Pan-cancer analysis of tumor metabolic landscape associated with genomic alterations

We compared the median values of metabolic signatures manually curated considering cancer-related pathways for each cancer type (Additional file 1: Figure S1A). Metabolic profiles of all cancers were visualized by t-SNE [7] (Additional file 1: Figure S1B). Metabolic profiles of cancer tissues were clustered according to each cancer type. Additionally, metabolic profiles were closely mapped with those of same histologic types (e.g. HNSC, LUSC, and CESC clusters.) Of note, metabolic profiles of TCGT and CHOL showed heterogeneous metabolic characteristics. TGCT showed two large clusters, which correspond to histologic subtypes, and CHOL showed more heterogeneous metabolic profiles than others (Additional file 1: Figure S2).

Metabolic signatures of all samples were ordered by TMB and presented by a heatmap (Fig. 1a). Carbohydrate metabolism showed the highest correlation with TMB followed by pyrimidine metabolism (Fig. 1b; p < 0.05 for all pathways). Of note, glycolysis, the well-known cancer metabolic feature, showed relatively weak correlation with TMB. LUSC showed the highest carbohydrate metabolism and TMB among 29 cancer types (Additional file 1: Figure S3). We could find a trend of increased carbohydrate and pyrimidine metabolism according to increased TMB across cancer types (Additional file 1: Figure S4). Of note, carbohydrate metabolism pathways included broad metabolic pathways related to energy metabolism including glycogen, glucose, disaccharides, pentose phosphate, and glycosaminoglycan, thus, the association with TMB suggested comprehensive changes in various energy metabolism pathways. We also investigated whether cancer types differ in the association between the metabolic landscape and TMB. Most cancer types showed a positive correlation between TMB and carbohydrate metabolism, while THYM and UVM showed a negative correlation (Additional file 1: Figure S5). Notably, TMB of these two cancer types was relatively lower than other cancer types.

Recently, TMB has been regarded as an important predictive biomarker for cancer immunotherapy, based on the idea that highly mutated tumors harboring more neoantigens can be targets of activated immune cells [8, 9]. In this regard, metabolic reconfiguration according to TMB can be an additional biomarker for cancer immunotherapy as tumor metabolism can be noninvasively and macroscopically estimated by PET. Moreover, the important metabolic signatures of each cancer type could help us find appropriate surrogate markers for PET radiotracers.

Several metabolic signatures were associated with overall survival across all cancer samples (Fig. 2a; p < 0.05 for all pathways). In general, carbohydrate and nucleotide metabolism were associated with poor prognosis. Specifically, poor prognostic metabolic signatures include carbohydrate metabolism, ribonucleotide synthesis, pyrimidine metabolism, purine metabolism, and glycolysis, whereas good prognostic signatures include peroxisomal lipid metabolism, fatty acid oxidation, and biological oxidation.

Metabolic signatures significantly associated with prognosis for each cancer type were presented in Fig. 2b. The most common metabolic signatures significantly associated with poor prognosis were ribonucleotide synthesis (7 of 29 cancers), followed by pyrimidine, purine, and carbohydrate metabolism (6 or 29 cancers) (Fig. 2c). The metabolic signatures significantly associated with good prognosis were fatty acid oxidation (6 of 29 cancers), peroxisomal lipid metabolism, and branched chain amino acids catabolism (5 of 29 cancers) (Fig. 2d).

In brief, the increased metabolic activity of carbohydrate and nucleotides in cancer was associated with poor prognosis as well as the progression of cancer in terms of TMB as found in previous results. Furthermore, our results were similar with the previous report, which showed that glycolysis, ribonucleotide, pyrimidine, and purine metabolism were associated with poor prognosis, and mitochondrial fatty acid oxidation and peroxisomal lipid metabolism were associated with good prognosis [2]. These results support the idea that the link between biological stepwise mutational progression, metabolic landscape reconfiguration, and clinical progression of cancer. Accordingly, aggressive malignant phenotypes of cancer cells obtained by accumulated mutations change metabolic phenotypes to demand energy source represented by carbohydrate and elements for proliferation represented by nucleotide.

As cancer cells exhibit various metabolic landscapes even in same cancer types, genetic alterations could underlie this variety. We tried to examine the genetic alterations related to the metabolic landscape of each cancer type to investigate whether there is a common feature in the metabolism-related genetic alterations across several cancer types. An example of carbohydrate metabolism-related genetic alterations of LGG is presented (Fig. 3a.) The tumors with high carbohydrate metabolism showed significantly more mutations of EGFR and fewer mutations of FUBP1, CIC, and IDH1. The genetic alterations of LGG related to other metabolic signatures are summarized in Fig. 3b. We defined differentially-mutated genes of tumors for at least one metabolic signature as the metabolism-related genes. All metabolism-related genes of each cancer type are presented in Additional file 1: Figure S6. Seven cancer types (CESC, DLBC, ESCA, KIRP, OV, PAAD, and READ) showed no metabolism-related gene.

We further examined whether the metabolism-related genes were cancer driver genes. All metabolism-related genes of 12 cancer types (ACC, BLCA, GBM, KICH, LIHC, LUSC, PCPG, PRAD, SARC, THCA, THYM, and UVM) were included in cancer drivers (Fig. 3c). In other cancer types, most of the metabolism-related genes overlapped with cancer drivers except STAD, COAD, and CHOL. The metabolism-related genes of CHOL did not overlap with cancer drivers. Notably, metabolic profiles of CHOL were highly heterogeneous (Additional file 1: Figure S2) as a previous result. STAD and COAD showed a large number of the metabolism-related genes than other cancer types. Since they are known to include tumors with MSI which is associated with hypermutation [10, 11], we examined the relationship between metabolic profiles of the two cancer types and MSI status. As shown in Additional file 1: Figure S7, tumors with high metabolic signatures including pyrimidine metabolism and glucose metabolism were clustered in those with MSI.

According to the results, cancer transcripts and protein networks consisting of metabolic pathways may be changed by cancer drivers instead of alterations of genes participated in the networks. It corresponds to the Darwinian selection of cancer cells which carry driver mutations facilitate cellular survival and growth by changing a favorable metabolic landscape [12]. The results also suggest that the metabolic change as a hallmark of cancer may be one of the tumorigenesis processes caused by genetic alterations rather than an independent process.