CircRNAs in cancer metabolism: a review

The modification of glucose metabolism is part and parcel of the most distinct differentiation in cancer. Normally, cells consume glucose to generate ATP through oxidative phosphorylation (OXPHOS) under the aerobic condition and channel into glycolysis only under the anoxic condition. However, cancer cells prefer glycolysis even in the normoxic environment. This phenomenon is known as “the Warburg effect” as Otto Warburg reported this phenomenon in the 1920s [9]. The Warburg effect is instrumental in malignancy. First, the change from OXPHOS to glycolysis relieves oxidative stress injury caused by mitochondria [10]. Second, excess lactic acid produced by the Warburg effect helps avoid immune surveillance. Besides, it is worth noting that recent profiles have proved that the Warburg effect does not account for the main energy supply due to the inefficiency of glycolysis in ATP synthesis. So it is hypothesized that the Warburg effect consumes a myriad of glucose to either create a hyperacid microenvironment or just starve normal cells in the vicinity [11]. The produced lactate can also stimulate vascularization via HIF1-α [12]. Specifically, the Warburg effect is triggered by abnormal expression of glucose transporters, metabolic enzymes, and oncogenes. Several important signaling pathways also participate in the regulation of this process. Hereinafter, we illustrate the effects of circRNAs on aerobic glycolysis of all these levels (Table 1).

As the carbohydrate metabolism is a major energy-providing pathway, a group of metabolic enzymes work together to maintain successful running of the process. Firstly, the glucose transporter (GLUT) takes charge of the import of glucide. To meet the excessive requirements of glucose, GLUTs ranging from GLUT1 to GLUT4 are all overexpressed in tumor cells. Then, three speed-limiting enzymes including hexokinase (HK), 6-phosphfructa-1-kinase (PFK), and pyruvate kinase (PK) work together to transform glucose into pyruvate. Lactate dehydrogenase A (LDHA), which executes the final step of aerobic glycolysis, can catalyze pyruvate into lactate. Finally, the monocarboxylate transporter (MCT) releases lactate into the extracellular matrix. In addition, pyruvate dehydrogenase kinase (PDK) blocks the traditional OXPHOS by countering pyruvate dehydrogenase (PDH) that catalyzes pyruvate into acetyl-COA [29]. PDK1 is a component of the PI3K/AKT pathway to affect cell growth, differentiation, and survival [30].

Some circRNAs can affect these enzymes. Some experimental studies reported that silencing of circHIPK3, which is abundant in pancreatic islets, decreased Slc2a2 expression that encodes GLUT2 [13]. CircHIPK3 could sponge miR-124, which represses the expression of several enzymes and transporters of glycolysis [31, 32]. Another circRNA circ-Amotl1 was reported to be able to physically bind to PDK1 and AKT1 and translocate them into the nucleus in order to antagonize apoptosis [14]. These circRNAs are likely to induce the Warburg effect in tumor tissues.

Metabolic changes in cancer are viewed as a subsequent step of the molecular transformation process. Distinctive transcription factors (TFs) also affect the Warburg effect. As a member of ncRNAs, circRNA regulates the mutant expression of TFs. So we pay attention to the selected TFs to elucidate the relationships between circRNAs and TFs (Fig. 2).

There is a tight association between hypoxia and the Warburg effect [33]. Solid tumors encountering hypoxia have to adopt glycolysis instead of oxidative metabolism. In response to the hypoxic condition, HIF-1 contributes to the overexpression of a series of carcinogenic genes. It was found that the lactate production rate increased proportionally with the level of HIF-1α, for it facilitates the transcription of genes related to GLUTs and glycolytic enzymes including HK, PKM, and MCT4 [11]. HIF-1 also enhanced the PDK1 function [34].

Because neovascularization frequently occurs at the periphery of tumors rather than within tumors, larger cancer tumors incline to suffer more severe hypoxia. Therefore, the tumor size and proliferation rate can often represent the extent of hypoxia damage. CircDENND4C, which was found to be upregulated by HIF-1α in breast cancer cells, aggressively promotes cell proliferation in a hypoxic environment [17]. Knowing that a larger mass consumes more energy and bears more extreme hypoxia, circDENND4C is supposed to enhance the Warburg effect. In human umbilical vascular endothelial cells (HUVECs), circ_0010729 coexists with HIF-1, especially in hypoxia-induced HUVECs, where circ_0010729 overexpressed and elevated the HIF-1α expression by sponging miR-186 [35]. With the support of thick vessels, more glucose is brought inside tumors. So circ_0010729 may also facilitate glycolysis by enhancing the progress of vascular endothelial cells.

c-myc, one of the most important oncogenic TFs, can also activate the Warburg effect. Ectopic c-myc can collaborate with HIF-1 to elevate the level of the glycolytic components GLUT1, HK2, and PDK1 [34, 36].

In bladder cancer, miR-145 could alleviate the Warburg effect by silencing c-myc [37]. To explain detailedly, polypyrimidine tract-binding protein1 (PTBP1) as the downstream of c-myc induces the up-regulation of PKM2 instead of PKM1. Meanwhile, miR-145 can counteract the function of the c-myc/PTBP1 axis to reduce the PKM2/PKM1 ratio [38]. Two circRNAs interplay with miR-145. One is circRNA_001569 sponging miR-145 in colorectal cancer, and the other is circBIRC6 [18, 19]. circBIRC6 also targets miR-34a, another miRNA repressing c-myc [19]. The above-mentioned Circ-Amotl1 can also assist c-myc in nuclear translocation in order to prevent the degradation of c-myc [15]. Besides c-myc, circ-Amotl1 could link with classic oncogenic proteins such as AKT and STAT3. Consequently, it is highly feasible that circ-Amotl1 enhances aerobic glycolysis to facilitate tumor development without working as a competing endogenous RNA (ceRNA). Intriguingly, circ-FBXW7 could encode protein FBXW7-185aa which was capable to competitively interact with the de-ubiquitinating enzyme USP28 [20]. So FBXW7-185aa could indirectly degrade c-myc. Circ-FBXW7 probably ameliorates the Warburg effect in this way.

Effective TFs work together to form a regulatory network that affects the Warburg effect. The network consists of different signaling pathways in which diverse circRNAs are involved. So reviewing the representative signaling pathways and associated circRNAs can help understand this metabolic mechanism more comprehensively.

The PI3K/Akt pathway plays a crucial role in motivating glucose uptake and glycogen synthesis. Akt can encourage HK2 and PFK2 to stimulate glycolysis [39]. MTOR complex1, as the downstream of Akt, can phosphorylate elf4E-binding proteins and support HIF-1α to promote GLUT1, HK2, and PFK2 [40, 41].

First of all, in gastric cancer, circNRIP1 could upregulate glucose uptake and induce lactification via sponging the miR149-5p that targets AKT1 [21]. Second, it is possible that miR-150 directly depresses the Warburg effect. Downregulation of miR-150 reinforced the resistance of ovarian cancer cells to pertuzumab by increasing p-Akt [22]. And in osteosarcoma, miR-150 might repress glycolysis by targeting the 3′-UTR of GLUT1 [42]. In Hirschsprung disease, the level of AKT3 scaled up with that of circ-ZNF609, while miR-150-5p expression was inversely associated with circ-ZNF609 and AKT3 [43]. Some experimental studies found circ-ZNF609 could harbor the target sites of miR-150-5p. So circ-ZNF609 might promote AKT3 by targeting miR-150-5p to improve the Warburg effect. More research is required to confirm its role in cancer metabolism. In osteosarcoma, circRNA_103801 was found to be associated with the HIF-1 and PI3K/AKT pathway by binding with miR-370 and miR-877 as predicted by bioinformatics tools [23]. In ovarian cancer, miR-29b targeted AKT2 and AKT3 to moderate the Warburg effect [44]. And in oral squamous cell carcinoma, circRNA_100290 sponged miR-29 family to downregulate CDK6 [24]. Similarly, miR-124 could target AKT1 and AKT2 to decrease GLUT1 and HK2 to relieve the Warburg effect in non-small cell lung cancer cells [31]. And in vascular smooth muscle cells, circWDR77 has been verified to sponge miR-124 [45].

KRAS mutations occur in a variety of cancers, especially in colorectal cancer, non-small cell lung cancer, and PCa. The alteration of KRAS upregulated GLUT1 and glycolytic enzymes to facilitate the Warburg effect [46]. In bladder cancer, ectopic circRNA-MYLK increased the expression of RAS and the downstream Raf/MEK/ERK [26]. CircRNA-MYLK can activate the vascular endothelial growth factor A to elevate the glucose brought by blood as a ceRNA for miR-29a. Additionally, depending on the ceRNA network, both lncRNA H19 and circRNA-MYLK were the ceRNA of miR-29a-3p in breast cancer [27]. And H19 could activate PKM2 and PKM2 polymers in hepatocarcinogenesis, with the upregulation of c-myc and HRAS [47]. Deductively, circRNA-MYLK might sponge miR-29a to discharge more H19 to increase PKM2. Similarly, circRNA_100290 sponged the entire miR-29 family, which is perhaps helpful for the expression of RAS.

The STAT3 pathway gives rise to the inflammatory environment in favor of cancerogenesis. Experiments in breast cancer cells have demonstrated that STAT3 could directly upregulate the expression of HK2 [48]. The aforementioned circ-Amotl1 could communicate with STAT3 to markedly increase cell proliferation [16]. By increasing the expression of STAT3 and inducing STAT3 nuclear distribution, circ-Amotl1 enhanced STAT3 expression. In osteosarcoma, hsa_circ_0009910 indirectly facilitated STAT3 with miR-449a as a mediator [49]. Circ_0009910 could sponge miR-449a which antagonized IL6R. IL6R increased BCL-2 and STAT3 in hepatocellular carcinoma (HCC). The deregulated circ_0009910 could cut down the levels of p-STAT3 and Bcl-2 [50].

The Wnt/β-catenin pathway can direct the Warburg effect by upregulating the transcription of MCT1 and PDK1 [12]. Mutant Wnt signaling can activate β-catenin leading to the overactive response of the downstream genes. Furthermore, c-myc has been ascertained as a Wnt/β-catenin target [28]. In papillary thyroid cancer, circ-ITCH could sponge miR-22-3p to upregulate CBL, an E3 ligase of nuclear β-catenin [51]. CBL, which is detrimental to tumor growth and development, could degrade β-catenin. So circ-ITCH might suppress the Warburg effect by impairing the activation of the Wnt/β-catenin pathway. Additionally, circRNA_102171, circ_0067934, and circRNA_100290 were in favor of the Wnt/β-catenin pathway by targeting CTNNBIP1, miR-1324, and miR-516b, respectively [25, 52, 53], while circ_0006427 inactivated the Wnt/β-catenin pathway by targeting miR-6783–3p [54].

Exacerbated lipid production is another hallmark of tumor metabolism, for cancer cell growth requires fatty acids (FAs) to replicate cellular membranes. Endogenous fatty acid synthesis (FAS) is triggered in various tumors [55]. Acetyl-COA can be converted from citrate catalyzed by ATP citrate lyase. The first step of FA biosynthesis is the transformation of cytosolic acetyl-COA to malonyl-COA via acetyl-COA carboxylase. Then, fatty acid synthase (FASN) catalyses malonyl-COA into long FA chains [2]. In contrast, altered lipid mobilization also has relevance with tumor progression. For example, increased lipolysis leads to a wasting syndrome known as cancer cachexia, which is characterized by acute fat loss [56]. Tumor cells can acquire FAs through lipolysis to perform FA β-oxidation (FAO) which provides energy for tumor proliferation. Due to the Warburg effect, less acetyl-COA is generated by glucose, while FAO can comprise this deficiency. Moreover, lipophagy, meaning autophagic degradation of lipid, is another pathway of lipid utilization [57]. Recent studies indicate that lipophagy is a well-conserved mechanism of lipid degradation. Lysosomes can digest lipid droplets with the help of phagophore to reproduce free FAs [58].

An inverse correlation between circRNA_0046367 expression and triglyceride (TG) level in HepG2 cell culture and liver tissues has been reported [59]. CircRNA_0046367 could sponge miR-34a to protect the peroxisome proliferator-activated receptor (PPAR) α from transcriptional repression. PPARα activates CPT2 and ACBD3 to degrade lipids [60]. So circRNA_0046367 might promote tumor progression partly through upregulating FAO. CircGFRA1 attracted miR-34a in triple negative breast cancer as a ceRNA to release the restrain of GFRA1, which can induce autophagy [61, 62]. More studies are required to confirm whether circGFRA1 concomitant with GFRA1 can promote lipophagy in cancer metabolism. On the contrary, circ-DB released by adipocyte exosomes also sponged miR-34 but appeared to raise the body fat ratio in HCC patients [63]. HCC patients with the higher body fat ratio incline to have more rapid tumor progression and poorer prognosis. In this circumstance, circ-DB seems to facilitate HCC progression via FAS. However, this mechanism is paradoxical to circRNA_0046367 whose target is miR-34a as well. Considering that the expression of miR-34a is increased in obesity causing the fat browning and the weight loss, the difference of circ-DB function may be caused by lipid in exosomes [64]. But the authentic mechanism remains elusive. After high fat stimulation, circRNA_021412 was downregulated in HepG2 cells causing reactivation of miR-1972 [65]. Increased miR-1972 reduced the level of LPIN1, which can coordinate with PPARα to recruit long-chain acyl-CoA synthetases. The inhibition of LPIN1 blocks FAO resulting in hepatic steatosis. The circRNA_021412/miR-1972/LPIN1 axis may also be involved in tumor progression. The expression of circ_0057558 was found to be positively associated with the level of TG in PCa [66]. PCa is a hormone-dependent cancer influenced by lipid metabolism abnormalities [67]. So circ_0057558 might regulate the level of lipid to promote PCa progression. CircFARSA, which was found to be highly expressed in lung adenocarcinoma, is supposed to interact with FASN through miR-330-5p and miR-326 in accordance with the in silico analysis [68]. And in laryngeal squamous cell carcinoma, hsa_circ_0033988 was lowly expressed and associated with FA degradation according to ceRNA network analysis and functional annotation analysis [69]. These results computed by special algorithms have yet to be further validated by experiments in vitro and in vivo (Table 2).

Since the 1950s, scientists have found that tumor cells consume much more glutamine than any other amino acid. Glutamine can be converted to glutamate catalyzed by glutaminase. Subsequently, glutamate can provide nitrogen for purine and pyrimidine biosynthesis, or participate in the TCA cycle by transforming to α-ketoglutarate. Glutathione, an important cellular antioxidant, also comes from glutamate. Thus, glutamine plays an essential role in tumor progression.

Despite the deficiency of direct evidence, scientists have investigated the impact of circRNAs on glutamine metabolism by comprehensive analysis (Table 3). In glioblastoma, functional analysis predicted that circRNAs may participate in glutamergic synapse and calcium signaling [70]. Computational analysis discovered the circRNA_002581/miR-122/Slc1a5 axis in non-alcoholic steatohepatitis [70]. Knowing that Slc1a5 is a kind of glutamine transporter, biological research is needed to evaluate the relationship between circRNA_002581 and glutamine metabolism [71].

Owing to the electron transport flux accumulated by proliferating cells and hypoxia in the tumor microenvironment, the constant generation of reactive oxygen species (ROS) is another attribute of cancer. Overexpressed ROS can develop into oxidative stress, eventually resulting in oncogene-induced cellular senescence. Although ROS overproduction is harmful for cell growth, a moderate amount of ROS is beneficial for the maintenance of a tumorigenic condition [72]. The tolerable level of ROS not only activates HIF1-α and NRF2 but inhibits protein phosphatases such as PTEN [2]. So reduction-oxidation (redox) homeostasis is a crucial component of cancer metabolism. Some studies demonstrated that several circRNAs were differentially expressed in substantia nigra and corpus striatum of NRF2 knock-out mice compared with the control group [73]. Increases CircNCX1 in response to ROS in cardiomyocytes could sponge miR-133a-3p to release the pro-apoptotic gene CDIP1 [74]. These circRNAs implicated with ROS may also regulate the redox equilibrium in tumor cells [Table 4].

The folate cycle plays a dominant role in the generation of NADPH, which effectively antagonizes ROS. Recent studies point out that folate is required for HCC and PCa cells growth [75, 76]. In PCa, circ_0062019 and its host gene SLC19A1 were significantly upregulated [66]. SLC19A1 can encode a membrane protein to transport folate. This result suggests that circ_0062019 may act as a booster to further PCa proliferation by the folate cycle.