Mutual regulation between N6-methyladenosine (m6A) modification and circular RNAs in cancer: impacts on therapeutic resistance

M6A modification, referring to the methylation of the sixth nitrogen (N) atom of RNA adenylate (A), is dynamically regulated by three important regulators consisting of m6A methyltransferases, m6A demethylases and m6A-binding proteins. M6A modification plays important roles in gene expression, thereby influencing many physiological and pathological processes such as cells differentiation, self-renewal and apoptosis, as well as the development of cancer, cardiovascular and metabolic diseases [18‐20].

M6A methyltransferases, also known as ‘m6A writers’, are involved in the composition of methyltransferase complex (MTC) to add m6A sites to RNA [21]. MTC is mainly composed of two methyltransferase-like proteins (METTL3/14) and Wilm’s tumor 1-associated protein (WTAP) [22]. METTL3, identified by Joseph et al. in 1997, is the most significant component protein in the complex, with the ability to catalyze the formation of m6A [23]. METTL14 has no catalytic activity, however, it can help METTL3 to recognize substrates by binding to METTL3 to form a heterodimer [24]. WTAP serves as a key component that recruits MTC to RNA targets [25]. Moreover, other m6A methyltransferases such as methyltransferase-like 16 (METTL16), zinc finger CCCH-type containing 13 (ZC3H13), vir-like m6A methyltransferase associated (VIRMA) and RNA-binding motif protein 15/15B (RBM15/15B) have been reported to exhibit indispensable roles in the process of RNA m6A methylation in recent years [26‐28].

In contrast to methyltransferases, m6A demethylases are vividly named as ‘m6A erasers’, and they act to remove methylation, thus making the m6A modification process dynamic and reversible. Until now, two FeII/α-KG-dependent dioxygenase AlkB family members, fat mass and obesity-associated protein (FTO) and alkB homolog 5 (ALKBH5), are identified as m6A erasers. In 2012, Jia et al. revealed that FTO was not only associated with human obesity [29], but also had oxidative demethylation activity of m6A residues in RNA [30]. In addition, ALKBH5 is the second identified m6A demethylase that can affect RNA nuclear export and metabolism. In male mice, the deficiency of ALKBH5 was found to increase m6A modification in mRNAs [31].

M6A modification can be recognized by specific m6A-binding proteins that serve as ‘m6A readers’. These proteins are able to bind to m6A sites or change RNA secondary structure to perform their biological functions [32]. The m6A readers mainly contain multiple YT521-B homology (YTH) domain family members (YTHDC1/2 and YTHDF1/2/3), heterogeneous nuclear ribonucleoproteins (HNRNPs) family proteins (HNRNPA2B1/C/G) and eukaryotic initiation factor 3 (eIF3), etc. [33]. They have a variety of functions, and participate in the regulation of RNA splicing, translation, transport, degradation, stability and processing. For example, YTHDC1 contributes to the RNA splicing and export [34, 35]. YTHDF2 recognizes m6A sites and drives RNA deregulation via two distinct pathways [36]. Recently, HNRNPA2B1 was reported to be inseparable from miRNA processing [37].

CircRNAs are generally derived from precursor mRNAs (pre-mRNAs) and possess covalently closed single-stranded structures, which facilitate their resistance to RNase [38]. CircRNAs are classified into four categories, including exonic circRNAs (ecRNAs), exon-intron circRNAs (EIciRNAs), circular intronic RNAs (ciRNAs) and tRNA intronic circRNAs (tricRNAs), and their formation mechanisms are distinct. To date, several mechanisms have been reported for the biogenesis of circRNA, including lariat-driven circularization, RNA-binding proteins (RBPs)-associated circlarization, intron pairing-driven circularization and the bulge-helix-bulge (BHB) motif of precursor tRNAs-depended circularization [39‐41]. Interestingly, circRNAs were initially regarded as useless by-products of pre-mRNAs. It was not until these years that people gradually uncovered that circRNAs are widely involved in the regulation of various aspects of many diseases, especially cancer [40, 42, 43]. The classic biological roles of circRNAs are that they can function as competing endogenous RNAs (ceRNAs) to sponge miRNAs and positively modulate downstream target genes expression. Specifically, circRNAs can competitively suppress miRNAs activity through miRNA response elements (MREs), leading to fewer miRNAs binding to 3′ untranslated regions (3′ UTRs) of target mRNAs, thereby relieving the inhibitory effects of miRNAs on mRNAs expression, and ciRS-7 was the first identified circRNA that act as a miRNA sponge [41, 44]. Furthermore, circRNAs also can act as regulators of their parental gene transcription, bind to proteins, and have unexpected translation potential [40].

Strikingly, extensive studies have confirmed that m6A plays an indispensable role in the metabolism and functions of circRNAs [17]. Currently, the roles of m6A in circRNAs are mainly concentrated in four aspects (Fig. 2): (1) Biogenesis of circRNAs. Timoteo et al. revealed that m6A regulated circZNF609 biogenesis in an METTL3/YTHDC1-dependent manner [45]. Similarly, the formations of circ-ARL3, circMETTL3 and circ1662 have been demonstrated to be closely associated with m6A modification in multiple tumors [46‐48]. However, the detailed mechanisms of m6A-dependent circRNAs biogenesis still need to be further explored. (2) Cytoplasmic export of circRNAs. Cytoplasmic export of circRNAs is essential for their functions. Some scholars have found that m6A-modified circNSUN2 in the nucleus could bind to YTHDC1, thus promoting its export to the cytoplasm [49]. Despite this breakthrough, more research about m6A-mediated cytoplasmic export of circRNAs is still required. (3) Degradation of circRNAs. CircRNAs are more stable than linear mRNAs due to their closed structure. Therefore, there are few reports on how circRNAs are degraded, and the specific mechanisms remain largely unclear. Recently, a new study revealed that circRNAs with m6A modification could be identified by YTHDF2, which participated in the formation of cleavage complexes with HRSP12 and RNaseP/MRP. Thus, the YTHDF2-HRSP12-RNase P/MRP axis exhibited a key role in circRNAs degradation [50, 51]. Additionally, m6A modification could inhibit the degradation of circRNAs, thus elevating circRNAs stability. For example, Wu et al. found that the half-life of m6A-modified circCUX1 was prolonged after treatment with Actinomycin D, indicating that m6A stabilized the expression of circCUX1 [52]. However, the molecular mechanisms by which m6A inhibits circRNAs degradation are unknown. (4) Translation of circRNAs. Most circRNAs are classified as ncRNAs, while a few circRNAs in the cytoplasm were found to possess peptides/protein-coding potential [53]. Because of the absence of 5′cap structure, the traditional cap-dependent translation model is not applicable to circRNAs. With continuous efforts, it is gradually known that circRNAs can be encoded via two cap-independent pathways, including the internal ribosome entry site (IRES) and m6A-driven translation [54, 55]. YTHDF3 and eIF4G2 are critical for the initiation of translating circRNAs containing m6A motifs. In addition, this m6A-mediated translation process can be inhibited by FTO and heightened by METTL3/14.

Conversely, aberrant expression of circRNAs also affect m6A modification. On one hand, circRNAs can indirectly modulate m6A regulators levels via sponging miRNAs. For instance, a recent data analysis suggested that circMAP2K4 interacted with miR-139-5p, which contributed to the target gene, YTHDF1, expression in hepatocellular carcinoma (HCC) [56]. Moreover, interacting with RBPs is also a significant function of circRNAs. Therefore, it is not surprise that some circRNAs, including circNOTCH1 and circPTPRA, can regulate the expression of target mRNAs through recruiting or competitively binding to m6A regulators [57‐59]. Taken together, m6A modification not only regulates biogenesis and function of circRNAs, but is also affected by circRNAs.

For the efficacy of anti-cancer drugs, it is of great significance to study the intracellular drug concentrations. Drug transport mediated by drug transporters is a pivotal process that influences drug concentrations in tumor cells, and increased drug efflux and decreased drug influx are leading reasons for cancer drug resistance [5, 78].

Many anticancer drugs, such as cisplatin and paclitaxel, are substrates of ABC transporter family proteins that take charge of drug efflux [79, 80]. In general, ABC transporter family proteins consist of transmembrane domain (TMD) and adenosine triphosphate (ATP) binding region located in the cytoplasm [81]. After being combined with ATP, the ATP binding region transports the corresponding substrates to the membrane by hydrolyzing ATP. TMD is responsible for binding to the transported substrates, forming a transmembrane channel and helping them to pass through the membrane [82]. Recently, extensive reporters have demonstrated that m6A modification is a key player in the expression of ABC transporter family, that are strongly associated with multidrug resistance and thus indirectly regulate therapeutic resistance in cancer. Sulin-like growth factor 2 mRNA-binding protein 3 (IGF2BP3), as a new m6A reader, enhances target mRNA stability and nuclear export to regulate gene expression [83]. A current study suggested that overexpression of IGF2BP3 in HCT8/T cell bound to the m6A site of ABCB1 and elevated ABCB1 expression, thereby reducing the sensitivity of colorectal cancer (CRC) cells to chemotherapy [60]. In addition to direct regulation of m6A, the expression of ABCB1 was also regulated by m6A-modified estrogen receptor related receptors (ERRγ). In chemoresistant cancer cells, m6A modification-induced upregulation of ERRγ promoted chemoresistance through elevating ABCB1 level [61]. Moreover, Liu et al. surprisingly found that YTHDF2 regulated the expression of a transcription factor, ATF3, which further interacted with the enhancer of ABCB1 to stimulate its expression, leading to tamoxifen resistance [62]. ABCC10 serves as an efflux transporter of gefitinib, and affects the concentration of gefitinib in cells [84]. FTO was found to be enriched in serum exosomes from gefitinib-resistant patients, and overexpression of FTO upregulated ABCC10 in an m6A-dependent manner, and conferred non-small cell lung cancer (NSCLC) to gefitinib resistance [12].

Apart from ABC efflux transporter family, the expression of some organic anion transporting polypeptides (OATP) transporters belonging to the solute carrier superfamily responsible for drug uptake, is also affected by m6A modification. It is acknowledged that there are 11 members in OATP transporter family. Among them, OATP1B1 and OATP1B3 are specifically expressed on the basement membrane of human hepatocytes, and mediate drug transport in the liver [85]. Sorafenib, a targeted anticancer drug for the treatment of inoperable and distant metastatic HCC and advanced clear cell renal cell carcinoma (ccRCC), is recognized as a substrate of OATP1B1 and OATP1B3 [86, 87]. In a recent study, decreased hepatocyte nuclear factor 3γ (HNF3γ) level was measured in HCC tissues compared with adjacent normal tissues, and it was induced by reduction of METTL14, which further facilitated sorafenib resistance by inhibiting OATP1B1 and OATP1B3 expression [73].

Autophagy is a conserved intracellular degradation process that is a critical factor in adapting to metabolic stress and maintaining intracellular homeostasis [88]. Under certain stress conditions, cells can timely self-degrade damaged organelles and aggregated or dysfunctional proteins through autophagy mechanism, which is beneficial for the degradation of cells to gain survival advantage [89, 90]. Accordingly, autophagy induction in tumor cells may favor their response to anti-cancer treatment, thus reducing cell death and hastening therapeutic resistance.

Growing studies confirm that m6A modification has a hand in autophagy activation and the formation of autophagosomes responsible for loading some degradative substances [91]. For instance, FTO-mediated reduction of m6A modification on unc-51-like kinase 1(ULK1) suppressed YTHDF2-driven degradation of ULK1 transcripts, thereby promoting the initiation of autophagy [92]. In another study, FTO deficiency led to the degradation of ATG5 and ATG7, which acted as significant regulators in autophagosome formation, thus weakening autophagy [93]. In addition, Li et al. assessed the correlation of YTHDF1 with hypoxia-induced autophagy in vivo and in vitro. Subsequently, they found that overexpression of YTHDF1 could enhance autophagy under hypoxia by facilitating translation of two other autophagy-related genes, ATG2A and ATG14 [94]. Meanwhile, multiple tumor cells were demonstrated to be resistant to drug treatment, based on the mechanisms of autophagy induction under hypoxia [95‐97]. Intriguingly, it was also revealed that omeprazole could augment m6A level in gastric cancer (GC) cells through eliminating FTO, and further studies found that reducing FTO attenuated pro-survival autophagy levels, thereby sensitizing GC cells to chemotherapy via boosting the activation of mTORC1 signaling pathway [63]. Hence, FTO is expected to be a potential therapeutic target to improve drug sensitivity in GC patients. METTL3 and METTL14, as important functional components of MTC, also play vital roles in the modulation of autophagy. A recent evidence shed light on the mechanism by which METTL3 reduced the sensitivity of seminoma to cisplatin by heightening autophagy [64].

Genomic instability caused by multiple factors, such as DNA damage, often leads to oncogenesis. Meanwhile, it also provides opportunities for cancer treatment [98]. The important ways of cancer therapy including chemotherapy and radiotherapy induce DNA damage so as to cause cell death. Nevertheless, tumor cells can activate their own DNA damage repair mechanisms, and ultimately become resistant to DNA-damaging drugs and radiation, resulting in therapy failure [99, 100]. DNA damage repair methods chiefly consist of nucleotide excision repair (NER), base excision repair (BER), homologous recombination repair (HRR) and non-homologous end joining (NHEJ), etc. [98].

It has been proved that some DNA damage repair proteins are essential players in chemoradiotherapy resistance. Moreover, the expression of these proteins is closely related to m6A modification. For example, excision repair cross-complementation group-1 (ERCC1), located on human chromosome 19, is a key gene in the NER pathway [101]. Its expression is positively correlated with cisplatin resistance in various cancers, including GC and cervical cancer [102, 103]. To broaden the horizons of FTO function in chemoradiotherapy resistance, Zhou et al. measured the expression of FTO and uncovered its regulatory role in cisplatin resistance and radioresistance. The results showed that FTO elevated the expression of β-catenin, which in turn caused the upregulation of ECCR1. Subsequently, increased ECCR1 gave rise to failure of cisplatin therapy and radiotherapy in cervical squamous cell carcinoma (CSCC) via activating NER [13]. DNA double strand breaks (DSBs) are the most deadly DNA damage, however, it can be repaired by HRR, which is a slow but highly accurate repair process [104]. Aberrant expression of key genes therein may result in altered sensitivity of tumor cells to anti-cancer therapy. One suitable example is that increased YTHDC1 mediated by epithelial membrane protein 3 (EMP3) downregulation exhibited a positive effect on the expression of BRCA1 and RAD51, thereby promoting DNA repair in breast cancer cells, leading to chemoresistance [65]. Additionally, VANGL1 was upregulated upon increased m6A levels in VANGL1 mRNA and miR-29b-3p depletion. The upregulated VANGL then inhibited the harmful influence of radiation on lung adenocarcinoma (LUAD) by activating BRAF/TP53BP1/RAD51 pathway correlated with DNA repair [71]. Interestingly, enhanced stability of TFAP2C induced by IGF2BP1 contributed to the activation of WEE1 and BRCA1 in cisplatin-resistant seminoma cells [66].

The TME is an indispensable soil for tumor growth and development. It is a complex and multicomponent system that contains several types of cells, such as cancer-associated fibroblasts and immune cells, as well as other biochemical factors such as cytokines and chemokines, etc. [105]. In recent years, emerging immunotherapy achieves the purpose of anti-tumor by stimulating or rebuilding immune system, which has brought tremendous progress in the treatment of malignant tumors [106]. With the increasing recognition of the complexity and variability of TME, people gradually realize the important role of TME in tumor immunotherapy. For example, immunosuppression in the TME is caused by multiple factors such as elevated immune checkpoints, recruitment of more immunosuppressive cells such as T regulatory cells (Tregs), myeloid-derived suppressor cells (MDSCs) and tumor-associated macrophages (TAMs), and increased production of immunosuppressive molecules limits the efficacy of immunotherapy, leading to therapeutic resistance [107‐110]. Tregs infiltration directly inhibits cytotoxic T cells proliferation and produces immunosuppressive cytokines such as IL-10, IL-35 and transforming growth factor β (TGF-β), thereby supporting tumor immune escape and potentiating tumor progression [111]. Another study exhibited that increased MDSCs and chemokines within the TME could promote the development of resistance to dual ICIs in BRAF inhibitor-resistant melanoma [112]. Not surprisingly, m6A modification can affect the sensitivity of various tumor cells to immunotherapy via taking part in the regulation of anti-cancer immune responses within the TME. Yin et al. revealed that METTL3 deficiency facilitated M1/M2-like TAM and Treg infiltration to attenuate the response to anti-PD-1 therapy [74]. Similarly, ALKBH5 knockout had a positive influence on anti-PD-1 treatment, relying on the inhibition of normal lactate transport and accumulation of Tregs and MDSCs [75]. Moreover, ALKBH5-mediated reduction of m6A modification of PD-L1 suppressed PD-L1 degradation, and its upregulation could improve the sensitivity of intrahepatic cholangiocarcinoma (ICC) cells to anti-PD-1/PD-L1 immunotherapy [76]. Recently, it was found that METTL3/14 knockdown elevated the efficacy of immunotherapy. Mechanistically, CD8+ T cells and chemokines including CXCL9 and CXCL10 were significantly upregulated upon METTL3/14 deletion, depending on enhanced IFN-c-Stat1-Irf1 signaling pathway in CRC and melanoma [77].

In 1982, the concept of EMT was firstly proposed by Greenberg et al. when they found that epithelial cells cultured in three-dimensional collagen gels lost their polarity and exhibited a mesenchymal phenotypes [113]. With further recognition of EMT, it has been observed that EMT not only occurs throughout embryonic development and wound healing, but also plays a key role in various aspects of tumor progression through enhancing tumor cells migration, invasion and anti-apoptosis abilities [114]. To date, plenty of studies have suggested that EMT is involved in therapeutic resistance of various tumors, including breast cancer, lung cancer and ovarian cancer [115‐117]. As a dynamic process, EMT-enhanced cells exhibit changes in molecular expression levels, such as reduction in epithelial markers (e.g. E-cadherin) and increase in EMT transcription factors (e.g. Snail, Slug and Twist). During EMT, m6A modification is also involved in the modulating these changes. A representative example is that overexpression of METTL14 prominently stimulated Twist expression and suppressed E-cadherin expression, which would lead to the enhancement of EMT [118]. Besides, Yu et al. analyzed the expression profile of m6A-related regulators acquired from TCGA database, and revealed that upregulated IGF2BP2 in head and neck squamous cell carcinoma (HNSCC) helped stabilize Slug mRNA, thus facilitating MET process and lymphatic metastasis of HNSCC [119]. Consistently, EMT regulated by m6A has also been observed in other cancers, including GC and breast cancer [120, 121]. Taken together, these studies imply that m6A modification may contribute to therapeutic resistance by influencing EMT process.

Strikingly, the activation of EMT is closely associated with the generation of cancer stem cells (CSCs) [122]. CSCs resemble embryonic stem cells with unlimited self-renewal and differentiation potential. These properties can confer tumor cells the ability to tolerate detrimental stress, such as damage caused by chemotherapy drugs and radiation [123]. M6A-induced upregulation of CBX8 could enhance stemness and chemoresistance of colon cancer (CC) through combing with KMT2b and Pol II to promote LGR5 transcription [67]. Moreover, m6A modification could facilitate the acquisition of CSCs properties via promoting Wnt/β-catenin signaling, thereby inhibiting chemosensitivity of CRC [68]. Similarly, CSCs generation and radiosensitivity in esophageal squamous cell carcinoma (ESCC) were also influenced by m6A methyltransferase METTL14 [72].

Indeed, there are several other mechanisms related to cancer therapeutic resistance, including alterations in drug-metabolizing enzymes, abnormal apoptosis, glycolysis, and so on. However, the role of m6A modification in these mechanisms is not fully understood and still under continuous investigation. Some chemotherapeutic prodrugs, such as cyclophosphamide and 5-fluorouracil (5-FU), need to be converted into active drugs by drug-metabolizing enzymes before binding to their targets, and drug-metabolism can also inactivate these anti-cancer drugs, indicating that drug-metabolizing enzymes are critical to drug efficacy [124]. An attractive report showed that mRNA levels of cytochrome (CYP) P450 family members (CYP1A2, CYP2B6, and CYP2C8) were remarkably elevated following treatment with RNA methylation inhibitor. Follow-up experiments demonstrated that METTL3/14 overexpression or FTO deficiency could facilitate CYP2C8 mRNA degradation, implying that m6A regulators may affect drug efficacy in cancer therapy by regulating drug-metabolizing enzymes [125].

Apoptosis, a controlled cell death process, participates both in regulating normal physiological processes and in the progression of some diseases [126]. Evasion of apoptosis often occurs in tumor cells and has become one of the important hallmarks of treatment failure. The proteins involved in apoptosis mainly include anti-apoptotic B-cell lymphoma-2 (BCL-2) family members (e.g. BCL-2, MCL-1 and BCL-XL), pro-apoptotic BCL-2 proteins (e.g. BAX and BAK) and inhibitors of apoptosis proteins (IAPs), etc. [127]. Interestingly, m6A modification was found to modulate anti-apoptotic BCL-2 family proteins expression. For example, the m6A writer METTL3 enhanced BCL-2 expression in breast cancer, leading to inhibition of apoptosis [128]. Notably, BCL-2 mRNA level was strongly associated with multiple drug resistance including paclitaxel, tamoxifen and trastuzumab [129‐131]. Likewise, overexpression of FTO could promote BCL-2 expression, thereby reducing the sensitivity of leukemia cells to tyrosine kinase inhibitors (TKIs) [14].

Altered energy metabolism is now known as one of the remarkable hallmarks of cancer. In 1924, Otto Warburg firstly uncovered that tumor cells preferentially relied on glycolysis for energy production, even under aerobic conditions [132]. This type of energy metabolism is called ‘aerobic glycolysis’ or ‘Warburg effect’. During aerobic glycolysis, tumor cells swiftly consume glucose to produce lactic acid and ATP. Although glycolysis produce much less ATP than oxidative phosphorylation, this quick energy production is more advantageous for rapid proliferation of tumor cells [133]. Additionally, plenty of lactic acid promote the formation of acidic TME, and glycolytic metabolic intermediates can be used as precursors for the synthesis of certain biomacromolecules, both of which contribute to tumor growth, distant metastasis and therapeutic resistance [134]. Recently, Yu’s group discovered that ALKBH5 depletion increased casein kinase 2 α (CK2α) expression, indicating that m6A modification promoted cisplatin resistance in bladder cancer (BC) by upregulating CK2α-mediated glycolysis pathway [69]. Similarly, lncRNA JPX facilitated aerobic glycolysis through the FTO/PDK1 axis, thus conferring resistance to temozolomide in glioblastoma multiforme (GBM) cells [70].

Accumulating studies have suggested that certain circRNAs can affect the levels of ABC transporters in drug-resistant tumor cells. For example, Zheng’s group revealed that circPVT1 could promote ABCC1 expression via binding to miR-145-5p, resulting in a reduced accumulation of cisplatin and pemetrexed in LUAD cells [138]. Consistently, ABCB1 expression in osteosarcoma cells was also modulated by circPVT1 [166]. One study involved circ_0007031, which contributed to 5-FU resistance of CRC cells through miR-133b/ABCC5 pathway [139]. In pemetrexed-resistant GC, overexpression of circMTHFD2 can sponge miR-124 and increase the expression of ABCB1 [140]. Additionally, the circSETD3/miR-520 h/ABCG2 axis was disclosed to be strongly associated with the sensitivity of NSCLC cells to gefitinib [161]. Unfortunately, studies on the roles of circRNAs in regulating drug uptake have not been reported.

DNA damage repair is an important mechanism of chemoresistance and radioresistance. Wang et al. uncovered that low-dose radiation was able to induce the secretion of exosomes rich in circ-METRN. The increased circ-METRN upregulated DNA damage molecule γ-H2AX expression, but γ-H2AX level returned to normal at 24 h, indicating enhanced DNA damage repair process in radioresistant glioblastoma cells [167]. Moreover, in cisplatin-resistant GC, the PI3K/AKT pathway, which positively regulates BRCA1 expression, can be activated by circAKT3 [141, 168]. Recently, circRNAs were found to regulate corresponding parent genes expression by binding to their cognate DNA locus to form R-loops [169]. This novel function of circRNAs has also been demonstrated in cancer. Namely, the interaction of circSMARCA5 with SMARCA5 could suppress the expression of SMARCA5, which is vital for DNA damage repair, and altered DNA damage repair modulated the sensitivity of breast cancer cells to cytotoxic drugs [142, 170]. In summary, these findings disclose the importance of circRNAs in cancer therapy by regulating DNA damage repair.

For cancer, most therapy strategies can efficiently eliminate part of tumor cells by inducing apoptosis. However, dysregulation of apoptotic signaling is instrumental to tumor cell survival and leads to therapeutic resistance. In recent years, evidence is accumulating to show that many circRNAs modulate apoptosis process in tumor cells via participating in the regulation of apoptosis-related proteins and signaling pathways. For examples, circCCDC66 was detected to be upregulated in cisplatin-resistant GC cells and tissues. CircCCDC66 depletion elevated cisplatin sensitivity due to its interaction with miR-618 to inhibit BCL-2 expression [143]. Liu’s group disclosed that loss of circDOPEY2 promoted translation of anti-apoptosis protein MLC-1 by decreasing ubiquitination and degradation of CPEB4. This function ultimately suppressed apoptosis and triggered chemoresistance of esophageal cancer (EC) cells [144]. BIRC3, also known as cellular IAP-2, belongs to the IAPs family and exerts a pro-survival and anti-apoptotic roles in tumors [171]. A recent study delineated that circDOCK1 could increase the expression of BIRC3 by sequestering miR-196a-5p, implying that circDOCK1 may be a significant biomarker and a valuable therapeutic target for oral squamous cell carcinoma in the future [172]. Besides, hsa_circ_0110757 could function by activating PIK/Akt pathway through hsa-miR-1298-5p/ITGA axis in glioma [173]. Noteworthily, previous reports have corroborated the close relationship between PIK/Akt signal pathway and BCL-2 expression, indicating that hsa_circ_0110757 might augment glioma resistance to temozolomide through inhibiting apoptosis [174, 175]. The tumor suppressor gene p53 is significant for apoptosis induction. Xu et al. uncovered that upregulated hsa_circ_0002874 could inhibit p53 expression via miR-1273f/MDM2 axis, leading to apoptosis inhibition and paclitaxel resistance in NSCLC [145].

Growing facts have elucidated that a plethora of circRNAs exhibit critical roles in multiple aspects of TME, such as angiogenesis, hypoxia, endothelial monolayer permeability (ECM) and immune surveillance [176]. Hypoxia is a typical feature of TME. The rapid proliferation of tumor cells promotes massive oxygen consumption and abnormal proliferation of tumor vascular structure, causing an imbalance between tumor oxygen supply and oxygen demand [177]. Hypoxic TME can confer greater viability of tumor cells during radiation and drug therapy by affecting drug efflux proteins expression, mitochondrial activity, apoptosis, autophagy and EMT [178]. It was reported that the expression of cZNF292 was strongly associated with hypoxia-induced radioresistance. Data from mechanism experiments suggested that cZNF292 deficiency elevated SOX9 nuclear translocation, which inhibited the activation of Wnt/β-catenin pathway, thereby enhancing the radiosensitivity of hypoxic HCC cells [159]. Consistently, some circRNAs, such as circNRIP1 and circASXL1, have been demonstrated to modulate hypoxia-associated resistance in chemotherapy [146, 147]. In addition, circRNAs can influence the efficiency of immunotherapy by regulating anti-tumor immunity [164, 165, 179]. For instance, Xu et al. disclosed that upregulation of circHMGCS1–016 promoted immunosuppression via sponging miR-1236-3 to increase CD73 and GAL-8 expression, which further contributed to the inhibition of CD8+ T and CD4+ T cells, thereby facilitating anti-PD-1 resistance in ICC [164]. Another example is that circFAT1 deletion improved the efficacy of anti-PD-1 immunotherapy through impairing cancer stemness and elevating CD8+ T cells infiltration into the TME. This effect was based on the role of circFAT1 in STAT3 activation [165].

Autophagy is an auto-catabolic process that can assist tumor cells to deal with therapeutic pressure and escape crisis [180]. To better address acquired resistance, having a clear understanding of the regulatory relationship between circRNAs and autophagy is of great importance. The regulatory roles of circRNAs in autophagy are mainly based on their function as miRNA sponges. CircPGAM1 was instrumental to drug resistance in laryngocarcinoma, whereas overexpression of miR-376a reversed this effect via decreasing ATG2A level [148]. CircCUL2 deficiency enhanced activation of autophagy by miR-142-3p/ROCK2 axis, thus promoting GC progression and cisplatin resistance [149]. Gao et al. have delineated that hsa_circ_0092276 was upregulated in doxorubicin-resistant breast cancer cells [181]. Another research team further uncovered that hsa_circ_0092276 increased ATG7 expression through binding to miR-384, leading to enhanced autophagy and doxorubicin resistance [150]. Moreover, the ULK complex made up of ULK1 or ULK2, mATG13 and FIP200 is important for autophagy activation [182]. A recent study found that ULK1 was highly expressed in imatinib-resistant chronic myeloid leukemia (CML) cells. The upregulation of ULK1 was mediated by the action of circ_0009910 in sponging miR-34a-5p [162]. Notably, further studies of circRNAs regulating autophagy-related therapeutic resistance in cancer through other biological functions are needed.

In recent years, increasing studies have suggested that abnormal expression of circRNAs can enhance EMT and endow tumor cells with therapeutic resistance. A typical instance is circ_0092367, whose depletion dampened epithelial splicing regulatory protein 1 (ESRP1) expression by acting as miR-1206 sponger. The downregulated ESRP1 promoted EMT and further contributed to chemoresistance to gemcitabine in PC [151]. Erlotinib has also been demonstrated to possess potential to treat PC patients. However, the development of acquired drug resistance would restrict its therapeutic efficacy. New evidence uncovered by Xu et al. show that the upregulation of circ_0013587 reversed erlotinib resistance in vitro and in vivo by binding to miR-1227 to increase E-cadherin expression [163]. Circ_0000079, derived from human USP1 gene, was markedly reduced in NSCLC patients resistant to cisplatin, which repressed EMT-induced chemoresistance by decoying FXR1 to block the formation of the FXR1/PRCKI complex in NSCLC [152].

Importantly, the resistance mechanisms associated with CSCs are also regulated by certain circRNAs. For instance, Zhao et al. found enhanced circRNA CDR1as and stemness markers, including SOX2, Nanog and OCT4, in cisplatin-resistant NSCLC cells. Mechanistically, CDR1as overexpression elevated stemness of cisplatin-resistant NSCLC cells through regulating miR-641/HOXA9 axis [153]. Moreover, circ_001680 upregulated the mRNA and protein level of BMI1 by sequestering miR-340, which further promoted the CSC population in CRC, thus accelerating cell proliferation, migration and irinotecan resistance [154]. Overall, clarification of the tight relationship between circRNAs and CSCs is vital in overcoming cancer therapeutic resistance.

The energy metabolism of tumor cells is significantly different from that of normal cells. For tumor cells, they prefer to generate ATP and lactic acid through the glycolytic pathway, which has been proved to induce cancer chemoradiation resistance. Extensive evidence has shown that some circRNAs are involved in glycolysis-induced chemoradiotherapy resistance in cancer. For instance, upregulation of circABCB10 augmented glycolytic metabolism to promote radioresistance by miR-223-3p/PFN axis in breast cancer. Breast cancer cells treated with glycolysis inhibitor 2-deoxy-D-glucose were found to be more sensitive to radiation than controls [160]. Notably, circGOT1 and its host gene GOT1 were identified to be upregulated in ESCC cells and tissues. Further mechanistic experiments showed that circGOT1 interacted with miR-606 to upregulate GOT1 expression, thereby facilitating aerobic glycolysis and cisplatin resistance [155]. Besides, lots of glycolysis-related genes expression were modulated by circRNAs, such as glucose transporter-1, lactic dehydrogenase A (LDHA) and hexokinases (HKs) [156‐158, 183]. Some time ago, Jiang et al. found that circARHGAP29 upregulated LDHA level via binding to and stabilizing c-Myc, resulting in increased glycolysis-mediated docetaxel resistance in prostate cancer cells [156]. HKs are involved in catalyzing the first step of glucose metabolism, and play essential roles in aerobic glycolysis in tumor cells. It was reported that exosome-delivered circDLGAP4 elevated HK2 expression by sponging miR-143, and the increased HK2 further enhanced glycolysis in doxorubicin-resistant neuroblastoma cells [157]. Consistently, the role of circ_0008928 in cisplatin resistance was positively modulating HK2 via miR-488 [158].

With the increasing understanding of circRNAs and m6A modification, it has been gradually uncovered that m6A is widely present in circRNAs, and involved in modulating their biogenesis, localization, degradation and translation. Since both circRNAs and m6A are key modulators in cancer therapeutic resistance, m6A modification is likely to affect the biological functions of circRNAs in cancer treatment resistance. Herein, we briefly summarized the recent research on the regulatory roles of m6A-modificated circRNAs in the treatment tolerance of malignancies such as HCC, hypopharyngeal squamous cell carcinoma (HSCC) and NSCLC (Table 3, Fig. 5).

As described above, m6A modification exerts a regulatory role in circRNAs. More noteworthily, abnormalities in circRNAs expression can take effects on m6A modification in cancer. In the past 5 years, these studies on m6A regulated by circRNAs have been accumulating, mainly focusing on CRC, HCC, BC, glioma and prostate cancer (Table 4, Fig. 6).