Epigenetic reprogramming induced by key metabolite depletion is an evolutionarily ancient path to tumorigenesis Open Access

Cancer is not a disease that only affects humans. The appearance of tumors seems to coincide with the origin of metazoans. Various tumor types have been documented in organisms ranging from hydra, clams, worms, flies and fish to mammals (Athena Aktipis et al., 2015; Scharrer and Lochhead, 1950). Cancers have even been identified in dinosaur fossil records (Ekhtiari et al., 2020; Rothschild et al., 1998, 1999). Tumors in Drosophila melanogaster were first reported in 1918 (Stark, 1918), and they represent one of the best-understood tumor models in species outside mammals (Villegas, 2019). For example, three conserved cell polarity genes – scribble (scrib), lethal giant larvae [lgl; also known as l(2)gl] and discs-large (dlg; also known as dlg1) – are collectively named as neoplastic tumor suppressor genes (Bilder, 2004; Bilder and Perrimon, 2000; Bilder et al., 2000; Gateff, 1978). When fly larvae are homozygous mutant for scrib, lgl or dlg, their imaginal discs and optical lobes, which are proliferative epithelial organs in larvae, grow into amorphous masses and kill the hosts within days (Figueroa-Clarevega and Bilder, 2015; Gateff, 1978). Even for well-studied fly tumors, it is unclear how closely they resemble human tumors, because these fly tumors reach immortality within days while carrying only one mutation. In contrast, human cancers typically develop over years and acquire multiple mutations to enter the path of carcinogenesis.

Cancer cells are generally thought as being in metabolic states distinct from those of normal cells (De Berardinis and Chandel, 2016; Finley, 2023). For example, most cells can use glucose as an energy source through conserved steps of the oxidation of glucose to pyruvate, the conversion of pyruvate to acetyl groups and the further oxidation of acetyl groups through the citric acid cycle (Nelson and Cox, 2017). Warburg showed that tumor tissue slices produced lactate from glucose even in the presence of ample oxygen, which is now known as a metabolic signature of cancer cells (the Warburg effect) (Finley, 2023; Heiden et al., 2009; Warburg, 1956). Genomics and functional studies have identified several metabolic enzymes frequently mutated in human, which support tumor growth. For example, mutations in isocitrate dehydrogenase genes (IDH1 and IDH2) are frequently observed in gliomas and acute myeloid leukemia (Mardis et al., 2009; Yan et al., 2009). IDH1 and IDH2 mutations lead to the accumulation of 2-hydroxygulatrate, which inhibits α-ketoglutarate-dependent dioxygenases and alters the histone and DNA methylation pattern (Chowdhury et al., 2011; Dang et al., 2009; Lu et al., 2012; Xu et al., 2011). Mutations in fumarate hydratase (FH) and succinate dehydrogenase (SDH), which lead to accumulation of fumarate and succinate, predispose patients to various cancer types (Baysal et al., 2000; Burnichon et al., 2010; Tomlinson et al., 2002; Toro et al., 2003). The accumulation of fumarate and succinate can suppress DNA repair pathway (Sulkowski et al., 2018, 2020) and inhibit phosphatase and tensin homolog (PTEN) through succination (Ge et al., 2022). Phosphoglycerate dehydrogenase (PHGDH) is frequently amplified in breast cancer and melanoma samples and increases serine flux to support tumor growth (Locasale et al., 2011; Possemato et al., 2011; Sullivan et al., 2019). Metabolic processes are deeply conserved across species, and it is unknown whether metabolic changes underlying tumorigenesis in different metazoan species share any commonalities.

In this study, we first noticed that a population of Lactate dehydrogenase (Ldh)⁺ cells emerges in fly tumors over time, resembling the Warburg effect observed in mammalian tumors. We further found that fly tumors exhibit major changes in glycolysis, tricarboxylic acid (TCA) cycle and oxidative phosphorylation processes, which lead to systemic depletion of acetyl-CoA. Consequently, we observed reduction in H3K9ac and H3K27ac levels and stochastic silencing of actively transcribed genes in these tumors. Moreover, these fly tumors also exhibit defects in the methionine metabolism process, which leads to systemic depletion of the methyl donor S-adenosyl methionine, reduction in histone methylation levels and de-repression of transposons. We then performed comparative studies between fly and human tumors using metabolic signatures and identified that human tumors with high similarity to fly tumors have a lower mutational load, younger patient age and lower DNA methylation levels.

The fly tumor models caused by the loss of conserved cell polarity genes, including scrib, dlg and lgl, are among the best-understood tumor models outside of mammals (Bilder, 2004; Bilder and Perrimon, 2000; Bilder et al., 2000). Note that the scrib and dlg mutants are phenotypically identical, as their encoded proteins function as a complex (Bilder and Perrimon, 2000; Bilder et al., 2000). We used the scrib and dlg mutant interchangeably in experiments, and the choice of which mutant to use was based on the ease of genetic combination. We have previously used single-cell transcriptomics to profile the fly scrib mutant wing disc tumors as they grew from 4 days to 14 days (Deng et al., 2019; Ji et al., 2019). From this dataset, we noticed that a population of Ldh⁺ cells emerged in the late-stage scrib mutant tumors (Fig. 1A). We further validated the existence of this Ldh⁺ cell population by examining the expression of a Ldh-GFP enhancer trap line (Wang et al., 2016) in the dlg mutant (Fig. 1B,C). Ldh⁺ cell populations were previously shown to be induced in fly wing disc tumors caused by active Pvr (Wang et al., 2016) or Hipk (Wong et al., 2019) signals. Similar to what has been shown in these contexts, the upregulation of the Ldh⁺ cell population in the scrib/dlg mutant tumors was also regulated by HIF-1 (also known as HIF) and PI3K/Akt signals (Fig. S1). Specifically, depletion of HIF-1 through RNA interference (RNAi) blocked the increase in Ldh⁺ cells (Fig. S1A), and activation of PI3K/Akt signaling through Pten RNAi promoted the expansion of Ldh⁺ cells (Fig. S1B).

We further examined whether the expansion of a Ldh⁺ glycolytic cell population led to elevation of lactate levels in dlg mutant tumors using a biosensor for L-lactate, R-iLACCO1 (Nasu et al., 2023). In the control wild-type wing discs, we observed weak fluorescence level at the hinge region (Fig. 1D). In the 14-day dlg mutant tumors, we observed bright red fluorescence signals (Fig. 1D). Notably, the R-iLACCO1 signal was frequently observed in cells adjacent to the Ldh-GFP⁺ cells. On average, only half of the R-iLACCO1⁺ cells were also positively expressing Ldh (Fig. 1D′), indicative of possible lactate transfer among these cells (Brooks, 2018). Together, these data suggest that the scrib and dlg mutant tumors exhibit characteristics of the Warburg effect as these tumors grow over time.

Glycolysis, the citric acid cycle (TCA cycle) and oxidative phosphorylation are tightly coupled processes for glucose breakdown and ATP production (Fig. 2A). In addition to the increase in Ldh⁺ cells in these tumors, we further noticed that genes responsible for the breakdown of glucose to pyruvate were collectively upregulated over time in the scrib mutant tumors (Fig. 2A; Fig. S2A). The pyruvate dehydrogenase complex genes, which convert pyruvate to acetyl-CoA for entry into the TCA cycle, were collectively downregulated over time in the scrib mutant tumors (Fig. 2A; Fig. S2B). Concurrently, TCA cycle genes were downregulated over time in the scrib mutant tumors (Fig. 2A; Fig. S2C). Moreover, genes encoding components of mitochondrial respiratory chain complexes I-V were collectively downregulated over time in the scrib mutant tumors (Fig. 2A; Fig. S2D). These data suggest that the breakdown of glucose through glycolysis, the TCA cycle and oxidative phosphorylation to generate ATP likely slows down at the step of pyruvate-to-acetyl-CoA conversion for entry into the TCA cycle in late-stage scrib mutant tumors.

We then compared the concentrations of metabolites between the dlg mutant larvae and control larvae through metabolomics profiling (Fig. 2A; Table S1). We observed elevated levels of metabolites in the TCA cycle – including citrate, cis-aconitate, isocitrate, succinate, fumarate and malate – in the dlg mutant larvae (Fig. 2A), likely indicative of an overall slowdown of metabolite flow through the TCA cycle. Notably, in human cancers caused by mutations in IDH1, FH or SDH, the accumulation of metabolites in the TCA cycle, such as succinate and fumarate, is also frequently observed (Chowdhury et al., 2011; Dang et al., 2009; Ge et al., 2022; Sulkowski et al., 2020; Xu et al., 2011). In fly brain tumors caused by the loss of Brat protein, the accumulation of TCA intermediates was also observed (Bonnay et al., 2020), highlighting that the accumulation of TCA intermediates might represent a highly conserved feature of tumors from different animal species.

Consistent with the downregulation of expression levels of PDH complex genes, which function to convert pyruvate to acetyl-CoA for entry into the TCA cycle, we observed a reduction in acetyl-CoA levels in the dlg mutant larvae (Fig. 2B). Acetyl-CoA not only serves as a metabolic intermediate for TCA cycle and lipid synthesis but also acts as the acetyl donor for protein acetylation, particularly histone acetylation (Diehl and Muir, 2020; Pietrocola et al., 2015; Shi and Tu, 2015). We then examined whether depletion of acetyl-CoA in the dlg mutant larvae affects protein acetylation levels, in particular, histone acetylation levels. Histone post-translational modifications H3K27ac and H3K9ac are classic markers for actively transcribed chromatin (Nègre et al., 2011; modENCODE Consortium et al., 2010; Tie et al., 2009, 2014). We observed that H3K27ac and H3K9ac levels were much lower in dlg mutant wing disc tumors than in wild-type wing discs (Fig. 2C,D). Removal of histone acetylation markers can potentially lead to the silencing of actively transcribed genes (Gallinari et al., 2007). To examine the consequences of an overall reduction in histone acetylation levels in dlg mutant tumors, we adopted a tubGal80-tubGal4 reporter system originally used to monitor loss of heterozygosity in aging intestinal stem cells (Siudeja et al., 2015). In this system, the inactivation of a tubulin (tub) promoter-driven Gal80 allows the expression of GFP activated by tub promoter-driven Gal4 (Siudeja et al., 2015). The inactivation of tubGal80 can result from either DNA mutations or deletions, or from epigenetic silencing in the tubGal80 locus (Siudeja et al., 2015). Using this system, we found that GFP⁺ cells were never observed in the control imaginal discs, indicating that the tubGal80-tubGal4 reporter system is not leaky (Fig. 2E). However, we frequently observed GFP⁺ clones in the 14-day dlg mutant tumors (Fig. 2E). We have previously performed whole-genome sequencing of dlg mutant tumors and did not detect further mutations except the original dlg mutation (Ji et al., 2019). Therefore, the stochastic loss of Gal80 likely results from the epigenetic silencing of the tubGal80 locus in association with the loss of histone acetylation markers.

We next examined additional metabolic changes observed in the dlg mutant larvae in comparison with the wild-type larvae and noticed a significant reduction in methionine cycle-related metabolites, such as S-adenosyl methionine, in the dlg mutant larvae (Fig. 3A,B). S-Adenosyl methionine is a major methyl donor. Therefore, we examined whether histone methylation status was affected in the dlg mutant tumors. We found that the level of H3K9me3, a well-recognized repressive chromatin marker (Nègre et al., 2011; modENCODE Consortium et al., 2010), was significantly reduced in the dlg mutant tumors in comparison with the control (Fig. 3C,D).

Removal of silencing markers such as H3K9me3 can potentially lead to transposon activation and loss of gene silencing in the heterochromatin region (Wood et al., 2016). We examined the transcript levels of multiple transposons in the scrib mutant tumors and observed de-repression of many transposons in the 14-day scrib mutant tumors (Fig. 3E; Table S2). Consistently, we performed whole-genome sequencing for the 4-day, 8-day and 14-day scrib mutant tumors, mapped transposition events and noticed that transposition events were more frequent in 8-day and 14-day mutant tumors in comparison with those in 4-day mutant tumors (Fig. 3F; Table S3). We then adopted a reporter for transposon activation, I-element-eGFP reporter, which only produces GFP signal upon retrotransposition (Wang et al., 2018). In the control wing imaginal discs, we never observed any GFP signal, indicative of tight repression of transposons in the wild-type somatic cells (Fig. 3G). In the 14-day scrib mutant tumors, we observed GFP⁺ cells in ∼50% of samples we examined (Fig. 3G), indicating that the reporter was not only reverse transcribed but its cDNA often successfully landed back into the genome (Wang et al., 2018). Together, these results suggest that the late-stage scrib/dlg mutant tumors likely harbor an increasingly diverse cell population with stochastic transposon activation. To determine whether depletion of S-adenosyl methionine is a driving force for tumorigenesis, we examined the functional consequences of perturbing enzymes in the methionine cycle in the dlg mutant tumors. Interestingly, we found that knockdown of AhcyL1 and AhcyL2, which potentially elevates S-adenosyl homocysteine hydrolase Ahcy activity (Parkhitko et al., 2016), elevated H3K9me3 levels and strongly inhibited dlg mutant tumor growth (Fig. 3H). Note that perturbation of AhcyL1 and AhcyL2 activity in normal wing discs did not affect normal developmental growth (Fig. S3).

Major metabolic pathways are highly conserved across animal species. Using Kyoto Encyclopedia of Genes and Genomes (KEGG) and Metabolic Atlas as references, we annotated 72 metabolic pathways conserved between fly and human (Table S4). We then use transcriptomics data from the scrib mutant tumor samples (Ji et al., 2019) and 10,501 human tumor samples in The Cancer Genome Atlas (TCGA) database to perform single-sample gene-set enrichment analysis (ssGSEA) for the 72 conserved metabolic pathways (Fig. 4A). Using metabolic pathway enrichment scores as a matrix, we further performed correlation analysis to identify human tumor samples with high metabolic similarity to fly tumors (Fig. 4A). In the top 7% human tumors with high metabolic similarity to fly tumors (Fig. 4B,C) (n=765/10,501), we identified younger patient age (Fig. 4D), a lower mutational load (Fig. 4E) and a lower level of DNA methylation (Fig. 4F), in comparison with tumor samples classified as non-similar to fly tumors. The human tumor samples that metabolically resemble fly tumors showed strong enrichment for mutations in IDH1, GTF2I and ATRX (Fig. 4G). GTF2I encodes a transcriptional factor with high mutation frequency observed in thymoma (Petrini et al., 2014). ATRX is a chromatin remodeler with high mutation frequency in glioma (Koschmann et al., 2016). Consistently, the tumor samples with high similarity to fly tumors were highly enriched in eight cancer types out of the total 33 cancer types, including B-cell lymphoma, glioblastoma, glioma, ovarian adenocarcinoma, testicular germ cell tumor, thymoma and uterine carcinosarcoma (Fig. S4). For the eight cancer types with a high percentage of tumors metabolically similar to fly tumors, the patient survival curves did not show significant differences between the two groups, except for glioma, for which patients with tumors metabolically similar to fly tumors had better survival outcomes (Fig. S5).

To further validate these results, we obtained an independent cohort of 185 Chinese Glioma Genome Atlas (CGGA) glioma tumor data and separated them into two groups based on similarity of their metabolic signature to that of fly tumors (Fig. S6A,B). Similarly, for this independent dataset, the patients in the similar group were significantly younger than those in the non-similar group (Fig. S6C). The human glioma samples that metabolically resembled fly tumors harbored a significantly lower number of mutations (Fig. S6D) and showed strong enrichment for IDH1 mutations (Fig. S6E), and the patients in the similar group had better survival outcome (Fig. S6F).

Together, these data show that human tumor samples sharing similar metabolic signatures with fly tumors harbor fewer mutations, which suggests that they may be less dependent on mutation accumulation over time for malignant growth.

Here, we demonstrated that epigenetic reprogramming induced by key metabolite depletion is an evolutionarily ancient driving force for immortal growth, lowering the barrier to tumorigenesis and the need for mutation accumulation over time. Interestingly, a recent study also demonstrated that the transient silencing of Polycomb complex is sufficient to drive malignant tumor formation in Drosophila (Parreno et al., 2024). Notably, even in human cancers, ∼5.3% of tumors do not harbor any known cancer-driver mutations (ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium, 2020). For example, mutations are rarely found in childhood brain tumors (Mack et al., 2014). Therefore, epigenetic changes induced by metabolic stress may represent an evolutionarily ancient route to malignant growth, which co-exist with the acquisition of multiple mutations in long-living species. For the portion of evolutionarily ancient human tumors, it will be interesting to use fly tumors as a model to test therapeutic interventions. For example, we have found that inhibition of AhcyL1 and AhcyL2, enzymes functioning in the methionine metabolic process, potently inhibits fly tumor growth. In the future, it would be interesting to test whether methionine metabolic intervention benefits patients with evolutionarily ancient tumors.

The metabolic network is largely conserved among animal species. The accumulation of TCA intermediates is observed in both human and fly tumors. Although many studies have focused on the accumulation of oncometabolites (Liu and Yang, 2021), our work here highlights the consequences of depletion of metabolites, including acetyl-CoA and S-adenosyl methionine, which are key substrates for histone post-translational modifications. Notably, we only examined a few general histone acetylation and methylation markers and observed a global reduction in these tumors. Whether specific histone modifications or specific gene loci are resistant to the depletion of metabolites is unclear at this stage. Similar to the concept of oncogenes and tumor suppressor genes, the depletion of metabolites is likely as important as the accumulation of oncometabolites in tumorigenesis processes. A global reduction in histone acetylation and methylation levels can lead to stochastic gene silencing and activation, the implications of which need to be further explored in human tumors.

Drosophila melanogaster stocks were maintained at room temperature, and fly crosses were raised in a 25°C fly incubator. The fly strains used in this study were as follows: scrib¹ FRT82B/TM6B (Bilder and Perrimon, 2000), dlg^GH19 FRT19A/FM7C (Ji et al., 2019), c855aGal4 [Bloomington Drosophila Stock Center (BDSC 6990)], Ldh-GFP (Wang et al., 2016), R-iLACCO1 (Nasu et al., 2023), I-element-eGFP-reporter (Wang et al., 2018), FRT40A tubGal80; tubGal4 UAS-GFP (a gift from the Xi laboratory, National Institute of Biological Sciences, Beijing, China), HIF-1 RNAi (BDSC 26207), Pten RNAi [Vienna Drosophila Resource Center (VDRC) 101475], AhcyL1 RNAi (BDSC 28523), AhcyL2 RNAi (BDSC 61913) and control strains (OreR or w1118).

Wing discs and tumors were dissected, fixed and stained following standard protocols. The primary antibodies used in this study were rabbit anti-H3K9ac (Abcam, ab4441, 1:1000), rabbit anti-H3K27ac (Abcam, ab4729, 1:1000) and rabbit anti-H3K9me3 (Abcam, ab8898, 1:1000). Secondary antibodies used were Goat anti-Rabbit IgG (H+L) Secondary Antibody, Alexa Fluor™ 546 (Invitrogen, A11035, 1:1000) and Goat anti-Mouse IgG (H+L) Secondary Antibody, Alexa Fluor™ 546 (Invitrogen, A11030, 1:1000). Hoechst 33342 (Invitrogen, H3570, 1:10,000) and Alexa Fluor™ 647 Phalloidin (Invitrogen, A22287,1:1000) were used to stain DNA and actin for cell outline, respectively.

Forty pairs of wing imaginal discs or tumors were dissected in PBS. Tissue samples were vortexed in RIPA lysis buffer for 1 min, followed by sonication. The sonication program settings included an amplitude of 50, a processing time of 15 min, a pulse-on time of 15 s and a pulse-off time of 10 s. After cell lysis, the samples were centrifuged at 18,400 g for 10 min. The supernatant was then measured for protein concentration using a Pierce BCA Protein Assay Kit. The primary antibodies used were rabbit anti-H3K9ac (Abcam, ab4441, 1:1000), rabbit anti-H3K9me3 (Abcam, ab8898, 1:1000), rabbit anti-H3K27ac (Abcam, ab4729, 1:1000) and mouse anti-gamma-tubulin (Abcam, ab27074, 1:1000).

Images were acquired on a Leica TCS SP8 confocal microscope. Wing discs and tumors were taken as z-stacks for volume quantification using ImageJ and CellProfiler (Stirling et al., 2021).

Bulk and single-cell RNA-sequencing data on scrib mutant tumors were obtained from our previous publications (Deng et al., 2019; Ji et al., 2019) (Gene Expression Omnibus accession numbers GSE130243 and GSE130566). The gene expression data for 10,501 tumors of 33 cancer types were obtained from TCGA Data Portal. The metabolic pathway annotation tables were manually curated from the KEGG database, the Human-GEM and the Fruitly-GEM Metabolic Atlas (Wang et al., 2021).

For a given tumor sample, an enrichment score for a specific metabolic pathway was calculated through ssGSEA based on gene expression data. The similarity between a given human tumor sample and fly scrib mutant tumor was then calculated by Pearson correlation test using the enrichment scores of 72 conserved metabolic pathways. We set a cutoff on adjusted P-values of correlation coefficients to classify human tumors into metabolically similar and non-similar groups. We then used available somatic mutation information from TCGA and applied Fisher's exact test to examine significantly enriched or lost mutation events in human tumors metabolically similar to fly tumors in comparison with human tumors metabolically non-similar to fly tumors. The processed pan-cancer expression data and somatic mutation information of TCGA were provided by University of California, Santa Cruz Xena. To verify our findings using the TCGA dataset, we applied the same methods on the expression and mutation data of glioma samples from CGGA. The gene expression data and somatic mutation information of CGGA are publicly available at the CGGA website.

For bulk RNA-sequencing analysis, the transcript abundance (as denoted in fragments per kilobase per million reads) of transposons was counted by using piPipes as described in Han et al. (2015). Gene expression heatmaps were generated using the pheatmap package, and principal component analysis was performed using FactoMineR and ggplot2 packages. Single-cell data merging and gene expression plot generation were performed using Seurat package (Hao et al., 2024).

The scrib mutant tumors of different stages were dissected in PBS. Genomic DNA was extracted using a DNeasy Blood and Tissue Kit (Qiagen, 69504). Paired-end sequencing was performed on an Illumina^® sequencing platform, with a read length of 150 bp at each end. We first mapped raw whole-genome sequencing data to the dm6 genome to estimate mappability via BWA v0.7.17 and randomly selected sequencing reads of each library by seqtk to obtain the same genome coverage for comparison. The transposition events of each sample were calculated by TEMP (Zhuang et al., 2014).

We thank the BDSC, VDRC, Dr Chris Doe, Dr Trudi Schupbach, Dr Rongwen Xi, Dr Zhao Zhang and Dr Robert Campbell for fly stocks.