Splice Variants of the Forkhead Box Protein AFX Exhibit Dominant Negative Activity and Inhibit AFXα-Mediated Tumor Cell Apoptosis

Background Loss-of-function in the apoptosis-inducing genes is known to facilitate tumorigenesis. AFX (FOXO4), a member of forkhead transcription factors functions as a tumor suppressor and has 2 isoforms, AFXα (505 a.a.) and AFXζ (450 a.a.). In human cancer cells, we identified an N-terminally deleted form of AFXα (α198-505), translated from a downstream start and 2 short N-terminal AFX proteins (90, and 101 a.a.) produced by aberrant splicing. Methods and Findings We investigated the expression and role of these AFX variants. Cell transduction study revealed that short N-terminal AFX proteins were not stable. Though α(198-505) protein expression was detected in the cytoplasm and nucleus, α(198-505) expressing cells did not show a nucleocytoplasmic shuttling mediated by PI3 kinase signaling. Whereas, we observed this shuttling in cells expressing either AFXα or AFXζ protein. AFXζ and α(198-505) lost the ability to transactivate BCL6 or suppress cyclin D2 gene expression. These variants did not induce cancer cell death whereas AFXα resulted in apoptosis. We found that AFXζ and α(198-505) suppress the AFXα stimulation of BCL6 promoter in a dose dependent manner, indicating dominant negative activity. These variants also inhibited AFXα induction of apoptosis. Conclusions Loss of function by aberrant splicing and the dominant negative activity of AFX variants may provide a mechanism for enhanced survival of neoplastic cells.

Introduction AFX (FOXO4) is a member of the class O (FOXO) subfamily of forkhead transcription factors that includes the functionally related proteins FOXO1 (FKHR), FOXO3a (FKHRL1), and FOXO4 [1,2]. FOXO transcription factors have important roles in metabolism, cellular proliferation, and apoptosis. Overproduction of FOXOs induces either cell cycle arrest or apoptosis. When relieved of FOXO activity, cells re-enter the cell cycle and start to proliferate [3]. A role for FOXO members in tumorigenesis was initially suggested by the observation that FOXO members are involved chromosomal translocations in certain types of tumors [4,5]. The PAX3-FKHR fusion product has been shown to transform cells in culture [6]. However, translocations involving FOXO proteins might also result in loss-of-function of a FOXO allele.
AFX is known to have two isoforms, AFXa (505 amino acids) and AFXf (450 amino acids) [7]. AFXf is a splice variant that encodes a shorter protein lacking amino acids 58-112, including the first 16 amino acids of the forkhead domain. AFXa is ubiquitously expressed, whereas AFXf is expressed predominantly in liver, kidney, pancreas, heart, and placenta. AFXf transcripts are not detected in ovary, testis, brain, prostate, colon, and leukocyte. The different tissue distributions AFXa and AFXf suggest distinct transcriptional regulation and actions on a different subset of target genes.
In this report, we describe novel spliced forms of the AFX transcript in human cancer cells and their effect on tumor apoptosis and growth.

AFX splicing variants
PCR amplification of reverse-transcribed cDNA samples from human cancer cell lines was performed using a primer set that anneals to both AFXa and AFXf cDNAs. The upper band (445 bp) corresponds to AFXa and lower band (290 bp) contains AFXf. There was a slight difference in the size of AFXf bands among the cell lines, prompting us to consider possible splicing variants (Fig. 1A). Direct sequencing of lower bands identified the AFXf isoform, as well as overlapping and mixed sequences near splice acceptor site of AFXf isoform (not shown). The lower bands were cloned and sequenced to investigate these sequences further. Two novel splicing variants were identified, in which the splice acceptance occurred aberrantly at 4 and 14 bases before splice acceptor site of the AFXf isoform (Fig. 1B, C). This aberrant splicing results in either a 4 or 14 base insertion, leading to a shift of the reading frame and producing premature stop codon. Splicing variants with the 4 or 14 base insertion encode predicted proteins of 90 (AFXtr1) or 101 (AFXtr2) amino acids, respectively (Fig. 1C). A variety of tumor cells showed different patterns of AFX transcripts (Fig. 1A). Either AFXtr1 or AFXtr2 were detected in most tumor cells. AFXf was detected in DOV13, OVCA429, CaCO2, HEK293, and 293FT cells. AFXa expression was absent in OVCA429 and 293FT cells. JEG3 cells express only AFXa. Percent ratio of AFXf, AFXtr1, and AFXtr2 was different each other according to tumor cells (Fig. 1A).
We became interested in the N-terminally truncated AFXa proteins because AFXtr1 and AFXtr2 are prematurely terminated in exon 1, and three potential alternative start codons (M1, M2, and M3) are present in exon 1b and 2 ( Fig. 2A). M3 is the methionine codon located at the end of forkhead domain. It contains a consensus Kozak sequence for initiation of translation and thus potentially encodes an N-terminally deleted AFXa. Western blot analyses using an anti-HA antibody from the lysate of 293FT cells transfected with C-terminal hemaglutinin (HA) epitope-tagged AFXa detected an additional band, suggesting a presence of N-terminally deleted AFX protein (Fig. 2B, arrow). In order to identify this product, we generated three N-terminally deleted AFX constructs tagged with C-terminal-HA epitope {a(131-505)-HA, a(178-505)-HA, and a(198-505)-HA}. Western blotting with anti-AFX antibody (C-terminal specific) after immunoprecipitation with anti-HA antibody of transfected cell lysates confirmed the presence of the a(198-505) protein (Fig. 2C, arrow). Expression and subcellular localization of AFX variants Cellular protein processing and localization of AFX variants was examined by immunofluorescence (IF) staining. JEG3 cells were transfected with the constructs carrying the N-or C-terminal HA epitope in-frame with AFXa-HA, AFXf-HA, or a(198-505)-HA. Immunofluorescence positivity was detected in about 5-8% of JEG3 cells. No protein expression was detected in cells transfected with HA-AFXtr1 and HA-AFXtr2 (Data not shown). To achieve high levels of gene expression, we generated adenoviral vectors carrying these constructs. After determining the dose of adenoviral vectors required to achieve 95-100% expression of the control b-galactosidase gene in HeLa cells, protein expression was still not detected for vectors containing HA-AFXtr1 or HA-AFXtr2 (data not shown), whereas 90-95% of cells expressed AFXa-HA, AFXf-HA, or a(198-505)-HA in either cytoplasms and nuclei (Fig. 3A). These results indicate that the short aminoterminal-AFX proteins are not appropriately processed or stable in the cells, although the product from the alternate translation start site a(198-505) is produced.
The effect of the cell signaling on subcellular localization was assessed by adding insulin, EGF, or LY294002 (a PI3 kinase inhibitor). Insulin and EGF treatment induced the excursions of AFXa and AFXf to cytoplasms and LY294002 pretreatment prevented these excursions, indicating that nucleocytoplasmic shuttling of AFXa or AFXf protein is occurred and regulated by PI3 kinase signaling pathway. Whereas a(198-505) expressing cells did not show this shuttling (Fig. 3B). This defect may be due to the fact that a(198-505) has lost two binding sites for 14-3-3 protein and half of the nuclear localization sequence (182-211 amino acids) [8,9,10,11].
AFX variants fail to transactivate the BCL6 gene or suppress the cyclin D2 gene The transcriptional activity of AFXf and a(198-505) was compared to AFXa using a luciferase reporter system. The amounts of plasmid vectors transfected were carefully optimized to minimize nonspecific effects [12] and luciferase values were normalized by protein measurements. AFXf and a(198-505) did not stimulate 3IRS-TATA-Luc, whereas AFXa activated this reporter about 15fold ( Fig. 4A). Two representative target genes, BCL6 and cyclin D2, were also examined. BCL6 is a transcriptional repressor of BCL-X L , an antiapoptotic protein, and the AFX protein is thought to induce activates apoptosis by stimulation of BCL6 [13]. The cell cycle inhibition by AFX protein involves down-regulation of cyclin D [14]. In 293FT cells, AFXf and a(198-505) did not stimulate BCL6p-Luc or suppress CD2p-Luc, whereas AFXa stimulated BCL6p-Luc 13-fold and suppressed CD2p-Luc by 75% (Fig. 4B, C). RT-PCR and Western blot analyses were used to investigate the effect of the AFX variants on the expression of BCL6 or cyclin D2 in cells infected with adenoviral vectors carrying AFXa, AFXf, or a(198-505). Whereas AFXa induced BCL6 expression in MCF7 cells, Ad-f or Ad-a(198-505) did not increase either BCL6 mRNA or protein (Fig. 4D). Ad-a suppressed cyclin D2 expression, however Ad-f or Ad-a(198-505) lost the ability to suppress its expression in MCF7 or OVCA429 cells (Fig. 4E). These results indicate that the AFX variants are loss-of-function mutants.

AFX variants have lost the ability to suppress tumor cell growth
HepG2, T47D, HeLa, OVCA429, and OVCA420 cells were infected with adenoviral vectors (from 2.5 to 10 PFU/cell ) expressing AFXa, AFXf, or a(198-505). Cell viability was assessed using the MTS assay at day 6 and the percent cell survival was calculated compared to Ad-Empty (Ad-E) infection. Infection of Ad-a suppressed 70-95% cell growth, whereas Ad-f or Ad-a(198-505) did not suppress cell growth (Fig. 5A), indicating that the AFX variants have lost the ability to suppress tumor cell growth.  Ad-E, an MOI (10 and 20 PFU/cell) that was equal to total amount of co-infected adenoviral vectors. Co-infection of Ad-a and Ad-E induced 30-40% more cell death compared to corresponding amounts of Ad-E infection. Co-infection of Ad-f, Ad-a(198-505), or Ad-a(1-200) prevented AFXa induction of cell death (Fig. 5B), suggesting dominant negative activity of AFX variants. Cellular examination was also performed using double immunofluorescent. HeLa cells expressing nucleic AFXa alone showed apoptotic nuclei, whereas HeLa cells co-expressing AFXa with AFXf or a(198-505) did not show apoptosis. Of note, AFXa was co-localized with AFXf or a(198-505) in nuclear speckles of these cells (Fig. 5C).

AFX variants inhibited AFXa induction of cell death by a dominant negative effect on the BCL6 gene
The mechanism of dominant negative activity of AFX variants was assessed further by examining effects on transcriptional regulation of 3IRS-TATA-Luc, BCL6p-Luc, and CD2p-Luc. Co-transfection of AFXa with equal or increasing amounts of the AFX variants was preformed in the context of these luciferase reporters. Co-transfection with the AFX variants reduced the AFXa-induced transactivation of 3IRS-TATA-Luc and BCL6p-Luc in a dose-dependent manner (Fig. 6A, B). The reduction by AFXf was comparable to that of a (1-200). However, the AFX variants did not reverse AFXa suppression of CD2p-Luc activity (Fig. 6C). These results suggest that the inhibition of AFXainduced apoptosis is mediated through dominant negative activity of AFX variants on AFXa in BCL6 gene expression.

Discussion
In this report, we identified spliced forms of AFX transcripts in multiple human cancer cell lines. Short aminoterminal AFX proteins (AFXtr1 and AFXtr2) produced by aberrant splicing were not stable, suggesting AFX inactivation by aberrant splicing. However, alternative splicing and translation produced AFXf and a(198-505), respectively. AFXf and a(198-505) lost the ability to transactivate BCL6 or to suppress cyclin D2 gene expression. Although inactive as individual transcription factors,  AFXf and a(198-505) exert dominant negative activity on AFXa stimulation of BCL6 gene. These variants also lost the ability to induce apoptosis but they inhibited AFXa induced apoptosis, presumably through dominant negative activity on the BCL6 gene. Inactivation of AFX by aberrant splicing and the dominant negative function of the AFX splicing variants could therefore provide a growth advantage during cancer progression. It has been suggested that AFXa and AFXf may antagonized each other because of their distinct transcriptional activity for different target genes. For example, AFXf stimulates PEPCK, G6Pase promoters, whereas AFXa does not activate these promoters [7]. We also observed distinct transcriptional regulatory function for AFXa and AFXf. The CD2 promoter was weakly stimulated by AFXf and repressed by AFXa (Fig. 4C). A novel finding in this study is the dominant negative function of AFXf or a(198-505) on AFXa regulation of the BCL6 promoter. A number of AFXa binding sites have been demonstrated in BCL6 promoter [13]. One AFXa binding site was also competed by a known forkhead-binding site (IRS) derived from the IGFBP-1 promoter [13]. Consistent with this, we also observed dominant negative activity by the AFX variants on AFXa stimulation of 3xIRS promoter construct. These results indicate that the dominant negative activity by AFX variants may be mediated through IRS (GCAAAACAA AC TTATTTTGAA).
The induction of BCL6 accounts for part of the apoptotic mechanism mediated by AFXa. FOXO dependent expression of IGFBP-1 [17], FasL [18,19] and Bim [20][21][22][23] have also been shown to promote apoptosis. Direct FOXO binding activity has been demonstrated in the FasL promoter (GTAAATAAATA) and the Bim promoter (GTAAACAC). It is possible that the AFX variants may also influence AFXa activation of these genes.
Of note, the AFX variants did not reverse AFXa suppression of the CD2 promoter. FOXO factors have been shown to elevate p27KIP1 expression and induce cell cycle arrest [24,25]. FOXO3a and AFX have also been shown to inhibit the cell cycle through downregulation of cyclin D by a p27KIP1-independent mechanism [2,14,26]. Chromatin immunopreciptation (ChIP) using the cyclin D1 [27] and cyclin D2 promoters [28] demonstrated FOXO binding on the cyclin D promoters. However, a recent study suggested that transcriptional repression of D-type cyclin may not involve direct binding of FOXO factors to cyclin D1 or D2 promoters [26]. These results suggest that transcription repressors activated by AFXa might play a role in downregulation of cyclin D.
An important question is whether there is a causal relationship between AFX splicing variants and cancer. For this to be feasible, the alternative splicing products would need to be expressed at significant levels compared with the normally spliced product [29]. We compared the pattern of AFX transcripts in two different cell lines (HEK 293 and 293FT) originated from same cells. By RT-PCR there was a decrease of the AFXa transcript in 293FT cells, whereas HEK293 cells still possessed it (Fig. 1A). HEK293 cells are transformed by the adenovirus E1 gene and 293FT cells were additionally transformed by simian virus (SV) 40 large tumor antigen. Loss of the AFXa transcript and the appearance of the AFX variants could contribute to the higher proliferation of 293FT cells relative to HEK293 cells (data not shown). However, the mechanism described in our study remains to be determined in primary tumors.
Defects in mRNA splicing occur frequently in human cancer cells [29] but the mechanisms leading to splicing defects in cancer are poorly understood. A point mutation in the genomic splice site or regulatory elements have been shown in selected splicing defects [30][31][32]. However, DNA sequencing of the intronic and exonic portions of AFX gene did not reveal mutations. Variations in the composition, concentration, localization, and activity of transacting regulatory factors may also modulate splice-site recognition and usage [33,34]. Activation of the oncogenic signal pathway also stimulates aberrant splicing [35]. It is of interest to better understand how splicing regulatory factors in cancer cells might affect AFX aberrant splicing and thereby cell survival.

RT-PCR and sequence analysis
Total RNA was extracted from each cells using TRIZOL reagent (Invitrogen, Carlsbad, CA) as described by the manufacturer. RNA (20 mg) was treated with DNAse-I (Promega, Madison, WI) for 30 min at room temperature. Random hexamers and AMV reverse transcriptase (RT) were used to synthesize cDNA using 10 mg of DNAse-I treated RNA. A portion (1/40) of the cDNA solution was used for amplification of AFX. Cycle conditions were: 2 min hot start at 96uC, followed by 35 cycles of 1 min at 94uC, 45 sec at 56uC, followed by 1 min at 72uC, and a final extension at 72uC for 5 minutes. An aliquot (50%) of each PCR reaction was resolved by electrophoresis on 1.8% agarose gels and DNA products were visualized with ethidium bromide. Oligonucleotides used for PCR amplification include forward 59-ACG TAT GGA TCC GGG GAA TG-39 and reverse 59-TCC ATC CTG CTG AGC TGT-39. PCR product give two bands; AFXa (445 bp) and AFXf (290 bp). The lower AFXf bands were excised and cloned into p-TOPO vector (Invitrogen). To identify splicing variants, 20 positive clones carrying the PCR segment from each cell line were sequenced using an automated DNA sequencer.

Plasmids and transfections
The cDNAs for AFXa, AFXf, AFXa(198-505), AFXa(1-200), AFXtr1, and AFXtr2 were amplified using KOD DNA polymerase (Novagen, San Diego CA) and the RT product of total RNAs from cancer cells and cloned into pCRH-Blunt (Invitrogen). After verification of sequencing, each cDNA was subcloned into pCDNA3. Fusion constructs were created by incorporating an N-or C-terminal influenza virus HA epitope (YPYDVPDYA) inframe with the AFXs in the pCDNA3 vectors.
Transfections of 293FT cells with 500 ng of luciferase constructs and 10 ng of each AFX construct were performed using the calcium phosphate method. Forty eight hrs later, cells were harvested 48h later and luciferase activity was assayed and normalized as described before [12]. Results were averaged from 3 independent experiments and are plotted as means6standard deviations for quadruplicated wells.

Adenoviral infection and cell proliferation assay
Recombinant adenoviral vectors {Ad-a, Ad-aHA, Ad-f, Ad-fHA, Ad-a(198-505), Ad-a(198-505)HA, and Ad-a(1-200)} carry-ing each of the AFX cDNAs, with or without HA epitope, were amplified, purified, titrated as described previously [37]. Cytotoxicity was assessed using a nonradioactive cell proliferation assay according to the manufacturer's protocol (Cell Titer 96 Aqueous Non-Radioactive Cell Proliferation Assay, Promega). The day after plating 3610 3 cells per well in 96-well plates, adenoviral vectors {Ad-a, Ad-f, and Ad-a(198-505)} were infected at various MOIs (2.5-10 plaque forming unit/cell). Empty adenoviral vector (Ad-E) was also infected as a control. To achieve 100% cellular expression of adenoviral transgene, MOIs for each cell line were determined by infection of an adenovirus carrying b-galactosidase [37]. Fresh medium (DMEM/F12 supplemented with 2.5% heat inactivated FBS) was added 8 h after infection and every 2 days thereafter. Cell viability was assayed 6 days after infection. Percent cell survival was quantitated relative to Ad-E infected or uninfected cells.