Multiple domains define the expression and regulatory properties of Foxp1 forkhead transcriptional repressors.

The Foxp subfamily of forkhead/HNF3 transcription factors has recently been recognized because of its involvement in autoimmune disease, speech and language disorders, and lung development. Domains unique to this subfamily include a divergent DNA-binding winged helix, a leucine zipper, a zinc finger, and a polyglutamine tract. Little is known about the properties of these proteins that are fundamental to their function as transcription factors nor how the Foxp sequence motifs regulate their transcriptional regulatory properties. We report here a structure/function analysis of the Foxp1 protein. We have analyzed the alternative splice isoforms 1A and 1C and also report the cloning and characterization of a novel isoform Foxp1D that lacks the polyglutamine domain. We have isolated the preferred DNA-binding sites for Foxp1 transcription factors. Foxp1A, C, and D isoforms and the related Foxp2 protein repress gene transcription via binding to this consensus site or to a naturally occurring site within the SV40 and the interleukin-2 promoters. In some cases the strength of Foxp1 repression is mediated by the polyglutamine domain. Unlike previously characterized forkhead factors, Foxp1 proteins can form homodimers or heterodimers with subfamily members. The dimerization domain was localized to an evolutionarily conserved C2H2 zinc finger and leucine zipper motif. Finally, we demonstrate that Foxp1, although broadly expressed, is further regulated by tissue-specific alternative splicing of these functionally important sequence domains. These results suggest that Foxp1 proteins have diverse functional roles in different cell and tissue types.

The Fox gene family encompasses a large number of transcription factors homologous to forkhead and HNF3. Fox transcription factors are identified by a conserved 110-amino acid DNA-binding (winged helix) domain (1,2). These proteins play important roles in regulating gene transcription during early development and organogenesis and control cell type-and tissue-specific gene expression (1,2). Among the Fox transcription factors, recent attention has been drawn to members of the Foxp subfamily (Foxp1, Foxp2, and Foxp3). The functional importance of the Foxp proteins is underscored by their spontaneous mutation in mouse and human diseases. Misregulation of FOXP1 expression is seen in a variety of tumors, leading to the speculation that FOXP1 is a tumor suppressor gene (3). Mutations in FOXP2 provided the first incidence for linkage of a single mutant allele to speech and language disorders (4). FOXP3 (JM2) mutations cause X-linked autoimmunity-allergic disregulation syndrome in humans (5), and the mouse orthologue of Foxp3 is responsible for the Scurfy mouse phenotype (6,7).
Despite the interesting physiological roles that have been uncovered by mutations in Foxp genes, little is understood about how Foxp transcription factors participate in signaling and gene regulation. Several candidate target genes with promoters containing forkhead binding sites have been proposed (8,9). Identification of a Foxp subfamily-specific DNA binding preference would be useful in validating current candidates and help identify new targets. Clues to the mechanisms by which Foxp proteins regulate transcription may lay in several interesting protein domains unique to the Foxp subfamily. Foxp transcription factors are among the most distantly related Fox subfamily members. Homology is limited to an 80amino acid region within the DNA-binding domain that is truncated at the carboxyl helices Wing 1 and Wing 2 (5,6,8,9). Such diversion from the HNF3 winged helix structure would predict significant differences in the preferred DNA-binding sequences for Foxp proteins. The HNF3 structure further indicated that Fox proteins bind DNA as monomers (10 -12). However, the observation that all characterized Foxp proteins contain a leucine zipper motif that resembles the N-myc zip dimerization domain questioned the generality of this conclusion. Mouse Foxp1 and Foxp2 and human FOXP1 also include an N-terminal polyglutamine stretch of 37-40 residues (9). Among transcription factor domains, polyglutamine has been identified as a common protein-protein interaction motif that can regulate transcriptional activity (13). The role of the polyglutamine domains in Foxp proteins has not been tested.
In this study, we provide evidence that various Foxp1 isoforms repress transcription via direct binding to Foxp consensus DNA-binding sites present in the promoter regions of SV40 and IL-2. 1 We show that the N-terminal polyglutamine domain strongly modulates Foxp1 repression activity on SV40 promoter. We demonstrate that a domain including leucine zipper and zinc finger motifs is both necessary and sufficient for specific dimerization between Foxp proteins. Finally, we show that Foxp isoform expression is highly regulated in tissues and cells.

Plasmid Constructions
Expression Vectors-Foxp1A was cloned into pCDNAI/NEO (Invitrogen) and is driven by cytomegalovirus immediate-early promoter. In our original cDNA cloning effort, we isolated a partial Foxp1A cDNA lacking exon 1. This cDNA encodes the same protein as Foxp1D and was cloned into pCDNAI/NEO. 5ЈHA-Foxp1A and 3ЈHA-Foxp1A were generated by introducing an HA epitope tag in frame of Foxp1A using PCR. Foxp1A mutants were generated using single-stranded site-directed mutagenesis (14). Expression vectors encoding full-length Foxp1C and Foxp2 were described previously (9) and kindly provided by Dr. Ed Morrisey (University of Pennsylvania School of Medicine).
Yeast Two-hybrid Constructs-pASICYH2-Foxp1FLЈ and pACTII-Foxp1FLЈ were generated by subcloning a Foxp1A (V543L,R544E) BstYI fragment encoding amino acids 146 -654 into the BamHI sites of pAS1CYH2 and pACTII in translational frame, respectively. Yeast two-hybrid vectors were kindly provided by Clarence Chan (University of Texas at Austin). pASICYH2-ab, a and b constructs were made by PCR amplifying the Foxp1A cDNA.
Transcription Reporter Constructs-1xb-pGL3 promoter was generated by subcloning a single Foxp1 DNA-binding site B upstream of the SV40 promoter into the BglII site of pGL3 promoter (Promega, Madison, WI). 3xbp-GL3 control was made by inserting tandemly ligated binding site B primer pairs into the BglII site of pGL3 control (Promega). IL-2-GL3 (NFAT-GL3) (15), which contains trimerized human distal NFAT site inserted into the IL-2 minimal promoter, was kindly provided by Dr. J. Crabtree (Stanford University School of Medicine). IL-2-mutGL3 was generated using a QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA).

Isolation of Foxp1A cDNA Clones
A single Foxp1A cDNA clone was isolated by using a (CAG) 12 repetitive oligonucleotide probe to screen a cDNA library prepared in Zap from poly(A) ϩ mRNA isolated from B-cell leukemia (BCL-1) cells. The BCL-1 cells were stimulated in culture by IL-5 and IL-2 to differentiate before harvesting of RNA. Its complete sequence was determined on both strands by the Exoquence method (8). The Foxp1 cDNA sequence was confirmed from independent cDNAs using fragments generated from the initial isolate (fragment 1, nucleotides 1-357; fragment 2, nucleotides 1516 -1978) as hybridization probes. Seven randomly picked clones were plaque-purified, and the cDNA inserts were excised as plasmids ( ZAP protocol; Stratagene).

RNA Analysis
RNase protection assays were performed using the RPAII kit according to manufacturer's protocols (Ambion, Austin, TX). 20 g of mouse total RNAs were used to assay for Foxp isoforms using either "5Ј probe" (nucleotides 93-425) or "3Ј probe" (nucleotides 1978 -2122). 2 g of mouse total RNAs were used for mGAPDH control. Either a DNA sequence ladder or radiolabeled PUC19 digested with Sau3AI was used as size marker for protected RNA bands. Mouse tissue Northern blot was hybridized with the 3Ј probe. The hybridization and washes were performed under high stringency (14) and exposed for 3 days on a Molecular Dynamics PhosphorImager.

Western Blot Analysis and in Vitro Co-immunoprecipitation
Monoclonal antibody 1G1 was generated against the forkhead DNAbinding domain of Foxp1. In vitro transcription and translation were performed using the TNT-coupled reticulocyte lysate system (Promega). Monoclonal anti-HA ascites 16B12 (BAbCO, Berkley, CA) was used in 1:1000 dilution, and 1G1 ascites were used in 1:5,000 dilution in 5% bovine serum albumin, phosphate buffered saline with 0.1% Triton X-100. Horseradish peroxidase-conjugated goat anti-mouse Ig (Amersham Biosciences) was used as secondary antibody at 1:1,500 dilution in 5% bovine serum albumin, phosphate buffered saline with 0.1% Triton X-100. The blots were visualized with the ECL kit (Amersham Biosciences). In cell line Western blots, 50 g of protein from various cell lysates were loaded. For immunoprecipitation, one-half of 35 S labeled in vitro translation lysates were incubated with 175 l of RIPA buffer (25 mM Tris, pH 7.8, 150 mM NaCl, 2 mM EDTA, 1% Nonidet P-40, 0.5% deoxycholate, 0.1% SDS, 100 M phenylmethylsulfonyl fluoride) containing 0.1 l of 16B12 at 4°C for 1 h. The antibody and lysate mix was incubated with a 50% protein A (Sigma)/RIPA slurry for 1 h at 4°C before being washed in 1 ml of RIPA three times. One-half of the final products were loaded on SDS-PAGE.

Yeast Two-hybrid Assay
We found that expression of Foxp1FLЈ fused to the GAL4 activation domain (pACTII-FLЈ) inhibited cell growth, perhaps because of disregulation of a yeast gene. For the purpose of yeast expression, we circumvented this problem by introducing into the forkhead DNA-binding domain two point mutations (V543L,R544E) that had been shown to eliminate DNA binding in other forkhead transcription factors (16,17). Yeast transformations were performed as described (18). Foxp1 pASI-CYH2 subclones were co-transformed with either pACTII-Foxp1FLЈ or pTD1-1 (pACTII-T-antigen from Clontech) into strain Y190. As a positive control, pVA3-1 (pASICYH2-P53) was co-transformed with pTD1-1. Protein-protein interactions were assayed by growth on selective media lacking histidine and confirmed by ␤-galactosidase activity assays using the 5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside (Xgal) filter assay (19).

Isolation of Foxp1-binding Sites
Foxp1 CAST experiments were carried out as described (20) with minor modifications. A random oligonucleotide pool (76 nucleotides) containing unique flanking sequences was rendered double-stranded by second strand synthesis. 2 l of C-terminally HA-tagged Foxp1A in vitro translation product and 0.5 l of anti-HA antibody 16B12 were used for binding and immunoprecipitation, respectively. PCR was carried out after phenol chloroform extraction and ethanol precipitation of the eluted oligonucleotides. One-fourth of the precipitant was used for subsequent PCR amplification. Four rounds of selection and amplification were carried out, and the final PCR product was digested with EcoRI and BamHI and subcloned into pBluescript II KS (ϩ) for sequencing.

Gel Mobility Shift Assays
For DNA binding assays, oligonucleotides A (CAAGGTAAACAAGA-CAACACAAATAA), B (AGGAATAAACACAACATAAATAATG), V1P (TCAAAAATATTGAAGTGTTATCACATACAC), SV40 (ATTTTTTT-TATTTATGCAGAGGCCGAGGGC), IL-2-wt (ATCGGAAAATTTGTT-TCATAATC), and IL-2-mut (ATCGGAAAAGAAGTTACATAATC) were annealed to their complementary strands and then gel-purified to eliminate nonhybridized oligonucleotides. Either in vitro translated protein, Escherichia coli expressed proteins or nuclear extracts were used as a source of Foxp1. Construction and expression of Foxp1 GST fusion proteins and electrophoretic mobility shift assays were previously described (8). Nuclear extracts of various cell lines were isolated by the method of Lee et al. (21). For competition experiments, competitors were allowed to bind with the GST fusion proteins for 5 min at room temperature before the labeled probe was added. The supershift assay was done by the method of Neufeld (22).

Dual Luciferase Assays in HEK 293T
Cells in a 6-well plate were co-transfected with 100 ng of reporter construct, 100 ng of pRL-TK (Promega) as a normalization control, and 800 ng of each eukaryotic expression vector using FuGENE 6 reagent (Roche Molecular Biochemicals). Dual luciferase assays were performed 48 h after transfection using a dual luciferase assay kit (Promega) according to the manufacturer's protocols.

Induction and Luciferase Assays in Jurkat
Cells grown in RPMI complete medium plus HEPES, pH 7.2, were harvested, washed with medium, and resuspended at 1.5 ϫ 10 7 /0.3 ml into electroporation cuvettes. IL-2-pGL3 (15 g), pRL-TK (5 g), and each expression vector DNA (5-20 g) or DNA of its empty vector (to normalize each sample to 40 g) were added. Following 10 min at room temperature, the DNAs were electroporated (Bio-Rad) at 240 V, 950 microfarads. The samples were left at room temperature for 10 min before being resuspended in 10 ml of Jurkat medium (RPMI with complete medium) and transferred to T25 flasks. 24 h after electroporation, the cells were treated with either phorbal ester phorbol 12,13dibutyrate (50 nM) and ionomycin (1 M) (induced) or Me 2 SO (uninduced). Following an 8-h incubation, whole cell lysates were prepared for dual luciferase assays and Western analysis.

Foxp1 Has at Least Four Splice Isoforms That Display Cell/
Tissue-specific Expression Patterns-Foxp1A, originally named glutamine-rich factor-1, was first isolated in a search for polyglutamine-containing transcription factors in B cell cDNA libraries (8). The cDNA isolated is identical to the reported Foxp1 isoform A (9) except that our cDNA encodes two additional glutamines at positions 172 and 203. The Foxp1 gene, assembled from our genomic clones of the locus and the public data bases, has 16 coding exons (Fig. 1A, top panel). Based on the Foxp1 genomic sequence and several unique Foxp1 expressed sequence tag clones (GenBank TM accession numbers BI649134, BG865862, BB858702, and BI697624), we have identified a novel Foxp1 isoform, Foxp1D (Fig. 1A, bottom panel). The Foxp1D cDNA contains a unique 5Ј-untranslated exon (exon 2b) and two in-frame stop codons that precede its putative initiating methionine in exon 3 (methionine 131 of Foxp1A). Foxp1D mRNA encodes a 576-amino acid protein that lacks the polyglutamine domain (Fig. 1B).
Previously, two additional isoforms, Foxp1B and Foxp1C, were isolated from mouse lung tissues (9) (Fig. 1B). Although their unique sequences suggest that different Foxp1 isoforms may possess distinct transcription regulatory properties, essentially nothing is known about how each isoform differs in their expression pattern and functional properties.
First we attempted to characterize tissue-specific Foxp1 isoform expression in mouse by Northern blotting with a probe (Fig. 1B, 3Ј probe) common to all Foxp cDNAs (Fig. 1C). Foxp1 messages of 3.2, 2.4, and 1.6 kb were detected. In brain and  (9). Foxp1 probes for RNA accumulation measurements are indicated beneath Foxp1A as lines. The 5Ј probe includes the 5Ј-untranslated region and spans amino acids 1-62. The 3Ј probe protects sequences encoding amino acids 581-628. C, Northern blot analysis of mRNA from the indicated tissues as detected by the Foxp1 3Ј probe indicated in B. D, RPA of Foxp1 expression in indicated adult mouse tissues. Total RNA was hybridized to the antisense strand of the 3Ј RPA probe. The loads were normalized to a housekeeping gene, mGAPDH (bottom panel). E, differential Foxp1 isoform mRNA expression in mouse adult tissues and B and T lymphocyte lines. Top panel, Foxp1A, Foxp1C, and Foxp1D mRNA hybridize to and protect 145 nucleotides of the 3Ј probe, whereas Foxp1B mRNA hybridizes to and protects 75 nucleotides of the same probe. Middle panels, Foxp1A and Foxp1B mRNA hybridize to and protect 203 nucleotides of the 5Ј probe, whereas neither Foxp1C nor Foxp1D mRNA hybridize to the 5Ј probe. Bottom panel, normalization of the amount of total RNA using a mouse housekeeping gene mGAPDH. thymus, the 3.2-and 2.4-kb transcripts are of similar abundance. In large intestine and testis, the 2.4-kb message is more abundant. The 1.6-kb message appears to be expressed in many tissues at low levels. Although we could not assign each messenger to a specific isoform, the blots demonstrate that tissues contain different relative levels of Foxp1 mRNA species. Northern analyses of a broad range of human tissues also indicated tissue-specific isoform expression for FOXP1 (data not shown). It is of interest that in humans, FOXP1 mRNA accumulated to highest levels in peripheral blood lymphocytes and in the caudate nucleus of the brain.
To distinguish different Foxp1 isoform mRNAs, we designed RNase protection assays using two unique probes: 5Ј probe and 3Ј probe (Fig. 1B). Our approach was most effective in distinguishing Foxp1B (protected band size, 72 nucleotides) from other Foxp1 isoforms (145 nucleotides), because the sequence of the 3Ј probe (Fig. 1B) is partially spliced out in Foxp1B. As shown in Fig. 1 (D and E), Foxp1B is weakly expressed in most tissues surveyed, with the highest expression observed in spleen, WEHI 231, and 70Z B cell lines. Based on the known expression level of Foxp1B, we can deduce specific expression of Foxp1A using the 5Ј probe. This probe overlaps a region that is shared by Foxp1A and Foxp1B but is missing in Foxp1C and 1D (Fig. 1B). Foxp1A expression can be estimated based on the signal differences of Foxp1Aϩ1B (Fig. 1E, middle panel) and Foxp1B (Fig. 1E, top panel). High levels of Foxp1A were detected in 70Z, BCL1, EL4, C2C12, fibroblasts, brain, and spleen. Finally, although individual expression levels of Foxp1C and Foxp1D cannot be distinguished by our assay, these two isoforms add up to high abundance in WEHI 231 and testis (Fig. 1E).
To study Foxp1 protein expression, we generated a monoclonal antibody (1G1) against the Foxp1 forkhead domain. The ability of 1G1 to detect Foxp1 proteins on Western blots was confirmed using a GST-Foxp1 forkhead fusion protein (data not shown) and in vitro translated Foxp1A ( Fig. 2A). Both 1G1 and the anti-HA mAb 16B12 recognized the full-length in vitro translated Foxp1A (ϳ95 kDa) with HA epitope-tagged either on its N-(5ЈHA) or C-terminal (3ЈHA) (ϳ70 kDa). Consistently, in vitro translation of Foxp1A produced a shorter translation product that is in frame with the full-length product (recognizable by 1G1). Because anti-HA mAb recognizes the truncated protein produced by 3ЈHA but not 5ЈHA, the truncation must occur at the N terminus. This property was later used in demonstrating that Foxp1 self-associates in vitro ( Fig. 3B and discussed below in "Results"). 1G1 showed no reactivity on Western blots toward in vitro translated Foxp2 or Foxp3/JM2 (data not shown), indicating that this reagent could be deemed subfamily-specific.
In various B and T cell lines (Fig. 2B) and in tissues (Fig. 2C), 1G1 detected at varying intensities three predominant proteins whose sizes (ϳ95, 70, and 50 kDa) were indistinguishable from those of in vitro translated Foxp1A, D, and C, respectively (Fig.  3E). Foxp1A is widely expressed, with highest levels in lung and B cell line 70Z and with moderate levels in brain, spleen, and other lymphoid lines (J558, WEHI 231, 33BTE, and 70 BET). Foxp1D expression is also tissue-and cell type-specific, with the highest levels in WEHI 231, 70Z, and lung. Moderate Foxp1D levels accumulate in brain, spleen, and testis and lymphoid lines M12.4, 33BTE, and 70 BET. Foxp1C protein is widely expressed in tissues, but its expression is somewhat limited in lymphocytes. Foxp1B was not observed in our Western blotting (with the possible exception of spleen), either because its expression level is low (consistent with our RPA results) or because 1G1 does not recognize its truncated forkhead domain with equivalent sensitivity. Foxp1 Undergoes Self-association via a Zinc Finger and Leucine Zipper-containing Domain-Forkhead transcription factors were not known to self-associate. However, the ability of FOXP3 to form homodimers was anticipated because it contains a leucine zipper domain (5). The importance of the leucine zipper domain was highlighted in some of the X-linked autoimmunity-allergic disregulation syndrome patients, who had developed this lethal disease simply because their FOXP3 alleles contained a single amino acid deletion within the leucine zipper domain.
The leucine zipper domain is shared by all three members of FIG. 2. Cell line-and tissue-specific expression of Foxp1 isoform proteins. A, monoclonal antibody 1G1 specifically recognizes in vitro translated Foxp1A. Foxp1A produces a full-length protein and a N-terminal truncated protein in reticulocyte lysates. The 5ЈHA and 3ЈHA constructs encode full-length Foxp1A proteins with either Nterminal or C-terminal HA epitope tags, respectively. In vitro translations of 5ЈHA and 3ЈHA templates were fractionated on SDS-PAGE and then Western blotted with an HA monoclonal antibody 16B12 or a Foxp1 monoclonal antibody 1G1. B, detection of Foxp1 protein isoforms using anti-Foxp1 mAb 1G1 on a Western blot of whole cell lysate prepared from the indicated B and T cell lines. C, 1G1/Western blot detection of Foxp1 protein isoforms in whole cell lysates prepared from adult mouse tissues.
the Foxp family (Fig. 3A). To test whether Foxp1 self-associates, we took advantage of the previous observation ( Fig. 2A) that in vitro translation of HA-tagged Foxp1A consistently produced both a full-length and an N-terminal truncated protein. For 3ЈHA in vitro translation products, anti-HA monoclonal antibody 16B12 recognized both the full-length protein (95 kDa) and the N-terminal truncation (70 kDa) on Western ( Fig. 2A) and efficiently immunoprecipitated both forms of proteins (Fig. 3B). For 5ЈHA in vitro translation products, the full-length protein (95 kDa) but not the N-terminal truncation (70 kDa) was recognized by 16B12 on Western (Fig. 2A). However, 16B12 immunoprecipitation of full-length 5ЈHA co-immunoprecipitated the nontagged truncated form (Fig. 3B), indicating that Foxp1A self-associates in vitro. A parallel 16B12 immunoprecipitation with a large excess of in vitro translated luciferase did not precipitate significant amounts of protein (Fig. 3B).
We next verified the self-association property of Foxp1 using the yeast two-hybrid system. Foxp1A lacking both its N-terminal polyglutamine and C-terminal acidic-rich domains (termed Foxp1FLЈ; Fig. 3C) was subcloned into GAL4-based activation and DNA-binding domain vectors. Co-transformation of yeast with pASICYH2-Foxp1FLЈ and pACTII-Foxp1FLЈ activated reporter expression, both for His complementation and ␤-galactosidase (Fig. 3D), indicating self-association of the protein.
Co-transformation of pASICYH2-Foxp1FLЈ and pTD1-1 (large T antigen fused to the GAL4 activation domain) did not activate reporter expression (Fig. 3D), indicating that the observed Foxp1 self-interaction was specific.
To map the self-association domain, the GAL4 DNA-binding domain was fused to various Foxp1A deletions (Fig. 3C) and co-transformed with pACTII-Foxp1FLЈ into yeast. The Foxp1 self-association domain corresponds to an evolutionarily conserved region (amino acids 338 -426; Fig. 3C, bar ab), including a C2H2 zinc finger and the leucine zipper motif. The leucine zipper motif alone (amino acids 362-426; Fig. 3C, bar b) is capable of mediating Foxp1 homodimerization (Fig. 3D). The C2H2 zinc finger (amino acids 338 -361; Fig. 3C, bar a) does not directly mediate Foxp1 homodimerization (Fig. 3D). However, weak interaction between the leucine zipper and T-antigen (Fig. 3D) suggests that the zinc finger contributes specificity to the Foxp1 self-interaction.
To test whether different Foxp family members interact physically via the conserved leucine zipper domain, a Foxp1A GST fusion protein of this region was used to pull down 35 Slabeled in vitro translated Foxp proteins. All of the Foxp proteins tested (Foxp1A, 1C, 1D, and JM2/FOXP3) but not the luciferase control were pulled down. We suggest that all Foxp proteins bearing the domain could interact physically with each other, although the extent to which this occurs in cells remains to be determined. These results suggest that homoand heterodimerization may provide an additional means to diversify Foxp1 regulation.
Selection of Foxp1 DNA Targets and Analysis of Binding-DNA binding of Foxp proteins had been previously performed on several promoters containing forkhead consensus sites (8,23). Because Foxp proteins have the most divergent forkhead domains within the family, we performed a CASTing experiment to determine the site-specific DNA binding preferences of Foxp1A. PCR primers with random 26-base core sequences were used for the first round of selection. In vitro translated C-terminally HA-tagged Foxp1A was used as a source of Foxp1A protein, and the 16B12 antibody was used to pull down the protein-oligonucleotide complex. The sequences of individual isolates cloned following four cycles of CASTing were determined and organized into groups (Fig. 4A). From 76 inde- pendent isolates, three unique sequences were isolated 25, 21, and 26 times, respectively, and the other four unique sequences were isolated once. The resulting Foxp1 consensus, TTATT-TRTRTTWKYKWYTWT, contains a 7-nucleotide (underlined) core, that is a subset of the 7-nucleotide core sequence shared among all forkhead binding sites: 5Ј-TRTTKRY (1). There are an additional 13 flanking bases preferred by Foxp1A. Because our CASTing only resulted in seven unique sequences from 76 independent isolates, all individual bases within the current consensus site might not be required for Foxp1 binding.
To confirm binding of Foxp1A and 1D to the isolates, electrophoretic mobility gel shift assays (EMSAs) of two of the most common sequences (Fig. 4A, lines A and B) were performed. Corresponding sense and antisense oligonucleotides were synthesized, annealed, and end-labeled. Duplexes were gel-purified for EMSA as previously described (8). In vitro translated Foxp1A and Foxp1D bound specifically to the two consensus oligonucleotides (Fig. 4B). Routinely we observed sharp yet nonspecific gel shifts (caused by binding of a protein(s) in the in vitro translation reticulocyte lysate), whereas the specific shifts corresponding to Foxp1A and 1D binding consistently gave multiple sizes that were difficult to resolve. This suggested that Foxp1A in vitro translation products included both full-length and N-terminally truncated products, both of which are capable of forming homo-and heterodimers. Foxp1 did not bind to an HNF3-specific binding site (V1P) contained in the immunoglobulin heavy chain variable gene (V H ) promoter V1P (8). The diffuse Foxp1A-DNA complexes, when bound by the 1G1 antibody, produced supershifts that were much sharper (Fig. 4D).
Several lines of evidence indicate that the interaction between the isolates and Foxp1 forkhead domain is specific. First, Foxp1A forkhead point mutants W539L and V543L,R544E did not bind the isolated oligonucleotides (Fig. 4, B and D), although similar amounts of wild type and mutant protein were used in the assay, as indicated by Western blot using anti-Foxp1A monoclonal antibody (Fig. 4C). Similar mutations within the forkhead domain of HNF proteins have been shown to abolish DNA binding (16,17). Second, double-stranded oligonucleotides A and B can compete for the binding of a GST fusion of Foxp1A to radiolabeled oligonucleotide A, but no com- petition was observed by V1P (Fig. 4E). Third, single-stranded oligonucleotides A and B were not bound by wild type Foxp1, nor did they compete with double-stranded annealed versions for binding (data not shown).
Foxp1 Isoforms Repress Transcription through Binding Sites within and Appended to the SV40 Early Promoter-To test whether Foxp1 can regulate transcription, we co-transfected Foxp1A or Foxp1D with a firefly luciferase reporter driven by the SV40 promoter, with or without the Foxp1-binding site B inserted upstream of the promoter (Fig. 5A; pGL3 promoter versus 1xb-pGL3 promoter). We employed a recipient cell line, human embryonic kidney cells (293T), which expressed no 1G1detectable levels of endogenous Foxp1A or Foxp1D (Fig. 5B). As shown in Fig. 5A (left panel), both Foxp1A and Foxp1D repressed transcription regardless of whether a Foxp1 consensus DNA-binding site was appended to the SV40 promoter. For Foxp1A, a 3-4-fold repression of transcription was observed with or without the addition of a synthetic binding site. However, for Foxp1D, repression was stronger than Foxp1A on the SV40 promoter alone (ϳ5-fold) and was significantly enhanced (ϳ25-fold) when a consensus Foxp1 site was placed upstream.
The same effects with essentially identical magnitudes of repression were observed in co-transfections utilizing a reporter with the same SV40 promoter but bearing a trimerized consensus Foxp1-binding site and an SV40 enhancer in a remote position (Fig. 5A, right panel). Foxp1A and Foxp1D were expressed about equally, as indicated by Western blots of transfected lysates (Fig. 5B, left panel). Thus, we conclude that the polyglutamine-devoid Foxp1D is a stronger repressor and that a synthetic Foxp1-binding site is not fully required to repress SV40 transcription (see below).
If the polyglutamine tract is inhibitory to repression, we would anticipate that Foxp1C, which eliminates this motif via alternative splicing, might also be a stronger repressor than Foxp1A. As summarized in Fig. 5C, each of the four reporters were repressed to nearly background levels in 293 cells in which the Foxp1C isoform was transiently overexpressed (data not shown). However, the polyglutamine inhibitory effect is not observed for the highly Foxp1-related (87% identical forkhead domain) polyglutamine-containing Foxp2 protein. Foxp2 repressed in our assay as strongly as Foxp1C (Fig. 5C) under conditions where both proteins were overexpressed equivalently to Foxp1A and D (data not shown). Because Foxp2 binds to the reporter consensus and SV40 sites with affinity similar to the Foxp1 proteins (data not shown), factors other than the polyglutamine track (e.g. posttranslational modifications; heteromerization) likely contribute.
The data of Fig. 5 also suggest that at least a component of Foxp1 repression may be mediated through a Foxp1-specific binding site within the SV40 promoter that had not been previously characterized. Experiments of Fig. 6 support this hypothesis. Scanning the sequence of the minimal SV40 promoter revealed a putative Foxp1 consensus TTATTTATG within the TATA-associated AT block (13) of the early region (Fig. 6A). Mutation of this site reduced luciferase activity of the resulting reporter about 30 -50-fold relative to wild type SV40 when transfected alone into 293 cells (data not shown). Co-transfection with full-length Foxp1A had no significant effect, whereas Foxp1D modestly activated the mutant promoter (Fig. 6B, left  panels). When a trimerized consensus Foxp1-binding site was appended to the mutant promoter, co-transfection with either form of Foxp1 repressed transcription 2-3-fold (Fig. 6B, right  panels). Given the magnitude of the repression observed by Foxp1D on 3Xb-pGL3 transcription (Fig. 5B), this relatively modest rescue suggests that the two sites might act synergistically in that configuration. No repression was observed for Foxp1A with a mutated forkhead domain (V543L,R544E) (Fig.  6B), even though the mutant protein expression was equivalent to that of wild type protein (Fig. 5B, right panel).
To test whether Foxp1 directly interacts with the putative forkhead site in the wild type SV40 promoter, EMSA analysis was performed, initially with in vitro translated proteins. Wild type Foxp1A interacts with the SV40 promoter, but no shift was observed when Foxp1A with point mutations in the forkhead domain (W539L) was employed (Fig. 6C). In whole cell lysates produced from J558 or M12.4 B cell lines, complexes formed with the Foxp1-binding site within the SV40 promoter were supershifted by the anti-Foxp1 monoclonal 1G1 (Fig. 6D), indicating that the SV40 promoter might be a natural target for Foxp1 proteins.
Foxp1 Isoforms Repress Transcription through Binding Sites within the IL-2 Promoter-We sought to extend the above results on synthetic and SV40 promoter repression to a more biologically relevant system. The importance of the Foxp family in immune regulation has been recently documented (24 -26). We noticed that the IL-2 promoter contains an uncharacterized sequence that can be aligned with our Foxp1 consensus (Fig.   FIG. 5. Foxp1 family members repress transcription through Foxp1-binding sites. A, Foxp1A and Foxp1D repress transcription through binding sites within and appended to the SV40 early region promoter. Transcription reporter constructs are pGL-3 promoter, a SV40 firefly luciferase reporter vector; 1xb-pGL3 promoter, pGL3 promoter with inserted Foxp1-binding site; pGL3-control, SV40 firefly luciferase reporter vector with SV40 enhancer sequences; and 3xb-pGL3-control, pGL3 control plasmid with Foxp1-binding sites. Cotransfections of 293 cells were performed with the empty mammalian expression vector pCEP4L or with pCEP4L expressing Foxp1A or Foxp1D. Firefly luciferase activities were measured and normalized to Renilla luciferase activities 48 h after transfection. All of the experiments were performed in triplicate. 7A). In EMSA experiments, a specific interaction (as judged by 1G1 supershift and data not shown) was observed between this IL-2 promoter Foxp1-binding site and in vitro translated Foxp1A (Fig. 7B, left panel). The Foxp1A forkhead mutant W539L, which does not bind to the consensus or SV40 Foxp1 sites (Figs. 4 and 6), bound very weakly to the IL-2 promoter (Fig. 7B, left panel). Nuclear extract prepared from Jurkat T cells, a line previously shown (27) to induce IL-2 expression following Ca 2ϩ influx and protein kinase C activation, contained Foxp1 binding activity for both the isolated Foxp1-binding site b and IL-2 sites (Fig. 7B, right panel). Mutation of several bases within the IL-2 site that are conserved within the Foxp1 consensus (IL-2-mut; Fig. 7A) eliminated both the in vitro interaction (Fig. 7B, middle panel) as well as Foxp1 binding activity in Jurkat (Fig. 7B, right panel).
Next we co-transfected Jurkat cells with various Foxp expression constructs along with a luciferase reporter carrying the distal IL-2 promoter region spanning the Foxp1 site inserted into the IL-2 minimal promoter (15). After 24 h, the cells were transferred into fresh medium and cultured for an additional 8 h in medium containing either Me 2 SO alone or ionomycin plus phorbal ester in Me 2 SO. Western blotting with mAb 1G1 confirmed that each of the Foxp1 constructs was expressed at levels estimated to be 3-5-fold above endogenous Foxp1 levels (data not shown). Consistent with previous reports (27), IL-2 promoter activity in the unstimulated cells was virtually negligible (data not shown). Upon stimulation, the robust luciferase activity induced by ionomycin and phorbal ester was repressed 3-5-fold in a dose-dependent manner by co-transfection of each of the Foxp1 constructs. Unlike the situation in 293T cells (Fig. 5), Foxp2 did not repress significantly in Jurkat cells, nor did we see significant differences in repression attributable to polyglutamine (e.g. Foxp1A versus Foxp1D or 1C). Lack of repression by the Foxp1A forkhead mutant W539L indicated that the activity was sequence specific. This was confirmed by the experiment of Fig. 7D, in which mutations of the IL-2 promoter that eliminated Foxp1 binding in EMSAs (Fig. 7B) also eliminated Foxp1-mediated repression of IL-2 promoter.
We conclude that Foxp1A, and likely the other forkheadcontaining Foxp1 members, bind specifically to a heretofore uncharacterized Foxp1 forkhead site within the IL-2 promoter. The data suggest that our isolated Foxp1 preferred binding sites of Fig. 4A can interact and mediate transcription repression by Foxp1 in a native promoter. These results further indicate that Foxp1 isoforms and Foxp2 may display different promoter-specific repression activities in the context of T cell activation.

DISCUSSION
In contrast with other forkhead subfamilies, domain homology within Foxp proteins extend beyond their forkhead DNAbinding domain. Distinguishing features of Foxp are the C2H2 zinc finger and leucine zipper motifs as well as a DNA-binding region truncated at the "wing" portions of the forkhead/winged helix domain. Foxp1A, Foxp2, and human FOXP1 also share a polyglutamine stretch at their N termini. Our functional studies demonstrated that each of these domains has an important role in defining the transcriptional regulatory properties of Foxp1.
To fully understand Foxp1 function and the mechanisms that underlie it, identification of physiological target genes would be required. Although previous studies (8,23) demonstrated interaction between Foxp and target sites identified for other forkhead proteins, isolation of the preferred DNA-binding sequence of Foxp1 constrains the search for target genes. In EMSA experiments, Foxp1 favors the isolated sites strongly over V1P (a prototypic HNF-3-binding site), indicating that the sequences we obtained are indeed highly preferred binding sites. Our casting results also support the proposed role for Foxp1 in regulation of mouse lung-specific CC10 promoter (9), because a Foxp1-binding site is contained within the 165-bp cis-regulatory region upstream of this promoter (TTATTT-GCTT). Despite the fact that Foxp forkhead domains are the most divergent among all family members, our isolated Foxp1binding sites, nonetheless, contain the conserved HNF core nucleotides. Thus, from a structure-function point of view, it appears that helices 1-3 of the Foxp1 forkhead domain are both necessary and sufficient for the direct base contacts within the core sequence (10 -12).
Ectopic overexpression of Foxp1 represses transcription of the SV40 promoter, an SV40 promoter bearing a preferred Foxp1-binding site, and the IL-2 promoter. Foxp1 repression of the SV40 promoter is mediated by binding to a region spanning its TATA box, which contains 9 bases of the Foxp1 consensus. When 3 base pairs of this sequence were mutated, transcription repression by Foxp1 was abolished. In contrast, previous studies showed that whereas deletion of this AT block interfered with the precision of the transcriptional initiation start site, the level of basal transcription was not affected (28,29). Several possibilities might explain the apparent discrepancy. First   FIG. 6. Foxp1 interacts with the SV40 promoter. A, sequence of SV40 early promoter. The six GC boxes are labeled, the putative Foxp1binding site is underlined, and the three nucleotides changed in the mutant promoter are indicated above the wild type sequence. B, no repression is observed for Foxp1A or Foxp1D on the mutant SV40 promoter. Repression is restored only for wild type Foxp1 when its binding site (3Xb) is inserted upstream of the mutant promoter. C, in vitro translated wild type Foxp1A but not its forkhead domain mutant W539L binds to the AT block of the SV40 promoter. Anti-Foxp1 mAb 1G1 supershifts the Foxp1A/AT block complex. D, 1G1 supershifts the SV40 promoter AT block complexed with endogenous Foxp1 protein from B lymphocyte cell lines J558 and M12.4. Preimmune control serum was employed at the highest concentration of 1G1. and foremost, the Benoist and Chambon (28) results were obtained in vitro. Second, because the AT block lies within the replication origin for SV40, our point mutation, in addition to eliminating a Foxp1-binding site, might have decreased replication of the reporter plasmids in 293T cells. Third, by mutating the AT block, we might have introduced a new binding site for an unknown factor that strongly represses reporter transcription. However, our detection of a Foxp1-specific protein complex with the SV40 promoter sequence within nuclear extracts of J558 and M12.4 B lymphocyte lines supports our contention that SV40 is a bona fide target of Foxp1.
We have also identified IL-2 as a putative physiological target for Foxp1. Several lines of evidence support this possibility, including direct and specific Foxp1-IL-2 promoter interaction, detection of Foxp1 mRNA and protein expression in T lymphocytes, and detection of a Foxp1-specific protein complex with the IL-2 promoter sequence in nuclear extracts of Jurkat T cells. Our data are consistent with previous observations that Foxp3 can also repress the IL-2 promoter (23) and that Xlinked autoimmunity-allergic disregulation syndrome patients produce high levels of IL-2 in their peripheral blood (5).
What is the mechanism for Foxp-mediated transcriptional repression? Previously it was shown that regions within Foxp1 and Foxp2 that span the zinc finger and leucine zipper domains repressed transcription when fused to the GAL4 DNA-binding domain (9). Here, we confirmed experimentally that the leucine zipper, in combination with the zinc finger domain, promotes self-association among different isoforms and subfamily members of Foxp proteins. Because both forkhead truncation and a single amino acid deletion within the leucine zipper domain cause loss of function for FOXP3 (5), Foxp-mediated repression may require both target DNA binding and dimerization. Whether specific dimerization partners are used to "fine tune" Foxp protein in their selectivity of target sites or whether dimerization is required for Foxp to recruit general transcriptional repressor(s) remains to be understood.
Although Foxp1A and Foxp1D repress the IL-2 promoter at similar levels in Jurkat cells, Foxp1D is significantly stronger than Foxp1A in 293T cells. Both isoforms are localized to the nucleus (data not shown), and their in vitro translated products appear to bind Foxp1 consensus DNA sites with comparable affinities. Differential interaction with common cellular proteins or with different cell type-restricted proteins could explain these differences. Even stronger repression in 293 cells was engendered by the Foxp1C isoform. Because the polyglutamine tract is missing from Foxp1C and 1D, it is an obvious candidate for a modulator of Foxp1A repression activity. In support of this idea, polyglutamine tracts are known to mediate FIG. 7. Foxp1 family members repress transcription in Jurkat T lymphocytes through a consensus Foxp1-binding site within the IL-2 promoter. A, sequence alignment of consensus, wild type (wt) IL-2, and mutated (mut) IL-2 Foxp1-binding sites. B, EMSA of Foxp1 binding. Left and center panels, in vitro translated wild type Foxp1A and Foxp1A W539L mutant with corresponding supershifts at increasing concentrations of 1G1 antibody (1:20 and 1:5, indicated by triangles) using wild type or mutant IL-2 Foxp1 duplex oligonucleotides of A as probes. Right panel, specific binding activity within Jurkat nuclear extract specific for consensus site B or IL-2 Foxp1 sites. C, Foxp1 isoforms, but not Foxp2, repress expression of induced IL-2-luciferase activity. Jurkat T lymphocytes were transfected with IL-2-luciferase, Renilla luciferase, and varying amounts (white, gray, and black boxes, 20, 10, and 5 g, respectively) of the following expression vector DNAs: empty pCEP4L, Foxp2, Foxp1A, Foxp1C, Foxp1D, and Foxp1A mutant W539L. Firefly luciferase activities were normalized to Renilla luciferase activities 24 h after transfection and 8 h after induction with phorbal ester, ionomycin, and Ca 2ϩ as detailed under "Materials and Methods." Shown are the average ratio values from two independent experiments. D, Foxp proteins do not repress transcription of the IL-2 promoter mutated within the Foxp1-binding site. Expression vector DNAs (white or gray boxes, 7 or 20 g, respectively) were co-transfected with IL-2_mut luciferase and Renilla luciferase reporters following induction. The ratios were normalized to the value given when 20 g of pCEP4L was co-transfected with wild type IL-2-luciferase (hatched box). protein interactions, either by forming multimers (30,31) or by interacting with polar amino acid-rich sequences (32). Selective expression of the Foxp1 isoform and/or its interaction partner(s) could regulate Foxp1 transcription activity as needed.
The reason(s) for the differential transcriptional activity of Foxp1 and Foxp2 is not structurally obvious. Foxp2 and Foxp1A are highly similar throughout and have conserved all functional domains, including the polyglutamine. Yet Foxp2 repressed at levels equivalent to those of Foxp1C in 293 cells but failed to induce significant repression in Jurkat cells. As with the case of Foxp1A and D, Foxp2 binds strongly in vitro to both the preferred Foxp1 casting sites and to the IL-2 site (data not shown). These differences underscore the functional complexities afforded by these proteins even when analyzed under controlled expression conditions in cultured cell lines.
Foxp1 expression shows tissue preference in human adults, accumulating to highest levels in peripheral blood lymphocytes and the caudate nucleus of the brain. 2 However, it is broadly expressed at moderate to low levels in many other adult tissues. The broad distribution of Foxp1 suggests it may have functional roles in diverse cell and tissue types. The combination of tissue-specific alternative splicing and subunit-subunit interaction may be important to tailor Foxp1 properties to the needs of individual cell types. During embryonic development, co-expression of Foxp1 and Foxp2 is found in lung epithelia and the developing nervous system (9). Foxp1 expression is detected in every lymphocyte cell line that we have analyzed as well as in spleen and thymus. Several of these lymphoid cells co-express Foxp3 (5). It is possible that one Foxp protein might affect the activity of the other by competing for target gene promoter binding or by acting either as a partner or as a negative regulator via direct protein interaction. Although redundancy among conserved family members often exists for transcription factors, Foxp1-targeted disruption in mice leads to an embryonic lethal phenotype. 2 Therefore, the potential redundancy apparent in the Foxp1 subfamily is insufficient to rescue Foxp1 loss, and at least in one or more of these tissues, Foxp1 has a critical and unique role.