Role for Homodimerization in Growth Deregulation by E2a Fusion Proteins

Abstract

The oncogenic transcription factor E2a-Pbx1 is expressed in some cases of acute lymphoblastic leukemia as a result of chromosomal translocation 1;19. The early observation that E2a-Pbx1 incorporates transcriptional activation domains from E2a and a DNA-binding homeodomain from Pbx1 inspired a model in which E2a-Pbx1 promotes leukemogenic transformation of lymphoid progenitor cells through transcriptional induction of target genes defined by the Pbx1 portion of the molecule. However, the subsequent demonstration that the only known DNA-binding module on the molecule, the Pbx1 homeodomain, is dispensable for the induction of lymphoblastic lymphoma in transgenic mice called into question the contribution made by the Pbx1 portion. In this study, we have used a domain swap approach coupled with a fibroblast-based focus formation assay to evaluate further the requirement for PBX1-encoded peptide elements in growth deregulation by E2a-Pbx1. No impairment of focus formation was observed when the entire Pbx1 portion was replaced with DNA-binding/dimerization domains derived from yeast transcription factor GAL4 or GCN4. Furthermore, replacement of Pbx1 with tandem FKBP domains that mediate homodimerization in the presence of a synthetic ligand led to striking growth deregulation exclusively in the presence of the dimerizing agent. N-terminal elements encoded by E2A, including the AD1 transcriptional activation domain, were required for dimerization-induced focus formation. We conclude that transcriptional target genes defined by heterologous C-terminal DNA-binding modules are not required in growth deregulation by E2a fusion proteins. We speculate that interactions between N-terminal E2a elements and undefined proteins that could function as components of a transcriptional coactivator complex may be more important.

The gene E2A encodes transcription factor proteins E12 and E47, which appear to function in the transcriptional regulation of tissue-specific gene expression. Particular roles in B-lineage lymphopoiesis, myogenesis, and pancreatic cell function have been demonstrated, but functions in diverse tissues probably exist (4). E2a proteins contain a C-terminal basic helix-loop-helix (bHLH) domain that functions in protein dimerization and DNA binding. Whereas the HLH portion mediates protein-protein interactions, the short adjacent run of basic amino acids is believed to mediate direct interaction with DNA (34). The bHLH domain is a feature shared by a large and diverse family of regulatory proteins, members of which have been identified in mammals, insects, plants, and yeast. Induction of target gene transcription by E2a proteins involves two activation domains, AD1 and AD2, located N-terminal to the bHLH (2).

An oncogenic role for E2A was first revealed by the demonstration that this locus at 19p13.3 is involved in a chromosomal translocation that is detectable in roughly 5% of cases of childhood acute lymphoblastic leukemia (ALL) (39). As a result of t(1;19), a C-terminal portion of E2a that includes the bHLH module is effectively replaced by most of the transcription factor Pbx1, encoded at 1q23, to create the chimeric protein E2a-Pbx1 (44). Alternative pre-mRNA splicing of the PBX1-encoded portion results in long (E2a-Pbx1a) and short (E2a-Pbx1b) protein products. In both splice variants the included portion of Pbx1 contains a homeodomain, a motif that possesses demonstrated roles in mediating DNA binding and protein-protein interactions (30). The observation that E2a-Pbx1 combines the AD1 and AD2 effector domains derived from E2a with a heterologous DNA-binding domain from Pbx1 has suggested a model in which leukemogenesis involves abnormal transcriptional regulation of key, unidentified target genes that are normally regulated by Pbx1 or related proteins.

E2a-Pbx1 has been shown capable of effecting neoplastic transformation in several experimental models. Forced expression of the recombinant oncoprotein causes focus formation or anchorage-independent growth in NIH 3T3 fibroblasts and blocks differentiation of cultured primary bone marrow cells (27, 41). In whole-animal studies, reconstitution of lethally irradiated mice with bone marrow transduced with E2a-Pbx1 retroviruses leads to the development of acute myeloblastic leukemia and E2a-Pbx1 transgenic mice succumb to T-lineage lymphomas (14, 25).

Structure-function correlative studies have examined the requirements for functionally defined domains within E2a-Pbx1 in transcriptional regulation and neoplastic transformation. Consistent with the transdominant model described above is the observation that transcriptional activation of Pbx1-dependent reporter constructs depends on inclusion within the E2a-Pbx1 molecule of the E2a activation domains and the DNA-binding homeodomain from Pbx1 (27, 41). Efforts to evaluate the roles of functionally defined domains in neoplastic transformation have often made use of fibroblast-based assays. Consistent with the transdominant model, these studies have demonstrated a requirement for at least the AD1 effector domain and portions of Pbx1 in cellular transformation (27, 41). Arguably the most surprising finding to emerge from this line of investigation, however, has been the consistent demonstration that the only known DNA-binding domain on the E2a-Pbx1 molecule, the Pbx1 homeodomain, is dispensable for transformation of either NIH 3T3 fibroblasts in tissue culture or lymphoid progenitors in transgenic mice (27, 41).

Pertinent to the role of the Pbx1 portion in E2a-Pbx1 leukemogenesis and potentially relevant to the question of homeodomain-independent neoplastic transformation are recent studies that elucidate participation by Pbx1 in protein-protein interactions. The affinity and sequence specificity of DNA binding by Pbx1 are markedly enhanced through direct interactions with the homeodomain-containing products of mammalian homeotic selector (Hox) genes (12, 35). This ability of Pbx1 to bind DNA in cooperation with other proteins has raised the possibility that a homeodomain-deleted E2a-Pbx1 mutant might be tethered to regulatory regions of target genes normally regulated by Pbx-Hox heterodimers through direct interactions with DNA-bound Hox proteins in the absence of any direct DNA contact by Pbx. In support of this notion, recent work has indicated that fusion to E2a of a small portion of Pbx1 from immediately C-terminal to the homeodomain that is required for optimal interactions with Hox proteins, termed the Hox cooperativity motif (HCM), is necessary and sufficient for neoplastic transformation (11). On the other hand, optimal cooperative DNA binding by Pbx1 and Hox in vitro requires the Pbx1 homeodomain in addition to the HCM (12). Nonetheless, the possibility that key intermolecular interactions could be more stable in vivo under physiological conditions exists. Clearly, questions as to the contribution of Pbx1-Hox interactions in transformation by E2a-Pbx1 remain.

In this study, we use a domain swap approach coupled with a fibroblast-based focus formation assay to evaluate further the requirement for specific Pbx1-defined target genes or protein-protein interactions in growth deregulation by E2a-Pbx1. Surprisingly, we observe that engineered E2a-Pbx1 mutants in which the entire Pbx1 portion has been replaced with a variety of heterologous peptide elements, including several not predicted to interact with DNA, retain the ability to induce growth deregulation in murine fibroblasts. Furthermore, we report that forced homodimerization of the portion of E2a present in E2a-Pbx1 is capable of inducing marked growth deregulation in the apparent absence of DNA binding. We conclude that, whereas downstream regulatory events defined specifically by heterologous, C-terminal DNA-binding domains appear to play some role in leukemogenesis, their participation is not required for growth deregulation in fibroblasts. We suggest that growth deregulation resulting from the physical approximation of E2a fusion proteins is mediated by specific protein-protein interactions involving N-terminal polypeptide elements within E2a.

MATERIALS AND METHODS

Construction of mutant cDNAs.

For the addition of a polypeptide sequence from GAL4, GCN4, a nuclear localization signal (NLS), or the FKBP_36V dimerization domains to the N-terminal portion of E2a, an artificial BglII restriction site was inserted immediately 3′ to the fusion point on the E2a cDNA by PCR using the oligonucleotide 5′-CCAGATCTAACACTGTAGGAGTCG-3′ as the downstream primer and an upstream primer overlapping the XhoI site in the E2a-coding region. The resulting product was ligated to the 5′ end of the E2a cDNA in pBluescript using the XhoI site. Sequences to be added to the 3′ end of the resultant E2a cDNA fragment were generated by PCR using upstream primers containing a BamHI restriction site and ligated to the E2a cDNA fragment using the compatible overhangs of the BglII and BamHI sites. The PCR template for the GAL4 additions was the plasmid pGAL4polyII, which contains a portion of cDNA encoding the first 148 amino acids of GAL4 with an internal XhoI site suppressed. The GCN4 cDNA fragments were amplified from genomic Saccharomyces cerevisiae DNA. For the addition to E2a of a synthetic NLS, the following sequence based on the consensus simian virus 40 (SV40) NLS was generated by PCR and ligated using the 5′ BglII site to the artificial 3′ BglII site on the E2a cDNA: 5′-AGATCTACTATGGCAAGCTTCCCAAAGAAGAAGAGGAAGCTCTAGAATTCC-3′. The insert encoding two tandem copies of FKPB_36V was generated by PCR from the plasmid pC₄M-Fv2E (kindly provided by ARIAD Pharmaceuticals, Inc., Cambridge, Mass.). E2a 1-562 was generated by PCR addition of a premature stop codon to the E12 cDNA. E2a-GCN4 250-281 1-273 was constructed by exchanging the 5′ end of E2a-GCN4 250-281 cDNA for that of the construct E2APBX1^Δ1-267 described in reference 41 using the XhoI restriction site within the E2a-encoding portion. To construct VP16-NLS-Pbx1a, the portion of the VP16 cDNA encoding amino acids 406 to 479 was amplified by PCR and ligated in frame using a HindIII site in the downstream primer to a synthetic oligonucleotide encoding an SV40 NLS which had previously been joined in frame to a portion of cDNA encoding the portion of Pbx1a C-terminal to the fusion site with E2a. The construction of E2a 1-483 has been described previously (41). Once constructed, all cDNAs were transferred into the unique EcoRI site of the MSV-tk-Neo retroviral backbone plasmid (kindly provided by Michael Cleary, Stanford University, although the vector is described in reference 42), in most cases with the use of synthetic linkers.

Retroviral gene transduction.

Retroviral constructs were transfected into the ΦNX-Eco packaging line (kindly provided by Garry Nolan, Stanford University) grown in 100-mm-diameter petri dishes in Dulbecco modified Eagle medium (DMEM) with 10% fetal calf serum by calcium phosphate precipitation using a standard protocol (31). Roughly 15 h after transfection, the culture medium was changed and the cells were transferred to a 32°C incubator for roughly 48 h, whereupon the retroviral supernatants were collected and frozen in aliquots at −80°C. The titer of infectious particles in the supernatants was determined using low-passage-number NIH 3T3 cells growing in six-well plates. Ten-fold dilutions of viral supernatant in DMEM–10% calf serum containing 5 μg of Polybrene/ml were layered onto subconfluent populations of cells. After 40 h, cells were placed under selection with G418 at 600 μg/ml. Selection was maintained with medium changes every 3 to 4 days until isolated drug-resistant colonies were readily apparent (roughly 2 weeks). Colonies were stained with Coomassie blue and counted. Viral titers were expressed as CFU per milliliter of supernatant. Typical titers obtained in this manner were between 10⁵ and 10⁶ CFU/ml.

Focus formation assays.

Focus formation assays were carried out in triplicate in 60-mm-diameter petri dishes. Low-passage-number NIH 3T3 cell derived from a single clone and obtained from the American Type Culture Collection (Manassas, Va.) were used in all experiments. For each construct, subconfluent monolayers of cells were infected with 10⁵ CFU of retroviral particles and then maintained for 13 days in DMEM–10% fetal calf serum with medium changes every 2 or 3 days. When used, the dimerizing agent AP20187 (supplied by ARIAD) was added to a final concentration of 100 nM. The resultant monolayers were then stained with crystal violet, and macroscopically visible transformed foci were counted.

Immunodetection of proteins.

For Western blots, NIH 3T3 cells grown in 60-mm-diameter petri dishes were infected with the various retroviral supernatants at a multiplicity of infection of 0.5. This inoculum was replaced with fresh medium roughly 16 h later. After a further 24 to 48 h, cells were washed in phosphate-buffered saline (PBS), lysed in 70 μl of protein sample buffer, and boiled. Fifteen micrograms of protein from each sample was fractionated on a 8% polyacrylamide gel under denaturing conditions and transferred electrophoretically to a nitrocellulose membrane. The membrane was blocked overnight in PBS containing 5% nonfat dried milk and 0.1% Tween 20 and then subjected sequentially to a 2-h incubation with primary antibody, three 5-min washes in PBS-Tween, and a 1-h incubation with horseradish peroxidase-conjugated, anti-mouse secondary antibody (Jackson ImmunoResearch Laboratories, West Grove, Pa.) at a 1:5,000 dilution. After a final wash, secondary antibody on the membrane was detected by autoradiography using chemiluminescence reagents (NEN Life Science Products, Boston, Mass.), according to the manufacturer's instructions. The anti-E2a monoclonal antibody YAE (Santa Cruz Biotechnology, Santa Cruz, Calif.) and the antihemagglutinin polyclonal antiserum Y-11 (Santa Cruz Biotechnology) each were used at a 1:2,000 dilution from stock. In the immunoprecipitation experiments, COS-7 cells were transfected by electroporation. Cells (1.6 × 10⁶ cells in 0.4 ml of complete culture medium) were placed in an electroporation cuvette (gap, 0.4 cm) with 20 μg of plasmid DNA and subjected to a pulse of electrical current (220 V, 1,050 μF) using a Gene Pulser II apparatus (Bio-Rad), after which they were maintained in complete medium for 2 days. The cells were then lysed in 500 μl of radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris [pH 8.0], 150 mM NaCl, 0.5% NP-40, 1 mM Na₃VO₄, 1 mM phenylmethylsulfonyl fluoride, 20 μg of aprotinin and 10 μg of leupeptin/ml). Lysates were precleared with protein G beads, 1 μl of the YAE monoclonal antibody was added, and the lysates were incubated overnight at 4°C with gentle agitation. This was followed by the addition of 10 μl of protein G beads and a further hour of agitation. The beads were collected by centrifugation, washed twice in RIPA buffer, and boiled in protein sample buffer in preparation for analysis by immunoblotting. In the in situ immunofluorescence experiments, NIH 3T3 cells were grown on sterile coverslips and infected with recombinant retroviral supernatant at a multiplicity of infection of 0.5. Two days after infection the coverslips were washed briefly in PBS, fixed for 10 min in 3.7% formaldehyde in PBS, and stored for various periods of time at 4°C immersed in 70% ethanol. Recombinant proteins were detected by indirect immunofluorescence using the YAE primary antibody at a 1:100 dilution and a fluorescein isothiocyanate-labeled secondary antibody (Jackson) at a 1:200 dilution according to a previously published protocol (32). Expression of the proteins E2A-GCN4 250-281 1-273 and VP16-NLS-Pbx1a was ascertained using monoclonal antibody clone G193-86 (Pharmingen, San Diego, Calif.) and a monoclonal antibody directed against the C terminus of Pbx1a (kindly provided by Michael Cleary, Stanford University), respectively.

RESULTS

We evaluated E2a fusion proteins for focus formation after retrovirus-mediated transduction into NIH 3T3 cells using a method based on one used previously by Kamps and colleagues (26). Western blotting of lysates from virus-infected NIH 3T3 cells using an antibody against the E2a portion of the proteins confirmed expression of recombinant proteins of the predicted size (Fig. 1A). Immunofluorescence studies with the same antibody indicated that all of the engineered proteins were localized preferentially to the nucleus (Fig. 1B).

FIG. 1.

Expression and nuclear localization of recombinant proteins. (A) Immunoblotting with an anti-E2a monoclonal antibody confirms that retroviral constructs confer expression of proteins of the expected sizes in NIH 3T3 fibroblasts. Endogenous E2a proteins are not detectable due to minimal reactivity of the antibody with murine E2a. (B) Representative images from immunofluorescence studies indicating that recombinant proteins appear localized exclusively to the nucleus.

As an initial test of the hypothesis that specific target genes determined by Pbx1 might not be required for transformation by E2a-Pbx1, we replaced the Pbx1 portion of E2a-Pbx1 with amino acids 1 to 148 of the S. cerevisiae transcription factor GAL4. This GAL4 domain was chosen for several reasons. (i) In common with Pbx1, GAL4 mediates sequence-specific DNA binding and protein-protein interactions. (ii) However, since yeast and humans are separated by a vast evolutionary distance and since GAL4 is not homologous to Pbx1, it seems unlikely that the two proteins would interact with similar partners. (iii) The properties of GAL4 as a transcription factor are extraordinarily well characterized. (iv) Expression of unfused GAL4 1-148 does not cause focus formation in fibroblasts (data not shown) (52).

Infection of NIH 3T3 cells with a retrovirus conferring expression of E2a-GAL4 1-148 resulted unequivocally in the formation of transformed foci (Fig. 2A). The number of foci produced was comparable to the number obtained with E2a-Pbx1a and well above background levels observed with vector alone. GAL4-encoded peptide elements make a required contribution to the transforming potential of the molecule, since neither full-length E12 nor E2a mutants with progressive, C-terminal truncations (E2a 1-562, E2a 1-483) were associated with focus formation above background levels in our assay. Furthermore, addition to the truncated E2a molecule of an NLS derived from SV40 (E2a-NLS) failed to reconstitute focus formation in this assay. Thus, the portion of E2a included in the naturally occurring leukemogenic proteins can be rendered oncogenic by the addition of several, but not all, heterologous peptide sequences.

FIG. 2.

Structure of engineered fusion proteins and results of focus formation assay. (A) Numbers indicate the average numbers of transformed foci per 60-mm-diameter petri dish, each of which was infected with 10⁵ CFU of retrovirus. Numbers in parentheses are standard deviations. For all constructs except E2a-GAL4 1-148, the results are derived from two independent experiments, from each of which two or three petri dishes were scored. The result for GAL4 1-148 is derived from three petri dishes generated in a single experiment. (B) Representative examples illustrating transformed foci produced by E2a fusion proteins.

GAL4 1-148 binds DNA as a homodimer, and the portions of the protein that mediate homodimerization and DNA binding have been mapped to separate domains (9, 28, 36). Thus, amino acids 11 to 38 constitute a zinc cluster motif that mediates DNA binding, whereas amino acids 65 to 94 constitute a module required for homodimerization. Despite requirements for both of these domains in the physiological function of GAL4, deletion mutants of E2a-GAL4 that lack either the zinc cluster (E2a-GAL4 40-148) or dimerization (E2a-GAL4 1-60) domain retain their ability to transform NIH 3T3 cells (Fig. 2A). These findings constitute a further argument against the relevance of mammalian GAL4 target genes in fibroblast transformation. Instead, they appear more consistent with the notion that the C-terminal addition of certain heterologous peptide elements, but not others, to truncated E2a alters cellular behavior through effects on the E2a portion of the molecule.

We further tested this idea by replacing the Pbx1 portion of E2a-Pbx1 with a basic leucine zipper (bZIP) domain derived from the carboxyl-terminal 55 amino acids of GCN4, another Saccharomyces transcription factor. GCN4 is homologous to the AP-1 family of mammalian transcription factors, which includes c-Fos and c-Jun. Like AP-1 proteins, GCN4 can bind AP-1 DNA elements and activate transcription from associated promoters (1, 54, 55). However, unlike AP-1 proteins, GCN4 is not transforming in fibroblasts but can be converted to a transforming protein by replacement of its amino-terminal activation domain by a corresponding domain from c-Fos or c-Jun (45). Expression of E2a-GCN4 227-281 in fibroblasts resulted in a large number of transformed foci (Fig. 2A). Thus, like effector domains from c-Fos and c-Jun but unlike that from GCN4, the activation domains of E2a are capable of carrying out functions beyond transcriptional activation that are required for neoplastic transformation. To investigate further the relationship between transcriptional activation and neoplastic transformation in the context of E2a-Pbx1, we replaced the E2a portion of the molecule with a potent activation domain derived from herpes simplex virus protein VP16, coupled with an NLS. Few foci were observed in transduced fibroblasts (Fig. 2A). These results suggest that transcriptional activation per se fails to account for focus formation by E2a fusion proteins.

The leucine zipper domain from GCN4, when isolated from the basic domain, is capable of mediating protein dimerization in the absence of DNA binding (33, 50, 56). Fusing the leucine zipper from GCN4 to truncated E2a (E2a-GCN4 250-281) produced more transformed foci than the transduction of E2a-Pbx1a, whereas fusion to E2a of a portion of GCN4 containing only the basic domain (E2a-GCN4 227-250) produced only background levels of focus formation (Fig. 2). Since the GCN4 bZIP domain is not capable of heterodimerizing with mammalian AP-1 proteins, transformation by the E2a-leucine zipper fusion is unlikely to have resulted from interactions with these (10, 50). Once again, neoplastic transformation by an E2a fusion protein appears to have occurred in the absence of DNA binding and seems more likely to have resulted from altered E2a function.

Previous studies have shown that AD1, at the N terminus of the protein, is capable of mediating transcriptional activation of reporter constructs and is required for neoplastic transformation by E2a fusion proteins (2, 27, 38, 41). In our study, deletion of an amino-terminal portion of E2a that included AD1 (E2a-GCN4 250-281 Δ1-273) abrogated transformation by the E2a-leucine zipper (E2a-GCN4 250-281) fusion protein (Fig. 2A). The observation that E2a-Pbx1 and DNA binding-incompetent mutants derived from it share a requirement for functional AD1 in fibroblasts transformation suggests that they may also share mechanisms of transformation that are similarly independent of DNA binding or Pbx1-defined target genes.

A recurrent theme among the yeast-derived, C-terminal additions associated with focus formation was the presence of domains capable of mediating homodimerization. Therefore, we made use of the FKBP system for regulated dimerization to investigate further a possible role for E2a homodimerization in growth deregulation by E2a fusion proteins (13). Portions of E2a were expressed as fusions with two tandem FKBP_36V domains, which homodimerize on interaction with the synthetic ligand AP20187 (Fig. 3A). Fibroblasts transduced with E2a_1-483-FKBP manifested striking growth deregulation exclusively when cultivated in the presence of the dimerizing agent (Fig. 3A and B). Deletion of E2a residues 1 to 273 abrogated this effect. Protein expression and AP20187-dependent dimerization were confirmed using coimmunoprecipitation and immunoblotting (Fig. 3C). These findings demonstrate that homodimerization of the portion of E2a present in E2a-Pbx1 is sufficient, in the absence of DNA binding, for growth deregulation in fibroblasts and that this promitotic effect requires polypeptide elements residing within amino acid residues 1 to 273 of E2a.

FIG. 3.

Induction of fibroblast growth by regulated dimerization of E2a_1-483. (A) Focus formation assay results. Growth deregulation is dependent on the presence of the dimerizing agent AP20187 and an N-terminal portion of E2a that includes AD1. (B) Representative petri dishes illustrating the effect of adding the dimerizing agent to fibroblasts expressing E2a_1-483-FKBP_36V. Fibroblasts in each dish were infected with 5 × 10⁵ CFU of retrovirus. (C) The anti-E2a monoclonal antibody YAE immunoprecipitates E2a_1-483-FKBP (solid arrow) but not E2a_274-483-FKBP (open arrowhead; lane 1) from lysates prepared from cotransfected COS-7 cells since the latter protein lacks the required epitope. However, the N-terminal-deleted protein coprecipitates with E2a_1-483-FKBP in the presence of the dimerizing agent AP20187 (lane 2), confirming AP20187-dependent dimerization.

DISCUSSION

The observation that an experimental mutant of E2a-Pbx1a lacking the Pbx1 homeodomain retains its ability to induce focus formation in NIH 3T3 fibroblasts and lymphomas in transgenic mice caused us to question the notion that specific Pbx1 target genes are required in neoplastic transformation by this protein. Our results demonstrate that the N-terminal two-thirds of E2a can induce focus formation in NIH 3T3 cells when fused to a variety of what initially were rather arbitrarily chosen, heterologous protein elements unrelated to Pbx1. In particular, forced homodimerization of this portion of E2a using the FKBP system produces striking growth deregulation.

Clinical observations indicate that Pbx1 is not unique in its ability to activate the latent oncogenicity of E2a when substituted at its C terminus. In t(17;19), a translocation seen in a small number of childhood ALL cases, the same C-terminal portion of E2a is replaced by a portion of the transcription factor Hlf (20, 22). E2a-Hlf includes an Hlf-derived bZIP domain capable of dimerization and sequence-specific DNA binding. Forced expression of E2a-Hlf produces a transformed phenotype in NIH 3T3 cells and confers cytokine-independent survival on otherwise interleukin 3-dependent, bone marrow-derived cell lines (21, 57). Hlf-defined target genes have been proposed as mediators of E2a-Hlf-dependent phenotypic changes (29). However, deletion of the entire HLF-encoded portion from the fusion protein fails to abrogate its antiapoptotic effect (23). Therefore, the antiapoptotic effect of E2a-Hlf appears not to require participation by Hlf-defined target genes.

A third leukemia-associated chromosomal rearrangement involving E2A was described recently. In a substantial proportion of cases of B-lineage ALL, an internal rearrangement of chromosome 19 results in the replacement of the same C-terminal portion of E2a lost in E2a-Pbx1 and E2a-Hlf by a portion of a novel protein of unknown function termed FB1 (7). A total of four ALL cases with translocations resulting in E2A-FB1 fusions were uncovered. In three of these, the translocations resulted in out-of-frame fusions with FB1, theoretically leading to truncation of the chimeric protein products within the FB1-encoded portion. Therefore, leukemogenesis by E2a-FB1 fusion proteins appears not to rely on induction of specific, FB1-defined target genes.

E2a fusion proteins could contribute to neoplastic transformation through dominant-inhibitory effects on wild-type E2a or related proteins. Abundant experimental evidence indicates that the E2a proteins E12 and E47 can function to promote cellular differentiation and inhibit proliferation, usually as heterodimers with tissue-specific bHLH proteins. For example, E2a-MyoD heterodimers promote differentiation and exit from the cell cycle in myogenic cells and forced expression of E2a in murine fibroblasts exerts an antimitotic effect by inhibiting progression through the G₁ phase of the cell cycle (43, 48). Intracellular signaling from the Notch family of transmembrane receptors inhibits lineage-specific differentiation, in part through repressing bHLH protein function (reviewed in reference 40). E47 function, which is required for B-cell differentiation, is inhibited by signaling from Notch1 or Notch2 (46). Compelling data suggest that the antimitotic effects of E2a may be mediated at least in part through direct transcriptional induction of the cyclin–cyclin-dependent kinase inhibitor p21/CIP1/WAF1 (49). Id proteins contain HLH domains but lack an adjacent basic region. Thus, they appear to act as naturally occurring dominant-inhibitory regulators of bHLH proteins by sequestering them in complexes that are unable to bind DNA. Id proteins appear necessary for cell cycle entry and promote cell cycle progression and cell proliferation (5, 48). Forced expression of Id3 produces a transformed phenotype in NIH 3T3 cells (15). In a manner at least superficially similar to that of Id proteins, the leukemogenic bHLH protein SCL/tal-1, overexpression of which is associated with T-lineage ALL, appears capable of sequestering E2a proteins in multimeric complexes and, depending to some degree on the cellular context, tends to promote cell proliferation and impair differentiation (17, 18, 47). Thus, available evidence justifies consideration of E2a proteins as potential tumor suppressors whose functional disruption may promote oncogenesis.

Also consistent with this view are results from whole-animal studies. Transgenic mice in which E2a-Pbx1 or E2a-Hlf transgenes are expressed under the control of an immunoglobulin heavy chain gene enhancer manifest similar phenotypic features, including disordered thymocyte maturation with thymic involution, a paucity of B-lineage lymphocytes, and a strong propensity to develop T-lineage malignant neoplasms (14, 19, 53). Most interestingly, these features resemble those observed in E2A-null mice, which manifest similar maturational abnormalities of thymocytes, develop T-progenitor malignant lymphomas at a high rate, and are completely deficient in B cells (3). The phenotypic resemblance of E2A-targeted mice to E2A-PBX1 and E2A-HLF transgenic animals is suggestive of a dominant-inhibitory mechanism of leukemogenesis by the chimeric gene products.

Fusion proteins could compete with wild-type bHLH or other growth-regulatory transcription factors for access to one or more effector proteins required for downstream antimitotic or transcriptional effects. In particular, the E2a domains AD1 and AD2 probably function as modules required for interactions with specific effector proteins. Although these interactions may be manifested experimentally as transcriptional induction of reporter constructs, their biological functions are likely to be more subtle and specific. Two observations from our study point to the existence of functional specificity in relation to E2a effector domains. First, although the yeast GCN4 transcription factor is not transforming in NIH 3T3 cells, replacement of its N-terminal activation domain with effector domains from E2a results in robust focus formation. Second, replacement of the E2a portion of E2a-Pbx1 with the potent activation domain from herpes simplex virus protein VP16 fails to reconstitute focus formation. These data are consistent with a model in which the effector domains from E2a and VP16 are capable of differentially influencing downstream transcriptional events through interactions with distinct protein partners.

A curious observation in studies evaluating the role of the Pbx1 portion in neoplastic transformation by E2a-Pbx1 has been that nonoverlapping portions of Pbx1, when fused to E2a, create proteins that transform fibroblasts. Thus, Kamps and colleagues showed that deletion from E2a-Pbx1 of the homeodomain and everything C-terminal to it enhanced, rather than impaired, focus formation and Chang et al. showed that a short polypeptide segment immediately C-terminal to the homeodomain is sufficient, when fused to E2a, for the induction of anchorage-independent fibroblast growth (11, 27). In suggesting that growth induction by E2a fusion proteins may result from functional alterations of the E2a portion that are independent of specific intermolecular interactions defined by heterologous C-terminal polypeptide additions, our results at least provide a novel conceptual framework within which this apparent paradox might be understood.

Our findings support the novel idea that homodimerization per se may play an important role in growth deregulation by E2a fusion proteins, independent of the role played by homodimerization in DNA binding. These observations may help explain the ability of a homeodomain-deleted E2a-Pbx1 mutant to induce lymphoblastic lymphoma in transgenic mice (41). It is well recognized that homo- or heterodimeric interactions mediated by the bHLH or bZIP modules are required for optimal DNA binding by wild-type E2a proteins and E2a-Hlf, respectively. Although the bZIP domain of Hlf can mediate heterodimerization with other bZIP transcription factors of the proline- and acidic amino acid-rich class, homodimerization of E2a-Hlf is sufficient for growth deregulation in fibroblasts (24). Recently, Calvo et al. showed that E2a-Pbx1 also can bind DNA as a homodimer and that homodimerization is mediated by polypeptide elements N-terminal to the homeodomain within the Pbx1 portion of the molecule (8). An additional independent dimerization module is predicted to lie within the homeodomain. The observation that the heterologous C-terminal polypeptide additions to E2a that result from leukemogenic chromosomal translocations can mediate homodimerization suggests that our findings derived from the addition of artificial homodimerization modules may be relevant to growth control in naturally occurring leukemia.

Coactivator proteins or complexes are implicated in transcriptional regulation by E2a. In electrophoretic mobility shift assays, a DNA probe containing tandem, but not single, E-box elements interacted with a stable protein complex containing E47 homodimers and the transcriptional coactivator protein p300 in a nuclear extract from B-lineage cells (16). Cotransfected p300 enhances reporter gene transactivation by either AD1 or AD2 in the context of fusion with the GAL4 DNA-binding module (51). Furthermore, the LDFS motif, located at the extreme N terminus of AD1 and required for transcriptional activation by AD1, mediates interaction with the SAGA coactivator complex in yeast (37). The ability of the viral oncoproteins E1A and SV40-T to induce cell proliferation and impair differentiation is mediated in part through interactions with the coactivator p300 and closely related CREB-binding protein (6). Our demonstration that the induced proximity of the N-terminal two-thirds of E2a causes pronounced growth deregulation in fibroblasts raises the possibility that misassembly, misdirection, or misfunction of an AD1-interactive coactivator protein or protein complex may play a role in neoplastic transformation by E2a fusion proteins.

Phenotypic differences between primary leukemic cells that express E2a-Hlf and those that express E2a-Pbx1 and between cultured cells forced to express these oncoproteins experimentally exist. Furthermore, although the Pbx1 homeodomain is not required for growth deregulation in fibroblasts or lymphomagenesis in transgenic mice, it is required for immortalization of primary murine myeloid cells (27). It seems likely, therefore, that although the particular heterologous C-terminal polypeptide additions derived from Pbx1 or Hlf are not required for growth deregulation in at least some cell types, they do play some role in oncogenesis. In indicating that the latent oncogenicity of E2a may be activated by homodimerization or, more generally, through C-terminal substitutions that have nothing to do with DNA binding, our results underscore the need to further delineate the relevance to cell cycle regulation of protein-protein interactions involving the N-terminal two-thirds of E2a.