Functional Interactions Among Members of the MAX and MLX Transcriptional Network During Oncogenesis

Abstract

The transcription factor MYC and its related family members MYCN and MYCL have been implicated in the etiology of a wide spectrum of human cancers. Compared to other oncoproteins, such as RAS or SRC, MYC is unique because its protein coding region is rarely mutated. Instead, MYC’s oncogenic properties are unleashed by regulatory mutations leading to unconstrained high levels of expression. Under both normal and pathological conditions MYC regulates multiple aspects of cellular physiology including proliferation, differentiation, apoptosis, growth and metabolism by controlling the expression of thousands of genes. How a single transcription factor exerts such broad effects remains a fascinating puzzle. Notably, MYC is part of a network of bHLHLZ proteins centered on the MYC heterodimeric partner MAX and its counterpart, the MAX-like protein MLX. This network includes MXD1-4, MNT, MGA, MONDOA and MONDOB proteins. With some exceptions, MXD proteins have been functionally linked to cell cycle arrest and differentiation, while MONDO proteins control cellular metabolism. Although the temporal expression patterns of many of these proteins can differ markedly they are frequently expressed simultaneously in the same cellular context, and potentially bind to the same, or similar DNA consensus sequence. Here we review the activities and interactions among these proteins and propose that the broad spectrum of phenotypes elicited by MYC deregulation is intimately connected to the functions and regulation of the other network members. Furthermore, we provide a meta-analysis of TCGA data suggesting that the coordinate regulation of the network is important in MYC driven tumorigenesis.

1. The MAX/MLX network: identification, evolution and organization of the network members

1.1. From the discovery of v-myc to the MAX/MLX network

Since its identification in the late 1970s the MYC gene has been the subject of intense research that has linked it to all major hallmarks of cancer. The ability of MYC to influence such a broad spectrum of cellular activities may be explained by the fact that its protein product, MYC, belongs to an extended network of transcription factors (referred to here as the MAX/MLX network) with diverse activities, which in turn regulate many biological processes (Fig. 1).

Figure 1. Diagram of the MAX/MLX network.

All members of the MAX/MLX network and their reciprocal heterodimerization partners (indicated by the two-headed arrows) are represented. Green arrows and red lines indicate transcriptional activation and repression respectively. E-Box, Enhancer-box; ChoRE, Carbohydrate response element.

1.1.1. MYC and MAX: enabling MYC activity

The oncogene v-myc was among the first retroviral oncogenes described. Initially identified as the common transforming genetic element shared by the retroviruses MC29, CMII, MH2, and OK10, it was later shown to be derived from a cellular oncogene, termed c-MYC, (from now on referred to as MYC) [1–3]. The prominent role of MYC in human cancer etiology was revealed by the findings that MYC overexpression, due to chromosomal translocation or gene amplification, was a common event in Burkitt’s lymphoma [4, 5] and myeloid leukemia, respectively [6]. Amplification of the subsequently identified MYC paralogs MYCN and MYCL were also reported in neuroblastoma and small cell lung carcinoma, respectively, firmly linking MYC family proteins to the etiology of human cancers [7–10].

Initial sequence comparisons of MYC family genes revealed they share a predicted basic Helix-Loop-Helix Leucine Zipper (bHLHLZ) domain also found in several transcription factors and known to be involved in dimerization and DNA binding [11], suggesting the possibility that MYC proteins may homo or heterodimerize. Initial experiments failed to establish homodimerization of MYC protein at physiological expression levels [12–17]. However, the search for a MYC heterodimerization partner led to the discovery of a novel bHLHZ protein, MAX (MYC interacting factor X) [18]. Importantly, MAX also recognizes and heterodimerizes with MYCN and MYCL, implicating MAX in mediating the function of the three MYC family proteins, which have distinct and largely mutually exclusive spatiotemporal expression profiles during development [18, 19]. In contrast with MYC family members, Max is an abundant and stable protein, expressed in both proliferating and in resting cells, regardless of MYC levels [20]. These properties of MAX, together with its ability to bind DNA and repress transcription as a homodimer, led to the first model in which gene expression of MYC targets was controlled by the shift between repressive MAX-MAX homodimers and activating MYC-MAX heterodimers [21–24].

1.1.2. MXD proteins: MYC antagonists

Because MAX was found to be present under conditions in which MYC proteins are not expressed, and because MAX homodimerizes poorly, if at all, in vivo [25] it was considered that MAX may bind to other bHLHLZ containing proteins. Indeed, two MAX interacting proteins MXD1 and MXD2 (formerly known as MAD1 and MXI1) were soon identified [26, 27]. MXD1 and MXD2 are structurally similar, sharing 43% of their overall protein sequence, and are evolutionarily related. Moreover, with the exception of the bHLHLZ domain, which mediates their specific interaction with MAX, they are distinct from MYC. In vitro DNA binding and protein interaction assays confirmed that both MXD1 and MXD2 are unable to homodimerize, yet efficiently compete with MYC for heterodimerization with MAX. More importantly, both MXD1-MAX and MXD2-MAX heterodimers recognize the same E-box containing DNA consensus sequence as MYC-MAX heterodimers, but function as transcriptional repressors (MXD1-MAX) [26] or are unable to activate transcription (MXD2-MAX) [27]. The discovery of MXD1 and MXD2 added a new level of complexity to the earlier model by showing that MAX acts to mediate the activity of both MYC and MXD proteins which appeared to antagonize each other. In addition, the data suggested that it is the relative abundance and regulation of MYC and MXD proteins that define the transcriptional and biological outcome of network activity [26].

Shortly after the discovery of MXD1 and MXD2, four additional MXD-like proteins members were added to the network (MXD3, MXD4, MNT and MGA) [28–30]. While MXD3 and MXD4 are highly homologous with the previously identified MXD1 and MXD2 proteins, MNT and MGA differ significantly. However, apart from MGA, which contains an additional T-box domain and can both activate and repress transcription, the similar functional properties of MXD1-4 and MNT proteins (as all included in the MXD family proteins), which act as transcriptional repressors, further strengthen the model in which MXD proteins antagonize MYC in regulating many aspects of cell biology.

1.1.3 The MLX module: the metabolite sensing arm of the network

The existence of a parallel network which resembles the MYC/MAX/MXD transcriptional network was proposed after the discovery of a MAX-like protein called MLX. Two research groups independently identified MLX as the binding partner of MXD1, MLX4, and MNT proteins , and showed that MXD/MLX complexes mediate transcriptional repression of E-box containing promoters [31, 32]. Subsequently two large bHLHZ proteins were found to heterodimerize with MLX and regulate transcription by binding to E-box containing promoters [33, 34]. MONDOA was identified as an MLX heterodimeric partner by a two hybrid screen, while MONDOB was identified by sequence homology analysis. MONDOB, which had previously been identified as CHREBP (Carbohydrate Response Element Binding Protein) was also known as the WBSCR14 gene, since it maps to chromosome 7q11.23, a region usually deleted in Williams-Beuren Syndrome (WBS) [33]. CHREBP was subsequently confirmed to be an MLX interacting protein [34]. In contrast with the nuclear localized MYC, MAX and MXD proteins, MLX and MONDO proteins localize principally in the cytoplasm and can shuttle between the two compartments in response to glucose and other metabolic stimuli [35–38]. Because most of the literature refers to MONDOB as CHREBP, we will also refer to it as CHREBP in this review.

1.2. Evolutionary conservation of the MAX/MLX network

Underscoring the fundamental importance of the MAX/MLX transcriptional network in regulating cell biology is the fact that it has been conserved throughout evolution. Orthologs of human MAX MYC, MXD, MLX, and MONDO genes have been identified in the Placazoan Trichoplax adhaerens, considered the simplest metazoan [39]. Moreover, MYC and MAX orthologs also exist in Capsaspora owczarzaki, and in the choanoflagellate Monosiga brevicollis [39–41]. The presence of MYC and MAX orthologs in these two unicellular organisms, which are the closest living organisms to metazoa, indicates the emergence of the MAX module of the network even predates the evolution of multicellularity and metazoans.

A comprehensive analysis of the evolutionary conservation of the MAX/MLX network in metazoa identified or predicted the existence of network members in almost all surveyed species (Fig. 2A) [42]. Although the composition and sequence of network members has changed considerably during evolution, a core network consisting of MAX, MYC, MXD, MLX, MNT, and MONDO orthologs predominates and is the basis from which other networks emerged. The MAX/MLX network in vertebrates has acquired additional complexity due to the presence of multiple paralogous genes, which have distinct heterodimerization preference, gene expression profiles and functions. Meanwhile, lower order organisms usually have fewer members and, in the case of nematodes and flies, the network appears to have a distinct, yet balanced structure (Fig. 2B).

Figure 2. The MAX/MLX network in the animal kingdom.

A) Species specific distribution of MAx/MLX network members. “X”, the bHLHZ was found within a protein or expressed sequence tag; “*”, part or all of the sequence was found within a genetic region; “0”, the protein is known to be absent. B) Schematic representation of the four main network topologies. Solid arrows indicate experimentally verified interactions, whereas dotted arrows indicate debated or unknown interactions. Figure created from data in McFerrin and Atchley (c) 2011, with permission from the authors.

The Diptera lineage possesses a minimal network consisting of single orthologs of human MYC, MAX, MNT, MLX and MONDO genes (Fig. 2b) [36, 43–45]. Drosophila myc (dmyc), previously known as diminutive (dm), and max (dmax) were the first members of the network identified in Drosophila [43, 46]. Since then, many laboratories have confirmed the high functional redundancy between human and Drosophila myc, max and mnt orthologs [for review 47, 48]. Moreover a recent report demonstrated Drosophila mlx (known as bigmax) and its Mondo-like dimerization partner mio are involved in sugar sensing and utilization. This suggests the parallel and balancing functions of the MAX/MLX network that control cell proliferation and metabolism, have been also maintained during evolution [49, 50].

1.3. Organization and Structure of MAX/MLX network members

All members of the MAX/MLX network are defined by the presence of a bHLHLZ domain which mediates heterodimerization between family members and sequence-specific DNA binding (Fig. 3) [42, 51–53]. MYC family members are characterized by having the bHLHLZ domain localized at the C-terminus of the protein, and a transcriptional activation domain (TAD) located near the N-terminus. Within the TAD, MYC family members contain two well defined and conserved domains termed MYC Box I and MYC Box II (MBI and MBII) both of which are functionally important. While MBI serves as phosphodegron and is mainly involved in the regulation of MYC protein stability [54]; MBII mediates the interaction of MYC with multiple proteins such as TRRAP, GCN5, TIP60 and TIP48 [55–58], all of which have been identified as components of histone acetyltransferase complexes [59, 60]. Deletion of MBII inhibits most MYC functions, consistent with the importance of histone acetyltransferase complexes recruitment in MYC’s normal and pathological activities [61]. Moreover, the TAD region of MYC also mediates its interaction with other effectors of MYC activity such as the bromodomain protein BRD4 and the P-TEFb transcriptional anti-pause complex [62–64]. Recent studies have proposed that MYC-mediated augmentation of RNA polymerase II (RNA Pol II) dependent elongation is the principal mechanism through which MYC increases expression of its target genes [65–67]. Two other conserved domains MBIII [68] and MBIV [69] have also been identified within MYC proteins. MYC mutants lacking these domains are partially defective in their transcriptional activity and have a reduced tumorigenic potential.

Figure 3. Domain organization of the MAX/MLX network members.

A representative member of each subfamily is included. Colored box indicates functional domains. MBI-IV, MYC box domains; NLS, nuclear localization signal; BHLH, Basic helix loop helix; LZ, leucine zipper; SID, SID3-interacting domain; DCD, dimerization and cytoplasmic localization domain; TAD, transcriptional activation domain; TRD, transcriptional repression domain. The Glucose sensing domain contains six conserved regions named MONDO Conserved Region (MCR). Proteins size is not represented to scale.

While MYC proteins possess a TAD which acts as binding sites for factors involved in transcriptional activation, MXD proteins possess a repression domain known as the SIN3 interaction domain (SID) [70]. The SID domain is localized near the N-termini of MXD proteins and mediates interaction with the paired amphipathic helix 2 (PAH2) domain of SIN3 proteins, which in turn recruits histone deacetylase containing complexes [71–73]. The loss of repression activity resulting from the inhibition of MXD-SIN3 interaction suggests that histone deacetylation is the main function contributing to transcriptional repression by MXD proteins [70]. Hence, in addition to competition for MAX and DNA binding, another level of MYC and MXD protein biological antagonism occurs as a consequence of molecular antagonism manifested by histone acetylation and deacetylation.

The antagonistic roles of these proteins also seem to be manifested at the level of histone methylation. High levels of trimethyl Histone H3 lysine 4 (H3K4me3) are associated with active promoters [74, 75]. MXD dependent recruitment of histone demethylase RBP2/KDM5a induces histone H3K4me3 demethylation at the TERT promoter and correlates with decreased expression of TERT during DMSO-induced differentiation of HL-60 Human promyelocytic leukemia cells [76]. Interestingly, Drosophila Myc interacts directly with the fly ortholog of RBP2, known as Lid, but this binding directly inhibits the demethylase activity of Lid [77, 78]. Moreover, MYC has been reported to interact with MEN1 [79], a subunit of the mixed-lineage leukemia-1 (MLL-1) methylation complex, which mediates H3K4 methylation. Consistent with this, MYC’s genomic binding profile strongly correlates with H3K4me3 levels [80, 81]. Several reports have shown that changes in global levels of histone H3K4me3 correspond with increasing or decreasing MYC expression [82, 83]. In the MYC inducible lymphoma cell line p493-6, however, activation of MYC was found to promote H3K4me3 methylation at only a subset of MYC target genes [84]. Overall, these data strongly suggest that the MYC and MXD proteins antagonize each other by promoting and inhibiting, respectively, both histone acetylation and methylation.

MAX and MLX apparently possess neither TAD nor SID domains and therefore are considered transcriptionally inert. However, MLX differs from MAX because, like MONDO proteins, it contains a dimerization and cytoplasmic localization domain (DCD domain) at its C-terminus [33–35]. MONDOA and CHREBP also contain five domains named Mondo Conserved Regions (MCRI-V) within their N-termini, which mediate glucose sensing and transcriptional activity [see reviews 36, 85]. Moreover the existence of a MCR6 domain in MONDOA protein, which represents a potential binding site for glucose-6-phosphate (G6P), has recently been proposed [86]. MONDOA associates to the outer mitochondrial membrane, where it can also potentially sense other metabolites such as glutamine (through α-ketoglutarate), adenine nucleotides and non-glucose sugars [37, 87–91]. Thus, MONDO proteins sense multiple metabolites to directly link MONDO transcriptional activity to the metabolic state of the cell. While the exact mechanism through which MONDO proteins regulates transcription is still not completely clear, it has been shown that ChREBP associates, through its TAD domain, with p300 and mediates histone acetylation [92].

MGA differs from all other members of the network because it contains an additional T-loop DNA binding domain and can either act as activator or repressor depending on its interaction with MAX. Apart from MGA belonging to a repressive complex that binds to MYC and E2F responsive genes in G0, very little is known about its biological functions [30, 93]. Very recently MGA, together with MAX and SMAD4, has been shown to be involved in BMP signaling during gastrulation in Xenopus embryonic development [94]. Interestingly MGA is recurrently inactivated in high risk chronic lymphocytic leukemia [95] and in lung cancer [96], suggesting that it may function as a tumor suppressor gene.

2. Functional interactions among network members

Many reviews have previously discussed the role of MYC in controlling multiple aspects of cell biology [97–104]. Here we will focus on how the activities of the two arms of the MAX/MLX network are functionally interconnected. In particular, we will discuss how a broad range of common MYC activities (Fig. 4A) are shaped by other network members (Fig. 4B) and the manner in which these biological functions are affected when deregulated MYC drives cell transformation. For example, it is becoming evident that MXD members oppose MYC pro-apoptotic functions, suggesting that some aspects of an intact MAX/MLX network may be required for MYC induced tumorigenesis.

Figure 4. Schematic representation of the biological activities regulated by the MAX/MLX network members.

Green arrows and red lines indicate positive and negative regulation respectively. Dotted red lines indicate a functional connection that has not been directly investigated or for which only a limited amount of data is available. A) MYC centered view. B) MAX/MLX network centered view.

2.1. Proliferation and transformation: in vitro and in vivo experimental models

2.1.1 MXD Proteins: inhibiting MYC induced proliferation and transformation

MYC proteins promote proliferation by interacting with MAX and altering the expression of thousands of genes. However, all MXD proteins, MXD1-4 and MNT, also heterodimerize with Max, bind to the canonical MYC/MAX E-Box consensus sequence and repress transcription from an E-Box containing promoter [26–29]. Competition between MYC and MXD proteins on each of these levels suggests that these transcriptional regulators act as functional antagonists.

Overexpression of all MXD proteins inhibits MYC/RAS induced co-transformation, confirming that MXD proteins can antagonistically inhibit MYC activities in vitro [28, 105–111]. Moreover, overexpression of MXD1 and MXD2 proteins inhibits proliferation and induces accumulation of cells in the G0/G1 cell cycle phase in multiple in vitro experimental systems [109, 110, 112–122]. The ability of MXD proteins to block MYC induced transformation requires the HLHLZ domain, the basic region and the SID domain, indicating that dimerizing with MAX, binding DNA, and interacting with the SIN3 containing histone deacetylase complex are required for inhibition of MYC activity [107, 108, 111, 123].

Consistent with MXD1 inhibition of proliferation in vitro, transgenic mice expressing MXD1 under the control of a constitutive and ubiquitous β-actin promoter have reduced body size and weight compared to their wild type littermates. Both embryonic fibroblast and hematopoietic progenitor cells derived from Mxd1 transgenic animals show reduced proliferation [120]. Specifically, mouse embryonic fibroblasts (MEF) overexpressing MXD1 were more sensitive to contact inhibition and underwent proliferative arrest at lower density [120]. However, overexpression of MXD1 under control of the β-actin promoter is not sufficient to completely inhibit proliferation as it does not preclude embryonic development [120]. Similar results were obtained when MXD1 was specifically overexpressed in T and B lymphocytes. Although in both cases mature T and B cells were obtained, neither cell types were able to efficiently proliferate in response to multiple mitogenic stimuli in vitro [124, 125].

These results are consistent with the phenotype of Mxd1^−/− mice where Mxd1 deletion delays cell cycle exit in granulocytes but does not completely block cell cycle arrest and differentiation [126]. Overall the analysis of Mxd1 transgenic and Mxd1^−/− mice suggests that additional pathways, besides those directly affected by Mxd1 overexpression, are likely required to fully promote cell cycle arrest in collaboration with Mxd1. Indeed, Mxd1/p27Kip1 double knockout mice show a much stronger phenotype than ether Mxd1^−/− or p27Kip1^−/− mice, with markedly impaired cell cycle arrest and differentiation of double null granulocytes. [127]. Finally, Mxd1^−/− mice appear normal and are not prone to tumor development, indicating Mxd1 is involved in, but not absolutely required for cell cycle arrest.

The involvement of MXD proteins in the regulation of cell cycle in vivo has been further strengthened by the analysis of Mxd2 knockout mice. Mice carrying homozygous deletion of Mxd2 gene show several tissue abnormalities, including prostate and spleen hyperplasia/dysplasia [128]. Moreover, Mxd2^−/− T lymphocytes show increased proliferation in response to anti-CD3 plus anti-CD28 antibody stimulation and Mxd2^−/−mice develop B cell lymphoma at low frequency and long latency. These data indicate that Mxd2 may be a potential tumor suppressor, although additional events may be required for tumorigenesis. A formal proof of this hypothesis came from experiments showing that Mxd2^−/− mice are more susceptible to developing squamous cell carcinoma of the skin and malignant lymphoma when treated with weekly topical application of the carcinogen 9,10-dimethyl-1,2-benzanthracene (DMBA). Consistently, loss of Mxd2 also accelerates the development of fibrosarcoma and lymphoma in an Ink4^−/− background [128].

2.1.2. MNT: the major MYC antagonist

In contrast with Mxd1, overexpression or deletion of Mnt in mice results in very strong phenotypes. Mnt transgenic mice die during embryonic development and are significantly smaller than their wildtype littermates [108]. Preliminary analysis of the nervous system indicated defective proliferation and differentiation. Mnt deficient mice develop normally, yet are subviable, as more than 95% of the Mnt^−/− animals die within a few days of birth [129, 130]. These broad organismal effects most likely reflect the ubiquitous expression of Mnt during development. [108, 129, 131–134].

Several lines of evidence point to MNT as a major MYC antagonist. First, primary MEFs lacking Mnt phenocopy Myc overexpressing cells. Mnt^−/− MEFs proliferate at an accelerated rate, grow to high density and prematurely enter S phase [129, 131]. Analogous results were obtained when Mnt expression was silenced by the use of a Mnt-specific siRNA [131]. Interestingly, the expression cyclin E1 (Ccne1) and cyclin-dependent kinase 4 (Cdk4), both previously identified as Myc responsive genes [129, 135–137], are elevated in Mnt^−/− cells. Hence, increased activity of CCNE1/CDK2 and CCND1/CDK4 complexes probably drives the hyper-proliferative phenotype of Mnt null cells. Moreover, Mnt^−/− MEFs are more sensitive to apoptosis, display reduced senescence, and are more susceptible to RAS induced transformation both in vitro and in vivo [129, 131].

The close functional antagonism between MYC and MNT is further underscored by the observation that, in Mnt deficient fibroblasts, MYC expression is drastically reduced, suggesting that cells compensate for the lack of MNT by decreasing the levels of MYC. Nevertheless, such cells are still capable of proliferation [129]. Notably, silencing of Mnt in rat Myc^−/− fibroblasts promotes both proliferation and apoptosis under conditions of serum starvation. These results suggest a model in which MYC, in response to mitogenic stimuli competes for MAX binding and alleviates MNT-mediated repression. The demonstration that MNT/MAX heterodimeric complexes are converted to MYC/MAX complexes when cells enter the cell cycle supports this model [138]. In addition Mnt deficiency rescues the proliferation defect caused by Myc deletion in primary MEFs [138]. This is consistent with data showing that in Drosophila while complete inactivation of dmyc is lethal due to arrested development at the early larval stage, double mutants flies lacking both dmyc and dmnt grow larger and advance further in development [139, 140]. Overall, the data obtained in both mammalian and Drosophila experimental systems provide a strong argument for considering MNT as the principal MYC antagonist.

In this regard it is not surprising that conditional disruption of Mnt in mouse mammary epithelium and T cells promotes the development of mammary adenocarcinoma and T cell lymphomas, respectively [129, 132, 141]. However, in both cases, the latency of tumor formation, ranging from 6 to 22 months, indicates that secondary mutations are likely to be cooperating with loss of Mnt in tumorigenesis, as proposed for Mxd2 knockout mice. In the case of Eε-Myc induced B-cell lymphomas such secondary mutations generally act to disable the pro-apoptotic machinery and are required for the development of the B-cell lymphoma in this mouse model [142, 143].However, recent reports demonstrate that loss of Mnt inhibits Myc driven tumorigenesis by promoting apoptosis. Therefore MNT may promote or inhibit tumorigenesis depending on the context [144] (see apoptosis section).

2.1.3. The MLX module: an underappreciated player in tumorigenesis

While much research on the functions of the MYC/MAX/MXD module has been carried out, and, as described above, the metabolic functions of MLX and MONDO proteins have been well defined [36, 145–148], comparatively little is known concerning the role of MONDO proteins in tumorigenesis and the possible interactions between the MAX and the MLX branches of the network, specifically in relationship to MYC activity.

It is important to note that, in addition to MONDOA and CHREBP, MXD1, MXD4, and MNT are also able to form heterodimers with MLX [31, 32]. In vitro assays show that both MXD1 and MNT maintain their function as transcriptional repressors when bound to either MAX or MLX. [31, 32]. Therefore, some aspects of the phenotypes observed when expression of MXD1 and MNT are altered, may be mediated by the heterodimerization with MLX. MONDO/MLX heterodimers have been considered the positive transcriptional arm of the MONDO/MLX/MXD module. Interestingly, initial experiments investigating the transcriptional capacity of MONDOA or CHREBP suggested they may possess both transcriptional activation and repression functions. MONDOA/MLX heterodimers activated an E-box driven artificial luciferase reporter, while CHREBP repressed the same promoter either when expressed alone or in combination with MLX [33, 34]. However, both MONDOA and CHREBP transcriptionally activate many genes involved in glucose and lipid metabolism [85]. Moreover, MONDOA may act either as an activator or as a repressor depending on nutrient availability [87]. Therefore, the ability of MONDO proteins to activate or repress gene expression may be context specific.

Considerable experimental evidence indicates that modulation of MONDO protein expression can affect tumor cell growth. MONDOA knockdown in HA1ER transformed embryonic human kidney epithelial cells increases both proliferation and soft agar colony formation compared to control cells [87]. CHREBP overexpression has also been shown to reduce the ability of U2OS cells to form colonies [34], suggesting that loss of MONDO proteins may promote tumorigenesis. However, silencing of MONDOA in Acute Lymphoblastic Leukemia (AML) and CHREBP in pancreatic beta cells and human colon carcinoma HCT116 or HepG2 hepatoblastoma cells, decreases cell proliferation and promotes apoptosis [149–151]. These contradictory results may simply be due to the use of overexpression versus silencing experimental strategies to investigate the role of the MONDO proteins. Alternatively, they may reflect cell type intrinsic functions of MONDO proteins. A recent review discussing MONDOA and MLX expression, as derived from available databases, showed that high MONDOA expression can correlate with either good or poor cancer prognosis [147]. It is important to note that because MONDOA transcriptional function depends on availability of metabolites and heterodimerization with MLX, the mRNA levels of MONDOA alone may not reflect its activity. Moreover, it is possible that MONDOA may act either as an oncogene or a tumor suppressor depending on the cellular context.

To date, the ability of MONDO/MLX heterodimers to promote or antagonize MYC dependent transformation in a defined mammalian system has not been fully investigated. However, dmyc and mio genetically interact in Drosophila, where double homozygosity for a hypomorphic dmyc diminutive allele (dm) and a mio mutant (dmon1) resulted in a synthetic lethal interaction [36]. This result implies that the two modules of the MAX/MLX network functionally interact in invertebrates. Overall these results point toward functional cross-talk between the two modules of the network and suggest a collaborative relationship between MYC/MAX and MONDO/MLX heterodimers.

2.2. Role of the network in differentiation

The expression patterns of individual network proteins are distinct and cell context dependent. In general, MAX and MNT are mainly unchanged during the cell cycle and are broadly expressed throughout development and adulthood. In contrast, MYC expression peaks upon cell cycle entry at the G0-G1 transition and is expressed at a constant level during the proliferating cell cycle [152]. Conversely, MXD1, 2 and 4 expression is increased during cell cycle exit and differentiation [153]. MXD3 differs from the other MXD proteins because it is expressed during S phase in proliferating cells [154, 155]. While CHREBP and MLX have both been shown to be induced after growth factor or serum stimulation [32, 150], MONDOA expression profiles have not been extensively studied during cell cycle, development or differentiation.

High levels of expression of MYC inhibit differentiation of many cell types and promote the reprogramming of somatic cells to induced pluripotent stem cells (iPS cells) [156–158]. In hematopoietic cells induced to differentiate, MYC levels decrease while MXD1 expression levels increase [159–162]. Analysis of hematopoiesis in Mxd1^−/− mice revealed a delay in granulocyte differentiation [126] and overexpression of MXD1 has been shown to be sufficient to induce erythroid differentiation of murine and human erythroleukemia cells [163, 164]. This suggests MXD1 plays an active role during differentiation. There is evidence indicating that the contribution of MXD1 to the differentiation process may primarily be due to its role as a MYC antagonist. Indeed, expression of a chimeric protein containing the MYC bHLHLZ domain fused to the MXD1 transcriptional repression domain represses MYC target genes and efficiently induces differentiation of the erythroleukemia cell line K562 [163].

Furthermore, Mxd1 knockout mice, as well as Mxd2, Mxd3 null mice develop properly and do not show any major differentiation defect [28, 126, 128]. While compensation among MXD proteins may partly explain the lack of a developmental defect in Mxd deficient mice, these results strongly suggest that, while MXD proteins may participate in the differentiation process, they are not individually necessary.

Interestingly, rat PC12 pheochromocytoma cells, which lack the Max gene [165] and therefore would be expected to be lacking any MYC or MXD dependent activities, are capable of differentiating in response to nerve growth factor (NGF) and RAS [166–168]. In addition, Max-null murine embryonic stem cells expressing an inducible Max gene cease proliferating, differentiate, and eventually undergo apoptosis when Max expression is repressed [169]. However, cell viability and proliferation can be rescued by treating these cells with MAPK and GSK3β inhibitors (2i condition) or overexpressing OCT4 [169]. The fact that both PC12 and max null mECS can undergo differentiation, strongly suggests that MXD proteins are not directly responsible for induction of differentiation, at least as heterodimers with MAX. Instead, these results imply that activation of MAPK and GSK3β signaling pathways are required for mESC differentiation. Interestingly in the 2i condition, MYC expression is strongly diminished; suggesting that in absence of cell cycle arrest or differentiation signals MYC expression can be reduced without impacting proliferation. Genetic studies in Drosophila have demonstrated that dmax loss of function mutations are not as severe as dmyc deletion, suggesting that dmyc possesses functions that are independent of dmax [170, 171]. The finding that loss of human MAX gene is strongly associated with heritable pheochromocytomas is consistent with MAX independent function of Myc in mammalian cells [172, 173]. Interestingly, MYC overexpression in PC12 cells inhibits Ras-mediated differentiation in the absence of MAX [174].

Both MLX and MONDO protein expression are developmentally regulated but their specific involvement in cellular differentiation has not been fully evaluated. However, Mlx (Carroll unpublished), MondoA (Don Ayer, personal communication) and Chrebp [175] knockout mice develop to adulthood without major developmental defects, indicating that the MLX module is not required for either cell viability or differentiation during embryonic development.

2.3. Apoptosis: a matter of balance

The induction of apoptosis by the deregulated expression of MYC has been well documented [for reviews see 97, 98, 176–179]. Deregulated expression of MYC elicits the activation of an intrinsic tumor suppressor program, which induces apoptosis. Therefore, the establishment of MYC driven tumors requires the inactivation of the apoptotic machinery, which can result from the loss of tumor suppressor genes such as TP53, or gain of function of anti-apoptotic genes [142, 143, 180, 181]. Moreover, high levels of MYC also sensitize cells to many different extrinsic apoptotic stimuli such as UV, ionizing irradiation, pro-apoptotic signals and nutrient or serum withdrawal [182–187].

Among the network members, MXD1, MXD4 and MNT have been shown to inhibit apoptosis in several different experimental systems. The first observation that MXD proteins are involved in apoptosis came from the analysis of Mxd1^−/− mice. Mxd1^−/−granulocytes are more sensitive to cytokine withdrawal and other apoptotic stimuli such as γ-irradiation and doxorubicin [126]. Conversely, overexpression of Mxd1 protects against apoptosis induced by tumor necrosis factor-related apoptosis inducing ligand (TRAIL), Fas ligand, serum withdrawal, UV and cysplatin treatments [113, 117]. Moreover, MXD1 is induced by hypoxia and mediates hypoxia-induced resistance to doxorubicin [188]. Importantly, MXD1 overexpression also specifically inhibits MYC dependent apoptosis upon serum withdrawal [113]. The mechanistic basis for this may be at least in part due to MXD1 mediated repression of PTEN expression and promotion of the PI3K/AKT pathway. Both MYC and MXD1 bind to the PTEN promoter, indicating that MXD may block MYC induced apoptosis at least in part by repression of a subset of MYC target genes [189]. However, in a set of experiments designed to evaluate the functional equivalence of MYC and MXD1 bHLHLZ domains it was shown that a chimeric protein containing the N-terminal domain of MYC and the bHLHLZ of MXD1 was unable to induce apoptosis upon serum withdrawal [190]. An interpretation of this result is that the bHLHLZ of MYC and MXD1 differ in their ability to interact with additional factors involved in apoptosis. Notably the MYC, but not the MXD1 bHLHZ domain mediates the interaction with the transcription factor MIZ1. Binding of MYC to MIZ1 inhibits expression of the MIZ1 targets p21^Cip1 and p15^ink4b. [191–193] and is required for MYC induced apoptosis of rat cells upon serum withdrawal [194].

Similarly, overexpression of MNT inhibits MYC transcriptional activity and oncogenic properties, while loss of MNT promotes proliferation but also sensitizes cells to apoptosis [108, 129, 131]. Surprisingly, RNAi mediated silencing of MNT effectively sensitizes primary MEFs, 3T3 immortalized mouse fibroblasts, and Rat Myc ^−/− fibroblasts to serum withdrawal induced apoptosis. This Indicates that loss of MNT mediated apoptosis is not solely MYC dependent [131]. In agreement with this interpretation, MYC inactivation only partially rescues MNT dependent apoptosis in MEFs [138]; hence inactivation of MNT promotes apoptosis, which is further exacerbated by activation of MYC.

Analysis of developing T-cells in a mouse model harboring a conditional deletion allele of Mnt confirmed that loss of Mnt promotes apoptosis in single (CD4+ or CD8+) and double positive (CD4+/CD8+) T-cells and also when MYC is ectopically expressed either in T cells or in primary and immortalized MEFs, suggesting that MNT provides a pro-survival signal and protects cells against MYC mediated apoptosis [144].

The precise molecular mechanisms through which loss of MNT sensitizes cells to apoptosis remain unclear. Nonetheless there are some clues. Inappropriate and increased levels of TP53 and ARF19 have been reported in primary Mnt ^−/− MEFs and Mnt KD cells compared to control cells [129, 131]. However, acute CRE-mediated deletion of Mnt in primary MEFs failed to confirm these results [138]. Moreover, Mnt-deficient thymocytes do not show abnormal TP53 expression levels. Notably, while BCL2 can rescue cells from MYC induced apoptosis, overexpression of BCL2 in either the presence or absence of MYC fails to rescue the Mnt ^−/− apoptotic phenotype, suggesting that MYC overexpression and loss of MNT promote apoptosis, at least in part, through two different pathways [144] .

Another possible mechanism related to loss of MNT induced apoptosis comes from a recent study reporting a high level of reactive oxygen species (ROS) in cells overexpressing MYC and lacking MNT. Furthermore treatment of Mnt null cells with antioxidants rescues cell viability, indicating that increased levels of ROS might be responsible for the reduced viability of Mnt null cells [144]. The fact that Mnt ^−/− cells also have reduced oxygen consumption indicates that such cells have impaired mitochondrial metabolism. Activation of MYC has been shown to promote mitochondria activity and ROS production [144]. Consistent with this, MYC overexpressing Mnt^−/− cells have higher ROS than either Mnt ^−/− or MYC overexpressing wild type cells. Similar to MNT, MXD4 has also been shown to protect cells from apoptosis during OX40 mediated activation of T cells [195].

Underscoring the close connection between oxidative metabolism and the MAX/MXD network, MXD2 has been reported to be a target of Hypoxia Inducible Factor-1 (HIF-1) [196]. HIF-1 is a heterodimer comprised of the constitutively expressed HIF1b subunit (also known as ARNT) and an oxygen regulated HIF-1a or HIF2a (commonly known as EPAS1) subunit. HIF1 regulates glucose metabolism and has been linked to tumor growth, metastasis and angiogenesis. In renal carcinoma cells, HIF1 is hyper-activated because the tumor suppressor Von Hippel-Lindau (VHL), which mediates its degradation, is lost [197]. In both renal cell cancer cell lines and primary tumors, MXD is highly expressed and antagonizes MYC transcriptional activity [198]. Surprisingly cells with reduced expression of MXD2 proliferate faster in vitro, yet require MXD2 to efficiently form tumors in vivo [198]. A possible explanation is that MXD2 expression partially antagonizes MYC activity and prevents MYC induced apoptosis under conditions of limited oxygen availability. Moreover, MXD2-mediated, HIF-dependent downregulation of mitochondrial metabolism may provide a survival advantage by reducing ROS levels [196, 198]. Indeed, MXD1 is also induced by hypoxia in colon cancer cells and mediates resistance to doxorubicin under hypoxia [188].

Overall MXD proteins can be considered to promote survival under stress conditions particularly when MYC is highly expressed. Therefore, although decreased MXD protein expression may be beneficial for tumor growth when expression of MYC is not deregulated, low levels or complete loss of MXD proteins becomes counterproductive and may promote apoptosis when MYC is overexpressed.

Very little is known about the role of MONDOA and CHREBP in apoptosis and more specifically in MYC mediated apoptosis. Silencing of MONDOA in acute lymphoblastic leukemia and MYCN amplified neuroblastoma cells promotes apoptosis suggesting that MONDOA may cooperate with MYC in tumorigenesis [151](Carroll, Diolaiti, McFerrin and Eisenman, unpublished).

CHREBP by contrast has been shown to mediate pancreatic beta cell sensitivity to high glucose (glucotoxicity). In response to sustained high levels of glucose, CHREBP induces oxidative stress and apoptosis through its target gene TXNIP which promotes oxidative stress by inhibition of thioredoxin [199]. CHREBP is also required for HCT116 colon cancer and HepG2 liver cancer cells survival [150], however how CHREBP activity affects MYC dependent apoptosis remains to be understood.

2.4. Growth and metabolism: does the MLX module enable nutrient addiction in Myc transformed cells?

Early studies defining MYC transcriptional targets provided a list of genes involved in several metabolic processes, indicating that MYC not only drives proliferation but also promotes the activity of both catabolic and anabolic metabolism [200–204] Recently the role of MYC as a master regulator of growth and metabolism has become widely accepted [102, 104, 205] MYC not only promotes the activity of all three RNA polymerases [206–208], but also directly promotes the expression of many genes involved in glucose and glutamine metabolism [209–216] as well as lipid and nucleotide biosynthesis and mitochondrial activity [104, 217–219] Moreover MYC regulates ribosome biogenesis and protein synthesis [220]. Overall, MYC activation promotes an extensive reprogramming of cellular metabolism in order to fuel anabolic pathways and provide the building blocks for cellular growth and thus, proliferation.

In contrast to the large amount of data available on MYC, the role of MXD proteins in regulating metabolism has not been systematically addressed. Interestingly MNT, as well as MXD1 and MXD4, can bind not only to MAX but also to MLX. Therefore, MXD proteins can potentially control cellular metabolism by influencing the transcriptional activity of both MAX and MLX modules. However, to date only MNT has been linked, albeit indirectly, to metabolism. CRE-mediated deletion of Mnt in T cells results in diminished oxygen consumption rate (OCR) and increased ROS production, implying an imbalance in cellular oxidative metabolism [144]. Moreover, miR-210, a hypoxia inducible miRNA, targets MNT to promote cellular proliferation under hypoxic conditions. Since both miR-210 overexpression and MNT silencing promote expression of many metabolic genes previously identified as MYC targets, it is possible that this effect is mediated indirectly by increased MYC transcriptional activity [221].

Both MONDOA and CHREBP play a crucial role in metabolism [36, 85, 222]. Both the subcellular localization and transcriptional activity of MONDO proteins are regulated by glucose. In general, upon glucose stimulation, MONDO/MLX heterodimers accumulate in the nucleus and activate transcription of their target genes. However, additional mechanisms control MONDO expression and transcriptional activity [223–225]. CHREBP and MONDOA are expressed in most tissues, but while CHREBP is enriched in liver [175], MONDOA is highly expressed in skeletal muscle [33]. This differential expression profile also appears to reflect the major biological functions of these proteins. Glucose induced CHREBP activation in the liver promotes the conversion of glucose to lipid. Indeed Chrebp^−/− mice exhibit hyperglycemia and diminished serum triglycerides [175]. Gene expression analysis and chromatin immunoprecipitation experiments have shown that CHREBP activates and directly binds to many glycolytic genes and to key regulators of lipogenesis such as Acetyl coA carboxylase (ACC), fatty acid synthase (FAS) and sterol-coA desaturase (SCD) [226–230]. MONDOA instead has been shown to regulate mainly glycolytic genes [37, 91]. Moreover both CHREBP and MONDOA regulate the common target gene TXNIP implying that they can mediate, at least in part, similar biological processes [37, 92].

As discussed in the next section, both MYC and MONDO proteins have been detected bound to the promoters of specific metabolic genes indicating that these proteins functionally cooperate to orchestrate cellular metabolism. [91, 231–233]. While MYC has been shown to be required for CHREBP and glucose-dependent activation of glycolytic genes [231, 232], it is not known whether MYC requires MONDO activity to properly reprogram metabolic pathways and drive proliferation. It is possible that MYC and MONDO may transcriptionally regulate common target genes; although it cannot be excluded that MONDO proteins may affect MYC biological activity by regulating a complementary subset of genes. MYC driven tumor cells have been shown to be dependent on high level of either glutamine or glucose [185, 186, 234] and inactivation of metabolic genes such as LDH-A (a MYC and MONDOA common transcriptional target) inhibits MYC dependent tumorigenesis [204, 235]. Interestingly MONDOA responds to both glucose and glutamine and has been proposed to coordinate glycolysis with glutaminolysis by directly sensing TCA cycle function to regulate TXNIP expression and modulate glucose uptake and glycolytic rate [87, 146]. Therefore, it is possible that altering MONDO function would negatively impact MYC oncogenic activity.

2.5. The target gene enigma

Based on the biochemical, transcriptional, and functional studies described above, it has been suggested that MAX/MLX network members may bind to common target genes. This model is also supported by the fact that MXD1 and MXD2 specifically inhibit MYC transforming activity but are less effective in blocking E1A mediated transformation. [105, 109]. Hence MXD proteins do not behave as general inhibitors of transformation, but specifically antagonize MYC activity. Indeed, there is compelling evidence that MXD1 and MNT bind to MYC target genes in vivo. However, very little genomic binding data are available for the other network members. MYC and MXD1 have been shown to directly bind and respectively activate and repress relevant genes such as TERT, CCND2 and PTEN [189, 236–243]. In each of these cases MYC/MAX heterodimers could be replaced by MXD1/MAX heterodimers, resulting in histone deacetylation and transcriptional repression through recruitment of an HDAC containing complex. In addition, MXD1 recruits RBP2 to the TERT promoter, which mediates histone H3k4me3 demethylation [76]. Other studies have reported an antagonistic role for MYC and MXD1 in controlling the expression of other important genes such as CDC25A [244, 245] and the three EIF4F complex subunits EIF4EF, EIF4EAI, EIF4EGI [246, 247]. However the direct binding of MXD1 to the promoters of these genes has not been examined.

Interestingly, the most comprehensive analysis of MYC and MXD target gene overlap has been performed in vivo. Overexpression of MXD1 in lymphocytes inhibits expansion, maturation, and growth of T cells. Gene expression analysis showed that over 80% of the genes repressed by MXD1 have previously been found to be induced by MYC [124]. These results strongly support the hypothesis that both MYC and MXD1 control the expression of a common set of target genes. Similar results were obtained by comparing the expression profiles of mammary adenocarcinomas resulting from either the overexpression of MYC or the conditional deletion of Mnt in murine mammary gland epithelial cells [132]. In this case the tumor gene expression profiles were basically indistinguishable and MYC and MNT genomic binding profiles were highly overlapping, strongly arguing that MYC and MNT regulate the same target genes in vivo.

While there is evidence showing that both MYC and MXD proteins reciprocally control the expression of a large set of common genes, it is not known whether MXD/MLX heterodimers share target genes with MXD/MAX or MONDO/MLX heterodimers. So far, the specific biological functions of MXD/MLX heterodimers, if any, are completely unknown. By contrast, the role of MONDOA and CHREBP in the regulation of glycolysis and lipogenesis has been well established [49, 85, 248, 249]. Surprisingly, MONDOA and CHREBP seem to bind to a much smaller number of genes than MYC. The fact that CHREBP binds to a tandem E-box spaced by 6 nucleotides (Carbohydrate response element or ChoRE) may in part explain this difference [250–252]. Moreover, while MYC target genes regulate many biological processes, most of the MONDO/MLX target genes are involved in metabolism. Specifically, when a constitutively transcriptionally active version of MONDOA definedΔN237NLS-MondoA was overexpressed in C2C12 myoblasts, only 81 genes were activated more than 1.8 fold. Among these, 53% and 29% of the MONDOA targets genes were respectively involved in metabolism and cell growth [91]. Among these genes, MONDOA was shown to directly bind and regulate the expression of LDHA and HKII, two previously reported MYC target genes, and PFK3FB [204, 213, 253]. Importantly, both MYC and MONDOA bind to the promoter of all three genes in C2C12 cells [91]. Although these results suggest a coordinate regulation of metabolic genes by MYC and MONDOA, a detailed analysis of the specific and relative contribution of MYC and MONDOA to the expression of these genes has not been evaluated.

MYC and CHREBP are both required for the glucose dependent activation of the Liver-type pyruvate kinase gene (PKLR). MYC and CHREBP bind to the PKLR promoter in presence of glucose. Silencing of MYC expression or inhibition of MYC/MAX heterodimerization decreases the binding of CHREBP/MLX to the PKLR gene promoter and its expression [233].These data suggest that MYC, through an unknown mechanism, facilitates the recruitment of CHREBP to the PKLR promoter. However the reciprocal analysis was not performed leaving it an open question as to whether CHREBP affects MYC transcriptional activity or how general this cooperation may be.

Overall, these data strongly argue for a crosstalk between MAX and MLX modules; however the full extent of the transcriptional and functional overlap between them has not been rigorously established. This will require detailed analyses of genomic binding, epigenetic changes, and gene expression for all network members in specific biological contexts.

3. The balance among network components determines distinct cellular states: a comprehensive model

Overall the results discussed so far can be summarized by the model shown in figure 5, where external signals define the balance between MYC, MXD and MONDO proteins and promote distinct cellular fates. High levels of MYC drive growth and proliferation while the switch from MYC to MXD proteins promotes cell cycle arrest and differentiation. However MXD family expression alone is not sufficient to completely arrest proliferation. Therefore, additional signals are necessary to inhibit MYC expression or promote MYC degradation in order to allow MXD proteins to efficiently counteract MYC activity. An apoptotic response occurs in situations with extremely high levels of MYC activity. MYC driven apoptosis may be further facilitated by loss of MXD, MNT or MONDO activities or under stress conditions such as fluctuation in nutrient availability, hypoxia or chemotherapy.

Figure 5. Comprehensive model of the MAX/MLX network and its role in determining cellular fates.

External signals (top) determine the expression levels of the network members, which in turn promote alternative cellular state (bottom). High levels of MYCc sensitize cells to pro-apoptotic stimuli. Loss of function of MXD1, MNT or MONDOA further promotes apoptosis in MYC driven tumor models. MGA has been reported recurrently lost in leukemia and lung cancers. Because MAX and MLX are transcriptionally inert and have multiple heterodimerization partners, they have been excluded from the representation for simplicity.

In order to support proliferation, MYC promotes the metabolic reprogramming of the cell. Therefore coordinate regulation of metabolic genes by MYC and MONDO proteins may be required to sustain tumor cell viability. In support of this model, MYC binds to the promoter and appears to regulate the expression of both CHREBP and MONDOA, and inhibiting MONDO protein expression promotes apoptosis in several experimental system [83, 146, 150, 151].

The prevailing view of MYC function has been that MYC/MAX heterodimers bind genomic DNA and recruit different transcriptional co-regulator complexes to modulate gene expression – in essence considering MYC as a standard transcription factor albeit with a very large number of gene targets [103]. However two recent reports have taken another view, namely that MYC acts as a universal amplifier of gene expression [66, 67]. The authors proposed a new model in which MYC, instead of acting as a molecular ON/OFF switch for a specific subset of target genes, globally amplifies the pre-existing gene expression program already in place in in a given cell. This general augmentation of gene expression occurs through MYC recruitment of P-TEFb, a complex that alleviates transcriptional pausing by RNA Pol II, and therefore promotes transcript elongation. While the “universal amplifier” model accounts for some properties of MYC, such as the wide variability of gene targets identified in different studies, its general validity is being hotly debated [254–257]. Moreover, whether and how the activities of the other members of the MAX/MLX network relate to this model remains an open question. Nonetheless, it is possible to imagine that, by competing with MYC for its genomic targets, MXD proteins may also globally counteract MYC’s anti-pausing activity. Alternatively MXD proteins may recruit pausing factors and inhibit RNA Pol II elongation. Thus, in the context of the “universal amplifier” model MXD proteins may be required to attenuate MYC pro-proliferative transcriptional activity, thereby weakening the ongoing transcriptional program and allowing other factors to promote cell cycle arrest or differentiation.

Interestingly, the amplifier model can also be considered in the context of cooperative regulation of genes involved in metabolism. MYC has been shown to regulate the expression of many metabolic genes, a subset of which are also targets of MONDOA and CHREBP. Perhaps MONDO proteins act as specific selectors of critical genes, with their expression mainly dependent on MONDO activity, while MYC binding would further amplify their expression through its RNA Pol II anti-pausing activity. This is consistent with the need to coordinate nutrient availably with mitogenic signals and with the ability of MYC to promote proliferation and cellular metabolic reprogramming. In this model, activation of MYC in the presence of nutrients would promote cellular metabolic reprogramming by co-opting and enhancing the MONDO-dependent transcriptional program. On the other hand, the absence of nutrients such as glucose would turn off MONDO transcriptional activity and prevent MYC driven metabolic reprogramming and proliferation, thus triggering apoptosis in a deregulated MYC state.

4. The MAX/MLX network in human cancer

Accumulating evidence indicates that the biological outcome of MYC deregulation is dependent on the expression and function of the other members of the network, which either antagonize or collaborate with MYC. Based on the organization of the network, it is reasonable to consider that changes in the expression of one network member may have a significant biological impact by modifying the relative abundance and activities of other functionally distinct network members. This notion is based on several considerations. First, MYC, MXD, or MONDO family proteins are expected to compete for binding to limiting amounts of MAX or MLX. Second, alternative heterodimers will bind the same or related DNA sequences and thus compete for binding sites, and third, network members impose different consequences in terms of gene target expression.

The importance of maintaining a homeostatic balance among network components is also suggested by their extremely tight regulation of expression. For example, MYC represses itself in a negative feedback loop which is often inactivated in tumor cells [258]. Also, MYC family members exhibit transcriptional crosstalk. For example overexpression of MYCN leads to downregulation of MYC in neuroblastoma and B cell lymphoma [259, 260]. In addition, loss of MNT is balanced by a decrease in MYC and MYCN expression [129, 130]. Moreover, MYC represses MXD4 expression by its interaction with MIZ1 [261, 262]. By repressing MXD4 expression MYC may not only promote its own association with MAX, but potentially influence the balance between MXD/MLX and MONDO/MLX heterodimers. Finally, MYC can also influence expression of the MONDO factors. For example MYC binds to the CHREBP gene promoter [ENCODE data and 83] and both MYC and MYCN overexpression enhances MONDOA expression and nuclear localization in several tumor cell lines (Carroll, Diolaiti, McFerrin and Eisenman, unpublished).

Interestingly, although MXD proteins antagonize MYC activity in many experimental systems, there are contradictory evidence that MXD family members are lost in human cancers [172, 173, 263–275]. As discussed above, it is generally accepted that MXD proteins are not frequently lost during oncogenesis because they are mainly expressed in differentiated tissues where MYC expression is low or absent. However, another explanation may be that MXD and MONDO proteins may be required to suppress MYC pro-apoptotic activity [276]. If correct it would be expected that MXD proteins should be widely expressed in tumor cells. Consistent with this hypothesis, MXD genes expression has been detected in several hematopoietic and solid tumors [263, 264, 266, 274, 277, 278]. However, a comprehensive analysis of MXD expression in human tumors has not been reported.

Even less is known about MONDO protein expression in human cancers. By sensing nutrient availability and regulating metabolic processes such as glycolysis and lipogenesis, Mondo proteins may profoundly impact tumor growth. MONDOA is overexpressed in acute lymphoblastic leukemia and its silencing promoted differentiation and apoptosis in leukemic cells [151] Recent data from our laboratory show that MONDOA silencing promotes apoptosis in cells expressing high level of either MYC or MYCN, suggesting that MONDOA may be required by MYC for properly reprogram cellular metabolism and for tumorigenesis (Carroll, Diolaiti, McFerrin, and Eisenman, unpublished).

4.1. The MAX/MLX Network as viewed from the TCGA

To evaluate the profile of the MAX/MLX network in human tumors, we analyzed the relative expression levels of the network members using the Pan-Cancer data from The Cancer Genome Atlas (TCGA) database (http://cancergenome.nih.gov/). This publically available dataset is mean normalized to reflect the variability of gene expression across 6609 samples representing twenty-four diverse tumor types, including both primary tumor and normal tissue where available. Using spearman correlation as a distance measure during hierarchical clustering [279], samples were largely clustered according to tissue and tumor type, with clusters of normal tissue being grouped with the corresponding primary tumor (Fig. 6A). Interestingly, it appears that most tumor types can be distinguished according to a specific expression pattern of MAX/MLX network members suggesting that each tissue has a distinct balance among members of the network. Lung and breast cancers are the exception, as the expression pattern of network members within these tissues is highly heterogeneous causing them to be broadly interspersed with the other tumor types and not distinctly clustered. For this reason, lung and breast samples were excluded from Figure 6A to avoid obfuscating the other tumor type clusters. The reason for such difference is not known, but it may in part reflect the great heterogeneity of lung and breast cancers.

Figure 6. Analysis of the MAX/MLX network in human cancer.

A) Relative distribution of the network members across TCGA tissue types, each represented by a different color (“tumor type column”). “Sample” column: Pink normal tissue; Blue, Primary Tumor; Red, Acute Myeloid Leukemia; Grey, Metastatic tumor; Green Recurrent tumor. Due to the heterogeneity in expression, breast and lung samples have been excluded from this representation to more readily identify the distinct expression patterns of the other tumors. B) Pearson correlation of network member expression levels among all samples (including lung and breast). Orange: positive correlation, blue: negative correlation. C) Median of expression values within each tumor type (colored coded squares) are shown for each network member. Black dashed line, median expression of all genes across tumor types; Grey dashed lines, top and bottom 25% of genes.

In general, MAX, MLX and MNT genes expression show minimal variability among tissues including either normal or tumor samples, as indicated by the almost homogeneous black color. This is consistent with the fact that they are expressed both widely and constitutively. In contrast, MYC family members show a variable expression profile being enriched or depleted dependent upon the tumor type. In particular, while MYC is expressed at high levels in a broad spectrum of cancers, MYCN and MYCL are highly expressed in a smaller number of tumors consistent with their more restricted tissue expression profile. Surprisingly, many cancers express high levels of multiple MYC family members (e.g. Ovarian cancer, Acute Myeloid Leukemia and Lower Grade Glioma).

MXD genes also show variable expression profiles. MXD2 has higher relative expression in kidney where it has been shown to be required for tumor growth [196, 198] Interestingly MXD3 seems to be more highly expressed in the same samples as MYC family members, consistent with the reported connection between MXD3 expression, proliferation and protection from DNA damage induced apoptosis [154]. Unexpectedly MXD1 is present at higher levels in a subset of tissues where MYC is also more highly expressed. Finally, both MONDOA and CHREBP show a tissue specific profile perhaps reflecting alternative metabolic functions and requirements of different tissues and related tumors. The observation that the network expression profile is tissue specific is important because it implies different tissues may have diverse thresholds of susceptibility to MYC deregulation and variable biological outcomes depending on the specific network structure.

Moreover, because it is known that the expression levels of the network members influence each other’s expression profile, we asked if there is any relationship between the expression levels of each network member by calculating the Pearson correlation between network genes using all available samples, including those from lung and breast cancer (Fig 6B). Intriguingly, five of the core network members (MAX, MYC, MXD1, MNT, MONDOA) as well as MGA are all positively correlated with each other and form a distinct clade (within dashed line). This again suggests their expression patterns are potentially linked and supports a model in which MXD proteins may be required to balance MYC pro-apoptotic activity, while MONDOA would collaborate with MYC to sustain the increased metabolic demand of cancer cells.

Interestingly, MYC drives the formation of tumors possessing different metabolic profiles when expressed in cells of different origins [280]. Conceivably, MYC driven metabolic reprogramming can vary depending on the cell type specific expression profile of MONDOA, CHREBP and other metabolic genes. Moreover, although MONDO proteins seem to mediate at least partially overlapping metabolic pathways, it is possible that they possess specific functions. As shown by figure 6A–C their expression, especially CHREBP, is highly variable among samples, therefore potentially influencing MYC-induced metabolic reprogramming in a tissue specific manner.

Overall, the analysis of TCGA data represented in fig. 6A,B suggests that MXD genes expression is not globally downregulated during oncogenesis. However, while the analysis highlights the tissue-specific distribution of the network members across tumor types, it does not provide comparative information regarding the relative expression level among network members within tumor types.

Figure 6C shows the relative log (base 2) gene expression profile of all network members within each tumor type. Although the median expression level is a simplified statistic and does not reflect the heterogeneity associated with each tumor type, the data shows that most network members have comparable ranges of expression and that their expression is, in general, greater than the overall median gene expression. The exceptions being MYCN, MYCL and ChREBP, which have more tissue specific expression, and MXD3, which is expressed at a lower level in all tissues. These data, together with the fact that loss of MXD or MONDO genes promotes MYC driven apoptosis supports a model where extreme network imbalance, as a result of the loss of a network member, may not be beneficial for tumor cells and suggests an intact and functional MAX/MLX network may be required for tumorigenesis.

5. Concluding remarks

In conclusion, while MYC undoubtedly is the major factor in driving tumorigenesis, the presence and activities of the other members of the network is likely to modulate the biological consequences of MYC deregulation. Understanding the specific biological functions of each network member in both normal and pathological conditions will lead to a deeper understanding of how deregulated MYC alters cellular behavior and potentially provide new opportunities for inhibiting tumorigenesis.

Highlights.

• A comprehensive model of the MAX/MLX intra-network interaction is proposed

• MAX/MLX network members are widely expressed in both normal and malignant tissues

• The MAX and MLX modules coordinate proliferation and metabolism

• Balanced expression among network members contributes to cell fate determination

• Different human cancers possess a distinct MAX/MLX network organization