Regulation of Tumor Initiation by the Mitochondrial Pyruvate Carrier

Summary

While metabolic adaptations have been demonstrated to be essential for tumor cell proliferation, the metabolic underpinnings of tumor initiation are poorly understood. We found that the earliest stages of colorectal cancer (CRC) initiation are marked by a glycolytic metabolic signature, including down-regulation of the mitochondrial pyruvate carrier (MPC), which couples glycolysis and glucose oxidation through mitochondrial pyruvate import. Genetic studies in Drosophila suggest that this downregulation is required since hyperplasia caused by loss of the Apc or Notch tumor suppressors in intestinal stem cells can be completely blocked by MPC overexpression. Moreover, in two distinct CRC mouse models, loss of Mpc1 prior to a tumorigenic stimulus doubled the frequency of adenoma formation and produced higher grade tumors. MPC loss was associated with a glycolytic metabolic phenotype and increased expression of stem cell markers. These data suggest that changes in cellular pyruvate metabolism are necessary and sufficient to promote cancer initiation.

Graphical Abstract

eTOC Blurb

A hallmark of cancer is altered metabolism in tumor cells, however, it is unclear whether cancer initiation also requires metabolic changes. Bensard, et al. demonstrate that constitutive enforcement of a glycolytic program through inactivation of the Mitochondrial Pyruvate Carrier (MPC), is necessary and sufficient to drive intestinal tumor formation.

Introduction

It is now nearly universally accepted that cancers exhibit a metabolic state that enables the tumor to build the biomass necessary for continuous proliferation, particularly in a microenvironment with limited oxygen and metabolites due to poor vascularization (Pavlova and Thompson, 2016; Vander Heiden and DeBerardinis, 2017). Indeed, it has been typically assumed that these metabolic phenotypes are adaptations to the unique requirements of the tumor microenvironment (Olson et al., 2016; Samanta and Semenza, 2018; Xie and Simon, 2017). Moreover, the atypical metabolic program of cancer cells is essential for tumor establishment and maintenance in animal models and is therefore the target of most metabolic experimental therapeutics (Bader et al., 2019; Hong et al., 2016; Schell et al., 2014; Sullivan et al., 2015; Yang et al., 2014). However, our understanding of cancer metabolism has been largely based on experiments comparing normal tissue to established tumors or cells derived from those tumors. In contrast, we have relatively little understanding of the metabolic underpinnings of tumor initiation.

If these metabolic “adaptations” are selected as a result of the constraints of the environment found in an established tumor, then one might hypothesize that they should only become prevalent or important well after the initial hyperproliferative events in tumor initiation. We set out to test this hypothesis by analyzing whether tumor initiation might also be subject to specific metabolic requirements and/or drivers using colon cancer as a model. Sporadic colon tumors are believed to follow a stereotypical progression wherein the initial tumorigenic event triggers intestinal stem cell hyperplasia, which leads to the formation of a benign adenoma (Sohaily et al., 2012; Vogelstein et al., 2010). This initial event is associated with hyperactivation of the Wnt/β-catenin pathway, typically through loss of function of the Apc tumor suppressor (Nusse and Clevers, 2017). Roles for Apc are conserved through evolution, with Apc mutations in Drosophila leading to increased intestinal stem cell proliferation and intestinal hyperplasia, similar to that seen in mammals (Cordero et al., 2012; Lee et al., 2009). Eventually, additional tumorigenic insults lead to the progression of this lesion into an invasive adenocarcinoma, which results in morbidity and mortality in human patients (Marley and Nan, 2016; Weinberg and Marshall, 2019).

Our studies found that the earliest transition, from normal human intestinal epithelium to a hyperplastic adenoma, is associated with a gene expression pattern that corresponds to an aerobic glycolytic metabolic program with low mitochondrial carbohydrate oxidation. To test the impact of this metabolic program in tumor initiation, we employed genetic manipulation of the Mitochondrial Pyruvate Carrier (MPC) as a means of modulating glucose metabolism in Drosophila and mice. The MPC is an evolutionarily conserved obligate heterodimer of two proteins, MPC1 and MPC2, that resides in the inner mitochondrial membrane and transports pyruvate into the mitochondrial matrix for oxidative metabolism (Bricker et al., 2012; Herzig et al., 2012). Consistent with this, the MPC is frequently deleted or under-expressed in many forms of cancer and re-expression of the MPC in colon cancer cells is sufficient to reduce their growth rate and tumorigenicity in mouse models (Rauckhorst and Taylor, 2016; Schell et al., 2014). We found that overexpression of the MPC is sufficient to fully block proliferation in intestinal stem cells that lack either Apc or Notch tumor suppressors. A similar effect is seen upon disrupting lactate dehydrogenase (LDH) function in oncogenic intestinal stem cells indicating that Apc-mediated tumor formation requires an appropriate cellular metabolic context. We also generated animals wherein intestinal stem cells are deleted for the MPC and, therefore, have a constitutively glycolytic metabolic program (Bricker et al., 2012; Herzig et al., 2012; Schell and Rutter, 2013). In multiple models of de novo tumor initiation, this manipulation was sufficient to significantly increase the frequency of adenoma formation and to result in the formation of higher-grade adenomas. This phenotype was accompanied by increased expression of canonical markers of Wnt/β-catenin signaling, suggesting that this pathway contributes to tumor formation. Our data suggest that a glycolytic metabolic program supports the initial steps of tumorigenesis, likely by driving a pro-proliferative gene expression program, establishing metabolism as a gatekeeper for intestinal tumor initiation.

Results

Modulating pyruvate metabolism predicts and predisposes for colon adenoma formation

In an effort to define the metabolic program that characterizes early colon adenomas, we interrogated two gene expression datasets generated from early stage human adenomas compared to normal colon (Sabates-Bellver et al., 2007; Skrzypczak et al., 2010). This analysis revealed robust differences in gene expression, including alterations in genes that control glucose metabolism (Figure S1A–B, Berg 2019). Lactate Dehydrogenase (LDH), which is required for the conversion of the glycolytic product pyruvate to lactate, was expressed more highly in adenomas (Figure 1A and S1C). This is consistent with the known role of LDH in diverting carbohydrate metabolites away from mitochondrial import and oxidation as part of aerobic glycolysis, which is often observed in established tumors and cells derived therefrom (Romero-Garcia et al., 2016; Sun et al., 2017; Vander Heiden et al., 2009). In contrast, we found that the genes that encode the first step in the oxidation of pyruvate, mitochondrial import, mediated by the Mitochondrial Pyruvate Carrier (MPC - encoded by the Mpc1 and Mpc2 genes) were both decreased in expression in adenomas (Figure 1B and S1D). Human colon adenocarcinomas, from TCGA and the above studies, also exhibit increased expression of LDHA and LDHB and decreased expression of MPC1 and MPC2, as previously observed (Schell et al., 2014) (Figure 1A, 1B, S1E). The fact that this gene expression phenotype is recapitulated in early stage adenomas implies that this transcriptional regulation of pyruvate metabolism might occur earlier than the establishment of an ischemic tumor microenvironment. We also found that the expression of both Mpc1 and Mpc2 are decreased in adenomas derived from two distinct mouse models of colon cancer, the azoxymethane/dextran sodium sulfate (AOM-DSS) model and a genetic model of heterozygous loss of Apc in intestinal stem cells (Figure 1C, 1D). This decreased expression is also observed at the level of protein abundance (Figure 2D, 3D).

Figure 1: Decreased MPC expression is a feature of and predisposes for tumor initiation.

(A-B) Normalized log₂ expression levels of LDHA, LDHB (A), Mpc1 and Mpc2 (B) mRNA relative abundance in human colon normal tissue (green), adenomas (orange), and adenocarcinomas (grey) from dataset GSE20916 (n=24–45 per group), ****FDR≤0.0001;***FDR≤0.001, FDR-corrected Wilcoxon rank-sums test, error bars SD. See also Figure S1A–D, Berg 2019. (C) 8 to 12 week-old transgenic mice were injected intraperitoneally three consecutive days with tamoxifen (TAM;20mg/day), then 30 days later, injected with azoxymethane (AOM; 10mg/mL/kg body weight). One week post-AOM, mice began three cycles of five days on 2% dextran sodium sulfate (DSS) in the drinking water, sixteen days on fresh drinking water. Tumor burden was assessed at 49 and 100 days post-AOM (red arrows). Schematics of the Mpc1^flox and Lrig1-CreERT2 alleles are shown. 8 to 12 week-old transgenic mice were injected intraperitoneally three consecutive days with tamoxifen (TAM; 20mg/day), then examined at 30, 60, or 90 days post-TAM (red arrows). Schematics of the Mpc1^flox, Apc^flox, and Lrig1-CreERT2 alleles are shown. 8 to 12 week-old transgenic mice were injected intraperitoneally three consecutive days with tamoxifen (TAM; 20mg/day), then examined at 60, 90, or 140 days post-TAM (red arrows). Schematics of the Mpc1^flox, Apc^flox, and Villin-CreERT2 alleles are shown. 8 to 12 week-old transgenic mice were injected intraperitoneally three consecutive days with tamoxifen (TAM; 20μg/day), then examined at 60, 90, or 140 days post-TAM (red arrows). (D) Normalized RNA counts for Mpc1 (left) and Mpc2 (right) of isolated adjacent normal colon crypts and adenomas from the two main tumor models, both control and Apc^Lrig1KO/+ crypts and adenomas, then Mpc1^Lrig1lKO and Apc^Lrig1KO/+Mpc1^Lrig1KO crypts and adenomas (normal crypts n=10–17, adenomas n=16–32), ****p≤0.0001, one-way ANOVA with multiple comparisons, error bars SD. (E) Anti-MPC1 staining of colon sections at 60 days post-TAM (40X, scale bars = 100μm). (F) At 100 days post-AOM, tumor burden (bins of lesion number per mouse) were assessed in Mpc1^Lrig1KO and control mice (n=28–45 mice). (G) At 90 days post-TAM tumor burden (bins of lesion number per mouse) were assessed in Apc^Lrig1KO/+Mpc1^Lrig1KO and Apc^Lrig1KO/+ mice (n=11–12). (H) At 140 days post-TAM, tumor penetrance (having one or more lesions per mouse) and burden (bins of lesion number per mouse) were assessed in Apc^VillKO/+Mpc1^VillKO and Apc^VillKO/+mice (n=18–28).

Figure 2: Loss of Mpc1 accelerates tumor initiation and dysplastic progression in the environmental tumor model.

(A) Macroscopic tumor burden was assessed in Mpc1^Lrig1KO, Mpc1^Lrig1KO/+, and control mice at 100 days post-AOM (n=28–45 mice), ****p≤0.0001, ***p≤0.001, one-way ANOVA with multiple comparisons. Representative gross colon tumor burden and distribution are shown (ruler tick = 1mm). (B) At 100 days post-AOM, microscopic tumor grade and size assessed by pathologic grading in colons of Mpc1^Lrig1KO, Mpc1^Lrig1KO/+, and control mice (n=10–12), n.s.=not significant, *p≤0.05, one-way ANOVA with multiple comparisons on number of microscopic tumors per mouse. (C) Hematoxylin/Eosin and anti-β-catenin staining of control and Mpc1^Lrig1KO adenomas at 100 days post-AOM, (100X, scale bars = 25μm). Black arrows indicate nuclear β-catenin. Green dashed circles indicate crypt necrosis. (D) Immunoblot of paired normal crypts (“C”) and tumor (“T”) samples from two control mice and two Mpc1^Lrig1KO mice for MPC1, MPC2, and actin. Densitometry for MPC1 and MPC2, normalized to actin is shown with relative values adjusted to 100 for each control mouse normal crypt sample. (E) 100 days post-AOM, Ki67-positive nuclei as a percentage of total adenoma nuclei of Mpc1^Lrig1KO and control adenomas were quantified on blinded high magnification images (n =5–7 mice, adenomas=10–20), **p≤0.01, unpaired two-tailed t-test. Representative images to the right (100X, scale bars=25μm). (F) 100 days post-AOM, cleaved caspase 3-positive nuclei of Apc^Lrig1KO/+Mpc1^Lrig1KO and Apc^Lrig1KO/+ adenomas (black arrows) were quantified on blinded high magnification images (n =5–7 mice, adenomas=13–15, 100X, scale bars=25 μm), n.s.=not significant, unpaired two-tailed t-test. Green dashed circle indicates crypt necrosis and these positive nuclei were omitted. (G) At 49 days post-AOM, regions of superficial inflammation, epithelial injury with active inflammation (AI), aberrant foci, and low-grade dysplasia assessed by pathologic grading in colons of Mpc1^Lrig1KO/+, Mpc1^Lrig1KO/+, and control mice (n=8–10 mice), n.s.=not significant, unpaired t-test on number of microscopic lesions per mouse.

Figure 3: Mpc1 deletion in the small intestine and colon increases dysplastic initiation in the Apc loss model of carcinogenesis.

(A) Macroscopic tumor burden was assessed in Apc ^Lrig1KO/+Mpc1^Lrig1KO, Apc ^Lrig1KO/+Mpc1^Lrig1KO/+, and Apc ^Lrig1KO/+ mice at 90 days post-TAM (n=11–12 mice), **p≤0.01, *p≤0.05, one-way ANOVA with multiple comparisons (B) Microscopic tumor grade and size assessed by pathologic grading in the small intestine (left) and the colon (right) for Apc^Lrig1KO/+ Mpc1^Lrig1KO and Apc^Lrig1KO/+ mice (n=5–7), n.s.=not significant, *p≤0.05, unpaired t-test on number of microscopic tumors per mouse. (C) Hematoxylin/Eosin, and anti-β-catenin staining of Apc^Lrig1KO/+ and Apc^Lrig1KO/+ Mpc1^Lrig1KO adenomas (100X, scale bars=25 μm). Black arrows indicate nuclear β-catenin. (D) Immunoblot of paired normal crypts (“C”) and tumor (“T”) samples from two Apc^Lrig1KO/+ mice and two Apc^Lrig1KO/+Mpc1^Lrig1KO mice for MPC1, MPC2, and actin. Densitometry for MPC1 and MPC2, normalized to actin is shown with relative values adjusted to 100 for each Apc^Lrig1KO/+ mouse normal crypt sample. (E) 90 days post-tamoxifen, Ki67-positive nuclei as a percentage of total adenoma nuclei of Apc^Lrig1KO/+Mpc1^Lrig1KO and Apc^Lrig1KO/+ adenomas were quantified on blinded high magnification images (n =3–5 mice, adenomas=10–23), n.s.=not significant, unpaired two-tailed t-test. Representative images to the right (100X, scale bars=25 μm). (F) 90 days post-tamoxifen, cleaved caspase 3-positive nuclei of Apc^Lrig1KO/+Mpc1^Lrig1KO and Apc^Lrig1KO/+ adenomas (black arrows) were quantified on blinded high magnification images (n =3–5 mice, adenomas = 15–16), n.s.= not significant, unpaired two-tailed t-test. Representative images to the right (100X, scale bars = 25 μm). Green dashed circle indicates crypt necrosis and these positive nuclei were omitted. (G) At 140 days post-TAM, macroscopic tumor burden per mouse in Apc^VillKO/+ and Apc^VillKO/+Mpc1^VillKO was assessed (n=18–27,28: one small intestine was damaged during dissection and removed from analysis). *p≤0.05, unpaired t-test. (H) Microscopic tumor grade and size assessed by pathologic grading in the small intestine (left) and colon (right) of Apc^VillKO/+ and Apc^VillKO/+Mpc1^VillKO mice (n=7–8), *p≤0.05, unpaired t-test on number of microscopic tumors per mouse.

Although we recognize that the phenotype of the established lesion, even if it is as early as the adenomas described, is not necessarily tied to the initial events of tumorigenesis, these data are consistent with the notion that tumor initiation is associated with the establishment of an aerobic glycolytic metabolic program with limited mitochondrial pyruvate uptake and oxidation. Therefore, we sought to test the hypothesis that constitutive enforcement of this metabolic program in normal intestinal stem cells might predispose those cells for hyperproliferation and adenoma formation. To do so, we employed a mouse model wherein we deleted Mpc1 selectively in adult LRIG1⁺ intestinal stem cells (Mpc1^Lrig1KO) (Figure S1F) (Powell et al., 2012). This approach is sufficient to cause loss of MPC1 and MPC2 proteins, due to their co-dependence for protein stability, and complete ablation of MPC activity (Flores et al., 2017; Gray et al., 2015; Grenell et al., 2019; Rauckhorst et al., 2017; Schell et al., 2017). Three consecutive days of tamoxifen injection led to a near complete loss of Mpc1 mRNA and MPC1 protein across the GI tract (Figure 1D–E). Otherwise untreated Mpc1^Lrig1KO mice exhibited overtly normal small intestine and colon morphology and length as well as unchanged body weight, body composition, food/water intake, metabolic rate, locomotor activity and intestinal barrier function at 120 days post-recombination (Figure S1G–M).

This genetic model was then subjected to an AOM-DSS treatment in order to elicit a tumorigenic environment (Figure 1C). Control mice exhibited a modest macroscopic tumor burden, with 55% of mice having fewer than 3 adenomas and only 10% having more than 5 adenomas at 100 days post-AOM (Figure 1F). In contrast, Mpc1^Lrig1KO mice exhibited an increased tumor burden compared to controls, with only 15% having fewer than 3 lesions and 60% having more than 5 (Figure 1F). To complement the environmental tumor model described above, we employed two genetic Apc^flox/+ models, one originating in the adult LRIG1+ intestinal stem cells (Apc^Lrig1KO/+) and a second targeting the entire Villin+ intestinal epithelium (Apc^VillKO/+) (Figure 1C) (Colnot et al., 2004; Marjou et al., 2004; Powell et al., 2014; Powell et al., 2012). In both models, tumor formation was elicited by sporadic loss of Apc heterozygosity after tamoxifen-induced inactivation of one allele (Barker et al., 2009; Beyaz et al., 2016; Colnot et al., 2004; Powell et al., 2014; Powell et al., 2012). Consistent with previous reports, Apc^Lrig1KO/+ mice exhibited a modest macroscopic tumor burden assessed from the duodenum to rectum, with 60% of mice having fewer than 10 adenomas and none having more than 24 at 90 days post-TAM (Figure 1G). The Apc^Lrig1KO/+Mpc1^Lrig1KO mice exhibited a higher tumor burden, with only 18% having fewer than 10 lesions and 45% having more than 25 (Figure 1G). Similarly, Apc^VillKO/+ mice exhibited a modest macroscopic tumor burden, with 72% of mice having fewer than 10 adenomas and only one mouse having more than 25 at 140 days post-TAM (Figure 1H). Homozygous loss of Mpc1 (Apc^VillKO/+Mpc1^VillKO) increased tumor burden dramatically such that only 44% had fewer than 10 lesions and 51% had more than 25 (Figure 1H). Interestingly, Mpc1 loss was associated with cellular hyperproliferation in the crypt at multiple timepoints in each of these models, despite otherwise normal crypt morphology and apoptosis (Figure S1N–S). In each of these analyses, heterozygous mice exhibited an intermediate phenotype. Apoptosis was unaffected by Mpc1 loss, with apoptotic events being restricted to the luminal surface of the colon, as expected (Figure S1T). These data demonstrate that a loss of MPC function and the concomitant impairment of mitochondrial pyruvate uptake is sufficient to promote intestinal tumor initiation in the background of multiple mouse cancer models.

Loss of Mpc1 promotes tumor initiation in an environmental tumor model

Mpc1^Lrig1KO mice exhibited a substantial increase in the incidence of macroscopic tumors compared to controls in the AOM-DSS experiment (Figure 2A). There was not a consistent gene dose-dependent difference in adenoma size across all genotypes (Figure S2A). The increased adenoma incidence was observed along the length of the colon, but was most pronounced in the proximal colon (15%, 12.8%, and 25.3% proximal distribution in control, Mpc1^Lrig1KO/+, and Mpc1^Lrig1KO respectively; Chi-square test p=0.0592) (Rosenberg et al., 2009). All genotypes exhibited a similar extent of AOM-induced DNA damage as assessed by quantifying nuclear positivity of phospho-histone H2A.X (Figure S2B). Moreover, the body weight response to DSS-induced colitis, body composition, and intestine length were similar across all genotypes (Figure S2C–E), suggesting that the observed tumor phenotype is unlikely to be the result of experimental or systemic parameters.

Colons from the AOM-DSS-treated mice were subjected to blinded histological examination. As expected from the macroscopic examination, Mpc1^Lrig1KO exhibited a higher overall tumor burden (Figure 2B, S2F). Whereas control mice developed only low-grade adenomas with limited focal high-grade features, loss of Mpc1 led to robust high-grade lesions, as noted histologically by fulminant crypt necrosis (shown) and cribriform glands (not shown) (Figure 2BC). This was particularly true upon homozygous Mpc1 deletion (Figure 2B). As expected, MPC1 protein was absent in Mpc1^Lrig1KO adenomas and this led to loss of MPC2 protein as previously observed (Figure 2D) (Schell et al., 2017; Gray et al., 2015). Consistent with our mRNA analyses, MPC1 protein expression was present in control adenomas, though strongly reduced compared to adjacent normal tissue (Figure 2D). Mpc1^Lrig1KO adenomas exhibited increased cellular proliferation, as assessed by increased Ki67 staining in grade-matched histological sections (Figure 2E), but no change in the frequency of apoptosis, as assessed by cleaved Caspase-3 positive nuclei staining (Figure 2F). While examination at 100 days post-AOM is typical in the AOM-DSS model, and the resulting data were supportive of the hypothesis that loss of Mpc1 led to accelerated tumor initiation, we also wanted to perform an earlier examination to capture the initial signs of adenoma formation. We therefore sacrificed mice at 49 days post-AOM and performed a histological analysis similar to that depicted in Figure 2C. At this time point, Mpc1^Lrig1KO and control colon epithelium showed similar residual epithelial injury and active inflammation, which is consistent with them having a similar direct response to the inflammatory and mutagenic burden (see Figure S2B–C). However, signs of hyperproliferative lesions, in the form of more and larger aberrant epithelial foci, were observed in Mpc1^Lrig1KO mice (Figure 2G, S2G). These data suggest that the higher grade of adenomas in the Mpc1^Lrig1KO mice at 100 days is unlikely to be explained simply by more rapid progression, and perhaps more importantly, that the earliest indicators of future tumor formation were more frequent upon Mpc1 deletion. Most likely, this is indicative of Mpc1 deletion rendering a higher number of cells susceptible to tumor initiation.

Loss of Mpc1 promotes tumor initiation in genetic tumor model

Unlike the AOM-DSS model, wherein adenomas were restricted to the proximal and distal colon, the Apc^Lrig1KO model exhibited adenomas throughout the small intestine and colon (Beyaz et al., 2016; Colnot et al., 2004; Johnson and Fleet, 2012). While the macroscopic tumor burden was much higher upon loss of Mpc1, the distribution of lesion site was not significantly different (p=0.0767) (Figure 3A and S3A–B). Blinded pathological grading of tumor histology showed a dramatic increase in the number of <1mm microadenomas in Apc^Lrig1KO/+ Mpc1^Lrig1KO intestine and colon as well as larger (>1mm) adenomas (Figure 3B). The lesions were almost exclusively low-grade dysplasia in both genotypes, although we observed rare high-grade lesions exclusively in the Mpc1-deficient setting (Figure 3B). All dysplastic lesions demonstrated nuclear β-catenin at multiple locations, which is consistent with complete loss of Apc function (Figure 3C). As with the AOM-DSS model, MPC1 protein (and MPC2 protein as a result) was absent in Mpc1^Lrig1KO adenomas and was decreased relative to adjacent normal tissue in control adenomas (Figure 3D). Similar to the AOM-DSS model, we observed no consistent gene dosage-effect of MPC loss on tumor size, intestine length, or body weight during the experiment (Figure S3C–E). In contrast, unlike the AOM-DSS model, loss of Mpc1 was not associated with increased intra-adenoma proliferation in the Apc^Lrig1KO model, as assessed by quantification of Ki67 staining (Figure 3E), consistent with the similar tumor sizes and grades across Mpc1 genotypes in this model. MPC1 genetic status had no effect on the incidence of apoptosis, as assessed by cleaved Caspase-3 positive nuclei staining in grade-matched histological sections (Figure 3F). The similarity between genotypes in tumor grade and proliferation in the Apc loss-driven tumor model, unlike that observed in the AOM-DSS model, raises the possibility of a genetic relationship between Mpc1 and Apc in tumor phenotype, which is discussed in more detail below.

To complement the above experiment using the Lrig1-CreER knock-in allele and to ensure that the observed phenotypes are not related to any peculiarities of this Cre driver (Powell et al., 2012), we performed a similar study of Apc and Mpc1 deletion employing tamoxifen-inducible Cre recombinase under control of the Villin promoter (Apc^VillKO/+Mpc1^VillKO) (Figure 1C) (Marjou et al., 2004). Gross assessment again confirmed that Apc^VillKO/+Mpc1^VillKO mice had higher tumor burden compared to Apc^VillKO/+ mice, and this was found throughout the intestine and colon (Figure 3G and S3F–G). As in the Lrig1-CreER model, adenoma size was not affected between Mpc1 genotypes nor was intestine length or body weight (Figure S3H–J). Histological examination revealed that, again, adenomas were almost exclusively low-grade regardless of Mpc1 genotype (Figure 3H). The increased proportion of low-grade microadenomas (<1mm) in Apc^VillKO/+Mpc1^VillKO mice reinforced the hypothesis that upon Apc inactivation, the loss of Mpc1 does not markedly affect tumor size or progression and that these two genes might share a specific genetic relationship (Dow et al., 2015; Sandoval et al., 2017; Schell et al., 2017).

Apc mutant tumor expansion in Drosophila is not affected by loss of mitochondrial pyruvate import

Our studies of MPC function in mouse adenoma formation depend on the clonal expansion of cells that undergo sporadic tumorigenic mutations in the AOM-DSS model or sporadic loss of the wild-type allele in the Apc^Lrig1KO/+ model. To confirm and extend this mammalian genetic approach, we turned to genetic manipulations in the Drosophila intestinal stem cell lineage where we can engineer a homozygous loss of Apc function in single stem cells in either the presence or absence of the MPC. As in mammals, the Drosophila intestinal epithelium is maintained by regulated intestinal stem cell proliferation and differentiation (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006). Our studies use MARCM clonal analysis to disrupt gene function in single intestinal stem cells and follow subsequent clonal expansion using a GFP lineage marker (Figure S4A) (Wu and Luo, 2006). In previous work, we showed that a loss of MPC function led to a two-fold increase in the size of Drosophila intestinal clones lacking Notch signaling, indicating that the MPC normally suppresses Notch enteroendocrine tumor expansion (Schell et al., 2017). To determine if a similar effect is seen on Apc tumor growth, we used MARCM clonal analysis to disrupt MPC function in intestinal stem cells that carry mutations in the two Drosophila Apc genes (Cordero et al., 2012; Lee et al., 2009). As expected, mutant clones lacking both Apc1 and Apc2 function are significantly larger than wild-type clones (Figure 4A, 4D, 4F) or clones that lack MPC function (Figure 4B, 4C, 4F). The size of Apc mutant clones, however, is unaffected by a loss of the MPC (Figure 4E, 4F). This result parallels that described previously in Apc^Lrig1KO/+ mice in which adenoma size is not affected by a loss of MPC function, suggesting that the interplay between pyruvate metabolism and Apc-mediated tumor initiation is conserved through evolution. In addition, these studies suggest that, unlike the case with Notch loss of function, Apc mutant tumors appear to be operating at a maximal glycolytic capacity that cannot be further enhanced by changes in mitochondrial pyruvate metabolism.

Figure 4. MPC overexpression arrests Apc mutant tissue hyperplasia.

Intestinal MARCM clones were generated using (A, G, L) an FRT82B wild-type chromosome (control), (B) a mutant chromosome for dMpc1, (C) a chromosome expressing Tub>dMpc1 RNAi, (D, H, M) a chromosome mutant for Apc1 and Apc2, (E) a chromosome mutant for Apc1 and Apc2 with Tub>dMpc1 RNAi (I, N), a Tub>dMpc1-dMpc2 overexpressing chromosome, and (J, O) an Apc1 and Apc2 mutant chromosome with Tub>dMpc1-dMpc2 overexpression. Clones were analyzed either 5 days (A-K) or 20 days (L-P) after induction. Quantitation of GFP+ clone size for each genotype is shown (F, K, P). Clones are marked by GFP (green) and nuclei are stained with DAPI (blue). The reduced effect of dMpc1 mutation alone or dMpc1 RNAi on clone size in panel F is less than that reported earlier because these clones were examined at five days after clone induction rather than 30 days (Schell et al., 2017). n≥30 intestines. ****p≤0.0001, ***p≤0.001, n.s.= not significant. Scale bars represent 50 μm (A), 20 μm (G), 40 μm (L).

Ectopically increased mitochondrial pyruvate metabolism blocks tumor formation

Overexpression of Mpc1 and Mpc2 is sufficient to reduce normal intestinal stem cell proliferation rates in Drosophila and increase mitochondrial pyruvate oxidation and proliferation in cancer cell lines (Schell et al., 2014; Schell et al., 2017). We thus exploited this genetic tool by engineering Drosophila intestinal stem cells that lack key tumor suppressors in either the absence or presence of MPC overexpression. This allows us to study the impact of increased mitochondrial pyruvate import at the earliest steps of oncogenesis. Accordingly, we examined the effect of dMpc1 and dMpc2 overexpression on the expansion of intestinal stem cells that lack Apc function. As expected, Apc mutant MARCM clones grow to more than twice the size of wild-type clones by five days after clone induction (Figure 4G, 4H, 4K). Intestinal stem cells that overexpress the MPC have reduced proliferation, consistent with our earlier observations (Figure 4I) (Schell et al., 2017). However, in contrast to the somewhat smaller Apc mutant clones that we expected to see upon MPC overexpression, we found a complete block in Apc clonal expansion under these conditions (Figure 4J, 4K). Antibody stains confirmed that these cells overexpress the MPC as expected (Figure S4B). Importantly, the same effect is also seen at twenty days after clone induction (Figure 4L–P). In addition, the number of Apc mutant clones with MPC overexpression does not change in intestines between five and twenty days demonstrating that MPC overexpressing cells are not more likely to undergo cell death (Figure S4C). Taken together, these results demonstrate that MPC overexpression does not affect the lifespan of the cells in which it is expressed, but it is sufficient to block the oncogenic effects of Apc loss-of-function.

The block in proliferation mediated by Apc loss might be due to a property of MPC overexpression other than its effect on metabolism. To test this possibility, we directed RNAi against LDH in Apc mutant clones. Reduced LDH activity leads to decreased aerobic glycolysis and increased mitochondrial oxidative metabolism in cancer cells (Fantin et al., 2006; Li et al., 2017; Schell et al., 2017; Xie et al., 2014). Consistent with these activities, expression of LDH RNAi in Apc mutant clones severely reduced their expansion, similar to the block in proliferation seen upon MPC overexpression (Figure 5A–E). Taken together with our studies of the MPC, these observations indicate that cellular metabolism acts downstream of the oncogenic pathways that are unleashed by Apc mutations in intestinal stem cells. Interestingly, this genetic placement is consistent with studies in zebrafish, which have shown that the MPC acts downstream from Apc in its roles in heart and intestinal development (Sandoval et al., 2017).

Figure 5. Increased mitochondrial pyruvate metabolism arrests Apc mutant tissue hyperplasia and blocks Notch enteroendocrine tumor expansion.

Intestinal MARCM clones were generated using (A) an FRT82B wild-type chromosome (control), (B) an Apc1 and Apc2 mutant chromosome, (C) a Tub>LDH RNAi chromosome (LDH RNAi), (D) an Apc1 and Apc2 mutant chromosome with Tub>LDH RNAi, and (F-H) a chromosome that expresses Tub>Notch RNAi either alone (F) or together with Tub>dMpc1-dMpc2 overexpression (G) or Tub>LDH RNAi (H). Quantitation of GFP+ clone size for each genotype is shown (E, I). Clones are marked by GFP (green) and nuclei are stained with DAPI (blue). n≥30 intestines. ****p≤0.000, n.s.= not significant. Scale bars represent 30 μm (A, F).

A loss of Notch signaling in intestinal stem cells results in a distinct type of tissue overgrowth in the intestine relative to Apc mutations, characterized by a restriction of the differentiated cell fate to the enteroendocrine (EE) lineage (Ohlstein and Spradling, 2006, 2007). We thus examined the effect of MPC overexpression on intestinal stem cells that lack Notch in an effort to determine if the resulting shift in metabolism is sufficient to block other types of intestinal hyperproliferative lesions. As expected, targeting Notch RNAi to intestinal stem cells results in significant clonal overgrowth (Figure 5F). This growth, however, is completely blocked by either MPC overexpression (Figure 5G, 5I) or LDH RNAi (Figure 5H, 5I), similar to our observations of Apc mutant clones. In addition, clones that lack Notch function in the presence of either MPC overexpression or LDH RNAi fail to stain for the Prospero marker for EE cells, indicating that these cells do not differentiate along the EE lineage as is usually seen in mutant Notch-derived tumors (Figure S4D–F). This observation, combined with the small size of the Notch RNAi cells, argues that they have maintained their precursor cell fate in spite of MPC overexpression or LDH RNAi. Taken together, these results suggest that reduced mitochondrial pyruvate metabolism is a common potentiator of hyperproliferation in oncogenic intestinal stem cell models in Drosophila.

MPC-mediated mitochondrial pyruvate oxidation is limited in intestinal stem cells and tumors

We returned to the mouse model to better characterize the metabolic consequences of MPC manipulation. To measure the contribution of glucose to the metabolic flux of intestinal crypt cells, we isolated crypts from control or Mpc1^VillKO mice and incubated them with ¹³C-glucose immediately ex vivo, followed by metabolite extraction and both GC- and LC-MS analysis. As expected, Mpc1^VillKO did not affect glucose uptake or glycolytic production of pyruvate (Figure 6A). However, despite this strong pyruvate labeling, little glucose carbon entered the TCA cycle in control crypts (Figure 6B). The direct labeling of representative TCA cycle intermediates citrate and malate and the TCA cycle derivative aspartate from ¹³C-pyruvate (the M+2 isotopomers of citrate, malate and aspartate) was between 5 and 15%. In Mpc1^VillKO crypts, this limited labeling from ¹³C-glucose into TCA metabolites was significantly reduced, consistent with a severe decrease in mitochondrial pyruvate uptake (Figure 6B). These data implied that other substrates might preferentially contribute carbon to the TCA cycle in intestinal crypts. We therefore set out to conduct analogous metabolic labeling experiments with ¹³C-labeled palmitate and glutamine. Exogenous palmitate is not typically added directly to cells in culture, beyond what is provided from serum. Therefore, before the palmitate tracing experiment, we assessed whether the addition of palmitate would significantly alter glucose contribution to the TCA cycle. Freshly isolated crypts were incubated with ¹³C-glucose and delipidated BSA with and without added palmitate. The contribution of glucose to mitochondrial TCA intermediates was similar between the two conditions, suggesting that the added palmitate did not significantly affect the flux of glucose oxidation (Figure S5A). We therefore conducted all isotope labeling experiments, including those with ¹³C glucose (Figure 6A, 6B), in the presence of delipidated BSA with palmitate.

Figure 6. Mpc1 loss influences fuel choice in intestinal crypts and adenomas.

Intestinal crypts or micro-dissected adenomas were freshly isolated and traced in parallel for each of three heavy labeled medias containing glucose, palmitate, and glutamine for 90 minutes. (A) Glucose (M+6) and pyruvate (M+3) labeling as measured by LC-MS from incubation with uniformly labeled ¹³C-glucose is not different in crypts isolated from control and Mpc1^VillKO mice (n=8–9) n.s.= not significant, unpaired t-test (B) ¹³C-Glucose labeling into TCA cycle products citrate, malate and aspartate is significantly reduced in Mpc1^VillKO crypts (n=8–9) ***p≤0.001, **p≤0.01, *p≤0.05, n.s. = not significant, n.d.= not determined, unpaired t-test. (C) ¹³C-Palmitate M+16 labeling is not different between control and Mpc1^VillKO crypts (left). ¹³CPalmitate labeling into TCA cycle products citrate, malate and aspartate is higher in Mpc1^VillKO crypts (n=8–9) (right). (D) Normalized contribution of each nutrient (palmitate, glucose and glutamine) to citrate (highlighted in pie chart) and malate production in control and Mpc1^VillKO crypts. For each metabolite, a normalized labeling fraction was calculated and then corrected for the abundance of the isotopic precursor present in the isolated crypts (n=8,9) ***p≤0.001, *p≤0.05, n.s. = not significant, unpaired t-test. (E) Labeling of TCA intermediates citrate and malate as determined by GC-MS from heavy glucose, palmitate or glutamine containing media in freshly isolated adenomas from Apc^Lrig1KO/+ or Apc^Lrig1KO/+Mpc1^Lrig1KO mice 90 days post-TAM. The glutamine labeling fraction into citrate and malate increases in Mpc1^Lrig1KO mice (n=3). It is noteworthy that while the Cre driver is different between the normal crypts and adenomas, it is unlikely to affect the results as they are both efficient at causing complete recombination throughout the intestinal epithelium within a week of tamoxifen treatment. (F) Glucose (M+6) and pyruvate (M+3) labeling with uniformly labeled ¹³C-glucose is not different in adenomas isolated from control and Mpc1^Lrig1KO mice (n=3 mice, pooled tumors >3 per mouse), n.s.= not significant, unpaired t-test.

We found that unlike glucose, both glutamine and palmitate contributed significantly to TCA cycle intermediates (Figure 6C, S5B). The contribution of ¹³C-glutamine was similar between control and Mpc1^VillKO crypts (Figure S5B). In contrast, ¹³C-palmitate labeling into TCA cycle intermediates citrate and malate was significantly increased in the Mpc1^VillKO crypts (Figure 6C). To quantify the specific contribution of the labeled nutrients to TCA metabolism, we calculated the normalized labeling fraction of each input towards either citrate or malate (Buescher et al. 2015). The palmitate labeling contribution increased to such a degree as to fully compensate for the loss of glucose flux into the TCA cycle in Mpc1^VillKO crypts (Figure 6D).

Considering that fully formed tumors might have a different metabolic profile than the stem cell-enriched crypts from which they arose, we isolated adenomas from Apc^Lrig1KO/+ and Apc^Lrig1KO/+ Mpc1^Lrig1KO mice at 90 days post-recombination and performed an immediate ex vivo isotope labeling experiment analogous to those described above for intestinal crypts. In the adenomas derived from Apc^Lrig1KO/+ animals, glucose, palmitate and glutamine all contributed to citrate, with glutamine being the dominant contributor to malate (Figure 6E). Interestingly, while labeling with ¹³C-glucose led to robust (~60%) labeling of the pyruvate pool as observed in the crypts (Figure 6F), labeling with ¹³C-glutamine led to a limited (~15%) labeling of the glutamate pool (Figure S5C), implying that glutamine contribution to the TCA cycle is much larger than depicted in the unnormalized measurements shown in Figure 6E. Glucose contribution to citrate labeling was diminished in adenomas from Apc^Lrig1KO/+ Mpc1^Lrig1KO mice. Unlike Mpc1-deficient crypts, however, the glucose-deficient labeling was not made up by increased palmitate-derived citrate. Instead, 13C-glutamine contribution to citrate was profoundly increased in adenomas derived from Apc^Lrig1KO/+ Mpc1^Lrig1KO mice (Figure 6E and S5D).

MPC loss elicits a more stem cell-like gene expression pattern in tumors

To begin to elucidate the mechanism underlying the effect of Mpc1 deletion on tumor initiation, we transcriptionally profiled size and grade-matched adenomas and adjacent normal tissue from AOM-DSS-treated and Apc^Lrig1KO/+ mice using a custom NanoString gene panel (Figure 7A, 7B, Berg 2019; Merlos-Suárez et al., 2011; Nusse and Clevers, 2017; Schell et al., 2017). We found that the adjacent normal tissue exhibited low expression of stem cell markers and high expression of differentiation markers in both models (Figure 7A, 7B: red and green). Interestingly, adenomas from control AOM-DSS mice clustered in distinct groups based upon the expression of stemness and differentiation marker genes. A subset of adenomas showed a similar pattern of gene expression to normal tissue (Figure 7A: orange with red bracket). The remainder of control adenomas displayed a more variable gene expression pattern, but was distinctly more similar to stem cells (Figure 7A: orange). Interestingly, every adenoma from Mpc1^Lrig1KO mice clustered with the stem-like group of control adenomas (Figure 7B: blue). This suggests that transformation and adenoma formation in control mice might be heterogeneous, resulting in both more differentiated lesions or more stem-like lesions. In contrast, the metabolic consequences of MPC loss resulted in all Mpc1^Lrig1KO adenomas attaining the more stem-like phenotype. Interestingly, adenomas from the Apc^Lrig1KO/+ model did not exhibit this Mpc1 genotype effect (Figure 7B). All adenomas clustered distinctly from the adjacent normal tissue, being characterized by enhanced expression of stemness markers, with no difference between Apc^Lrig1KO/+ and Apc^Lrig1KO/+ Mpc1^Lrig1KO mice (Figure 7B). This is consistent with the fact that all adenomas in the Apc^Lrig1KO/+ model result from the homozygous loss of Apc and constitutive activation of the pro-stemness Wnt/β-catenin pathway, which appears non-additive with the prostemness metabolic consequences of MPC loss. Once Apc is lost, there appears to be no additional impact of MPC loss on the stemness gene expression program, which is consistent with previous results employing pharmacological modulators of the Wnt/β-catenin pathway and the MPC in intestinal organoids (Sandoval et al., 2017; Schell et al., 2017; van der Flier and Clevers, 2009).

Figure 7. Mpc1 deletion elicits enhanced stemness in adenomas and Mpc1 anti-correlates with stemness in mouse and human adenomas.

(A-B) Adenoma tissue (orange and blue; n=16–32) and normal uninvolved colon epithelium (green and red; n=3–10) were isolated from AOM-DSS day 100 mice (A) and Apc^LrigKO/+ and Apc^LrigKO/+Mpc1^LrigKO/+ mice at 90 days post-TAM (B). RNA was extracted and evaluated on the 22-gene custom gene set including stemness and differentiation markers (top) and abundance displayed according to the heat map shown. Samples were clustered via calculating the Euclidean distance between centroids. See Berg 2019 for raw data and analysis code. (C) Linear least squares regression of Mpc1 gene expression was performed using the adenoma and normal colon samples from GSE201916 and the resulting derived gene list showing Mpc1 anti-correlation (r<−0.6; p-value < 0.01) was analyzed by PANTHER (v13.1). Negative log₁₀(p-values) are reported for each pathway with the number of pathway genes identified in parentheses. The dotted line marks the significance of pathway overrepresentation (−log₁₀(0.05) = 1.3). See Berg 2019 for associated tables. (D) GSE20916-derived joint plots of relative gene expression for Mpc1 versus Wnt pathway genes Axin2 (left, p=1.25E-12) and Sox9 (right, p=3.12E-10). The r-value and the fitted regression line with the slope’s 95% confidence interval for each correlation are shown.

To interrogate the relevance of our findings in mouse models to human colon cancer, we queried two datasets that contain normal and adenoma samples (Sabates-Bellver et al., 2007; Skrzypczak et al., 2010). As described previously, we found that Mpc1 and Mpc2 gene expression was lower in adenomas compared to normal human colon (see Figure 1B and S1D). As with our murine studies, we found that expression of the same panel of stemness and differentiation genes distinguished normal colon from adenomas in both datasets (Figure S6A–F). We further queried these human datasets for associations between Mpc1 and Mpc2 gene expression and all other genes. Filtering with a correlation threshold of r > 0.6 and p-value < 0.01, we employed both PANTHER (v13.1) and DAVID (v6.8) tools to identify known pathways that were enriched in genes whose expression correlates with Mpc1 or Mpc2 (Berg 2019; Huang et al., 2009a; Huang et al., 2009b; Thomas et al., 2003; Thomas et al., 2006). We found that Mpc1 expression inversely correlated with several cell proliferation and biosynthetic metabolism pathways, including purine biosynthesis, cell cycle and stem cell regulation (Figure 7C and S6G). Interestingly, both analyses also identified an inverse correlation between Mpc1 gene expression and the Wnt signaling pathway (Figure 7C and S6G). Upon deeper interrogation, Mpc1 expression negatively correlated with canonical colonic Wnt/β-catenin target genes, as shown for Axin2 (r= −0.73, p= 1.25E-12), and Sox9 (r= −0.67, p= 3.12E-10), (Figure 7D, S6H–I, Berg 2019; Nusse and Clevers, 2017). This relationship between Mpc1 gene expression and activity of the Wnt/β-catenin pathway is reminiscent of the mouse adenoma data described above and is suggestive of regulatory interplay between Mpc1 and the Apc repressor of Wnt/β-catenin pathway activity.

Discussion

We set out to characterize the metabolic hallmarks of tumor initiation and determine if they play an instructive role in promoting tumorigenesis (Pavlova and Thompson, 2016). In support of this concept, we found that some of the earliest molecular features of hyperproliferative colon adenomas in both mice and humans are suggestive of a metabolic program of aerobic glycolysis, characterized by low mitochondrial import and oxidation of the glycolytic product pyruvate. Consistent with these correlative data, we found that intestinal crypts, which are rich in intestinal stem cells, the cell of original for colon tumors, exhibit limited glucose contribution to mitochondrial oxidation. Instead, palmitate and glutamine are more significant mitochondrial substrates in these cells. These data led us to hypothesize that constitutive enforcement of a glycolytic and non-oxidative program of carbohydrate metabolism might predispose these crypt cells, which are believed to be the tumor initiating cells, for tumor initiation in the colon. Using genetic ablation of the Mitochondrial Pyruvate Carrier (MPC), which leads to constitutively impaired mitochondrial pyruvate oxidation, we demonstrated that the establishment of a metabolic program of glycolysis and its derivative pathways is sufficient to promote tumor formation in multiple mouse intestinal cancer models. We also tested the hypothesis that decreased mitochondrial pyruvate oxidation might be required for tumorigenesis. Using two distinct Drosophila models of clonal tumorigenic hyperproliferation, we found that constitutive enforcement of high mitochondrial pyruvate import was sufficient to completely block tumor formation.

The mammalian data presented herein are suggestive of a genetic and regulatory relationship between the Wnt/β-catenin pathway, which is critical in maintaining stem cell identity, and mitochondrial pyruvate import and metabolism (Sandoval et al., 2017; Schell et al., 2017). First, as early as we can detect the hyperproliferative lesions resulting from homozygous loss of Apc in the Apc^Lrig1KO/+ model, those lesions exhibit low expression of Mpc1 and Mpc2. These data are consistent with the hypothesis that loss of Apc enforces decreased expression of the MPC genes and establishment of a glycolytic metabolic state. Second, we found that activity of the Wnt/β-catenin pathway, as assessed by expression of its target genes, is inversely correlated with Mpc1 gene expression in human normal and adenomatous colon. Third, while loss of Mpc1 changes the characteristics of adenomas generated in the sporadic AOM-DSS model, adenomas generated in response to Apc loss are essentially identical regardless of Mpc1 status. This is most clearly demonstrated by comparison of gene expression across Mpc1 genotypes in the AOM-DSS versus Apc deletion models. Control adenomas in the AOM-DSS model exhibit heterogeneity in their expression pattern of stemness and differentiation genes. Loss of Mpc1 leads to a profound constraint on that heterogeneity, whereby every adenoma clusters in the more “stem-like” group. On the other hand, when tumors are initiated by loss of Apc, additional loss of Mpc1 has no significant effect on expression of these genes. This same concept is recapitulated by analysis of intra-adenoma proliferation. Both loss of Mpc1 (in the AOM-DSS model) and loss of Apc lead to increased cellular proliferation, but combining Mpc1 loss with the Apc model has no additional effect. Similarly, loss of Mpc1 leads to a striking increase in the frequency of high-grade features in adenomas from the AOM-DSS model. However, in the Apc model, adding the genetic deletion of Mpc1 has essentially no effect. Interestingly, loss of Mpc1 does increase the frequency of adenoma formation in the Apc model, even if those adenomas are not different. We believe that this is best explained by loss of Mpc1 leading to increased frequency of Apc loss of heterozygosity, which is the precipitating genetic lesion for tumorigenesis in this model. Indeed, we observed increased cellular proliferation in normal crypts upon Mpc1 deletion without a matched increase in apoptosis, which would be predicted to increase the frequency of replication errors that would result in inactivation of the remaining wild-type Apc allele (Barker et al., 2009; Dow et al., 2015; Sansom et al., 2007; Tomasetti et al., 2017).

A genetic relationship between Apc and Mpc1 is further supported by our studies in Drosophila. As in the mouse, loss of either the MPC or Apc leads to hyperproliferation of Drosophila intestinal stem cells, however, there is no additive effect of combining these genetic disruptions. More importantly, we find that the hyperproliferation elicited by a deletion of Apc is completely suppressed by ectopic expression of the MPC. This is reminiscent of a similar effect reported in the zebrafish intestine, wherein MPC1 expression rescues the differentiation defects imposed by a loss of Apc (Sandoval et al., 2017). Taken together, the data presented in the mammalian and fly intestine and that from published work in zebrafish support the hypothesis that Mpc1 is genetically epistatic to Apc. In the context of cancer, the MPC thus functions downstream from traditional tumor suppressor genes to regulate tumor initiation and growth.

We propose that the Mpc1-dependent effects described herein are explained by a stem cell autonomous mechanism. In Drosophila, the MARCM clonal analysis used for our studies effectively targets genetic manipulations to single intestinal stem cells and marks their lineage. Similarly, in the mouse, we see no evidence, either in the intestine or elsewhere, of a defect that would elicit signaling to intestinal stem cells to promote their hyperplasia or transformation. Moreover, we observe basal hyperproliferation of intestinal stem cells in the context of an otherwise completely normal intestinal epithelium. This is consistent with the previously published observation that low activity of the MPC in intestinal stem cells is critical for maintaining the markers and phenotypes of stemness, including in the simplified ex vivo organoid system where cell non-autonomous effects are minimized (Schell et al., 2017). Altogether, we propose the hypothesis that loss of MPC activity elicits a cell autonomous metabolic program that is pro-stemness and pro-proliferative, perhaps similar to the metabolism enforced by the Wnt/β-catenin pathway, and this metabolic milieu promotes the earliest steps of tumor initiation.

We do not yet know why Mpc1 loss leads to the observed proliferative and pro-tumorigenic phenotype. Our data, however, strongly suggest that it is related to the blockade of mitochondrial pyruvate oxidation and the concomitant establishment of a constitutive aerobic glycolysis program. Indeed, we demonstrate that loss of Mpc1 leads to a profound loss of glucose contribution to TCA cycle intermediates in intestinal crypts. In addition to loss and gain-of-function for the MPC, we showed in Drosophila that inactivation of lactate dehydrogenase in stem cells is sufficient to replicate the effects of MPC overexpression. In all cases, our results were consistent with the metabolic fate of pyruvate playing a decisive role in setting the proliferative tone of intestinal stem cells. These effects of perturbed carbohydrate metabolism are analogous to the observation that increased fatty acid availability and metabolism via a high-fat diet leads to enhanced proliferation and tumorigenesis (Beyaz et al., 2016). Carbohydrate and fatty acid oxidation have a well-described antagonistic relationship and it has been repeatedly observed that loss of MPC activity (and carbohydrate oxidation) leads to increased fatty acid oxidation and vice versa (Goetzman and Prochownik, 2018; Hue and Taegtmeyer, 2009; Schell et al., 2017). Indeed, we found that loss of the MPC in intestinal crypts leads to increased palmitate contribution to TCA cycle flux in ex vivo intestinal crypts. It is therefore likely that the published effects of fatty acid abundance and the effects of altered carbohydrate metabolism described herein are two manifestations of a common underlying mechanism (Cheng et al., 2019; Mihaylova et al., 2018). For example, it is possible that loss of Mpc1 increases tumorigenesis via increased fatty acid uptake and metabolism (Beyaz et al., 2016). Interestingly, we find that ex vivo adenomas also have limited glucose contribution to TCA cycle intermediates, with the majority being derived from glutamine. Moreover, unlike Mpc1-deficient crypts, adenomas from Mpc1-deficient mice appear to compensate for the loss of glucose-derived TCA cycle flux by increased glutamine oxidation as seen previously in cultured cancer cells (Yang et al., 2014).

The above data demonstrate that decreased mitochondrial pyruvate metabolism, enforced by elimination of MPC activity, is sufficient to increase the oncogenic susceptibility of the fly and mouse intestinal tract. Therefore, constitutive enforcement in intestinal stem cells of the metabolic program found in hyperproliferative colonic lesions predisposes those stem cells to adenoma formation. Conversely, prevention of that metabolic program prevents proliferation caused by the loss of two distinct tumor suppressors. “Driver” mutations in oncogenes or tumor suppressors have traditionally been thought to be sufficient to direct cell transformation and tumor initiation (Hanahan and Weinberg, 2011; Vogelstein and Kinzler, 2015). In this view, cellular metabolism is seen as providing an important permissive environment that generates the metabolic intermediates required for cell growth and proliferation. Our studies place metabolism in a different context – as a critical gatekeeper that acts downstream from oncogenic signal to drive cancerous initiation and growth.

Limitations of the Study

The most obvious limitation of this study is that the tumor models that we have used are not faithful representations of the typical process of tumor formation in humans. We have used genetic manipulations or treatment with chemical carcinogens to induce tumor formation in mice and flies. While these manipulations are the best available tools for tumor formation in animal models that are broadly accessible, they generate a scenario where essentially all animals exhibit tumor formation within a relatively short timeframe. Future studies will likely address the degree to which a change in cellular metabolic profile predisposes it for tumor formation in more natural settings.

STAR METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Jared Rutter (rutter@biochem.utah.edu). All unique/stable reagents generated in this study are available from the Lead Contact without restriction.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animal Care:

All IACUC guidelines were followed and the protocol for mouse experiments carried out in this study was submitted to the University of Utah Institutional Animal Care and Use Committee. Mice were maintained on a Teklad global soy-protein free diet (2920×030917M). Home cages with pelleted paper bedding with nest building enrichment of irradiated paper and tissue strips were placed on ventilated racks with automated water taps. Food and water were available ad libitum, and food was only withdrawn if required for an experiment. Mice were co-housed with at least one to four other mice when possible in a temperature-controlled facility with 12-hour light/dark cycles. Mice were checked daily by facility technicians for signs of distress and bedding changes occurred bimonthly. No formal randomization was performed for mouse studies. Mice were used as they became available, paired with littermate controls where possible. Male and female mice C57BL6J at 8–12 weeks of age were used, and both genders were included in each analysis, except for the metabolic cage measurements in which only male littermates were used to avoid differences due to dyssynchronous estrus cycles. Heterozygous Lrig1^CreER mice were obtained from Jackson Labs (Powell et al. 2012). Mpc1^flox/flox mice were generated as previously described (Schell et al. 2017). Apc^{loxp exon 14} (Apc^flox/+) have been described and were a gift from Ömer Yilmaz (MIT) (Beyaz et al. 2016; Powell et al. 2014; Colnot et al. 2004). Villin^CreER mice were also gifted from Ömer Yilmaz (MIT).

Rodent:

Control mice consisted of Lrig1^CreERand Mpc1^flox/flox without Cre for chemical carcinogenesis and homeostatic experiments and are combined under the name “Control”. Lrig1^CreER X Apc^flox/+ (Apc^Lrig1KO/+) and Villin^CreER X Apc^flox/+ (Apc^VillKO/+) were control mice for genetic carcinogenesis experiments. Breeding lines were maintained and outcrossed Mice in carcinogenesis protocols were inspected daily by lab personnel and weighed at a minimum once a week. Mice showing signs of distress including overgrooming-related sores, weight loss greater than 20% of starting body weight, anal prolapse or frank rectal bleeding were removed from the study and not included in any subsequent final data analyses. The total number of animals removed was 11 animals (7 males, 4 females) with no significance towards genotype or tumor induction model. All compounds used for mouse studies are detail in KEY RESOURCES TABLE. All male and female mice were treated at 8 to 12-weeks of age, receiving three intraperitoneal injections of tamoxifen,100μL 20mg/mL in peanut oil on consecutive days in the mornings between 7:00AM and 11:00AM in home cages.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-Pros	Developmental Studies Hybridoma Bank #MR1A	RRID:AB_2722117
Rabbit Drosophila MPC1 and MPC2	Gift from William McDonald. Metabolic solutions development company. Kalamazoo, MI.	Colca et al., 2013
Chicken GFP	Abcam #13970	RRID:AB_300798
Alexa anti-chicken 488- conjugated secondary	Abcam #150169	RRID:AB_2636803
anti-mouse Cy3-conjugated secondary	Jackson Immunoresearch #715– 165151	RRID:AB_2315777
anti-rabbit Cy3-conjugated secondary	Jackson Immunoresearch #715– 165152	RRID:AB_2307443
MPC1 (IHC) 1:200	Sigma Aldrich HPA045119	RRID:AB_10960421
MPC1 (Immunoblots) 1:1000	Cell Signaling #14462	RRID:AB_2773729
MPC2 (Immunoblots) 1:1000	Cell Signaling #46141	RRID: AB_2799295
Ki67 (IHC) 1:600	Abcam, rabbit polyclonal, ab15580	RRID:AB_393778
b-catenin (IHC) 1:200	BD Biosciences #610153	RRID:AB_397554
Actin (immunoblots) 1:1000	Millipore MAB1501	RRID: AB_2223041
Cleaved Caspase 3 (IHC) 1:100	Cell Signaling #9664	RRID: AB_2070042
H2A.X (IHC) 1:500	Cell Signaling #2595	RRID: AB_10694556
Phospho-H2A.X (IHC) 1:500	Cell Signaling #5438	RRID: AB_10707494
Biotin-SP-AffiniPure Donkey Anti-Rabbit IgG (H+L) antibody 1:100	Jackson Immunoresearch #711-065-152	RRID:AB_2340593
Biotin-SP-AffiniPure Donkey Anti-Mouse IgG (H+L) 1:100	Jackson Immunoresearch #715–065150	RRID:AB_2307438
Biotin-SP-AffiniPure Donkey Anti-Rat IgG (H+L) 1:100	Jackson Immunoresearch #712-065-153	RRID:AB_2315779
Goat Anti-Mouse IgG (H&L) Antibody Dylight™ 800 Conjugated (immunoblots)	Rockland #610-145-002-0.5	RRID:AB_10703265
Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 680 (immunoblots)	Invitrogen #A10043	RRID:AB_2534018
Chemicals, Peptides, and Recombinant Proteins
Tamoxifen	Sigma Aldrich T5648	N/A
Azoxymethane	Sigma Aldrich A5486	N/A
Dextran Sodium Sulfate	USB/Affymetrix 14489	N/A
Fluorescein-isocyannate-dextran	Sigma Aldrich 46945	N/A
VectaStain ABC Elite	VectorLabs PK6100	N/A
DAB substrate	VectorLabs SK4100	N/A
hematoxylin	Sigma Aldrich GHS316	N/A
Permount	Fisher Chemical SP15	N/A
5-bromo-2’-deoxyuridine	Invitrogen b23151	N/A
Vector Antigen Unmasking	VectorLabs H3300	N/A
Solution
Donkey Serum	Jackson Immunoresearch #017-000-001	N/A
AffiniPure Fab donkey antimouse fragments	Jackson Immunoresearch #715-007-003	N/A
Mammalian Protease Inhibitor Cocktail	Sigma Aldrich P8340	N/A
PhosSTOP	Roche Molecular 04906845001	N/A
Peanut Oil	Sigma Aldrich P2144	N/A
Phosphate Buffered Saline	Invitrogen 10010049	N/A
U-13C-L-Glutamine	Cambridge, CLM-1822	N/A
U-13C-Glucose	Cambridge, CLM-1396	N/A
U-13C -Palmitate	Cambridge, CLM-409	N/A
L-Glutamine	Sigma Aldrich G3126	N/A
D-Glucose	Sigma Aldrich G8270	N/A
Sodium Palmitate	Sigma Aldrich P9767	N/A
Bovine Serum Albumin (Ultra Fatty Acid Free)	Roche, CN 03 117405 001	N/A
DMEM (base powder), supplemented with Amino Acids	US Biological D9800–27	N/A
Murine EGF	Peprotech #315–09	N/A
Murine Noggin	Peprotech #250–38	N/A
Murine Wnt-3a	Peprotech #315–20	N/A
CHIR99021	Tocris #4423	N/A
N-acetyl cysteine	Sigma Aldrich A7250	N/A
L-carnitine	Sigma Aldrich C0283	N/A
Matrigel	Fisher Scientific #356321	N/A
Vectashield mounting medium	Vector, Laboratories. USA H1200	N/A
Formaldehyde 16% Ultra pure EM grade	Polysciences Inc, #18814–10	N/A
Critical Commercial Assays
Qiagen FFPE Rneasy kit	Qiagen cat: 73504	RRID:SCR_008539
nCounter Elements 60-Plex TagSet (ELE-060-P1TS)	Nanostring Technologies// Schell et al. 2017 Nat Cell Bio	N/A
Deposited Data
Nanostring Mouse Colon crypt and adenoma RNA data	GSE136710	N/A
Experimental Models: Cell Lines and Drosophila lines
UAS-dMpc1 RNAi	Vienna Drosophila resource center	VDRC: 102734
UAS-LDH RNAi	Vienna Drosophila resource center	VDRC: 102330
UAS-Notch RNAi	Bloomington Drosophila Stock Center	BDSC:7078
UAS-dMpc1-P2A-dMpc2	Generated in the Thummel Lab, (Schell et al 2017)	N/A
w[*]; P{ry[+t7.2]=neow[*]; P{ry[+t7.2]=neoFRT}82B sr[1] e[1] Apc2[N175K] Apc[Q8]/TM3, P{w[+mC]=ActGFP}JMR2, Ser[1]FRT}82B sr[1] e[1] Apc2[N175K] Apc[Q8]/TM3, P{w[+mC]=ActGFP}JMR2, Ser[1]	Bloomington Stock center	BDSC: 7211
hs-flp, UAS-GFP; + ; Tub-GAL4, FRT82B, TubGal80/TM6C	Washington University in St. Louis, Gift from Craig Micchelli	N/A
Mpc1flox/flox mice (C57bl6)	Rutter Lab	N/A
VillinCreER mice (C57bl6)	Yilmaz Lab	N/A
LrigiCreER mice (C57bl6)	Jackson Laboratory Stock No: 018418	RRID:IMSR_JAX:018418
Apcloxp exon 14 (C57bl6)	Yilmaz Lab	N/A
Software and Algorithms
XPRESStools v.0.0.2b1	DOI: 10.5281/zenodo.2581696	N/A
Python Software Foundation. Python Language Reference, version 3.6.	Available at http://www.pvthon.org	RRID:SCR_008394
Eric Jones and Travis Oliphant and Pearu Peterson and others, SciPy: Open source scientific tools for Python, 2001	http://www.scipy.org/	RRID:SCR_008058
AltAnalyze	http://www.altanalyze.orq/	RRID:SCR_002951
GEOparse v1.1.0Other	https://github.com/guma44/GEOparse	N/A

Colon and Small Intestinal Crypts:

Small intestinal and colon crypts were isolated as similarly as described previously (Schell et al. 2017). Mice were euthanized by cervical dislocation, and then gastrointestinal tract dissected from pylorus to anus. The mesentery was removed with scissors. The entire colon and small intestine were opened longitudinally with ball-tip scissors and washed with ice cold PBS. Mucus, fecal remnants, and small intestinal villus were scraped away using a glass coverslip. At this stage, gross specimens for preservation were fixed in buffered 10% formalin (Fisher Scientific 23-245-685) for 24 h, transferred to 70% ethanol and paraffin-embedded for sectioning and staining. Alternately, specimens for crypt isolation were placed in a 2.5mM EDTA-PBS solution incubated at 4°C for 20 minutes with gentle agitation. Tubes were then vigorously shaken by hand 10–20X and the crypt-containing supernatant was transferred into a fresh tube. Crypts were washed in PBS and collected by centrifugation (3 × 300g for 2 minutes at 4°C). At this stage, aliquots were taken and snap frozen in liquid nitrogen for protein analysis or RNA extraction. Viable crypts were resuspended in ice cold DMEM/F12 prior to final resuspension in isotopic labeling media.

Drosophila stocks:

Drosophila stocks were maintained on standard food containing 3% sucrose, 6% glucose, 8% yeast, and 1% agar in a 25°C incubator. Transgenic lines are as follows: UAS-dMpc1 RNAi (Bricker et al., 2012), UAS-LDH RNAi (VDRC# 102330), UAS-Notch RNAi (a gift from C. Micchelli), and UAS-dMpc1-P2A-dMpc2 for MPC overexpression (Schell et al., 2017). The UAS-LDH RNAi transgenic line has been used previously for functional studies of this enzyme (Charlton-Perkins et al., 2017; Slaninova et al., 2016). A double mutant for Apc1^Q8 and Apc2^N175K was used to remove Apc function (Bloomington 7211). A hs-flp, UASGFP; +; Tub-GAL4, FRT82B, TubGal80/TM6C stock (a gift from C. Micchelli) was used to introduce different transgenes and mutants for clonal analysis.

METHOD DETAILS

Characterization of MPC1 deletion in mouse intestine and colon:

For experiments relevant to loss of MPC1 in intestinal and colonic homeostasis, 8 to 12-week-old mice were injected intraperitoneally with 100μL of 20mg/mL tamoxifen (Sigma Aldrich T5648) dissolved in peanut oil (Sigma Aldrich P2144) for three successive days and euthanized 30, 60, 90, or 120 days later. BrdU (Invitrogen b23151) was dissolved in PBS (Invitrogen 10010049) to 10mg/mL and 100μL was injected intraperitoneally 4hrs prior to euthanasia to label dividing cells. Injections were carried out in home cages between 7:00AM and 11:00AM and tissue was collected 4-hours later +/−10 min from the time of injection. Dissections of the intestine and colon either were prepared for histological examination or for epithelial collection, described below.

Mouse chemical induction of carcinogenesis:

For chemical induction of carcinogenesis (AOM-DSS), control and experimental mice were treated with tamoxifen as described above. 30 days post tamoxifen injection, control and Mpc1^Lrig1KO mice were injected with 10mg/mL/kg Azoxymethane (Sigma Aldrich A5486) dissolved in PBS (Invitrogen 10010049). 7 days following AOM injection, mice were subjected to 2% Dextran Sodium Sulfate MW 14000 (USB/Affymetrix 14489) in the facility-provided drinking water for 5 days, followed by facility-provided water for remaining 16 days in a 21-day cycle. Mice underwent 3 cycles of DSS-induced colitis in series. Mice were monitored daily for weight change and removed from the study following greater than 20% body weight loss, extreme diarrhea, or severe rectal bleeding (<5 mice removed total). Mice were euthanized 49 or 100 days following AOM injection. BrdU was administered as described above, though not to the 100 day cohorts. Dissections of the intestine and colon either were prepared for histological examination or for epithelial collection. Tumor burden for tumorigenesis experiments was manually counted using Olympus SZ61 dissecting microscope and imaged via GelDoc Imaging platforms.

Mouse genetic susceptibility to carcinogenesis:

For heterozygous Apc model of carcinogenesis, control and experimental mice were tamoxifen as described above. Apc^Lrig1KO/+, Mpc1^Lrig1KO/+Apc^Lrig1KO/+, Mpc1^Lrig1KO/Apc^Lrig1KO/+ mice and Apc^VillKO/+, Mpc1^VillKO/+Apc^VillKO/+, Mpc1^VillKO/Apc^VillKO/+were monitored every 2–3 days for any signs of intestinal or colonic distress. Mice were euthanized 30, 60, 90, or 140 days (Apc^VillKO/+ only) following tamoxifen injection. BrdU was administered as described above. Dissections of the intestine and colon either were prepared for histological examination or for epithelial collection. Tumor burden for tumorigenesis experiments was manually counted using Olympus SZ61 dissecting microscope and imaged via GelDoc Imaging platforms.

Mouse Intestinal Barrier integrity:

Control and Mpc1^Lrig1KO male mice were fasted for 12 hours then gavaged with 100μL/kg fluorescein-isocyannate dextran (FITC-dextran, Sigma Aldrich 46945) dissolved in PBS (background control mice were gavaged with 100μL PBS), and 20μL of corn oil to lubricate the gavage needle in their home cage. Food was then replaced in the cage. 6 hours post-gavage, mice were euthanized and serum obtained via cardiac puncture. The most distal fecal pellet was obtained from the rectum and FITC-dextran extracted via a PBS maceration and centrifugation to clear the debris. Serum and fecal supernatant were diluted 1:10 with sterile deionized water and assayed for fluorescence on Varioskan Flash Spectrophotometer. Background serum and fecal samples were obtained as above to identify background fluorescence from each sample type. All experimental results were analyzed relative to the background adjusted to a value of 1. No subjects or data was excluded from analysis.

Whole mouse body composition and gross metabolism:

Control, Mpc1^Lrig1KO/+, and Mpc1^Lrig1KO male and female mice were weighed once per week beginning with tamoxifen administration. Total body composition was assayed by nuclear magnetic spectroscopy (Bruker MiniSpec) prior to tamoxifen administration and again prior to euthanasia 60 days post tamoxifen and 100 days post azoxymethane. Paired control and Mpc1^Lrig1KO male littermates were metabolically profiled 60 days post-tamoxifen for 72 hours in metabolic cages. Data from the final full 24 hours was analyzed. All metabolic phenotyping profiles were visualized using Prism GraphPad (v7). Work was performed by the Metabolic Phenotyping Core, a part of the Health Sciences Cores at the University of Utah. No subjects or data was excluded from analysis.

Mouse immunoblotting:

Three biological replicate crypt isolations were lysed in RIPA with mammalian protease and phosphatase inhibitors (Sigma Aldrich P8340, Roche Molecular 04906845001). Samples were sonicated and centrifuged at 10,000g for 10 minutes, and the supernatant retained. Protein concentration was determined by BCA Assay (Thermo Scientific 23225) and diluted for equivalent protein loading volumes. Samples were resolved by SDS-PAGE and analyzed by Licor Odyssey. Primary antibodies included MPC1 (Cell Signaling #14462), MPC2 (Cell Signaling #46141), and actin (Millipore MAB1501). Secondary antibodies were Goat Anti-Mouse IgG (H&L) Antibody Dylight™ 800 Conjugated (Rockland #610-145-002-0.5) and Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 680 (Invitrogen #A10043).

Mouse immunohistochemistry:

All tissue processing, staining, and analysis was blinded and scored in duplicate. Immunohistochemical staining was performed on 4-micron thick sections of formalin-fixed, paraffin-embedded tissues. Sections were air-dried and then melted in a 60°C oven for 30 minutes. Slides were loaded onto the Ventana Discovery automated staining instrument (Ventana Medical Systems, Tucson, AZ), de-paraffinized with the EZ Prep solution. Antigen retrieval was performed with CC1 (Cell Conditioning 1, pH 8.5) for 32–60 minutes at 95°C depending on the primary antibody. For host mo use antibodies, mouse Detective (mouse protein blocker, BioCare Medical) is applied for 40 minutes at 35 degrees C. Primary antibody incubations (β-catenin: BD Biosciences cat. 610153, Cleaved-Caspase-3: Cell Signaling #9664, Ki-67: ab15580) was applied for 1 hour at 35–37°C — dilutions specified in the KEY Resources Table. A secondary antibody reagent (goat anti-rabbit: Sigma cat. #B8895, rabbit anti-mouse F(ab’)2: Dako cat. #E0413) was applied for 24–60 minutes at 37°C. Positive signal was visualized using the DAB Map detection kit, which is a goat anti-mouse/anti-rabbit secondary HRP biotinylated detection system, utilizing DAB (3–3’ diaminobenzidine) as the chromogen. Tissue sections were counterstained with hematoxylin for 12 minutes. The slides were removed from the immunostainer and placed in a dH2O/DAWN ™ mixture. The sections were gently washed in a mixture of de-ionized water and DAWN™ solution to remove any coverslip oil applied by the automated instrument. The slides were gently rinsed in deionized water until all of wash mixture was removed. The slides were de-hydrated in graded ethanols, cleared in xylene and then coverslipped.

Slides were scanned with the Zeiss AxioScan Z1 at the Fluorescence Microscopy Core Facility, a part of the Health Sciences Cores at the University of Utah for 20X images and measurements made using ZEN 2 and ZEN lite software. Microscopy equipment was obtained using a NCRR Shared Equipment Grant # 1S10RR024761–01. 4X, 10X, 20X and 40X representative images were obtained on an Olympus IX71 microscope, using MicroSuite software. Images were processed in Adobe Photoshop, with exposure times and adjustments identical between genotypes and tumor model groups.

Pathologic grading and quantification of proliferation.

Histological grading was performed by a board-certified pathologist with subspecialty training in gastrointestinal pathology (KJE) on representative H&E-stained (H&E) slides from the two models of carcinogenesis, AOM-DSS and Apc LOH, at the following respective time points: days 49 and 100 (all mice) and days 90 (all mice) and 140 (Apc^VillKO/+ only). H&E slides were blinded as to time-point, experiment, and genotype, and each tumor was assessed for size and grade. Photographs of slides, stained for Ki67 as described above, were taken of distal colon adenomas in the area of highest labelling by KJE. Printouts of photographs were blinded for quantification of proliferation by Ki67+ nuclei. Five researchers performed counts for both positive and negative nuclei in adenoma tissue from two tumor models and both Mpc1-deleted and control mice, excluding non-epithelial nuclei. Results were tabulated by image, then unblinded for comparison. All specimens generated were included in analysis. No subjects or data was excluded from analysis.

Drosophila MARCM clonal analysis:

The MARCM system was used to generate GFP-marked clones in the adult intestine, as described (Wu and Luo, 2006). Four to five day old adult female flies of the following genotypes were subjected to one 38°C heat pulse for 60 minutes to generate GFP-marked clones in the intestine. Control: y, w, hs-FLP, UAS-GFP / w ; +; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B; dMpc1 mutant: y, w, hs-FLP, UAS-GFP / w; +; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B dMpc1¹; dMpc1 RNAi: y, w, hs-FLP, UAS-GFP / w; + / UAS-dMpc1 RNAi; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B; dMpc1/2 overexpression: y, w, hs-FLP, UAS-GFP / w; + / UAS-dMpc1-dMpc2; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B; LDH RNAi: y, w, hs-FLP, UAS-GFP / w; + / UAS-LDH RNAi; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B; Apc1 Apc2 mutant: y, w, hs-FLP, UAS-GFP / w; +; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B Apc1^Q8 Apc2^N175K Apc1 Apc2 mutant with dMpc1 RNAi: y, w, hs-FLP, UAS-GFP / w; + / UAS-Mpc1 RNAi; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B Apc1^Q8 Apc2^N175K; Apc1 Apc2 mutant with dMpc1/2 overexpression: y, w, hs-FLP, UAS-GFP / w; + / UAS-dMPC-dMpc2 ; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B Apc1^Q8 Apc2^N175K; Apc1 Apc2 mutant with LDH RNAi: y, w, hs-FLP, UAS-GFP / w ; + / UAS-LDH RNAi ; Tub-GAL4, FRT82B, Tub-Gal80 / FRT82B Apc1^Q8Apc2^N175K. Clones were analyzed at 5 days (most experiments) or 20 days (Figure 2F–J) after induction. The number of nuclei in GFP-expressing clones was used to quantify clone size. GFP was detected by fluorescence and nuclei were detected by DAPI stains.

For Notch RNAi studies, intestines were analyzed 7 days after clone induction. Animals of the following genotype were used: Notch RNAi: y, w, hs-FLP, UAS-GFP / w ; UAS-Notch-RNAi; Tub-GAL4, FRT82B, Tub-GAL80 / FRT82B Notch RNAi with dMpc1/2 overexpression: y, w, hs-FLP, UAS-GFP / w; UAS-Notch-RNAi/ UAS-dMpc1-dMpc2; Tub-GAL4, FRT82B, Tub-GAL80; Notch RNAi with LDH RNAi: y, w, hs-FLP, UAS-GFP / w; UAS-Notch-RNAi/ UAS-LDH RNAi; Tub-GAL4, FRT82B, Tub-GAL80.

Drosophila histology and immunostaining.

Intestines were dissected in 1xPBS and fixed with 4% formaldehyde (Polysciences Inc, EM grade) overnight at 4°C. Tissues were washed four times with 0.1% Triton, 1xPBS (PBST) and incubated with mouse anti-Pros antibodies (DSHB) diluted 1:100 in PBST or 1:50 diluted rabbit antibodies directed against dMPC1 and dMPC2 at 4°C overnight (Colca et al., 2013). The ti ssues were then washed and incubated for 3–4 hours with either donkey anti-mouse Cy3-conjugated secondary antibodies (Jackson 715-165-157) or donkey anti-rabbit Cy3-conjugated secondary antibodies (Jackson 715– 165–152). Samples were mounted using Vectashield (Vector, USA) with DAPI. Images were acquired using an Olympus FV1000 confocal microscope and assembled into Z stack projections for the figures. Quantitation of the number of cells per clone is shown as a scatter plot with each data point representing the clone size in a single intestine.

Mouse intestinal crypt and tumor isotope labeling:

_ For isotope labeling metabolomics, U-¹³C-Palmitate, U-¹³C -Glucose, and U-¹³C -Glutamine labeling was performed in parallel on both normal crypts from control and Mpc1^Lrig1KO mice and on tumors from Apc^Lrig1KO/+ and Apc^Lrig1KO/+ Mpc1^Lrig1KO were sacrificed 30 or 90 days following tamoxifen administration, respectively. Crypts were isolated and tumors dissected as described above. Normal crypts were counted, and 10,000 crypts were seeded per well of a 6 well plate coated in a 10% Matrigel/90%DMEM-base mixture. Crypts were overlaid with 2mL DMEM media containing growth factors (Wnt-3a, EGF, Noggin, R-spondin), 5 μM CHIR99021, 1 mM NAC, 1 μg/ml L-carnitine, and 150 M U-¹³C - Palmitate (Cambridge, CLM-409) conjugated to fatty acid-free BSA, or 25 M U-¹³C-Glucose (Cambridge, CLM-1396), or 5 M U-¹³C-L-Glutamine (Cambridge, CLM-1822). In each heavy labeled media type, the other carbon sources in identical ¹²C were added, with the exception of the experiment to determine the effect of exogenous palmitate on glucose utilization. Crypts were incubated for 90 minutes and then spun down in ice cold labeling media to remove Matrigel and resuspended in 100ul of extraction buffer (40% methanol, 40% Acetonitrile, 20%water, all LC/MS grade with 0.5% formic acid). Samples were vortexed and placed on ice. Within five minutes, samples were neutralized by adding 15% ammonium bicarbonate (amount determined before each experiment), vortexed again and returned to ice for 10 minutes. Samples were cleared in a cold centrifuge at max speed for 10 minutes. For tumor tracing, 3–5 tumors were placed per well of a 6 well plate in 2mL DMEM media as described above. After 90 minutes, tumors were washed in warm saline (0.9M NaCl), weighed and snap-frozen in liquid nitrogen. For the tumor tissue extraction, each sample was transferred to 2.0ml ceramic bead mill tubes (bioExpress). To each sample was added cold 90% methanol (MeOH) solution to give a rough final concentration of 80% MeOH to each tissue sample. The samples were then homogenized in an OMNI Bead Ruptor 24. Homogenized samples were then incubated at −20°C for 1 hr. After incubation the samples were centrifuged at 20,000 × g for 10 minutes at 4°C. The supernatant was then transferred from each bead mill tube into a labeled, fresh micro centrifuge tubes. Pooled quality control samples were made by removing a fraction of collected supernatant from each sample. Process blanks were made using only extraction solvent and went through the same process steps as the real samples. All samples, crypt and tumor, were then dried en vacuo.

LC-MS tracing analysis for crypts & tumors:

Extracted polar metabolite samples were analyzed by LC-MC. Separation was achieved by hydrophilic interaction liquid chromatography (HILIC) using a Vanquish HPLC system (ThermoFisher Scientific). The column was an Xbridge BEH amide column (2.1 mm × 150mm, 2.5 μM particular size, 130Å pore size, Waters Co.) run with a gradient of solvent A (20 mM ammonium hydroxide, 20 mM ammonium acetate in 95:5 acetonitrile:Water, pH 9.5) and solvent B (100% acetonitrile) at a constant flow rate of 150 uL/min. The gradient function was: 0 min, 90% B; 2 min, 90% B; 3 min, 75% B; 7 min, 75% B; 8 min, 70% B; 9 min, 70% B; 10 min, 50% B; 12 min, 50% B; 13 min, 25% B; 14 min, 25% B; 16 min, 0% B; 20.5 min, 0% B; 21 min; 90% B; 25 min, 90% B. Autosampler temperature was 4°C, column temperature 30°C and injection volume 2 μL. Samples were injected by electrospray ionization into a QExactive HF orbitrap mass spectrometer (ThermoFisher Scientific) operating in negative ion mode with a resolving power of 120,000 at m/z of 200 and a full scan range of 75–1000 m/z. Data were analyzed using the MAVEN software package and specific peaks assigned based on exact mass and comparison with known standards (Melamud et al., 2010). Extracted peak intensities were corrected for natural isotopic abundance using the R package AccuCor (Su et al., 2017). No subjects or data was excluded from analysis.

GC-MS analysis for crypts & tumors:

All GC-MS analysis was performed with an Agilent 5977b GC-MS MSD-HES and an Agilent 7693A automatic liquid sampler. Dried samples were suspended in 40 μL of a 40 mg/mL O-methoxylamine hydrochloride (MOX) (MP Bio #155405) in dry pyridine (EMD Millipore #PX2012–7) and incubated for one hour at 37 °C in a sand bath. 25 μL of this solution was added to auto sampler vials. 60 μL of N-methyl-N- trimethylsilyltrifluoracetamide (MSTFA with 1%TMCS, Thermo #TS48913) was added automatically via the auto sampler and incubated for 30 minutes at 37 °C. After incubation, samples were vortexed and 1 μL of the prepared sample was injected into the gas chromatograph inlet in the split mode with the inlet temperature held at 250°C. A 1:1 split ratio was used for this analysis. The gas chromatograph had an initial temperature of 60°C for one minute followed by a 10°C/min ramp to 325°C held fo r 5 min. Then a 100°C/min ramp to 375°C and a final hold time of 5 minutes. A 30-meter Zebron ZB-5HT with 5m inert guard capillary column was employed for chromatographic separation. Helium was used as the carrier gas at a rate of 1 mL/min. Peak picking was performed using Agilent MassHunter software and isotopically corrected using a custom version of AccuCor that incorporated silicon. No subjects or data was excluded from analysis.

Mouse tissue gene expression profiling.

Whole crypts samples were selected either as untreated normal crypts or as adjacent normal crypts selected at random with inclusion criteria as an adenoma from the same mouse included. Whole crypts were extracted for RNA using QIAGEN RNeasy kit. Two methods were used to obtain tumor tissue for molecular analysis. A subset of tumor tissue was micro-dissected from formalin-fixed paraffin embedded colon samples, and RNA was extracted using Qiagen FFPE RNeasy kit (cat: 73504). A second set of tumor tissue was dissected from the normal mucosa under an Olympus SZ61 dissecting microscope in ice cold PBS, then snap frozen in liquid nitrogen. RNA was extracted using the Qiagen RNeasy (cat: 74104). FFPE extracted RNA and frozen tissue RNA were comparable in quality and concentration, and used in parallel in subsequent studies. Gene expression was profiled by transcript abundance using the NanoString Elements platform according to the manufacturer’s protocols. The custom 22 gene probe set, described previously, includes intestinal specific stem and differentiation markers, MPC1 and MPC2 (Schell et al. 2017). Using the NanoString nSolver 4 analysis software, raw counts were corrected by background thresholding with geometric mean factor normalization, and genes were mean centered for heatmap generation, and two adenomas from AOM-DSS experiments were excluded for raw and normalized gene expression with geometric means deviating 2-fold from the dataset average. Plots were generated for principle component analysis using XPRESSplot (Berg 2019 bioRxiv). Gene normalizations were performed with scikit-learn (v0.21.3) (Buitinck et al. 2013), and other statistical measures were calculated using SciPy (v1.3.0) (Jones et al. 2019) and StatsModels (v0.10.1) (Seabold et al. 2010) within the Python (v3.7.3) environment. A perpetually-available archive of the data and code used for pertinent analyses and figures not generated by Prism GraphPad (v7) is available at https://doi.org/10.5281/zenodo.3377989 (Berg 2019).

Human Datasets for normal mucosa, adenomas, adenocarcinoma:

Human data was accessed from the GEO database (GSE20916 and GSE8671) under inclusion parameters of normal tissue and non-cancer adenoma tissue. Unified GTEx and TCGA data for colon cancer (COAD) were obtained from (Wang et al. 2018). For microarray datasets, multimapping probes were dropped and probe sets mapping to the same gene were collapsed. For the GTEx/TCGA data, expression levels were mean-centered (z-score) and any z-score > 5 for the analyzed gene set was thresholded at 5 to prevent high signal from concealing expression data falling within a more “normal” range. This can be explored more in the interactive notebook (Berg 2019). Plots were generated and principle component analysis were performed using XPRESSplot (Berg 2019 bioRxiv). Gene normalizations were performed with scikit-learn (v0.21.3) (Buitinck et al. 2013), and other statistical measures were calculated using SciPy (v1.3.0) (Jones et al. 2019) and StatsModels (v0.10.1) (Seabold et al. 2010) within the Python (v3.7.3) environment. A perpetually-available archive of the data and code used for pertinent analyses and figure generation not generated by Prism GraphPad (v7) is available at https://doi.org/10.5281/zenodo.3377989 (Berg 2019).

The custom pyruvate metabolic gene set was composed of the KEGG pyruvate metabolism annotation with select genes included from KEGG Glycolysis/Gluconeogenesis, KEGG TCA cycle, KEGG Monocarboxylate transporters, and glucose transporters. This list thus includes relevant enzymes and transporters, core and offshoot pathways of pyruvate metabolism to better address whether the pyruvate metabolic gene program shifts during tumor initiation. A complete gene list as well as the origin lists are available at https://doi.org/10.5281/zenodo.3377989 (Berg 2019).

A linear least-squares regression was used to calculate correlation coefficients for Mpc1 or Mpc2 against every gene and r >/= 0.60 (p-value < 0.01) was used as a cut-off. Correlated and anti-correlated gene lists were analyzed by DAVID (v6.8) and PANTHER (v13.1) to determine enriched pathways within these gene cohorts. −log₁₀(p-values) were reported. All code necessary to reproduce the Figures and select tables can be found at https://doi.org/10.5281/zenodo.3377989 (Berg 2019). Resulting pathway lists were parsed for pathways relevant to stem cell maintenance, growth signaling, cancer pathways, and colon-specific pathways, with a p-value <0.05. Correlation Gene Lists with Mpc1 and Mpc2, and table summaries from DAVID and PANTHER are available at https://doi.org/10.5281/zenodo.3377989 (Berg 2019).

QUANTIFICATION AND STATISTICAL ANALYSIS

Prior to rodent studies and in accordance with IACUC protocols, we performed a power analysis for both the AOM-DSS and Apc loss of heterozygosity mouse experiments to determine cohort sizes of 30 and 10, respectively including both male and female mice. All mouse studies were performed in smaller rolling cohorts of 5–7 mice per time-point and no cohort to cohort variation could account for the phenotypes observed. All assays and studies, excluding the NanoString gene expression assay, were performed with technical replicates (n=3) or with multiple blinded researchers where appropriate. In the NanoString gene expression assays, a minimum of biological replicates (n=3) were examined. All statistics and significance unless otherwise stated were performed in Prism GraphPad (v6–7), using unpaired student t-test’s, Chi-square tests, and one-way ANOVA with multiple comparisons with Sidak multiple test correction where appropriate. Other analysis archived at https://doi.org/10.5281/zenodo.3377989 used an FDR-corrected Wilcoxon rank-sums test for statistical significance. The lines and error bars depict the average ± the standard deviation (Berg 2019).

DATA AND CODE AVAILABILITY

The code and data are available at https://doi.org/10.5281/zenodo.3377989 (Berg 2019). This study did not generate the human data, which was accessed from the GEO database (GSE20916 and GSE8671) and unified GTEx and TCGA data for colon cancer (COAD) were obtained from (Wang et al. 2018). The Nanostring mouse adenoma and crypt data generated by this study are available on the GEO database (GSE136710).

Highlights.

• Intestinal tumors exhibit low Mitochondrial Pyruvate Carrier (MPC) expression

• MPC inactivation is sufficient to promote intestinal tumor formation in mice and flies.

• MPC overexpression in the fly is sufficient to prevent oncogene-induced tumorigenesis.

• MPC expression correlates negatively with expression of Wnt/β-catenin target genes.

Context and Significance.

The scientific community has clearly demonstrated in recent years that cancer cells exhibit atypical metabolic characteristics and can, in many cases, become dependent on those features. Here, researchers at the University of Utah aimed to determine whether cancer initiation also involves a change in cellular metabolic state. Indeed, the authors found that enforcing a more “tumorigenic” metabolic behavior led to an increased incidence of tumors in both mouse and fly models. Moreover, they found that prevention of this “tumorigenic” metabolic profile can block cancerous growth in response to otherwise strongly tumorigenic stimulation. These results demonstrate that the cellular metabolic program is necessary and sufficient to impact the incidence of tumor formation and lays the foundation for future work to determine how changes in this program can decrease human tumor formation.