HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Boyd, K. E. Articles by Farnham, P. J. Search for Related Content PubMed PubMed Citation Articles by Boyd, K. E. Articles by Farnham, P. J.

---

Review Article

Identification of Target Genes of Oncogenic Transcription Factors

McArdle Laboratory for Cancer Research, University of Wisconsin Medical School, Madison, Wisconsin 53706

	Abstract

Top Abstract Introduction Oncogenic Transcription Factors... Oncogenic Transcription Factors... Methods That Can Aid... Conclusions References

 
Disregulation of many transcription factors is associated with the development of human neoplasia. Transcription factors regulate cell growth, differentiation, and apoptosis by binding to specific DNA sequences within the promoter regions of growth-regulatory genes and modulating expression of these genes. This simple model is complicated by the fact that mammalian transcription factors are often members of large protein families that bind to similar DNA sequences. This raises the question as to whether members of a particular family regulate expression of overlapping or unique sets of genes. This review is focused on addressing this question using the Ets, Myc, and E2F transcription factor families as examples. Deregulated activity of some, but not all, members of these families is observed in cancer. Here, we summarize the data illustrating the concept that binding of individual members of these families of factors can result in promoter-specific responses and review the studies that have provided some insight into how target gene specificity is achieved. Since, for all of these oncogenic transcription factors, it remains unclear exactly which target genes are important in neoplasia, we have also reviewed the many approaches researchers are using to identify target genes of the various Ets, Myc, and E2F family members.

	Introduction

Top Abstract Introduction Oncogenic Transcription Factors... Oncogenic Transcription Factors... Methods That Can Aid... Conclusions References

 
Most nuclear oncogenes have been identified as promoter-specific transcription factors that can cause neoplastic transformation when their activity is increased to inappropriately high levels. These promoter-specific transcription factors regulate the activity of the basal transcription complex (which includes RNA polymerase II and initiation factors TFIIA, TFIIB, TFIID, TFIIE, TFIIF, and TFIIH) formed near the transcription start site of the gene. Promoter-specific transcription factors can control the switch from inactive to active chromatin, stabilize the transcription complex on the promoter DNA, recruit other factors to the complex, or mediate the transition from initiation to elongation. Promoter-specific transcription factors usually have two modular domains, a DNA binding domain and a regulatory domain. Several classes of site-specific transcription factors have been identified based on their DNA binding and/or dimerization domains. Commonly found dimerization structures include helix-loop-helix motifs and leucine zippers. DNA binding domains include basic regions, which are often found in combination with helix-loop-helix and/or leucine zippers, homeobox domains, which consist of a helix-turn-helix structure, and the zinc finger motif that consists of pairs of cysteine-cysteine or histidine-cysteine residues that bind zinc and form finger-like structures that interact with specific DNA sequences (1). There are multiple examples of transcription factors that use each type of DNA binding motif.

Increased activity of promoter-specific factors is achieved in many ways in human cancers (Fig. 1). For example, leukemia-associated chromosomal rearrangements can result in expression of a novel transcription factor (2); many of these rearrangements involve members of the Ets family of transcription factors. Other cancers, such as Burkitt's lymphoma and neuroblastoma, are thought to be due to increased levels of structurally normal members of the Myc family of transcription factors (3, 4). Finally, in certain cases such as retinoblastomas, mutations in upstream regulators result in increased activity of members of the E2F family of transcription factors (5). Each of these three families of transcriptional regulators, Ets, Myc, and E2F, is composed of multiple members, and all members of a given family contain highly related DNA binding domains. In general, in vitro DNA binding assays indicate that factors having similar DNA binding motifs bind to similar DNA sequences. In fact, it is this commonality of DNA binding motifs that creates the problem that will be discussed in this review. If a group of different transcription factors can all bind to the same sequence, how does one determine which member of the family regulates a particular target gene? Of particular importance to cancer is the corollary question, how can increased levels or activity of one member of a family of transcription factors cause neoplastic transformation whereas high levels of other, very similar, family members do not lead to the loss of cell growth control?

View larger version (29K): [in this window] [in a new window]   Figure 1.    The activity of oncogenic transcription factors can be modulated in a variety of ways in human tumors. (A) Many leukemias and solid tumors are characterized by the presence of a chromosomal translocation that creates a novel fusion protein. In the example shown, a member of the Ets family, the FLI-1 protein, is fused to the EWS protein to create a novel transcription factor that retains the ability to bind to Ets sites. (B) The increased levels of transcription factors in tumor cells has been shown to be due to gene amplification. The extra copies of the genes encoded in the amplicons can be found as extrachromosomal elements called double minutes or as integrated chromosomal structures called homogeneously staining regions (HSRs). In the example shown, the HSRs contain the N-myc gene and surrounding chromosomal DNA. In neuroblastomas, more than 100 copies of the N-myc gene have been found. (C) Functional loss of a negative regulator can increase the activity of pre-existing transcription factors. In the example shown, the activity of pre-existing E2F family members is increased due to loss of the negative regulator RB or due to the increased phosphorylation state of RB caused by alterations in the abundance or activity of cyclin-dependent kinases.

	Oncogenic Transcription Factors are Members of Multigene Families

Top Abstract Introduction Oncogenic Transcription Factors... Oncogenic Transcription Factors... Methods That Can Aid... Conclusions References

 
The Ets Family of Transcription Factors.
Members of the Ets family of transcription factors play important roles in organogenesis, hematopoiesis, B-cell development, and signal transduction (for a review see Ref. 6). These proteins all bind to a purine-rich consensus sequence (C/A GGA A/T) and transcriptionally activate a number of genes that contain these Ets binding sites. The Ets DNA binding domain is composed of 85 amino acids; structural analysis indicates the presence of three helices and a four-stranded ß sheet similar to structures of helix-turn-helix motifs found in several other mammalian and bacterial transcription factors. Disregulation of Ets family members is often associated with neoplasia. For example, v-ets, the first member of the Ets gene family to be identified, is co-transduced with the v-myb gene in the E26 retrovirus that expresses a GAG-MYB-ETS fusion protein. Similarly, a newly identified Ets protein, ESX, becomes overexpressed at an early stage of human breast cancer development (7). Members of the Ets family of transcription factors are often found as fusion proteins in leukemias and other solid tumors. In accordance with a postulated role for Ets proteins in regulating cell growth control, overexpression of various Ets proteins (e.g. Ets1 and Ets2) can lead to neoplastic transformation of cells in culture and can cause tumors in nude mice.

Ets family members provide many interesting examples of how chromosomal rearrangements can lead to neoplastic transformation. In particular, almost all leukemias are characterized by a particular DNA rearrangement that results in fusion of a transcription factor to another protein. In some cases, the second protein is also a transcription factor; in other cases, the function of the fusion partner is as yet unknown. An Ets family member that is a common target of gene rearrangements in human leukemias is a site-specific DNA binding transcription factor called tel (translocation, ets, leukemia). Tel is rearranged in acute myelogenous leukemia (AML), chronic myelomonocytic leukemia (CMML), myelodysplastic syndrome (MDS), and acute lymphoblastic leukemia (ALL). There are now more than 20 different chromosomal translocations in human leukemias involving the tel gene. One aspect of the tel gene fusions is that different regions of the tel gene are retained in the different leukemias. For example, in the TEL/MN1 fusion, the TEL DNA binding domain is fused to the MN1 protein. This is thought to result in deregulation of Ets target genes. In contrast, in the TEL/AML1 fusion, the N-terminal helix-loop-helix domain of TEL (but not the DNA binding domain), is fused to a nearly complete AML1 protein. The TEL/AML1 fusion has an extremely high occurrence frequency in ALL, the most common malignancy of childhood (8); thus, the TEL/AML1 fusion is the most common gene rearrangement in any pediatric malignancy. AML1 normally binds to the DNA sequence TGTGGT, which is found in the transcriptional regulatory region of the T-cell antigen receptor and in a variety of cytokines and their receptors. Thus, for the TEL/AML1 fusion protein, neoplastic transformation is thought to be due to deregulated transcriptional activity of AML1 target genes. It appears as if the fusion protein does not activate transcription, but rather inhibits basal promoter activity (9). Thus, it has been proposed that the TEL/AML1 fusion protein is a dominant interfering protein that can efficiently downregulate all AML-regulated target genes and that it is the loss of AML-activated gene expression that causes leukemia.

EWS/FLI-1 and EWS/ERG are fusion genes characteristically found in Ewing's sarcomas and primitive neuroectodermal tumors of childhood. EWS/FLI-1 and EWS/ERG are chimeric proteins consisting of the amino terminus of EWS, a putative RNA-binding protein, fused to the carboxyl terminal DNA binding domain of one of two different Ets family members, FLI-1 or ERG. The mechanism of transformation by fusion proteins that incorporate Ets DNA binding domains is not known, but likely includes inappropriate activation of genes containing Ets binding sites in their promoter regions. Evidence in support of this hypothesis includes the finding that all tumors containing an EWS/FLI-1 fusion have an intact FLI DNA binding domain and deletion of this Ets DNA binding domain abolishes the ability of EWS/FLI-1 to transform NIH 3T3 cells. There are many Ets family members in a cell. Although the EWS/FLI-1 rearrangement provides one additional factor that can bind to an Ets site, the other normal Ets family members are still present. Therefore, one might assume that the fusion protein will be in competition with other normal Ets proteins. It is unclear how simply increasing the amount of one particular DNA binding factor, in the continued presence of the multitude of other factors present in the cell that bind to the same sequence, can dramatically influence gene regulation. However, studies suggest that EWS/FLI-1 and FLI-1 may regulate distinct subsets of target genes although they have the same DNA binding domain (10). These experiments, as well as others that suggest that target gene specificity among the Ets family members (both normal and oncogenic varieties) does exist, are discussed below.

The Myc Family of Transcription Factors.
There are three very similar mammalian myc genes, c-myc (which can encode three separate proteins called Myc1, Myc2, and MycS), N-myc, and L-myc (11. All Myc proteins share a carboxyl terminal basic region helix-loop-helix-leucine zipper (bHLH-LZ) domain that mediates sequence-specific DNA binding and dimerization. Myc dimerizes with Max, another bHLH-LZ protein, and these heterodimers bind to the core DNA sequence CACGTG. Two predominant Max proteins exist that differ by a nine amino acid insertion amino-terminal to the basic domain. Other minor splice variants of Max have also been identified. Although Max protein levels are constitutive under all growth conditions, Myc protein levels are high in proliferating cells and low in quiescent or differentiated cells. The Myc proteins (except for MycS) have a transactivation domain at the amino terminus. Max does not contain a transactivation domain and thus Max/Max homodimers bind to, but do not activate transcription from, the same sites as Myc/Max heterodimers. Another protein, called B-Myc, is thought to be encoded by a duplicated second exon of the c-myc gene. This protein contains the transactivation domain but is missing the third exon of c-myc and thus does not have a nuclear localization signal or a DNA binding/heterodimerization motif (12).

The c-myc gene is probably the most extensively studied nuclear oncogene. The oncogenic potential of myc can be activated by chromosomal translocation, retroviral insertion, or gene amplification. For example, rearrangements of the c-myc gene, to a locus downstream of the immunoglobulin enhancer, is found in every case of Burkitt's lymphoma. One of the earliest discoveries of the role of c-Myc in carcinogenesis was the finding that avian leukosis virus (ALV) inserts near the c-myc gene in chickens and controls its aberrant expression, leading to the development of lymphomas. Amplified c-myc is found in carcinomas of the colon, lung, breast, and ovary, as well as other tumor types. The effects of over- and underexpression of c-Myc on cell growth have been examined in several experimental systems. Overexpressed c-Myc protein can immortalize primary cells and transform established cell lines. Increased expression of c-Myc can also induce apoptosis in cells growth-arrested by a variety of means and at various points in the cell cycle. Several studies have employed c-Myc knockout cell lines and c-Myc antisense RNA to examine the effects of inhibiting myc expression. In general, artificially raising the amount of c-Myc can transform a cell and lowering the amount of c-Myc slows cell growth and delays entrance into S phase. The N-myc gene is expressed in cells of primarily neuronal lineages during embryogenesis 13. N-Myc heterodimerizes with Max and binds to the same DNA sequence element as c-Myc/Max heterodimers. However, unlike c-myc which is expressed in many cell types and displays deregulation in a variety of tumors, the oncogenic activation of N-Myc occurs only in neuroblastomas and in other types of neuroectodermal tumors. In neuroblastoma, the most common alteration of N-Myc levels is mediated by gene amplification that can be correlated with a poor prognosis. In some cases, the N-myc gene has been found to be amplified more than 100-fold. Fewer studies have been performed on the L-myc gene. Amplification of L-myc has been seen in cases of small cell lung cancer. Due to the high degree of homology between c-myc and L-myc, it is generally assumed that overexpressed L-Myc will function in a manner similar to overexpressed c-Myc in a transformation assay; however, L-Myc has been shown to co-transform rat embryo cells with a lower efficiency than c-Myc (14).

Although it is clear that overexpression of Myc family members can lead to neoplastic transformation, the mechanism by which Myc mediates this response is controversial. One hypothesis put forth is that transformation by Myc is regulated by interaction between Myc and other cellular proteins. The basic-HLH region of Myc has been shown to interact with a variety of proteins (other than the heterodimeric partner, Max) such as BRCA1, YY1, AP2, NMI-1, and Miz1. Two other protein interaction domains that are found in the N-terminus of all Myc family members, called Myc Box 1 (MB1) and Myc Box 2 (MB2) have been implicated in Myc transforming activity. MB2 is a contact site for a cellular protein called TRRAP that participates positively in Myc-mediated neoplastic transformation (15). A cellular protein called BIN1 has recently been shown to bind to MB1 (16). This protein can inhibit transformation mediated by Myc in colony formation assays. It is likely that the oncogenic potential of Myc family members is related to their ability to regulate gene expression. The identification of target genes for the Myc family of transcriptional regulators has been the focus of much study for almost two decades. To date, two dozen or so genes have been suggested to be regulated by c-Myc (11). For most, it is unclear how increased or decreased levels of the putative target protein are critically important for Myc's oncogenic capacity. It has been proposed that a Myc target gene should a) have a binding site for Myc that is crucial for transcriptional regulation; b) show a pattern of growth regulation similar to that of the Myc protein (or opposite that of Myc for a Myc-repressed target gene); and c) encode a protein that, when deregulated, could contribute to neoplasia. Some genes that have been suggested to be activated by Myc include cad (required for pyrimidine biosynthesis), cdc25a (a cell cycle phosphatase), odc (required for polyamine biosynthesis), elF-4E (a translation factor), and ISGF3 (a transcription factor). Evidence also exists to suggest that Myc represses target genes that negatively influence cell proliferation; such genes may include C/EBP (a transcription factor), and gadd153/CHOP (a transcription factor). The molecular mechanisms by which c-Myc mediates transcriptional repression are not well defined; however, it is believed that c-Myc interacts with proteins other than Max to repress transcription and that DNA binding sites distinct from Myc/Max binding sites may be required. It was recently demonstrated that expression of many of these putative c-Myc target genes remains unchanged in c-Myc null cells, with the exception of cad and gadd45 (17). Therefore, it is likely that many true transcriptional targets of c-Myc have yet to be identified.

The relative importance of Myc as a transcriptional regulator in mediating neoplastic transformation is highly debated. Much of this controversy comes from the difficulty in clearly defining a true Myc target gene. Since other proteins that can bind to the same DNA element (which is called an E box) as the Myc/Max heterodimers also exist in cells, it has been difficult to classify cellular promoters as being activated specifically by Myc. A group of proteins that includes Mad, Mxi1, Mad3, and Mad4 can also dimerize with Max and bind to E boxes. Mad/Max heterodimers are repressors of transcription due to interactions with the mSin3 family of transcriptional repressors. It is postulated that Mad/Max complexes function in differentiation pathways by inhibiting transcription from Myc-activated promoters. Yet another partner of Max called Mnt (or Rox) has recently been cloned. Mnt/Max also functions as a transcriptional repressor in a complex with mSin3 proteins, and it has been postulated that Mnt/Max complexes may be involved in repression of transcription in quiescent cells. Other proteins such as USF1 and USF2, as well as TFE3 and TFEB, can bind to and transactivate via an E box. Overexpression assays often show that a putative Myc target is also regulated by USF and Mad family members. In addition to being in competition with other E box-binding proteins, Myc can also bind to another sequence element called an initiator. This element is located at the transcription start site of some cellular genes and is thought to be a site at which both positive and negative regulatory proteins can interact with the basal transcriptional machinery. Myc has been shown to negatively regulate some promoters that have particular initiator elements (18) and to interact directly with another protein called TFII-I that also binds to initiator elements (19). It is proposed that Myc heterodimerizes with TFII-I instead of Max to mediate transcriptional repression. Several experiments have highlighted the significance of transcriptional repression by Myc. For example, Myc proteins that have lost the transcriptional repression function due to mutation of a region of the N-terminus also lose the ability to transform cells (18). Although this suggests that the repression function of Myc is critical for transformation, others have shown that these same Myc constructs have also lost transcriptional activation at certain promoters (20). Recent studies of MycS, a naturally occurring N-terminal-truncation of c-Myc which can only repress transcription, may provide some insight into this problem (21). MycS can mimic Myc in certain assays, such as stimulation of proliferation and allowing for anchorage-independent growth. However, unlike full length Myc, the N-terminally truncated MycS protein cannot transform primary cells in cooperation with Ras. These results suggest that the repression function of Myc may be important in imparting some, but not all, of the biological phenotypes seen in tumors having high levels of Myc. It is likely that critical Myc target genes will include both those that are activated by and those that are repressed by Myc family members.

As with the Ets family, it is unclear why deregulation of Myc family members is commonly associated with human cancer when in the same cells there exists highly abundant factors such as USF that can bind to and activate transcription from the same DNA elements. Studies of mice that have been engineered to lack different USF factors show that whereas USF1 and USF2 may encode redundant functions, mice lacking both USF1 and USF2 are embryonic lethal. Thus, Myc family members cannot substitute for USF family members. Similarly, mice lacking either N-Myc or c-Myc die in utero (mice lacking L-Myc are normal), suggesting that N-Myc and c-Myc may have nonredundant transcriptional activities and that the abundant and ubiquitous USF proteins cannot substitute for either c-Myc or N-Myc (22-27). Another unsolved question is why deregulation of different Myc family members is associated with different types of tumors. c-Myc, N-Myc, and L-Myc can all cooperate with a mutated Ras to transform cells neoplastically in culture and transgenic mice overexpressing c-Myc, N-Myc, or L-Myc via an immunoglobulin enhancer develop maligancies (28-30). However, N-Myc is associated only with neuroblastomas and L-Myc with small cell lung cancers, whereas c-Myc is associated with a variety of human tumors. Does N-Myc activate a distinct set of target genes whose deregulation is only critical in certain cells or is N-Myc overexpression detrimental to all cells in the body except for neuronal cells? It is likely that the answers to all of these questions will require an understanding of target gene specificity of these DNA binding transcription factors. A review of our general understanding of Myc family member target gene specificity is provided below in Section III.

The E2F Family of Transcription Factors.
The E2F family is a collection of transcription factors that function as heterodimers that bind to and regulate transcription from a consensus sequence (TTTSSCGC, where S is C or G) known as an E2F site (31-33). To date six different mammalian E2Fs (E2F1, E2F2, E2F3, E2F4, E2F5, and E2F6) have been cloned; each of them can heterodimerize with either DP1 or DP2. The E2F components contain a central DNA binding and dimerization domain and (except for E2F6) a C terminal transactivation domain. A subset of the E2Fs (E2F1, E2F2, and E2F3) also contains a conserved N-terminal protein interaction domain. One protein that can bind to the N terminal domain is cyclin A. Studies have shown that cyclin A-dependent kinase activity, but not cyclin B-, E-, or D-dependent kinase activity, can negatively regulate the in vitro DNA binding activity of heterodimers containing E2F1, E2F2, or E2F3. The DP proteins contribute to the DNA binding activity of the heterodimer, but do not contain a transactivation domain. Nested within the transactivation domain of the E2F proteins is a protein interaction domain that mediates contact with the Retinoblastoma (Rb) family of proteins. E2F1, E2F2, and E2F3 all bind preferentially to Rb; E2F4 and E2F5 bind preferentially to the Rb-related proteins p107 and p130. E2F1, E2F2, and E2F3 have nuclear localization signals and are predominantly nuclear at all times. However, E2F4 and E2F5 lack nuclear localization signals; they are brought into the nucleus via interaction with either DP2 or a splice variant of DP1, both of which have nuclear localization signals. E2F4 and E2F5 can also be brought into the nucleus by interaction with p107 or p130.

Rb, p107, and p130 can all repress transcription of a promoter that contains an E2F site. However, it is clear that the binding of the Rb family of proteins to E2F family members does not inhibit their DNA binding activity. Therefore, the Rb family of proteins must influence E2F transcriptional activity. The E2F transactivation domain has been shown to interact with several components of the basal RNA polymerase II transcriptional machinery, such as TBP, TFIIH, and CBP. Since E2F cannot bind to both Rb and the other transcription factors at the same time, it is believed that Rb may repress E2F-mediated transcriptional activation by interfering with functional transcription complex formation. Because Gal4/Rb fusion proteins can block transcription when artificially brought to the DNA of a variety of promoters (34), Rb must also have a functional repression domain that is distinct from the E2F binding domain. Insight into the mechanism by which Rb can repress transcription comes from the recent findings that Rb, p107, and p130 can all associate with a histone deacetylase (35). It is postulated that the recruitment of the histone deacetylase to E2F-regulated promoters represses transcription through changes in chromatin structure. Rb (or p107 or p130) must be released from the E2F/promoter complex for the gene to be transcribed. The disruption of the E2F/Rb protein complex is thought to be primarily due to the action of cell cycle–regulated kinases. Each of the Rb family members contains multiple sites for phosphorylation by these kinases. The current model is that cyclin D-dependent kinases initiate the phosphorylation events and that cyclin E- and cyclin A-dependent kinases complete the hyperphosphorylation. Hyperphosphorylated Rb (as well as hyperphosphorylated p107 and p130) cannot bind to E2F proteins. Thus, the increased cyclin-dependent kinase activity that results as cells progress through G₁ into S phase of the cell cycle results in release of Rb from E2F and activation of E2F target genes. One of these E2F target genes is E2F1 that leads to a positive autoregulation loop. The action of the cyclin-dependent kinases is kept under control by cdk inhibitors of the p16 and p21 families. In turn, p21 is controlled by the activity of the p53 tumor suppressor protein. Deregulation of the different components of this pathway (e.g., by decreasing levels of Rb, p53, or the cdk inhibitors or by increasing the levels of the cyclins) can result in upregulation of E2F activity.

Unlike members of the Ets and Myc families of transcription factors, E2F family members have not been found to be mutated in human cancers. However, the factors that control the activity of E2F proteins are mutated in a large number of different types of human tumors. Many studies have shown that Rb and p53 are lost in a variety of human tumors and that cyclin D1 and cyclin E are upregulated in human tumors. The loss of Rb and the increased cyclin-dependent kinase activity (with the resultant increased phosphorylation of Rb) both result in the conversion of the E2F/Rb transcriptional repressor complex to the E2F transcriptional activator. The frequent deregulation of the Rb/cyclin signal transduction pathway in human cancers suggests that E2F activity should be increased in tumor cells. Several recent studies have indicated increased E2F activity in tumors having mutated Rb protein (5, 36). These studies, taken in combination with the fact that known E2F target genes encode proteins whose activities are required for DNA synthesis (e.g., dihydrofolate reductase, DNA polymerase-, and thymidine kinase) or cell cycle progression (e.g., cyclin E and B-myb), suggest that overexpression of E2F proteins should provide an environment conducive to DNA replication and neoplastic transformation. Accordingly, several studies have demonstrated the ability of E2F family members to transform cells grown in culture (37). Also, recent studies have shown that E2F1 can cooperate with Ras to induce tumors in keratinocytes in transgenic mice (38, 39).

Several strains of mice have been created that are nullizygous for a particular E2F family member. In general, these mice are viable for at least several weeks after birth (40-42) suggesting that E2F family members can at least partially substitute for each other in the regulation of many E2F target genes. However, E2F1 null mice develop tumors in a limited number of organs after about a year, and E2F5 null mice develop hydrocephalus caused by excessive secretion of cerebrospinal fluid, suggesting that these E2Fs are not completely redundant with the other E2F family members. Additional evidence in support of the hypothesis that different E2F complexes regulate different target genes will be described below. The apparent contradiction between the overexpression studies of E2F1 (which suggests it is an oncogene) and the nullizygous phenotype (which suggests that E2F1 may be a tumor suppressor) might be explained by the complex regulation of E2F activity. The increased amounts of E2F1 upon overexpression in tissue culture and in transgenic mice will change the ratio of E2F1/Rb complexes versus free E2F in the cell. If enough E2F1 is provided, it can sequester the Rb protein, then E2F target promoters will be occupied by free E2F and derepression can occur. The loss of E2F1 can also result in removal of Rb from E2F1-specific promoters, again leading to derepression of certain E2F target genes. It is also possible that it is the loss of activation of a subset of E2F target genes in the E2F1 null mouse, rather than a loss of repression, that leads to tumor formation in certain tissues. Similar to the situation with the Myc oncogene, it is likely that the set of important E2F target genes will include genes that are activated by E2F family members as well as genes that are repressed by E2F family members.

	Oncogenic Transcription Factors Show Target Gene Specificity

Top Abstract Introduction Oncogenic Transcription Factors... Oncogenic Transcription Factors... Methods That Can Aid... Conclusions References

 
As described above, families of oncogenic transcription factors are grouped according to their ability to bind to the same DNA consensus element; accordingly, the DNA binding domains in the various members of a family of transcription factors are highly conserved. In vitro analyses have shown that, in general, similar DNA binding domains will recognize similar DNA sequences. However, there is biological evidence suggesting that different Ets, Myc, and E2F family members have different functions. This indicates that there must be some target gene specificity among different, yet highly similar, family members. This apparent contradiction between initial in vitro DNA binding studies and biological phenotypes can be reconciled by a variety of mechanisms. For example, it is possible that some DNA binding specificity can be determined by subtle differences in the structure of two highly related factors. This may allow two factors to bind to certain common elements but also allow a subset of sites to be bound only by one of the factors. In this case, the specificity is likely to be retained in assays that use isolated DNA binding sites. Another mechanism to engender target gene specificity would be to allow for the context of the binding site within a promoter region to influence the affinity or activity of some, but not all, members of a family of transcription factors at a particular promoter. In this case, specificity may only be observed when longer DNA fragments are used and when additional proteins are included in the assay. As described below, both of these modes of specificity are used to provide target gene specificity for members of the Ets, Myc, and E2F families of oncogenic transcription factors.

Differences Can Be Determined by the Specific Target Site.
As noted above, levels of c-Myc are quite low in a normal cell but are increased in human tumors. This has led to the hypothesis that the increased levels of c-Myc lead to neoplastic transformation of cells by deregulating transcription of Myc target genes. However, this hypothesis does not take into account the fact that other proteins that bind to the same sequence, in particular USF1 and USF2, are ubiquitous and highly abundant in both normal and tumor cells. One possible hypothesis to explain why Myc, but not USF, can transform cells is that USF might bind to only a subset of the E boxes that are bound by Myc. Therefore, several studies have focused on determining if Myc and USF binding sites are identical or if the related proteins bind overlapping subsets of sites. In general, two different types of experiments have been performed that are directed toward a more detailed understanding of Myc versus USF binding sites. In the first type of experiment, binding sites are selected from a pool of random sequences using the DNA binding properties of purified protein. The development of a consensus sequence allows for identification of a high affinity site, whereas the obvious exclusion of a particular nucleotide at a particular position from all captured sequences suggests the identification of a disfavored nucleotide. Another type of experiment involves direct mutational analysis of the sequences flanking the consensus hexamer CACGTG of a particular promoter. Using a selection procedure, Solomon et al. (43) found that a c-Myc/Max heterodimer fails to bind to a CACGTG hexamer when the core is flanked by a 5' T or a 3' A. In contrast, Bendall and Molloy (44) selected for USF binding sites and found little preference for particular sequences flanking the core binding site. A comparison of the ability of USF and c-Myc to bind to the selected USF binding sites showed that a T at the -4 position was inhibitory to Myc binding. Similarly, direct mutation of the sequences flanking the CACGTG hexamer in the hamster cad promoter indicated that both USF and Myc could bind to the wildtype E box in the promoter but that mutation of the flanking sequences could abolish Myc, but not USF, binding (45). Thus, several studies have shown that although USF is relatively insensitive to the specific sequences flanking the CACGTG of the E box, alterations of one or two base pairs adjacent to the core hexamer can greatly influence the affinity of Myc for the binding site. Although most of the analysis of the different E boxes has been performed using in vitro DNA binding assays, several studies have assayed for E box function within the cell. For example, when the sequences flanking the E box in the hamster cad promoter were altered to contain a 5' T and a 3' A, transcriptional regulation of the promoter was abolished in mouse cells (45). Similarly, E boxes containing a 5' T and a 3' A were also unable to support activation by c-Myc in yeast cell assays (46).

In summary, USF appears to bind to a wider variety of E boxes than does c-Myc. However, the collection of sites to which a factor binds may be more complex than initial studies indicate. For example, although in vitro selection of Myc binding sites identified the CACGTG consensus as a high affinity site, a series of non-canonical binding sites was also obtained (47). Furthermore, Hann et al. (48) have shown that Myc1 (a slightly longer form of c-Myc that is abundant in certain growth-inhibited cells) can bind to and transactivate promoters that contain a C/EBP consensus element (TTATGCAAT). Such studies, which indicate that a factor can bind to a variety of different sites, taken in combination with the multitude of experiments indicating that promoter context can influence binding (see below), suggest that a better approach for identification of target genes would be to identify sites to which a transcription factor is bound in genomic DNA. These methods are discussed below.

There is also much evidence that different Ets family members have different biological properties. Similar to the studies of Myc and USF binding, a comparison of DNA binding properties of different Ets proteins has revealed that not all Ets proteins bind to exactly the same sequences. For example, binding site selection using the Ets1 DNA binding domain identified a consensus sequence that differs from the binding site consensus for the Ets domain protein E74A (49). Similarly, a comparison of protein binding to two Ets binding sites in the interleukin-2 promoter indicated that Elf-1, but not Ets1 or Ets2, could bind to these particular Ets sites (50). Some of the documented differences in target gene specificity between the various Ets family members are due to the slight differences in the two DNA binding domains. For example, DNA binding site selection indicates that the Ets domain in Elk-1 exhibits a more stringent DNA-binding site specificity than does the corresponding domain of Sap-1 (51). Elk-1 selects sites conforming to the consensus sequence ACCGGAAGTR, whereas Sap-1 selects the more degenerate sequence A(C/t)CGGA(A/t)(G/a)(T/c)N. Thus, the Ets domain of Sap-1 can bind to a series of sites to which the Ets domain of Elk-1 cannot bind. Further studies have identified the regions of the Elk-1 and Sap-1 DNA binding domains that mediate the differences in target site selection. Mutational analyses have shown that two amino acids found at particular positions in Sap-1 allow a greater degree of flexibility in binding to DNA than do the two amino acids found in similar positions in Elk-1 (51).

In addition to amino acid changes in DNA binding domains conferring target gene specificity, DNA binding specificity at isolated Ets consensus sites can also be modulated by other regions of the Ets protein. One such example comes from a comparison of FLI-1 and EWS-FLI-1 (52). Both of these proteins contain the exact same Ets DNA binding domain, however, they demonstrate different DNA binding specificity. FLI-1 can bind to the Ets site adjacent to the serum response element of the fos promoter, but only in the presence of another protein called serum response factor (SRF). However, the EWS-FLI-1 fusion protein binds to the same site in the absence of other protein-protein interactions. Deletional analysis of FLI-1 revealed the presence of an inhibitory domain in the N terminus of FLI-1 (which is missing in the fusion protein) that prevents autonomous binding of FLI-1 to the Ets site in the SRE. Thus, differences in protein structure outside of the highly conserved DNA binding domains of family members can contribute to target gene specificity. In summary, at least some of the target gene specificity observed with the Ets family of transcription factors can be explained by specific differences in various domains of particular Ets proteins. However, as will be described below, target gene specificity in the Ets family can also be conferred by methods other than selective DNA binding.

The DNA binding domains of all the E2Fs are very similar, and in vitro DNA binding assays suggest that most, if not all, of the E2F/DP heterodimers bind with similar affinity to the same collection of target sites. These in vitro studies are complicated by the fact that E2F4 is very abundant in comparison to the other E2Fs and comprises the majority of the E2F activity in nuclear extracts of most types of cells. Therefore, most in vitro DNA binding assays are mainly a monitor of the ability of E2F4 to bind to an E2F site. However, there is one report that demonstrates differential binding of different E2Fs to a particular site. Liu et al. (53) have reported that a site in the cyclin A promoter is weakly recognized by E2F4, but clear binding is seen by E2F1 and E2F3. In contrast to most of the in vitro assays that show little specificity of different E2Fs, a comparison of the ability of the different E2Fs to activate a panel of genes upon overexpression in quiescent cells using an adenoviral construct suggests that target gene specificity does exist in vivo (54). Further analysis suggests that some of these differences may be due to the fact that the E2F/DP heterodimers also interact with members of the Rb family. In vitro casting experiments suggest that the optimal consensus site for the trimolecular complex E2F1/DP1/Rb is different from the optimal consensus for E2F1/DP1 (55). For example, the trimolecular complex containing Rb preferentially bound to two inverted, overlapping E2F sites (TTTc/gGCGCg/cAAA) whereas the E2F1/DP1 dimer preferred a single site (TTTCCCGC). A requirement for binding to an inverted, overlapping site (as opposed to a single E2F site) would greatly restrict the number of target genes that could be regulated by E2F1 when bound to Rb. Therefore, the Rb/E2F1/DP1 complex may regulate only a subset of genes that can be bound by E2F1/DP1. All combinations of Rb family members with different E2F/DP heterodimers have not yet been tested for DNA site specificity. However, it is possible that the addition of Rb family members to the E2F/DP heterodimers may cause protein conformational changes that allow a divergence of binding specificities of the different E2Fs.

In summary, members of a family of transcription factors recognize a very similar set of DNA binding sites. However, subtle differences inherent in the structure of the different family members and/or conformational changes caused by interaction with another protein can result in different family members binding to overlapping, but not identical, sets of DNA sequences. As shown in Figure 2, USF is allowed a greater flexibility in the sequences that flank an E box than is Myc. Similarly, Sap-1 binds to a less stringent consensus sequence than does Elk-1. Finally, the interaction of E2F1/DP1 heterodimers with Rb can restrict the subset of possible E2F sites available to E2F1.

View larger version (21K): [in this window] [in a new window]   Figure 2.    Sequences flanking core consensus elements can determine family member specificity. Shown are examples of how alterations in the nucleotides flanking Myc, Ets, and E2F binding sites can affect the binding affinity of specific members of a family of transcription factors (see text for details). For the top and middle panels, the lower oligonucleotide represents a less stringent consensus that can be bound by some, but not all, of the DNA binding protein complexes that represent a particular family of transcription factors. For the bottom panel, the inclusion of a third protein in the E2F1/DP1 complex alters the DNA binding preference of the E2F family member. In each case, the sequence identical in the pairs of binding sites is underlined.

Although studies of the effects of Myc and USF on the transcriptional activity of putative target genes have been difficult due to the modest in vivo transcriptional activity of these proteins, the amino terminal transactivation domains of USF and Myc have been shown to display some target gene specificity. Evidence that the differences observed in transcriptional assays can be mediated via mechanisms other than differential binding affinity for specific E boxes comes from several different experiments using Gal4 fusion proteins. Boyd et al. (58) have shown that USF fused to the Gal4 DNA binding domain cannot activate the cad promoter containing a Gal4 site. In contrast, the Myc transactivation domain can increase cad promoter activity when fused to Gal4. Since the Gal4 DNA binding domain is mediating the protein/DNA interactions in both cases, the transactivation domain of Myc may make a critical protein-protein contact with some component of the transcriptional machinery with which USF cannot interact. Other evidence suggests that the USF activation domains are very sensitive to core promoter structure. Luo and Sawadogo (59) have shown that a domain called USR that is well conserved in USF1 and USF2 can only activate transcription in the presence of both a TATA box and an initiator element. Although the cad promoter lacks a TATA box, it does contain a consensus initiator. Therefore, one might predict that the addition of a consensus TATA box would alter the cad promoter such that it could now be activated by USF. However, the addition of a TATA box does not convert the cad promoter to a promoter that can be activated by USF (Boyd, unpublished results). Evidence supports the hypothesis that not all initiator elements are bound by the same proteins. Replacement of the cad initiator (which has 93% homology to the consensus initiator element) with two different sequences, each having about a 90% homology to the consensus, indicated that one, but not the other, replacement initiator could direct transcription from the cad promoter in vivo (60). Therefore, it remains possible that Myc and USF are active only in the context of specific initiator elements. If so, then interaction between Myc or USF with distinct initiator binding proteins may provide some aspects of target gene specificity. Other experiments also suggest that Myc may make specific interactions with some component of the basal transcription complex that cannot be reproduced by USF. Studies of Desbarats et al. (20) suggest that Myc can activate gene expression over a longer distance than can USF. It is possible that strong protein-protein interactions between Myc and a component of the basal machinery allow this long-distance activation via a DNA-looping mechanism. It has been proposed that the position-independent activation properties of Myc can allow it to activate a larger set of target genes than can USF. Other studies have shown that different cellular promoters will display different extents of position independence in their activation by Myc. For example, the dhfr promoter is very sensitive to the position of the bound Myc but the cad promoter is less sensitive (58). Therefore, it is likely that a combination of DNA binding specificity and protein-protein interactions combine to create different subsets of Myc and USF target genes.

Most E2F transactivation studies have indicated that overexpression of any of the E2Fs can lead to activation of a particular target gene. Some of this lack of target gene specificity may be due to the fact that several of the E2F genes are themselves regulated by E2F sites. Therefore, introduction of a single E2F into a cell may lead to increased levels of several different E2Fs. It is also likely that overexpression of a single E2F can obscure promoter specificities that do occur under normal physiological conditions. One of the few cases in which target gene specificity has been observed when different E2Fs are used in transactivation assays is a study involving the cyclin D1 promoter. Watanabe et al. (61) found that E2F1 negatively regulates the cyclin D1 promoter but that E2F4 is an activator of the same promoter construct. Interestingly, they also showed that a nearby Sp1 site contributed to the repression mediated by E2F1. Others have shown a direct interaction between Sp1 and E2F1, E2F2, or E2F3, but not E2F4 or E2F5 (62). This raises the possibility that the observed target gene specificity between E2F1 (which represses the cyclin D1 promoter) and E2F4 (which activates the cyclin D1 promoter) may be due to E2F1 recruitment of Sp1. Although specific E2Fs were not tested, two others experiments have found functional relationships between Sp1 and E2F sites. For example, the murine thymidine kinase promoter contains an Sp1 binding site spaced seven base pairs upstream from an E2F binding site. It was shown that mutation of either of the two sites in the context of a stably integrated promoter construct abolished the in vivo footprint at both sites. Functional analyses indicated that increasing the separation of the Sp1 and E2F sites by an additional 20 base pairs also abolished the cell cycle stage–specific promoter activity, suggesting that an interaction between the proteins binding to these two sites was critical (62). Thus, a prediction of this work is that the murine thymidine kinase promoter would be a target gene of only a subset of E2Fs that bind Sp1 (i.e., E2F1, E2F2, and E2F3). The mouse dhfr promoter is also extremely sensitive to the position of the E2F site. The mouse dhfr promoter contains four binding sites for the transcription factor Sp1 located from -50 to -210 and an E2F site located at the transcription start site. Mutational analysis indicates that the E2F site is the critical determinant in specifying cell cycle stage–specific transcriptional regulation of this promoter (63). The positional requirements for the E2F site were investigated by inserting DNA fragments just upstream of the E2F site. Shifting the E2F site downstream of the start site by about 66 base pairs abolished the increase in transcriptional activity that normally occurs from this promoter as cells enter the S phase (64). These studies did not identify the reason why movement of the E2F site was detrimental to transcriptional regulation of the mouse dhfr promoter. Other studies have shown that E2F1 can cooperate with Sp1 to activate the hamster dhfr promoter in cotransfection assays performed using Drosophila cells. This cooperation required the region of the N terminus of the E2F1 protein that is conserved in E2F2 and E2F3 (65), again supporting the hypothesis that promoters containing Sp1 binding sites may be regulated by a subset of E2Fs. However, the Sp1 sites in the mouse dhfr promoter are fairly far apart (50 base pairs), leaving open the possibility that it is the proximity to another promoter element that is critical for regulation of the mouse dhfr gene.

In summary, promoter context can be a critical determinant of the degree to which a particular transcription factor can regulate a given promoter. As shown in Figure 3, members of the Ets, Myc, and E2F families can be influenced both by the presence of other site-specific DNA binding proteins and by core promoter elements. For example, certain Ets factors require cooperation with proteins bound to a nearby AP1 site to achieve transcriptional activation. Likewise, in certain promoters the E2F site must be located within a very close proximity to an Sp1 site in order for E2F proteins to regulate transcription. In contrast, specific activation by Myc versus USF appears to be determined by the precise arrangement of core promoter elements and the bound transcription factors.

View larger version (23K): [in this window] [in a new window]   Figure 3.    Promoter context can influence the activity of oncogenic transcription factors. (A) Shown are schematics that represent a promoter (e.g., collagenase) that contains both an Ets and an AP1 binding site versus a promoter (e.g. stromelysin) that does not contain an AP1 site adjacent to the Ets site. Although both of the Ets family member proteins Erg and Ets2 can activate the collagenase promoter, only Ets2 can activate the stromelysin promoter. This is due to the requirement for Fos and Jun binding to the AP1 site to recruit Erg to the Ets site (see text for details). (B) Shown are schematics of a minimal promoter containing a Gal4 binding site that has been placed just upstream of the core promoter and a minimal promoter containing a Gal4 site that has been placed several kilobases downstream of the core promoter. Although both Gal4USF and Gal4Myc fusion proteins can activate the top construct, only Gal4Myc can activate the bottom construct from the downstream Gal4 site (see text for details). (C) Shown is an example in which movement of an E2F site relative to other promoter elements can alter transcriptional regulation. Insertion of 20 bp between the Sp1 site and the E2F site in the tk promoter can abolish the increase in transcriptional activity that is normally observed in the S phase (see text for details). One hypothesis that may account for these results is the requirement for a physical interaction between Sp1 and an E2F family member. Since only a subset of E2F family members interacts with Sp1, promoters containing adjacent E2F and Sp1 sites may be regulated by only a subset of E2F family members.

	Methods That Can Aid in the Identification of Target Genes

Top Abstract Introduction Oncogenic Transcription Factors... Oncogenic Transcription Factors... Methods That Can Aid... Conclusions References

View larger version (27K): [in this window] [in a new window]   Figure 4.    Identification of target genes with Expression Profiling. Shown is a schematic of the DNA Array and Filter Hybridization approaches that can be used to identify previously cloned genes as direct and indirect targets of oncogenic transcription factors. For both techniques, Poly(A)+ messenger RNA is isolated from control and experimental cells and converted into cDNA. Probes are then generated by labeling the cDNA populations with fluorescent dye or 32P. (Left) DNA arrays that contain thousands of cDNA sequences are hybridized with both probes (labeled with two different dyes) simultaneously. Hybridized signals, detected by scanning microscopy, fluoresce in a range between the two dyes indicating how much of either probe is hybridized (e.g., white boxes indicate cDNAs that are represented in the control sample, black boxes indicate cDNAs that are represented in the experimental sample, and gray are common to both). (Right) Hybridization to the cDNA filters, containing hundreds of cDNAs, can be carried out in parallel using the control and experimental probes on two identical filters. Relative signal intensities can be compared between control and experimental filters to define cDNAs that increase or decrease in response to treatment.

View larger version (46K): [in this window] [in a new window]   Figure 5.    cDNA-based methods used to identify target genes. Shown is a schematic of the Serial Analysis of Gene Expression (SAGE) and Representational Difference Analysis (RDA) approaches that can be used to clone both direct and indirect target genes of oncogenic transcription factors. For both techniques Poly(A)+ messenger RNA is isolated from both control and experimental cells. (Left) In SAGE, mRNAs (only shown for experimental sample) are converted into double-stranded cDNAs using biotin-labeled (small black circles) oligo(dT) primers. Labeled cDNAs are digested with a restriction endonuclease and then bound to strepavidin beads (large gray ovals). The sample is divided in half and ligated to two different linkers (gray and striped lines) containing a recognition site for a type IIS restriction endonuclease (which cut at a defined number of bases downstream of the recognition site); this serves to tag a short fragment of each cDNA present in the population. Following digestion with the type IIS enzyme, the cDNA fragments are blunt-ended, ligated together, and then amplified by PCR. The PCR products are digested with the original restriction enzyme to remove the linkers. Digested products are pooled and ligated into a sequencing vector. Clones (usually containing multiple tagged inserts) are sequenced to identify each tag. Computer analysis is performed to match sequence tags to known cDNAs or ESTs and to catalogue the expression frequency of each sequence tag within a population. (Right) In the RDA approach, control and experimental mRNA populations are converted into cDNAs and amplified by PCR. To find those mRNA species that are increased in the experimental sample (shown in the schematic), the experimental cDNAs are ligated to PCR linkers, denatured, and then hybridized in the presence of excess control cDNA. Cross-hybrids between control and experimental cDNAs (i.e., mRNA common to both populations) will not become amplified. Resulting hybrids that contain linkers on both strands (i.e., mRNAs that are more abundant in the experimental sample) will become amplified by PCR. Amplified cDNAs, called the first round difference product, are then used in two to three more rounds of hybridization and amplification. To find mRNAs that are decreased in the experimental sample, control cDNAs are ligated to PCR linkers and then hybridized in the presence of excess experimental cDNA (not shown). The final difference products are visualized on an agarose gel and subcloned. Not shown in this figure is the differential display technique that is clearly outlined in a previous report (73).

View larger version (31K): [in this window] [in a new window]   Figure 6.    Cloning genomic fragments bound by transcription factors through ChIP and DIP techniques. Shown in this figure is a schematic of the chromatin immunoprecipitation (ChIP) and DNA immunoprecipitation (DIP) approaches that can be used to identify and clone direct binding targets of oncogenic transcription factors. (Left) For the ChIP procedure, intact cells are treated with formaldehyde or UV light to covalently cross-link DNA-bound factors to their cognate binding site. The cells are then lysed, and the nuclei are sonicated to fragment the cross-linked chromatin. These DNA fragments contain nucleosomes (gray ovals), other transcription factors (squares and triangles), and the transcription factor of interest (black circles). (Right) For the DIP procedure, genomic DNA is purified from cells and digested with a restriction enzyme. Prepared DNA is then incubated with a preparation of the purified protein of interest (black circles). For both ChIP and DIP, protein-DNA complexes are collected by immunoprecipitation with an antibody specific for the protein of interest. Immunoselected chromatin or DNA fragments are then purified and cloned into a vector (to identify novel binding targets) or analyzed by PCR using gene-specific primers (to verify binding to a putative target gene).

Arrayed filters, containing DNAs complementary to hundreds of mRNAs, have been applied to the search for both c-Myc and N-Myc target genes. Although investigators are limited to examining the expression levels of those mRNAs represented on the commercially available filters, the variety of such filters is increasing. For example, it is now possible to purchase filters containing cDNAs from human, mouse, and rat cells. Specialty filters and custom filters are also becoming available; there are now filters containing genes relevant to cancer and to apoptosis. It is likely that the utility of this method will increase in the next several years when investigators have a wider variety of different filters from which to choose. Using currently available filters, several groups have investigated the effects of Myc family members on gene expression. For example, liver-specific overexpression of c-Myc in transgenic mice results in the rapid development of hepatocellular adenomas. Hybridization of radiolabeled probes from wild type and c-Myc transgenic liver mRNA to a pair of Clontech Atlas Mouse cDNA filters identified nine genes that are differentially expressed in the liver of Myc transgenic animals (S. Kim, personal communication). Northern analysis revealed that these nine genes were expressed at moderate to high levels in the cell, suggesting that the original profiles could not detect low abundance mRNAs. This group hypothesized that a probe obtained after a single round of representational difference analysis (see below and Figure 5 for details of this technique) would enhance detection of low-level messages. Accordingly, an additional nine differentially regulated genes were identified using the new probes. Clontech's Atlas Human Array and Cancer Array filters have been used by another group to screen for N-Myc targets using a neuroblastoma cell line having a tetracycline-regulated N-Myc construct (S. Mac, personal communication). Preliminary results from this study identified 10 mRNAs that were increased in the presence of high levels of N-Myc and five that were decreased (e.g., -prothymosin was increased up to 6-fold in cells that overexpress N-Myc).

Microarray technology has been developed by a number of laboratories for a more comprehensive analysis of the expression of known genes. These arrays typically consist of high-density grids of unique cDNA clones (or ESTs) or oligonucleotides that correspond to thousands of different transcripts (68, 69). Similar to the cDNA filter arrays, the relative expression levels of a large number of mRNA species can then be assessed in parallel by generating cDNA probes from a population of mRNA obtained from control and test samples, followed by hybridization of these probes against duplicate arrays. From the hybridization signals, the relative expression levels of all genes represented on the array can be ascertained. Although this technique has several advantages over the filter arrays (e.g., the number of genes examined can be in the thousands), problems remain. For example, each of the two different microchip methods requires expensive machinery, software, and reagents. This limits the access of these techniques to only a handful of investigators. Second, each of the two microarray techniques requires that the sequence of a gene be in the database, as either a cDNA or an expressed sequence tag, before it can be monitored in this assay. Thus, if the key target genes of a particular transcription factor have not yet been cloned, this type of analysis may be both expensive and nonproductive.

Although very few experiments using microarrays have been reported, one experiment has provided data concerning the c-myc oncogene. Primary human fibroblasts containing inducible Myc protein were used to examine gene expression in quiescent cells and in cells progressing toward S-phase after c-Myc activation. Of approximately 6,000 genes present on the chips, 27 showed consistent upregulation and 9 downregulation in three separate experiments (in addition, a number of genes showed induction in two out of three experiments). Among the known Myc target genes, odc showed the most robust and reproducible changes. The others all represent new potential Myc target genes, and several were confirmed by Northern blot analysis (H. Coller, C. Grandori, R. Eisenman, and T. Golub, personal communication). The relatively small number of affected genes observed in these experiments suggests that many c-Myc target genes are not yet represented on the chips or that many Myc targets are low abundance messages that undergo only slight changes in response to c-Myc and are difficult to detected in this assay.

In summary, several investigators are attempting to identify Myc target genes using DNA arrays. There are currently ongoing attempts to use both filters and microchips to identify N-Myc (B. Carroll, personal communication), Ets (C. Denny, personal communication) and E2F (J. Nevins, personal communication) target genes. Many of the preliminary results obtained to date have not been extremely encouraging. First, there is a high rate of false positives using these methods. It is extremely important to use a different method (e.g., Northern analysis, RNAse protection, or RT-PCR) to confirm the differential expression seen using the arrays. Second, in the c-Myc and N-Myc studies, several of the genes identified using the filters were already previously known to be regulated by Myc family members. For example, one of the mRNAs that was dramatically unregulated by c-Myc using DNA microarrays was odc, and the gene that was the most dramatically regulated by N-Myc using filter arrays was found to be -prothymosin. Both of these genes had previously been reported to be regulated by both c-Myc (70, 71) and N-Myc (72). However, despite the many difficulties of these new approaches, a handful of novel c-Myc target genes have been identified in a fairly rapid fashion using DNA microarrays.

cDNA Cloning Methods.
As noted above, the methods using DNA filter arrays and DNA microchips have been slow in providing new insight into the identification of target genes of Myc, E2F, or Ets family members. One different approach that researchers are taking to identify such target genes is to clone those mRNA species that are increased or decreased in response to deregulated activity of a particular member of these families. Methods based on this approach include subtractive hybridization, Differential Display (DD), Representational Difference Analysis (RDA), and Serial Analysis of Gene Expression (SAGE). For these techniques (outlined in Fig. 5), mRNAs isolated from control and experimental cells are used to generate cDNAs that are then compared to each other in a variety of ways. The obvious benefit of any of these cDNA cloning techniques over the current cDNA arrays is that they can identify both known and novel genes as targets of oncogenic transcription factors.

The general principle behind subtractive hybridization is that cDNAs common to both control and experimental samples are removed or suppressed to enrich for differentially expressed mRNA species. cDNAs generated from the experimental sample are hybridized to control cDNA, duplexes are removed, and the remaining nonsubtracted cDNA clones are screened to identify those clones that display differential expression. This approach has proven to be difficult, time consuming, and not highly sensitive since low abundance mRNAs are not easily cloned. To attempt to circumvent the sensitivity problems inherent in subtractive hybridization methods, a PCR-based approach was developed. Differential Display is a PCR-based approach that allows for a rapid, broad search of large expression differences. With this technique, subsets of cDNAs from both the experimental and control samples are amplified using a set of random primer pairs, and the PCR products are compared directly on a sequencing gel (73). Although this technique is less time-consuming, it still lacks sensitivity and can have a high false positive rate. Therefore, RDA, which combines both subtractive hybridization and PCR amplification, is now a commonly used approach. RDA allows for the rapid identification of both slight and significant changes in a broad range of message levels. By performing successive rounds of hybridization between experimental and control cDNA populations followed by selective PCR amplification of unique DNA sequences, one can generate PCR products corresponding to differentially expressed sequences (74, 75). These difference products can be visualized easily in an ethidium-stained agarose gel, isolated, and cloned. In contrast to RDA, which selects out differentially expressed mRNAs for analysis, the SAGE approach directly catalogues all messages present in a population of cells by tagging, cloning, and sequencing short segments of the 3' end of all cDNAs generated from an mRNA sample (76). SAGE provides a highly sensitive measure of the relative expression levels of mRNA species under specific cellular conditions. However, the labor-intensive nature of this technique, which requires extensive sequencing in the range of tens of thousands of sequencing reactions, and the advanced computer analysis required to interpret the sequencing results is likely to limit its application.

Several cDNA cloning approaches have been used successfully to discover a handful of c-Myc target genes. For example, Benvenisty et al. (77) identified ECA39 as a Myc-regulated gene using a subtractive hybridization approach. This group reasoned that c-Myc target genes that are critical for tumorigenesis will be deregulated in a variety of c-Myc induced tumors. Accordingly, cDNA prepared from a brain tumor cell line derived from a c-Myc transgenic mouse was subtracted with normal brain mRNA to yield a library of potential Myc target genes. In a second step, the library was rescreened with cDNA probes from both c-Myc induced brain and lymphoma cell lines to identify common clones between the two tumor lines. The first subtraction refined the search from 25,000 phage down to 500 phage, and the second round of screening resulted in a total of 9 positive cDNA clones. One such clone, named ECA39, contained an E box element and was characterized as a direct target of c-Myc through promoter studies. Unfortunately, the function of ECA39 in cell growth remains elusive. Additional c-Myc target genes have been identified through screening a subtracted library that was enriched for genes that exhibit mid-G1 serum response kinetics (78). Using labeled cDNAs derived from cells expressing a conditionally inducible form of c-Myc, clones corresponding to ornithine decarboxylase (odc), lactate dehydrogenase (ldh-a), and one novel sequence were defined as Myc-responsive genes. Another group studying Myc's role in tumorigenesis used the RDA approach as a screen for genes that contribute to the anchorage-independent growth phenotype of Ratla-Myc cells (79). RDA was performed with template mRNA prepared from both parental and Myc-transformed Ratla cells grown under nonadherent conditions. Following two or three rounds of RDA, differentially expressed products were verified by performing Southern slot blot analysis with the original Ratla and Ratla-Myc cDNA populations. Further analysis of the confirmed clones using Northern blots revealed that, of 23 positive clones examined, 20 displayed differential expression with a range from 2–20-fold. Revealed in this screen were several previously identified Myc target genes including odc, -tubulin, ldh-a, and two collagen genes, as well as five novel sequences. Importantly, one of the novel genes, called rcl, was shown to be sufficient to induce anchorage-independent growth and was shown to be a direct target of the Myc-ER fusion protein, suggesting rcl may be a critical link between c-Myc and tumorigenesis. In a follow-up study, Shim et al. (80) demonstrated that the ldh-a promoter is a direct target of Myc activity and that ldh-a contains two E box elements. It was also established that ldh-a plays an important role in transformation since ldh-a antisense RNA reduces clonogenicity of c-Myc transformed lymphoblastoid cells.

cDNA cloning strategies have also been applied in studies investigating the downstream cellular target genes of the Ets family of transcription factors. In one such study, Robinson et al. (81) performed differential display analysis to identify Ets1 and Ets2 targets. Expression of either Ets1 or Ets2 has been shown to transform 3T3 cells. Therefore to screen for target genes related to this phenotype, differential display PCR was performed with RNA prepared from parental NIH 3T3 cells or cells transfected with either an Ets1 or Ets2 expression vector. From eight different primer sets, the authors identified 82 differentially expressed cDNA bands. Strikingly, many of the clones were false positives since only 16 of the clones showed reproducible differential expression by Northern analysis. Of these clones, only three were known genes: cbf (CArG box binding factor), pla2p (phospholipaseA2 activating protein), and egr1. Interestingly, pla2p was only differentially expressed in the presence of Ets2 overexpression and egr1 was only differentially expressed in the presence of Ets1 overexpression. These results suggest that Ets1 and Ets2 may activate unique downstream targets. The authors established that egr1 is a direct target of Ets1 activity through an Ets1 binding site screen (discussed in the next section) and mutational analysis of the promoter. Other groups interested in the role of Ets domain fusion transcription factors in neoplasia have also used cDNA cloning methods to search for downstream target genes. For example, Lawlor et al. (82) performed differential display analysis to identify target genes of the EWS-ETS fusion protein through a comparison of pNET (primitive neuroectodermal tumor) cell lines that are characterized by an EWS-ETS chromosomal translocation and other small round cell tumor lines. One differentially expressed band, confirmed by Northern analysis, corresponded to the Gastrin-releasing peptide (ghp) gene. Although elevated expression of ghp was observed in all pNET cell lines and primary tumors that were tested, ghp is not a direct transcriptional target of the EWS-ETS fusion protein in transfection assays. In another approach, Braun et al. (10) used RDA as a screen for EWS/FLI target genes. Overexpression of the EWS/FLI fusion protein, but not wild type FLI-1, can transform NIH 3T3 cells. Therefore, to identify unique EWS/FLI targets, RDA was performed on mRNA harvested from NIH 3T3 cells stably expressing either FLI-1 or the EWS/FLI fusion protein at both high constitutive levels and under conditional regulation. The authors characterized eight upregulated transcripts and two downregulated transcripts in the presence of EWS/FLI. Of these, a rapid increase in expression of stromelysin following conditional EWS/FLI expression suggested that it may be a direct transcriptional target. Other EWS/FLI activated transcripts identified in similar screens have been characterized as the human homolog of the Drosophila manic fringe gene, mfng, and the cyclin ubiquitin conjugating enzyme related gene, mE2-C (83, 84). The observation that expression of mfng significantly increases the tumorigenicity of NIH 3T3 cells when injected into SCID mice provides a link between EWS/FLI and transformation. Upregulation of mE2-C by EWS/FLI also represents another pathway that could impact on the cell cycle. However, mfng and mE2-C, like many other genes identified in these screens are not direct downstream targets of EWS/FLI (C. Denny, personal communication).

In summary, investigators have used cDNA cloning methods to identify Myc and Ets target genes; similar methods are currently being used to identify E2F target genes (N. Heintz, personal communication). However, these techniques are not problem-free. Although in certain cases, a low rate of false positives has been obtained (79), cDNA cloning methods such as RDA often yield a high rate of false positives. Even though RDA and differential display can identify genes that display altered expression upon deregulation of a transcription factor, these techniques are more qualitative than quantitative. Therefore, for each of the