletion of either PINK1 or PARK2 was shown to lead to shorter survival and increased invasiveness of tumor cells, whereas high mRNA expression of PRKN/PARK2 correlates with improved survival in human patients (40). Cancer cells from KC mice lacking PINK1 or PARK2 demonstrate increased mitochondrial iron accumulation and increased lactate production in a HIF1α-dependent manner both in vivo and in vitro. Similarly, deletion of BNIP3 in breast cancer cells leads to upregulation of glycolysis and promotes tumor progression (41). Studies examining the role of NIX in cancer have also suggested a putative role as a tumor suppressor in certain contexts (18, 42–44). Instead, we find that in KRAS-mutant pancreatic cancer cells, NIX serves to suppress mitochondrial ROS production by decreasing the mitochondrial network, augmenting redox robustness, and allowing for increased glycolytic metabolism of glucose to lactate—similar to the metabolic switch observed in red blood cell maturation (39). These activities promote development of PDAC from PanIN lesions. In the KC and KPC mouse models, the role of NIX is protumorigenic, as NIX ablation significantly delays tumor progression and extends survival. In addition, elevated expression of NIX mRNA within a cohort of patients with PDAC from the TCGA correlates with significantly worse survival. Together, our results demonstrate how NIX-mediated programmed mitophagy may play a distinct and possibly opposing role to the traditional damage-control selective autophagy response (i.e., through PINK1/PARK2 or BNIP3) to mediate metabolic transformation and promote pancreatic cancer.

Intriguingly, we observed that the mitochondrial content of tumors in NIX knockout mice at clinical endpoint was comparable to that within KPC controls, as was also seen in the organoids derived from these tumors after culture in vitro. The observed normalization of mitochondrial content in vivo and in vitro paralleled a compensatory upregulation of BNIP3 or general autophagy flux in organoids derived from NIX-deficient tumors, presumably to maintain a preferred reduced mitochondrial set point. In addition to this apparent compensation, it is also possible that some of the observed equalization of mitochondrial content may represent expansion of the mitochondrial network during disease progression within the KPC (NIX WT) tumors themselves. Increased mitochondrial biogenesis, for example, driven by MYC and other oncogenic transcription factors, has been shown to be engaged to support increased metabolic demands during tumor development (45) and might act to counter mitochondrial suppression as KPC tumors progress.

We found that loss of Nix restores functional mitochondria to pancreatic cancer cells, leading to increased respiratory capacity in these cells and a reduced ability of resultant PanIN lesions to progress to cancer in vivo. Our findings are in line with a recent analysis of an siRNA screen that found increased expression of general autophagy pathways and NIX among KRAS-dependent human cancer cell lines in comparison with their KRAS-independent counterparts (46). In this analysis, the KRAS-dependent cell lines tended to be more glycolysis-dependent, supporting the notion that NIX may be important for driving KRAS-mediated metabolic transformation (46). Decades ago, a debate arose questioning whether cancer cells that metabolize glucose through aerobic glycolysis exhibit “impaired respiration” and whether this metabolic shift is an initiating event in tumorigenesis (47–49). More recently, several studies have demonstrated that KRAS-mutant cells do indeed require mitochondrial function to metabolize glucose, glutamine, and lactate, and to support tumorigenesis (3, 7, 50, 51). Despite our observation that conditional deletion of NIX initially delays PDAC development, we find that, consistent with those previous studies, conditional NIX knockout mice still eventually succumb to malignant disease. In addition, our findings suggest that therapeutic targeting of mitophagy in combination with ROS-generating compounds may lead to new possibilities for the treatment of PDAC and other KRAS-mutant cancers.

A fragment of the murine Kras genomic locus harboring Exon1 was cloned in pBluescript. Next, a Gly (GGT) to Val (GTT) substitution was introduced in codon 12. Finally, a FRT-LoxP-STOP-LoxP cassette was cloned upstream the Exon1^G12V, and an additional FRT site was cloned downstream of Exon1^G12V. TL1 ES cells (52) were electroporated with the linearized pBluescript-FRT-LSL-Kras^G12V-FRT–targeting construct, and correctly targeted puromycin-resistant clones were identified by Southern blot. Two positive clones exhibiting a normal karyotype were used to generate chimeric mice by microinjection into C57BL/6 blastocysts. Germline transmission of the targeted allele was confirmed by Southern blot analysis of tail DNA from the agouti offspring.

All animal experiments were reviewed and approved by the IACUC of Cold Spring Harbor Laboratory. The LSL-Kras^G12D; LSL-Trp53^R172H; Pdx1-Cre and Nix^FL strains have been described previously (13, 37, 53). Animals were maintained on a C57BL/6J background back-crossed at least 10 generations. The number of animals used in each experiment is stated in the figure legends.

Resected pancreatic cancer specimens were obtained as excess tissues from Johns Hopkins University (Baltimore, MD) following an Institutional Review Board–approved protocol. Tissue was fixed in formalin and processed for IHC as described below.

Tissues were fixed in 10% neutral buffered formalin for 24 hours before paraffin-embedding and sectioning. Slides cut from paraffin blocks were deparaffinized and rehydrated. Antigen retrieval was performed in 10 mmol/L citrate buffer (pH 6) for 6 minutes in a pressure cooker. For IHC staining, endogenous peroxidase activity was quenched by incubation in 3% hydrogen peroxide for 15 minutes followed by rinsing in water and blocking in 2.5% normal horse serum. Slides were incubated with primary antibody diluted in blocking solution overnight at 4°C. Secondary antibody (Vector ImmPRESS) was used according to manufacturer's instructions followed by development using DAB (Vector Laboratories). Slides were dehydrated and mounted with coverslips and imaged using a Zeiss microscope. For immunofluorescence staining, following antigen retrieval, sections were blocked in 2.5% normal horse serum before incubation in primary antibody diluted in 5% BSA in TBST overnight at 4°C. Secondary antibody and DAPI were diluted in 1% BSA in PBS and added to sections for 1 hour in the dark before mounting with coverslips. The following antibodies were used: NIX (#12396, Cell Signaling Technology), p62 (Enzo), pERK1/2 (#4370, Cell Signaling Technology), Tom20 (#42406, Cell Signaling Technology), Ki-67 (RM-9106, Thermo Fisher Scientific), and BNIP3 (Sigma).

Blocks were serially sectioned for 80 sections and hematoxylin and eosin (H&E) staining was performed on every fifth and sixth section. H&E sections were evaluated for highest grade lesion present in a blinded fashion. H&E sections of pancreata from 9- to 12-month-old KC mice were scanned using an Aperio scanner, and image analysis was performed using AperioImageScope v11.2.0.780. To determine the area of normal pancreas, the Positive Pixel Count v9 macro was used with the following parameters: Hue value (0.006), Hue width (0.5), Color Saturation Threshold (0.27), Upper pixel threshold value (220), Lower pixel threshold value (0). To find the fraction of normal pancreas, the total number of positive pixels was divided by the total number of pixels (positivity output).

PanIN lesions were imaged on an LSM 710 Confocal Microscope (Zeiss) at 63× magnification. Tumor sections were imaged on a SP8 confocal microscope (Leica) at 40× magnification. ImageJ v.1.51n was used for all image analyses. To find the total mitochondria area, images were first processed by applying a Gaussian blur (sigma = 0.5). To separate the true mitochondrial staining from background, the images were then made binary by using the same pixel intensity threshold for all images within an experiment. To calculate the total area of all mitochondria, the “Analyze particles” function was used to count all objects. Nuclei were counted using the “Analyze particles” function and objects of size 50 to infinity were counted. Manual segmentation of adjacent nuclei was done as needed.

MEFs were generated and maintained as described previously (12). MEFs were used at early passage for every experiment but were not tested for Mycoplasma at any point. Human PDAC FA6 (CVCL_4034) and Suit2 (CVCL_3172) cells were originally obtained from Clare Hall Laboratories (CRUK). Human cell lines were periodically tested for Mycoplasma and confirmed as Mycoplasma-negative using the MycoAlert Mycoplasma Detection Kit (Lonza LT07-318), most recently after resubmission experiments. Cell line authentication was not performed. Two-dimensional FPC cancer cell lines were generated from FPC mice as described previously (54). Mouse organoids and 2-D mouse cancer cell lines were tested for Mycoplasma at least once after isolation or thawing and were retested prior to performance of metabolomics and flow cytometry experiments using the MycoAlert Mycoplasma Detection Kit (Lonza LT07-318). All experiments were performed on organoids/cell lines cultured for fewer than 15 total passages.

MEFs, human PDAC cell lines (Suit2, FA6), and mouse cancer cell lines were maintained in DMEM (Gibco, 21969) containing 25 mmol/L glucose and supplemented with 2 mmol/L glutamine, 1% penicillin/streptomycin, and 10% FBS. When indicated, 2-D experiments were performed in low glucose medium (glucose-free, pyruvate-free, glutamine-free DMEM (Gibco A1443001) supplemented with 10% FBS, 0.5 mmol/L glucose, 2 mmol/L glutamine, and 1% penicillin/streptomycin). MEK inhibitor experiments used AZD6244 at a concentration of 5 μmol/L for 24 hours. Galactose proliferation experiments were performed in galactose media (glucose-free, pyruvate-free, glutamine-free DMEM (Gibco A1443001) supplemented with 10% FBS, 5 g/L galactose (Sigma), 2 mmol/L glutamine, and 1% penicillin/streptomycin). For induction of mitochondria-targeted ROS, mitoPQ (Abcam) or carrier (control) was added to media at 5 μmol/L for 20 minutes.

Organoids were generated and cultured as described previously (14). Briefly, organoids grown in Matrigel Growth Factor Reduced (Corning) were removed from Matrigel by incubating with 2 mg/mL dispase at 37°C. Isolated organoids were dissociated using TrypLE Express (Gibco) and the resultant single cells were counted and plated for experiments as described elsewhere. When indicated, organoid experiments were performed in low-glucose medium: glucose-free, pyruvate-free Advanced DMEM/F-12 supplemented with 2 mmol/L glucose and growth factors as described previously with the exception of R-spondin–conditioned medium, epidermal growth factor, and N-acetyl-cysteine, which were excluded (14). For autophagy flux experiments in organoids, chloroquine (Sigma, 25 μmol/L) or vehicle control was added to media for 24 hours.

Retroviral and adenoviral infections were performed on MEFs and organoids as described previously (12, 14). Briefly, dissociated organoids were resuspended in Advanced DMEM/F-12 (plus 1% penicillin/streptomycin, 2 mmol/L GlutaMAX, and 1 mmol/L HEPES pH 7.2–7.5; “+++ medium”) containing retrovirus or adenovirus (500 pfu/cell) and centrifuged for 1–2 hours at 800 × g. Antibiotic selection of retrovirally infected organoids (puromycin 2 μg/mL) was performed for at least 48 hours.

shRNAs were chosen from a published library and cloned into the LEPG and RT3GEPIR vectors as described previously (55). Cells infected with hairpins cloned into the tet-inducible RT3GEPIR vector were treated with doxycycline (1 μmol/L) for 72 hours prior to experimental analysis unless otherwise stated. The following shRNA target sequences were used:

• Renilla Luciferase 713: TGCTGTTGACAGTGAGCGCAGGAATTATAATGCTTATCTA

• TAGTGAAGCCACAGATGTATAGATAAGCATTATAATTCCTATGCCTACTGCCTCGGA

• Mouse NIX 447: TGC TGT TGA CAG TGA GCG ATC AGA AGA AGA AGT TGT AGA ATA GTG AAG CCA CAG ATG TAT TCT ACA ACT TCT TCT TCT GAC TGC CTA CTG CCT CGG A

ON-TARGETplus SMARTpool siRNA (Horizon Discovery/Dharmacon/Thermo Fisher Scientific) was used for all siRNA knockdown experiments. Cells were transfected with the selected ON-TARGETplus SMARTpool siRNA constructs (25 nmol/L final concentration), DharmaFECT 1 reagent (Thermo Fisher Scientific), and Opti-MEM (Life Technologies) according to manufacturer's recommendations.

The following Dharmafect siRNA targets were used:

• Nontargeting control (catalog no. D-001810-10-05)

• Human KRAS (catalog no. L-005069-00-0005)

• Human NIX/BNIP3L (catalog no. L-005069-00-0005)

• Human MAP2K2/MEK2 (catalog no. L-003573-00-0005)

• Human p62/SQSTM1 (catalog no. L-010230-00-0005)

• Mouse NIX/BNIP3L (catalog no. L-058953-00-0005)

MEFs were infected with adenoviral control or CRE at least 72 hours prior to transfection with siRNA. At the start of the experiment, both MEFs and human cells were seeded into three wells each on duplicate 6-well plates per condition (1 × 10⁵ cells per well for MEFs and 7.5 × 10⁴ cells per well in human cells). The following day, cells were transfected with indicated siRNA constructs and left in 25 mmol/L glucose DMEM overnight. At this point, one plate of the triplicate wells was counted to determine day 0 counts. On the second plate, media were either replenished with 25 mmol/L glucose DMEM or switched to low-glucose DMEM. Media was changed for a second time on day 2. These plates were counted either 3 or 4 days later as indicated, and the resulting ratios to the day 0 counts compared.

For the hypoxia proliferation comparison in Supplementary Fig. S8, we utilized FA6 cells that stably express pBABE-iRFP-puro, generated as described previously for HCT116 cells (56). Otherwise, the cells were treated the same as in other experiments, shifted to ± 1% oxygen conditions at 24 hours post-transfection (when they were also shifted to low-glucose media). Instead of counts, the day 4 iRFP intensity values were compared across conditions after confirming that the day 0 initial scans of each well all had similar values.

Dissociated organoids were plated as 10,000 single cells in 2.5 μL domes of Matrigel (Corning) in XF96 assay plates (Agilent Technologies) and cultured in low-glucose medium for 48 hours prior to assay. Domes were overlaid with 180 μL of low-glucose medium. Adherent cells were first transfected with indicated siRNA constructs and cultured for 48 hours before being seeded at 2 × 10⁴ cells/well in XF96 plates and cultured for an additional two days in low-glucose medium. Mito Stress Test assays were performed according to manufacturer's instructions using oligomycin (1 μmol/L), FCCP (1 μmol/L), and rotenone/antimycin A (0.5 μmol/L/0.5 μmol/L) with 3 to 4 measurement cycles before and after each injection.

Human cells were transfected with indicated siRNA constructs and MEFs infected with adenoviral control or CRE, and cells were cultured for an initial 48 hours postinfection/transfection in 25 mmol/L glucose DMEM medium. MEFs and human 2-D cells were then cultured in low-glucose (0.5 mmol/L) DMEM for 48 hours prior to fixation (96 hours postinfection/transfection). Cells were washed in ice-cold saline (0.9% NaCl) and fixed in an ice-cold solution of 2% glutaraldehyde and 2% formaldehyde in 0.5 mol/L sodium cacodylate buffer (pH 7.4) for 4 hours at 4°C. The cells were then washed in 0.1 mol/L sodium cacodylate buffer five times, treated with 1% osmium ferricyanide at 20°C for 2 hours, washed five times in deionized water, and treated with 2% uranyl acetate in 0.05 mol/L maleate buffer at pH 5.5 for 2 hours at 20°C before embedding in Quetol epoxy resin and sectioning. Images were taken on an FEI Tecnai G2 transmission electron microscope operated at 120 kV using an AMT XR60B digital camera running Deben software.

TEM was performed on organoids as described previously (54). Briefly, organoids grown in Matrigel domes were fixed directly in 24-well culture plates overnight at 4°C using 2% glutaraldehyde/2% paraformaldehyde in 1× PBS. Following fixation, domes of Matrigel containing organoids were lifted from the plate and post-fixed in 1% osmium containing 1.5% potassium ferrocyanide before embedding in resin and sectioning. Sections were imaged using a FEI BioTwinG2 transmission electron microscope. At least 20 individual cells were imaged for each genotype.

To determine the mitochondria volume fraction of MEFs, human cancer cell lines, and organoids, at least 20 randomly obtained images per sample were taken at a direct magnification of 6,500× (2-D cell lines) or 11,000× or higher (organoids). For 2-D cell lines, a 12 × 10 grid was placed over each image and for organoids, a 16 × 16 grid was placed over each image and scaled using the STEPanizer stereology tool (https://www.ncbi.nlm.nih.gov/pubmed/21375529; ref. 57). The mitochondrial fraction was determined by computing the ratio of grid intersections that fell within mitochondria to the sum of all grid intersections that fell within a cell (mitochondrial + nonmitochondrial cellular vertices).

Cristae per mitochondrion were quantified manually from a minimum of 70 mitochondria in total per condition, taken from the TEM images of three independent experiments obtained as described above.

One hundred thousand single cells of dissociated mT organoids were plated in 300 μL of Matrigel in prewarmed 6-cm plates overlaid with 4 mL of low-glucose medium containing 1 μmol/L doxycycline and 2 μg/mL of puromycin and cultured for 72 hours. Medium was replaced with ¹³C₆-glucose (Cambridge Isotopes, used at 2 mmol/L) or ¹³C_5,¹⁵N₂-glutamine media (Cambridge Isotopes, used at 2 mmol/L) containing medium and organoids were collected at two time points: 0 minutes, 45 minutes. Organoids were collected by quickly removing medium, washing cells in 1× PBS, and scraping cells with Matrigel on ice into 1.2 mL of cold 30% acetonitrile: 50% methanol. Samples were rocked for 5 minutes at 4°C before being centrifuged at 15,000 × g for 10 minutes at 4°C. The supernatants were transferred to new tubes and stored at −80°C before being used for LC/MS analysis. Resulting pellets were lysed in 500 μL of RIPA buffer and protein concentration was measured using the DC Protein Assay (Bio-Rad). Domes of Matrigel (300 μL) only were plated in 6-cm plates in triplicate, overlaid with medium, and collected at the two timepoints as a control.

For 2-D cells, LC/MS sample preparation was performed broadly as described previously (58, 59). Briefly, cells were incubated in assay medium containing low-glucose DMEM supplemented with 2 mmol/L uniformly labeled ¹³C₆-glucose, 1-¹³C₁-glucose, or ¹³C₅, ¹⁵N₂-glutamine (all from Sigma) in lieu of the unlabeled glucose or glutamine (for ¹³C_5,¹⁵N₂-glutamine experiments) for 45 minutes (¹³C₆-glucose, ³C_5,¹⁵N₂-glutamine) or 3 hours (1-¹³C₁-glucose). Metabolites were extracted by rapidly removing cell medium, washing wells once with ice-cold PBS, and lysing cells in ice-cold methanol/acetonitrile/H₂O (50:30:20) at volumes scaled on the basis of cell counts of the counting plate to 2 × 10⁶ cells per mL extraction buffer. Sample plates were shaken at 4°C for 10 minutes before the extraction buffer was collected from each well, spun for 15 minutes at 16,000 × g in a chilled (4°C) centrifuge, and then analyzed by LC/MS. For analysis of NADPH of 2-D cells, cells were incubated in fresh medium containing or lacking 20 mmol/L menadione for 3 hours prior to metabolite extraction.

For all LC/MS samples, metabolite analysis was performed as described previously (60). Briefly, a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific) was used together with a Thermo Ultimate 3000 HPLC system to analyze prepared samples. The HPLC setup consisted of a ZIC-pHILIC column (SeQuant, 150 × 2.1 mm, 5 μm, Merck KGaA), with a ZIC-pHILIC guard column (SeQuant, 20 × 2.1 mm) and an initial mobile phase of 20% 20 mmol/L ammonium carbonate, pH 9.4, and 80% acetonitrile. Cell and medium extracts (5 μL) were injected and metabolites were separated over a 15-minute mobile phase gradient, decreasing the acetonitrile content to 20%, at a flow rate of 200 μL/minute and a column temperature of 45°C. The total analysis time was 23 minutes. All metabolites were detected across a mass range of 75 to 1,000 m/z using the Q Exactive mass spectrometer at a resolution of 35,000 (at 200 m/z), with electrospray ionization and polarity switching to enable both positive and negative ions to be determined in the same run. Lock masses were used, and the mass accuracy obtained for all metabolites was below 5 ppm. Data were acquired with Thermo Xcalibur software. The peak areas of different metabolites were determined using Thermo TraceFinder 4.0 software where metabolites were identified by the exact mass of the singly charged ion and by known retention time on the HPLC column. Commercial standards of all metabolites detected had been analyzed previously on this LC/MS system with the pHILIC column.

For gas chromatography/mass spectrometry (GC/MS) analysis of steady-state TCA cycle pools, cells were cultured in low-glucose medium for 48 hours. Cells were then washed with ice-cold PBS, detached with a cell scraper on ice, and pelleted by centrifugation. Metabolites were extracted using methanol-chloroform (2:1) and metabolic profiling was performed as described previously (61, 62). Briefly, the aqueous metabolites were derivatized in a solution of 20 mg/mL methoxyamine hydrochloride in pyridine for 17 hours and then silylated with N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA, Sigma) for 1 hour as described previously (63). The derivatized samples were diluted in hexane before being injected into a Thermo Scientific TRACE Ultra Gas Chromatograph (Thermo Fisher Scientific; injector temperature: 220°C, helium carrier gas flow rate: 1.2 mL/minute). The column effluent was introduced into a TRACE DSQ Quadrupole Mass Spectrometer (Thermo Fisher Scientific; transfer line temperature: 250°C, ion source temperature: 220°C, electron beam: 70 eV). The detector was turned on after a solvent delay of 120 seconds, and data were collected in full scan mode using 3 scans/second across a mass range of 50 to 650 m/z. GC/MS chromatograms were analyzed using Thermo Xcalibur software (Thermo Fisher Scientific). Metabolites were identified using the National Institute of Standards and Technology (NIST) database of mass spectra. Peaks were integrated individually in Xcalibur and then peak intensity data were further analyzed using Microsoft Excel.

Organoids were dissociated into single cells and resuspended in phenol red-free +++ medium. Adherent 2-D cells were labeled for 30 minutes in serum-free and phenol red–free DMEM. The following dyes were used: MitoTracker Green (40 nmol/L), MitoSOX Red (5 μmol/L), DiIC₁(5) (25 nmol/L), or DCF-DA (10 μmol/L; Life Technologies). Cells were incubated with dyes in a 37°C, 5% CO₂ incubator for 30 minutes before washing with 1× PBS and resuspension in phenol red–free +++ medium (dissociated organoids) or 1× PBS+ 2% FBS (2-D cells). 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI; Sigma Aldrich) was added to a final concentration of 1 μg/mL to each sample and was used to identify viable cells for analysis. Stained cells were analyzed on a BD LSR-II or BD Fortessa flow cytometer using unstained cells and single color–stained cells as controls. At least 10,000 events were collected for each sample and results were analyzed using FlowJo Software. Unless otherwise stated, median fluorescence intensity values relative to control cells are reported for each probe.

Suit2 and FA6 cells were retrovirally infected with pBABE.mCherry-GFP-FIS1_101–152 (mitoQC, obtained from Dr. Stephen Tait's laboratory at CRUK) and selected with hygromycin (500 μg/mL; ref. 21). Single-cell suspensions of stably infected human cancer cells treated with siRNA (NT, NIX, or p62) were analyzed by flow cytometry on a BD Fortessa flow cytometer based on the protocol developed for the tandem LC3-GFP-mCherry autophagy reporter (22). The median-derived mCherry/GFP ratio parameter was compared between samples.

RNA was isolated from cultured cells and organoids using TRIzol followed by Pure Link RNA isolation kit with DNase treatment according to the manufacturer's instructions. Reverse transcription was performed on 1 μg of total RNA using TaqMan Multiscribe Reverse Transcription reagents (Applied Biosystems). Quantitative real-time PCR was performed on a QuantStudio 6 Flex Instrument (Applied Biosystems) using TaqMan probes (listed below). Relative gene expression was calculated using the ΔΔC_t method with either Actin/ACTIN (MEFs, KPC tumor tissue, and human PDAC) or Hprt (organoids and FPC cells) used as an internal control.

• Hs99999903_m1 Human ACTIN

• Mm00607939_s1 Mouse Actin

• Hs00364284_g1 Human KRAS

• Hs01087963_m1 Human NIX/BNIP3L

• Mm00786306_s1 Mouse Bnip3l

• Mm00448091_m1 Mouse p62/Sqstm1

• Mm00446968_m1 Mouse Hprt

• Mm01135606_m1 Mouse Nrf1

• Mm00489774_m1 Mouse Acsm3

• Mm01208835_m1 Mouse PGC-1alpha

• Mm00450187_m1 Mouse Park2

• Mm00550827_m1 Mouse Pink1

• Mm00448091_m1 Mouse p62/Sqstm1

• Mm00517492_m1 Mouse Kras

• Mm00447485_m1 Mouse Tfam

Total DNA was extracted from cells using a DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturer's recommendations with modifications as described previously (17). For DNA extraction from pancreas tissue, the Qiagen user-developed protocol “Purification of total DNA from soft tissues using the TissueLyser and the DNeasy Blood & Tissue Kit (DY11 Aug-06)” was used. Total DNA was then quantified, diluted to 3 ng/μL, and used to determine the relative copy number of mitochondrial DNA (COX2) and nuclear DNA [Amyloid precursor protein (APP)] by quantitative PCR using a 7900HT real-time PCR machine (Life Technologies) and the primer/probe combinations listed below:

• Mouse APP forward: CGGAAACGACGCTCTCATG

• Mouse APP reverse: CCAGGCTGAATTCCCCAT

• Mouse APP probe: TCGCTGACGGAAACCAAGACCC (FAM and TAMRA)

• Mouse COX2 forward: GAGCAGTCCCCTCCCTAGGA

• Mouse COX2 reverse: GGTTTGATGTTACTGTTGCTTGATTT

• Mouse COX2 probe: AAACTGATGCCATCCCAGGCCGA (FAM and TAMRA)

• Human APP forward: TTTTTGTGTGCTCTCCCAGGTCT

• Human APP reverse: TGGTCACTGGTTGGTTGGC

• Human APP probe: CCCTGAACTGCAGATCACCAATGTGGTAG (FAM and TAMRA)

• Human COX2 forward: CGTCTGAACTATCCTGCCCG

• Human COX2 reverse: TGGTAAGGGAGGGATCGTTG

• Human COX2 probe: CGCCCTCCCATCCCTACGCATC (FAM and TAMRA)

For mT8 organoids and 2-D cell lines, the NADPH/NADP⁺ ratio was determined from the NADPH and NADP⁺ pools measured by LC/MS during metabolomic analysis. For mT5 organoids, the NADPH/NADP⁺ ratio was measured by first isolating organoids from Matrigel using Cell Recovery Solution (Corning) and washing with ice-cold 1× PBS. Pelleted organoids were evaluated using the NADP/NADPH Assay Kit (Abcam ab65349) according to the manufacturer's instructions.

The isolation of mitochondrial-enriched fractions from mouse tissue and cells was performed as described previously (64, 65) with modifications as detailed below. For pancreas tissue, sections (approximately 60 mg) were collected in ice-cold PBS and finely diced. Cells were incubated in 1 mg/mL collagenase V (Sigma) at 37°C for 30 minutes with mixing, pelleted by centrifugation (200 × g), resuspended in 0.25% Trypsin/EDTA (Life Technologies). Cells were then incubated at 37°C for 10 minutes with mixing followed by washing with DMEM + 10% FBS. Cells were pelleted by centrifugation, washed in ice-cold PBS, and resuspended in ice-cold isotonic HIM buffer (200 mmol/L mannitol, 70 mmol/L sucrose, 1 mmol/L EGTA, 10 mmol/L HEPES, pH 7.5) containing phosphatase inhibitors and protease inhibitors (Roche). Cultured cells were detached from cell culture dishes using 0.25% Trypsin/EDTA (Life Technologies) and pelleted.

Pelleted single cells were then homogenized in an Isobiotec cell homogenizer using a steel bead to allow 18 μm of clearance through the homogenizer aperture (Isobiotec). Unbroken cells and nuclei were removed as pellets by consecutive centrifugation (at 350 × g and 800 × g) 4°C. The supernatant was centrifuged at 9,000 × g for 10 minutes at 4°C to obtain cytoplasmic (supernatant) and mitochondrial-enriched (pellet) fractions. The mitochondrial pellet was washed before lysis in RIPA buffer containing phosphatase inhibitors and protease inhibitors (Roche) and the cytoplasmic fraction was clarified by centrifugation. Both fractions were sonicated for 3 minutes at 4°C to shear any contaminating genomic DNA.

Standard procedures were followed for Western blot analysis. MEFs and human PDAC cells were prepared and analyzed as described previously (60) using LI-COR secondary reagents and a LI-COR Odyssey Infrared Scanner. Organoids and FPC cells were collected and lysed in 0.1% Triton X-100, 15 mmol/L NaCl, 0.5 mmol/L EDTA, 5 mmol/L Tris, pH 7.5 with the addition of Mini-Complete protease inhibitors (Roche) and a PhosSTOP phosphatase inhibitor cocktail (Roche). Protein lysates were clarified and separated using 4% to 12% Bis-Tris NuPAGE gels (Life Technologies) and then transferred onto polyvinylidene difluoride membranes (Millipore). Membranes were blocked with 5% nonfat dry milk or 5% BSA in TBST (0.1% Tween 20 in tris-buffered saline), and incubated with primary antibodies overnight rocking at 4°C. Membranes were incubated with horseradish peroxidase (HRP)–conjugated secondary antibodies for 1 hour at room temperature and detected by ECL (GE Healthcare). Primary antibodies used: CoxIV, Nix, p62, Vinculin, LC3-B, HIF1α, BNIP3 (rodent specific) and BNIP3 (for human samples), HSP60, p62, HSP90 (used for human PDAC and MEF Western blots), phospho and total p44/p42 (all from Cell Signaling Technology); CV-5/ATP5A (Abcam); Hsp90 (Millipore; for organoid and tissue extract Western blots). IRDye 800CW and 680LT raised in Donkey (anti-rabbit and anti-mouse) secondary antibodies were used for MEF and Human PDAC Western blots (LI-COR) and HRP-conjugated secondary antibodies were used for protein detection in organoids and mouse cancer cell Western blots (Jackson Immunoresearch).

We obtained TCGA expression data by accessing the publicly available harmonized cancer datasets hosted on the National Cancer Institute GDC data portal (available at: https://portal.gdc.cancer.gov/). Patients whose tumors were determined to be non-PDAC by histopathologic analysis were excluded (2). Patients were divided into high (upper 50th percentile) and low (lower 50th percentile) BNIP3L expression and Kaplan–Meier survival analysis was performed.

All data were plotted using Prism 7 (GraphPad). Statistical analysis for each experiment was performed using the tools within Prism 7, the indicated tests, and multiplicity-adjusted P values. Figures were prepared using Illustrator (Adobe). All data represented as means and SE (error bars) unless otherwise indicated. Asterisks denote P value as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

All data are presented in the main text and Supplementary Materials.

K.H. Vousden is a chief scientist at CRUK; a member of the scientific advisory board of The Ludwig Institute; a member of the board of directors of BMS; and an honorary chair at the Institute of Cancer Research, Queen Mary University, and Imperial College; reports receiving commercial research grants from AstraZeneca and Astex; has received speakers bureau honoraria from Nestle, Novartis, Genentech, GlaxoSmithKline, and many academic talks; has ownership interest (including stock, patents, etc.) in BMS and Grail, Inc.; and is a consultant/advisory board member for PMV Pharma, Raze Therapeutics, CNIO, IRB, MRC Unit Dundee, Abramson Cancer Centre, Infosys Foundation, Frankfurt University Cancer Centre, Oncode Institute, The Gurdon Institute, ISREC, BACR, EMBO Council, Royal Society, and Vallee Scholar Committee. D.A. Tuveson reports receiving commercial research grants from Fibrogen and ONO, has ownership interest (including stock, patents, etc.) in Leap Therapeutics and Surface Oncology, and is a consultant/advisory board member for Leap Oncology and Surface Oncology. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.J. Humpton, B. Alagesan, D. Lu, K.H. Vousden, D.A. Tuveson

Development of methodology: T.J. Humpton, B. Alagesan, D. Lu, J.N. Skepper

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.J. Humpton, B. Alagesan, D. Lu, G.N. Yordanov, C.S. Leonhardt, M.A. Yao, P. Alagesan, M.N. Zaatari, Y. Park, K.F. Macleod, P.A. Perez-Mancera

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.J. Humpton, B. Alagesan, G.M. DeNicola, D. Lu, G.N. Yordanov, M.N. Zaatari, M.P. Murphy

Writing, review, and/or revision of the manuscript: T.J. Humpton, B. Alagesan, G.M. DeNicola, D. Lu, C.S. Leonhardt, K.F. Macleod, P.A. Perez-Mancera, M.P. Murphy, G.I. Evan, K.H. Vousden, D.A. Tuveson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Alagesan, J.N. Skepper

Study supervision: G.I. Evan, K.H. Vousden, D.A. Tuveson

We thank Dr. Gerald Dorn for kindly providing the Nix^FL mouse strain. We thank Dr X. Zou for assistance in generating the FRT-LSL-KrasG12V-FRT mouse. We thank Dr. Stephen Tait for providing the mitoQC construct and Dr. Richard Hartley for providing the mitoPQ control compound. We also acknowledge the Cold Spring Harbor Laboratory Animal and Genetic Engineering, Animal and Tissue Imaging, Microscopy, and Flow Cytometry core facilities, which are funded by the NIH Cancer Center Support Grant 5P30CA045508. We thank Susan Van Horn and the Stony Brook Central Microscopy Imaging Center for assistance with electron microscopy of organoids. We also acknowledge the Metabolomics facility and Cancer Metabolism Research Unit at the CRUK Beatson Institute and thank G. Mackay and D. Sumpton for their assistance and advice. We thank Dr. L. Baker for her critical reading of the manuscript. We thank members of the D. Tuveson Lab, N. Sodir and other members of the G. Evan lab, and C. Labuschagne, A. Hock, M. Yang, and other members of the K. Vousden lab for their assistance and advice. D.A. Tuveson is a distinguished scholar of the Lustgarten Foundation and Director of the Lustgarten Foundation–designated Laboratory of Pancreatic Cancer Research. D.A. Tuveson is also supported by the Cold Spring Harbor Laboratory Association, the V Foundation, the Cold Spring Harbor Laboratory and Northwell Health Affiliation (Project Lazarus grant), and the NIH (NIH 5P30CA45508, 5P50CA101955, P20CA192996, 1U10CA180944, U01CA224013, U01CA210240-01A1, 1R01CA188134, and 1R01CA190092). B. Alagesan is supported by NCI 5F30CA200240. Y. Park is supported by R50CA211506. G.N. Yordanov is supported by a Boehringer Ingelheim Fonds Fellowship. K.F. Macleod is supported by NIH/NCI RO1 CA200310 and RO1 CA216242. M.P. Murphy is supported by Medical Research Council UK (MC_UU_00015/3) and by a Wellcome Trust Investigator award (110159/Z/15/Z). T.J. Humpton and K.H. Vousden are supported by Cancer Research UK and by ERC grant 322842-METABOp53. T.J. Humpton was also supported by the Gates Cambridge Trust. G.I. Evan and D. Lu are supported by Cancer Research UK Programme Grant A12077 (principal investigator: G.I. Evan): “Deconstructing Myc oncogenesis.”