|Home | About | Journals | Submit | Contact Us | Français|
Poorly organized tumour vasculature often results in areas of limited nutrient supply and hypoxia. Despite our understanding of solid tumour responses to hypoxia, how nutrient deprivation regionally affects tumour growth and therapeutic response is poorly understood. Here, we show the core region of solid tumours displayed glutamine deficiency compared to other amino acids. Low glutamine in tumour core regions led to dramatic histone hyper-methylation due to decreased α-ketoglutarate levels, a key cofactor for the Jumonji-domain containing (JmjC) histone demethylases (JHDMs). Using patient-derived V600EBRAF melanoma cells, we found that low glutamine-induced histone hyper-methylation resulted in cancer cell de-differentiation and resistance to BRAF inhibitor treatment, which was largely mediated by methylation on H3K27, as knockdown of the H3K27-specific demethylase KDM6B and methyltransferase EZH2 respectively reproduced and attenuated the low glutamine effects in vitro and in vivo. Thus, intra-tumoural regional variation in the nutritional microenvironment contributes to tumour heterogeneity and therapeutic response.
Glutamine is one of the major carbon and nitrogen sources to support cancer cell survival and proliferation1. Glutamine catabolism is required to maintain pools of TCA cycle intermediates. For example, glutamine can be converted by glutaminase (GLS) to glutamate, which can be further converted to α-ketoglutarate in the TCA cycle2. Similar to glucose metabolism, increased glutamine uptake is controlled by oncogenes, such as c-MYC and KRAS3–5. On the other hand, as tumours grow, increased glutamine catabolism may deplete the local supply and lead to periods of glutamine deprivation. This is supported by in vivo studies wherein numerous tumours, including hepatomas and sarcomas, glutamine falls to almost undetectable levels relative to normal tissues6. A recent study using metabolomics analysis comparing paired pancreatic tumour patient samples with benign adjacent tissue specimens revealed that glutamine is one of the most strongly depleted metabolites in tumours7. However, how low glutamine levels in solid tumours affects tumour growth and therapeutic response remains largely unknown.
Reversible histone lysine methylation controlled by a variety of histone methyl-transferases (HMTs) and histone demethylases (HDMs) modulates chromatin structure and thereby contributes to a variety of cellular processes such as transcription, replication and repair8. A subset of HDMs named Jumonji-domain containing HDM (JHDM) utilizes α-ketoglutarate (αKG), oxygen and Fe (II) as cofactors and releases succinate and formaldehyde as by-products9. The requirement for αKG in mediating histone demethylase activity implied a potential interplay between metabolism and epigenetic modification. For example, a recent report demonstrated that glutamine deprivation in ES cells leads to increased methylation on H3K27me3 and H3K9me310. In addition, mutant forms of the metabolic enzymes isocitrate dehydrogenase 1 (IDH1) and IDH2, which display neomorphic functions by producing 2-hydroxyglutarate (2HG) from αKG11, can increase histone methylation and block tumour cell differentiation via inhibiting JHDM activity, as 2HG is a structural analog of αKG12. However, although αKG is a metabolite generated from glutamine, whether low glutamine levels in solid tumours affect the activity of these αKG-dependent JHDM and hence modulate histone methylation remains unclear.
In this study, we measured the extent of glutamine heterogeneity regionally within solid tumours, dissected the impact of low glutamine on tumour cell de-differentiation and drug sensitivity, and mechanistically linked specific histone demethylation to the impact of microenvironmental glutamine levels on tumour cell plasticity. Our study provides important evidence that regional glutamine deficiency leads to de-differentiation and drug resistance via inhibition of histone demethylation.
To assess potential differences in glutamine levels in the periphery vs. core region of tumours, we dissected different tumour xenografts and measured the glutamine concentrations in these distinct intra-tumoural regions (Supplementary Fig. 1a). We found that glutamine concentrations were consistently and significantly lower in the tumour core regions compared to the tumour periphery (Fig. 1a). We next measured the αKG level in the tumour tissues by liquid chromatography mass spectrometry (LC-MS) analysis. As with glutamine, compared to the periphery regions, αKG levels were significantly decreased in the tumour cores (Fig.1b). Moreover, αKG level was largely restored by injecting glutamine into the tumour cores, suggesting glutamine is sufficient to maintain tumour αKG levels (Fig. 1c). Since αKG is an essential co-factor for the JHDMs, we tested whether low glutamine levels in the tumour core regions correlated with differential histone methylation levels. At all the histone H3 methylation sites, which are the targets of JHDMs13–15, histone methylation levels dramatically increased in the core region (Fig. 1d). HIF-1α was used as a “marker” of the core tissues as it is induced in the core region of solid tumours due to hypoxia16. In addition, KRAS expression was used to distinguish tumour tissues from tumour adjacent normal tissues. We further examined histone methylation using melanoma tumours spontaneously developed from mice carrying a Braftm1Mmcm/Ptentm1Hwu allele17. Similarly, increased histone methylation was found in the core regions compared to the periphery in tumours from transgenic mice (Fig. 1e). In addition, histone hyper-methylation in the tumour core regions was also confirmed by immunohistochemistry staining using antibody against H3K27me3 or H3K9me3 (Fig. 1f). Moreover, we directly visualized H3K27me3 levels in intact tumours using the recently established technology for intact, slice-free, whole tissue imaging and phenotyping (PACT-defined as Passive CLARITY)18 (Supplementary Fig. 1b). In agreement with Western blot and IHC analysis, H3K27me3 staining significantly increased in the melanoma tumour core regions (Fig. 1g). Furthermore, HIF-1α and H3K27me3 staining strongly overlapped (Supplementary Fig. 1c). Thus, low glutamine and high histone methylation levels commonly co-exist in the core regions of tumours. Tumour cores often contain extensive dead cells. Despite the increased cell death (Supplementary Fig. 1d,e), there were still a certain amount of live cells in the tumour cores as cells in tumour cores could be stained by Ki67 antibody and displayed a negative staining of cleaved caspase-3 (Supplementary Fig. 1c,d).
Next we determined if increased histone methylation in tumour core regions was induced by glutamine deficiency. We found that low glutamine was sufficient to induce histone hyper-methylation, particularly on K9, K27 and K36, in both a time and dose-dependent manner (Fig. 2a,b). Moreover, histone hyper-methylation required 0.2 mM for melanoma cells (Fig. 2b), which was similar to the glutamine concentration in tumour core regions (Fig. 1a). Furthermore, cell permeable dimethyl-αKG was sufficient to prevent low glutamine-induced histone methylation in all tumour cell types tested (Fig. 2c), suggesting that increased histone methylation induced by low glutamine levels resulted from low αKG. As a control, intracellular glutamine and αKG under these conditions were also measured (Fig. 2d). Furthermore, melanoma cells treated with glutaminase inhibitors L-DON and compound 96819, 20 also displayed increased histone methylation (Fig. 2e). As expected, DON treatment led to an 80% decreased αKG level (Fig. 2f). Interestingly, we found that the glutamine level decreased by 40% upon DON treatment. In addition, to understand whether cell proliferation rate had any effect on histone methylation, we examined H3 lysine methylation in confluent cells and observed no increase in histone methylation as compared to cells cultured in low glutamine (Supplementary Fig. 2a). Furthermore, cell number and cell survival were assayed under these conditions, but no correlation between cell proliferation and histone methylation was found (Supplementary Fig. 2b,c).
To assess the extent of selectivity of nutritional deficiency in the core regions of tumours, we measured the concentration of all amino acids in the core vs. peripheral regions by LC-MS. Interestingly, only the levels of five amino acids (including glutamine) were significantly lower in the core relative to the peripheral regions (Table 1), and this result was confirmed again by another independent metabolomics analysis (Supplementary Table 1). We further determined which relative amino acid deficiency was most responsible for increased histone methylation using medium that contained the corresponding concentration of each decreased amino acid (although the levels in cell culture may drop quickly due to consumption) and found only low glutamine levels resulted in increased histone methylation (Fig. 3a). In addition, we found histone methylation was only induced when glutamine was low, but not under a combinatorial deficiency of the other four amino acids (Arg, Ser, Asp, Asn) (Supplementary Fig. 3a). Moreover, other forms of metabolic stress had no significant effect on histone methylation levels (Fig. 3b and Supplementary Fig. 3b). Interestingly, the combination of low glutamine and hypoxia resulted in greater histone methylation than either alone, particularly on H3K4me3, suggesting that hypoxia is not the only mechanism contributing to this phenomenon in solid tumours (Fig. 3b).
Moreover, we found that glutaminase inhibitor, compound 968, led to dramatic histone methylation in the tumour periphery, with levels similar to the core region of control treated tumours (Fig. 3c). Using PACT, we found that, in contrast to the untreated tumour in Fig. 1g, 968-treated tumour displayed higher levels of H3K27me3 methylation in both tumour periphery and core regions (Fig. 3d). In addition, increased histone methylation was largely attenuated by intra-tumour injections of glutamine, particularly on H3K9 and H3K27 (Fig. 3e). Taken together, these data demonstrate that low levels of glutamine are the major contributor to increased histone methylation in the tumour core regions.
Previous reports indicate that histone hyper-methylation induced by IDH mutants prevents cell differentiation via inhibiting αKG-dependent histone demethylases12. Therefore, we examined whether decreased αKG in low glutamine conditions can affect cell differentiation using two adipocyte models, human adipocyte derived stem cells (ADSC) and 3T3-L1 cells. Interestingly, we found that cells cultured in low glutamine failed to differentiate compared to cells cultured in complete medium (Fig. 4a). In melanomas, markers for de-differentiation have been identified such as CD271, CD133, and ABCB521–24. Differentiation markers in melanomas have also been reported, such as ERBB3, KIT, PMEL, and GJB25–29. Next, we performed RNASeq and compared the genes related to melanoma de-differentiation and differentiation in core vs. periphery regions. Consistent with the result that low glutamine inhibited adipocyte differentiation, we found that de-differentiation related genes were up-regulated while differentiation genes were down-regulated in the tumour core regions compared with matched peripheral regions (Fig. 4b). This finding in xenograft tumours was further corroborated by the up-regulation of a panel of de-differentiation genes in melanoma cells cultured under low glutamine conditions (Fig. 4c,e), while the differentiation genes are simultaneously down-regulated (Fig. 4d & Supplementary Fig. 4a). In addition, low glutamine-induced temporal CD271 expression paralleled the dose response and time course of low glutamine-induced histone hyper-methylation (Fig. 4g vs. vs.2b,2b, and Fig. 4h vs. vs.2a).2a). To further distinguish whether low glutamine leads to de-differentiation or selectively expands a small subpopulation of cells that already carry the signatures, we sorted and collected CD133/CD271 double negative (CD133-/CD271-) cells (Supplementary Fig. 4b,c). We then cultured these cells in low glutamine followed by FACS analysis and found a large portion of the previously CD271-/CD133- cells now expressed CD271 and CD133 (Fig. 4f). Consistently, H3 lysine methylation and CD271 expression level increased in these cells after glutamine starvation (Supplementary Fig. 4d). In addition, CD271 expression was specifically induced by deficiency of glutamine but not other amino acids (Fig. 4i), which could be reversed by adding dimethyl-αKG or glutamine back in the medium (Fig. 4j). As expected, the expression of CD271 was increased in the core region of solid tumours, and compound 968 treatment induced CD271 expression also in the periphery of tumours (Fig. 4k,l).
We then assessed whether glutamine deficiency-induced de-differentiation would impact responses to targeted therapies. First, we found that V600EBRAF melanoma cells maintained over 80% viability in low glutamine (0.1 mM) or low glucose (5–30 % of complete medium) (Fig. 5a & Supplementary Fig. 3c). Compared to melanoma cells in complete medium, cells proliferated at a slower rate in medium containing 0.1mM glutamine or in 30% glucose (Fig. 5b). Interestingly, we found that melanoma cells cultured in low glutamine prior to BRAF inhibitor, PLX4032, treatment were less sensitive to the drug compared to those cultured in complete medium (Fig. 5c). Although cells displayed a similar proliferation rate under low glutamine or low glucose conditions, we found only the low glutamine condition resulted in drug resistance, but not the low glucose condition (Fig. 5d). These data correlated with the previous observation that only low glutamine, but not low glucose conditions, induced hyper-methylation of histones (Fig. 3b).
We next investigated whether low glutamine-induced cell de-differentiation is mediated by histone hyper-methylation. We found that global histone methylation inhibitors, including Adox and DZNep30, 31 reversed low glutamine-induced CD271 expression (Fig. 6a). Furthermore, an inhibitor targeting H3K27me3, EPZ00568732, suppressed the low glutamine-induced CD271 expression to a similar extent as the global inhibitors. In contrast, H3K9 inhibitors BRD477033 and UNC063834 had no effect on low glutamine-induced CD271 expression (Fig. 6a and Supplementary Fig. 5a,b). These data suggest that tri-methylation on H3K27 plays a major role in low glutamine-induced melanoma tumour cell de-differentiation. Consistently, the H3K27me3 methylation inhibitor EPZ005687 prevented low glutamine-induced methylation on H3K27 and CD271 expression in a dose-dependent manner (Supplementary Fig. 5c) as well as the expression of de-differentiation genes in general (Fig. 6b). To further confirm that changes in CD271 expression were dependent on H3K27 methylation, we knocked down KDM6B, a H3K27-specific demethylase, and observed dramatic induction of H3K27 methylation and CD271 expression in cells cultured in complete medium (Fig. 6c). Next, we knocked down EZH2, the H3K27 methyltransferase in melanoma cells and found that cells cultured in 0.1mM glutamine medium with EZH2 knockdown failed to induce methylation on H3K27me3 and CD271 expression (Fig. 6d and Supplementary Fig. 6a,b). Importantly, the knockdown effect of EZH2 shRNAs could be rescued by over expression of an shRNA-resistant EZH2 cDNA in the melanoma cells (Supplementary Fig. 6a,b). Consistently, we found the induction of de-differentiation genes by low glutamine was dramatically blocked by EZH2 knockdown (Fig. 6e). Because H3K27me3 functions to suppress transcriptional activation, we next examined whether H2K27me3 plays a role in inhibiting expression of differentiation genes. First, suppression of differentiation markers in low glutamine medium was reversed by treating cells with H3K27me3 methylation inhibitor EPZ005687 (Supplementary Fig. 5d,e) or knocking down EZH2 (Fig. 6f). Chromatin Immunoprecipitation (ChIP) analysis using antibodies against H3K27me3 revealed that differentiation genes are directly repressed by this H3K27me3 methylation marker at the promoter regions (Fig. 6g).
Besides JmjC family histone demethylases, there are other αKG dependent enzymes like HIF1-α prolyl-hydroxylase (HIF-PH) and TET enzymes that promote DNA demethylation. Knockdown of HIF-1α had no effect on histone hyper-methylation in low glutamine conditions (Supplementary Fig. 7a,b). Moreover, whole genome DNA methylation sequencing found no significant differences between complete and low glutamine cultured melanoma cells (Supplementary Fig. 7c), and DNA methylation inhibitors 5-Azacytidine and 5-Aza-2′-deoxycytidine failed to inhibit CD271 induction in 0.1mM glutamine medium (Supplementary Fig. 7d).
We next tested if low glutamine-induced resistance to BRAF inhibitor, PLX4032, was mediated by hyper-methylation of H3K27. Melanoma cells cultured in low glutamine displayed resistance to PLX4032, but this effect was prevented when global histone demethylase inhibitor DZNep or H3K27 methylation specific inhibitor, EPZ005687, were supplemented in the low glutamine condition (Fig. 7a). In addition, knockdown of the H3K27 specific demethylase, KDM6B, which resulted in increased CD271 expression (Fig. 6c), led to resistance to PLX4032 treatment compared to the control cells (Fig. 7b). On the other hand, the protective effect against PLX4032 treatment in low glutamine conditions was reduced after knockdown of the H3K27 methyltransferase EZH2 (Fig. 7c). Finally, we performed a mouse xenograft experiment using melanoma cells and treated the mice with PBS control, DZNep, PLX4032, or the combination of DZNep and PLX4032. We found that treatment with PLX4032 alone inhibited tumour growth, especially at early time points; however, when combined with DZNep, PLX4032 treatment led to dramatic decreases in tumour size (Fig. 7d). As expected, histone methylation was attenuated in the DZNep and PLX4032 combination treated tumour core regions (Fig. 7e). Interestingly, no impairment in tumour growth was found upon DZNep treatment, suggesting it may take a long time (more than 10 weeks) to see the significant effects on tumour growth35. We also observed similar results on tumour growth and core histone methylation when we replaced DZNep with H3K27me3 specific methylation inhibitor EPZ005687 (Fig. 7f-h).
It is well established that the core of solid tumours is highly hypoxic and hypoxia is considered to be a major contributor to drug resistance36. However, whether nutrients are also depleted in regions of solid tumours similar to oxygen is not clear. Here, we found that glutamine is dramatically decreased in the tumour core regions compared to the periphery (Fig. 1a and Table 1). Interestingly, a few other amino acids, including serine, asparagine, arginine, and aspartate, are also significantly decreased in tumour core regions. A recent report demonstrated that glutamine, serine, and asparagine are of the most strongly depleted metabolites in human pancreatic tumors7. This result is highly consistent with our data that those amino acids are significantly decreased in the tumour core probably due to the heavy usages and poor tumour blood supply (Table 1, Supplementary Table 1). It supports the hypothesis that amino acids that are consumed by multiple anabolic processes (such as glutamine and serine) become depleted in the tumour core regions relative to those used mainly for protein synthesis.
It is worth mentioning that, different from cultured cells, whether glutamine is a major source for αKG in vivo is likely to depend on both the tumour genotype and tissue of origin, similar to glucose metabolism37. For example, 13C-glutamine tracing resulted in increased levels of TCA cycle metabolites in Myc-driven liver tumours38. In addition, analysis of NSCLC tumour metabolism in patients has indicated both glucose and glutamine can be important for TCA anaplerosis39, 40. Nonetheless, a recent report demonstrated that glutamine is not a major source for αKG in KRas-driven lung tumours41. Here, we demonstrated that in V600EBRAF melanoma tumours, that glutamine is sufficient to provide αKG in vivo (Fig. 1). Interestingly, we found that the decrease in αKG levels in melanoma tumour core regions are more extensive than the decrease in glutamine levels (Supplementary Table 1), suggesting that other sources beside glutamine could also contribute to αKG levels in vivo, such as low glucose or hypoxia in tumour core regions.
How does methylation on H3K27me3 promote de-differentiation? Although H3K27me3 is associated with repressed transcription, accumulating evidence support that H3K27me3 is essential for maintaining the self-renewal capability of embryonic and adult stem cells8, 42. In fact, it has been reported that critical genes involved in differentiation, such as HOX genes, are targeted for repression by H3K27 methylation43, supporting the hypothesis that H3K27me3 contributes to de-differentiation by inhibiting transcription of differentiation-related genes. Consistent with this model, we found that neural crest differentiation genes are directly repressed by this H3K27me3 methylation marker at promoter regions in melanoma cells (Fig. 6g). This is consistent with the previous study showing adipocyte differentiation genes such as Adipoq and Cebpa are suppressed by the binding of H3K9me3 and H3K27me3 at promoter regions12. In addition, several reports demonstrated the H3K27me3 specific methyl-transferase, EZH2 promotes cancer stem-like cell expansion44, 45. In agreement, our data demonstrate that EZH2 is required for low glutamine-induced expression of de-differentiation related genes (Fig. 6e).
Distinct tumour microenvironments may influence therapeutic responses mediated by inducing cancer cell de-differentiation46. For example, depletion of oxygen in tumours contributes to cancer stem-like cells and promotes drug resistance, where it is usually located at tumour ‘core’ or regions with a lack of blood vessel47–51 . Consistently, our data showed another example that regional glutamine deficiency can also induce cancer cell de-differentiation and lead to drug resistance (Fig. 5). It is worth noting that other studies suggested that cancer stem cells that promote metastasis are located at the tumour fronts or vascular niche52–54. It is not clear whether low glutamine-induced dedifferentiation enhances metastatic potential. However, it is possible that low glutamine/hypoxic regions will eventually result in VEGF expression and vascular formation over time, which will further provide other critical factors for cancer stem cell maintenance and metastasis, such as β-catenin, VEGF, Nrp1, and TGF52–54. In support of this, in the invasive front of pancreatic tumours, a distinct subpopulation of cancer stem cells was identified that determines the metastatic phenotype, but not the tumourigenic potential55.
In the past few years, targeting EZH2 as a potential cancer therapy has been explored. However, understanding where and how EZH2 inhibitors will be useful in cancer therapy is critical for the use of these drugs56. Our results provide the molecular basis for an important strategy to combine cancer drugs, such as BRAF inhibitor, with EZH2 inhibitor to achieve a more promising treatment efficacy.
We thank members of the Kong laboratory for helpful comments on the manuscript. This work was supported by National Institutes of Health (NIH)/National Cancer Institute (NCI) grants R01CA183989 (to M.K.), American Cancer Society Research Scholar RSG-16-085-01-TBE (to M.K.) and Stand up to Cancer Philip A. Sharp Innovation in Collaboration Award. M.K. is the Pew Scholar in the Biomedical Sciences and the V scholar in Cancer Research. X.H.L. is supported by DNA Damage Response and Oncogenic Signaling (DDROS) Training Program at City of Hope. Research reported here includes work carried out in Core Facilities supported by the NIH/NCI under grant number P30CA33572.
AUTHOR CONTRIBUTIONSM.P. designed and performed most of the experiments, analyzed and interpreted the data and wrote the manuscript. M.K. conceived and supervised this study, designed experiments and wrote the paper. M.A.R. and X.H.L. helped to measure metabolites and assisted with mouse experiments. R.P.K. and V.G. performed PACT experiments. T.Q.T. assisted with flow cytometry experiments. Y.Y. assisted with qPCR experiments. J.E.H and K.K.R. helped set up melanoma cell culture. W.H., C.S. and R.S.L. provided patient-derived melanoma cells and conceptual advice on melanoma de-differentiation. X.X. assisted with IHC experiments. D.E.S assisted with ChIP experiments and H.L. performed the bioinformatics analyses. D.K.A provided conceptual advice on hypoxia and metabolism experiments. X.L. and J.W.L performed and helped to analyze the metabolomics experiments.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.