The LIM protein FHL3 binds basic Krüppel-like factor/Krüppel-like factor 3 and its co-repressor C-terminal-binding protein 2.

The ability of DNA-binding transcription factors to recruit specific cofactors is central to the mechanism by which they regulate gene expression. BKLF/KLF3, a member of the Krüppel-like factor family of zinc finger proteins, is a potent transcriptional repressor that recruits a CtBP co-repressor. We show here that BKLF also recruits the four and a half LIM domain protein FHL3. Different but closely linked regions of BKLF mediate contact with CtBP2 and FHL3. We present evidence that CtBP2 also interacts with FHL3 and demonstrate that the three proteins co-elute in gel filtration experiments. CtBP and FHL proteins have been implicated in both nuclear and cytoplasmic functions, but expression of BKLF promotes the nuclear accumulation of both FHL3 and CtBP2. FHL proteins have been shown to act predominantly as co-activators of transcription. However, we find FHL3 can repress transcription. We suggest that LIM proteins like FHL3 are important in assembling specific repression or activation complexes, depending on conditions such as cofactor availability and promoter context.

The Sp/Krü ppel-like factor (KLF) 1 family of mammalian DNA-binding proteins consists of the Sp1-related proteins (Sp1-6) and a subfamily termed Krü ppel-like Factors (KLF1-17), consisting of erythroid Krü ppel-like factor (EKLF/KLF1), lung Krü ppel-like factor/KLF2, basic Krü ppel-like factor/ KLF3, and others (1)(2)(3)(4). The Sp/KLF proteins play diverse roles in regulating gene expression during development. For example, EKLF/KLF1 is important for ␤-globin gene expression (5,6), and LKLF/KLF2 plays critical roles in lung development, T cell maturation, and in endothelial cells (7)(8)(9). Sp/ KLF proteins contain a characteristic DNA binding domain (DBD) at or near their C terminus that consists of three Krü ppel-type Cys 2 His 2 zinc fingers that bind GC and CACCC boxes in regulatory elements of genes. Different members of the family exhibit similar DNA-binding specificity, but in general the Sp1-like subgroup has a higher affinity for GC boxes and the KLF subgroup proteins bind more strongly to CACCC sequences.
Once bound to DNA the proteins are thought to regulate transcription by recruiting co-regulatory molecules. The Nterminal domains of different Sp/KLF family members exhibit little or no homology, and differential cofactor recruitment by these divergent domains may in part explain why some members behave as activators, whereas others act as repressors.
The situation is complex as other regions of the proteins, such as the zinc finger domain also mediate protein-protein interactions, and individual proteins may recruit both co-activators and co-repressors. For instance, EKLF/KLF1 has been shown to activate transcription by recruiting the histone acetylase proteins p300/CBP and P/CAF through its N terminus and the chromatin remodeling complex E⅐RC1 through its zinc finger domain (10). The zinc finger domain can also recruit a histone deacetylase complex to silence gene expression (11).
The CtBP co-repressors bind numerous other regulatory proteins, including many conventional DNA-binding proteins, and accessory molecules, such as polycomb and the viral protein E1A (13). The mode of action of CtBP is yet to be fully elucidated. CtBP proteins can recruit histone deacetylases (16,17), however, deacetylase-independent repression has also been observed. CtBP interacts with polycomb group proteins (18) and proteins such as Ikaros (19) that are contained in chromatin remodeling complexes, suggesting that CtBP proteins may also participate in regulation of gene expression through the noncovalent modification of chromatin structure.
We have found that BKLF/KLF3 is a potent transcriptional repressor, and during our analysis of transcriptional repression by BKLF (14), we noted that abrogation of CtBP recruitment does not entirely abolish this function. This suggested that BKLF may recruit one or more additional cofactors to regulate transcription. A two-hybrid screen against BKLF identified the LIM-only protein FHL3 as another BKLF partner protein.
FHL3 is a member of the recently recognized four and half LIM domain (FHL) family, which consists of FHL1-4, ACT, and KyoT1 (20 -26). These proteins are made up of four LIM domains, plus one N-terminal "half " LIM domain whose amino acid sequence resembles that of a single GATA type zinc finger. LIM domains are composed of a double zinc finger motif that co-ordinate two zinc ions and are primarily thought to mediate protein-protein interactions (27). The term LIM originates from the isolation of three Caenorhabditis elegans transcription factors, LIN-11, Isl-1, and Mec-3, proteins in which this domain was first described (28). FHL1 and 2 are highly expressed in skeletal muscle and have been observed to be present in the cytoplasm. Consequently FHL proteins were originally hypothesized to play a role in cytoskeletal function in muscle cells (24,25). Recently, however, FHL family members have also been detected in the nucleus, and it now appears that in this compartment several FHL family members function as transcriptional co-regulators. The testis-specific protein ACT serves as a co-activator of CREM and CREB (21), and its restricted expression pattern helps explain how testis-specific gene activation by CREM is mediated. FHL2 has been shown to act as a specific co-activator of the androgen receptor (AR) (23) and the zinc finger protein Wilms tumor-1 (WT1) (29) and as a co-repressor of the multi-zinc finger protein PLZF (30). Sub-cellular partitioning also appears to be important for the regulation of FHL function. FHL2 has been observed in both nuclear and cytoplasmic fractions, and its translocation to the nucleus in response to Rho signaling can lead to potent stimulation of ARmediated gene expression (31). Additionally, ACT interacts with the kinesin KIF17b, and this interaction is thought to regulate its nuclear/cytoplasmic partitioning (32). To date, the FHL proteins described have been found to behave as conventional transcriptional co-regulators, raising the possibility that they may be involved in recruiting the basal transcriptional machinery or chromatin remodeling enzymes, but their precise mechanism of action has not yet been defined.
Here we show that FHL3 binds BKLF and acts as a corepressor. We also demonstrate that FHL3 contacts CtBP2, a separate BKLF co-repressor. Because sub-cellular distribution of FHL factors may affect their function, we were interested in examining the cellular location of FHL3. We present evidence that FHL3, CtBP2, and BKLF proteins exist in a large complex only in the nucleus and that FHL3 is significantly enriched in the nucleus only when it is co-expressed with both BKLF and CtBP2. We suggest that FHL3, and other members of the FHL family, operate as linking modules that stabilize multiprotein transcriptional complexes in the nucleus in addition to their reported roles in the cytoplasm.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid Screening and Assays-The Clontech two-hybrid system was used according to the manufacturer's instructions. A human erythroleukemia (K562) cell cDNA library in the gal4 activation domain (AD) fusion vector, pGAD10, was transfected into the yeast strain HF7c harboring the gal4DBD-BKLF-(1-268) fusion protein expressed from pGBT9. Candidate interactors were re-tested by co-transfecting gal4DBD-bait proteins and gal4AD-prey proteins into HF7c. Transformants were selected on Trp/Leu-deficient plates and patched onto Trp/Leu/His-deficient media. Growth was scored after 30 h. Controls containing bait and prey alone were also conducted and were negative for growth up to more than 72 h.
In Vitro Binding Assays-Full-length BKLF and FHL3 were cloned into the vector pcDNA3 (Invitrogen), and 35 S-labeled in vitro translated proteins were generated using T7 polymerase and the TNT system from Promega. GST fusion proteins were prepared and binding assays were carried out as previously described (33).
Northern Blotting-A 32 P-labeled 300-bp fragment from the coding sequence of FHL3 was used to probe two commercial filters, a multiple tissue Northern blot containing mRNA from human adult tissues, and a blot containing mRNA from various cancer cell lines, both from Clontech. Hybridization was performed at high stringency, using Ex-pressHyb (Clontech). These membranes were stripped and re-probed using a human ␤-actin probe supplied by Clontech.
Mammalian Cell Transfections and Reporter Assays-NIH-3T3 cells were cultured and transfected, using the calcium phosphate method, as described previously, as were chloramphenicol acetyltransferase and growth hormone reporter assays (14). COS cells were cultured and transfected, using the DEAE-dextran method, as previously described (15). K562 stable cell lines expressing BKLF were generated as described (34), using the expression vector pEF1␣.neo.BKLF.
Nuclear and Cytoplasmic Extract Preparation-Nuclear extracts were prepared as previously described (35). Cytoplasmic extracts were prepared during the procedure for preparation of nuclear extracts as above.
Antibody-Bead Cross-linking and Co-immunoprecipitation-To generate antibody-bead conjugates, 100 l of pre-immune or immune sera was incubated with 100 l of protein-A-agarose beads (Roche Molecular Biochemicals) to a final volume of 1 ml with PBS, for 60 min at 4°C. Beads were then washed twice with 10 volumes of 200 mM sodium borate, pH 9.0, and were resuspended in 10 volumes of 200 mM sodium borate, pH 9.0, with dimethylpimilimidate (Sigma) at a final concentration of 20 mM. Cross-linking was performed for 30 min at room temperature, and the reaction was stopped by washing once with 10 volumes of 200 mM ethanolamine, pH 8.0, and then incubation with 200 mM ethanolamine, pH 8.0, for 2 h at room temperature. Beads were then washed twice with 10 volumes of 100 mM glycine, pH 2.8, to remove non-cross-linked antibody, washed once with 10 volumes PBS, and resuspended in PBS.
For immunoprecipitation of nuclear extracts, 80 -100 l of extract from transfected COS cells were diluted 1:3 with Nonidet P-40 buffer (0.5% Nonidet P-40/Igepal, 150 mM NaCl, 50 mM Tris, pH 8.0, 1 g/l leupeptin, 1 g/l aprotinin, and 1 mM phenylmethylsulfonyl fluoride). The solution was pre-cleared with 10 l each of protein-A and protein-G beads at 4°C for 30 min. The supernatant was then immunoprecipitated with 10 l of anti-BKLF or anti-FHL3 beads, or the corresponding pre-immune beads, with incubation at 4°C for 1 h. The beads were washed four times with Nonidet P-40 buffer and were resuspended in SDS-PAGE loading buffer with 100 mM dithiothreitol. The samples were treated at 60°C for 5 min, separated by SDS-PAGE, and transferred to nitrocellulose overnight at 100 mA. Blots were then probed by standard Western blotting techniques with the appropriate primary antibody then bound with the appropriate secondary antibody linked to horseradish peroxidase and were detected using the PerkinElmer Life Sciences Chemiluminescence Reagent Plus kit. For immunoprecipitation of cytoplasmic extracts, a higher salt buffer was used during the immunoprecipitation step to maintain the salt concentration equivalent to that used for nuclear extracts. Subsequent washes were performed with the 150 mM NaCl Nonidet P-40 buffer.
Anti-FHL3 Antibody Generation-A fusion protein between GST and full-length FHL3 was prepared as described (14). The GST fusion protein was then eluted from the GSH-agarose beads using 25 mM glutathione (GSH) in elution buffer (100 mM Tris-HCl (pH 7.5), 120 mM NaCl) for 1 h at 4°C with rotation. The GSH-agarose beads were pelleted and re-extracted. The eluates were pooled, and the protein concentration was determined. The protein was lyophilized, and four 300-g doses were used as antigens for rabbit inoculation. Inoculation and sera collection were performed by the Veterinary Services Division of the Institute of Medical and Veterinary Sciences (101 Blacks Rd., Gilles Plains, S.A., 5086, Australia).

RESULTS
BKLF Interacts with the LIM Protein FHL3-We have previously shown that BKLF can recruit the co-repressor protein CtBP2 to repress transcription (14). A short amino acid motif in BKLF, Pro-Val-Asp-Leu-Thr (PVDLT), is required for the interaction between BKLF and CtBP2, and when this motif is disrupted, repression by BKLF is compromised. However, some repression by BKLF is retained even when the PVDLT motif is mutated (14). This suggested that BKLF might recruit additional cofactors to repress transcription. We thus carried out a further two-hybrid screen to identify additional BKLF cofactors.
The bait used in this screen consisted of the gal4DBD fused to amino acids 1-268 of BKLF (encompassing the repression domain of BKLF (14)). We screened a human erythroleukemia cell (K562) cDNA library. K562 cells express embryonic and fetal globins and are frequently used as a model of erythroid cells at the fetal stage of development (36,37). This library was used as BKLF is highly expressed in erythroid cells (35) and is believed to play a role in hematopoiesis (38,39). We screened 4 ϫ 10 5 clones: 132 primary His ϩ colonies were isolated, 10 of which were also positive for the lacZ reporter. Sequencing revealed two clones that carried an 899-bp cDNA insert that matched the human gene termed four and a half LIM protein 3 (FHL3) (25). As a first test for the specificity of the interaction, these isolates were tested against the gal4DBD alone and negative control baits, including gal4DBDs fused to p53, the N finger of GATA-1, and lamin C protein. No interactions were observed (data not shown). We also swapped FHL3 into the bait plasmid and tested the ability of a gal4DBD-FHL3 bait to interact with a gal4AD-BKLF prey. As shown in Fig. 1A the interaction is again observed, whereas there is no interaction with the negative control prey, gal4AD alone.
The clones we isolated lacked half of the C-terminal LIM domain of FHL3. To isolate the full-length clone, primers were designed to the 5Ј-and 3Ј-ends of the coding sequence (Gen-Bank TM accession number HSU60116), and the K562 cDNA library was used as a template in a polymerase chain reaction. The resulting product was cloned, and sequencing confirmed the presence of a full-length cDNA encoding human FHL3.
To verify this interaction in another system, a GST-FHL3 fusion protein was produced by expression in Escherichia coli. The protein was immobilized on GSH-agarose beads ( Thus, BKLF can interact with FHL3 in both a yeast two-hybrid assay, as either bait or prey, and in GST pull-down assays. FHL3 mRNA Expression in Cell Lines and Adult Tissue-Because FHL3 was first identified as a gene that was highly expressed in muscle cells, we wished to test whether it was expressed at appreciable levels in K562 and other selected cells. We probed a Northern blot containing RNA from various cancer cell lines, including K562 cells, with a probe generated from a portion of the coding sequence of FHL3 ( Fig. 2A). FHL3 message was detected in the K562 sample, and low level expression was also found in HeLa (fibroblast), SW480 (colorectal The input (lane 1) contains 20% of the radiolabeled BKLF used in the assay. Equivalent amounts of GST proteins, shown in the lower panel by Coomassie Blue staining, were used in each case. Lane 1 in the lower panel contains molecular weight markers (MW). C, mapping of the domain in BKLF that binds FHL3. Yeast were co-transformed with the gal4AD-FHL3 plasmid and the gal4DBD fusions with the BKLF deletion constructs shown. The black box represents the PVDLT motif in BKLF, and the 1-268mut construct carries a mutation in this motif that abolishes CtBP binding. Interactions, as measured by growth of yeast carrying the plasmids on SD-H-L-T medium, are shown in the right column.

FIG. 2. FHL3 mRNA is expressed in K562 (erythroleukemia) cells and is highly expressed in adult heart tissue. A, FHL3
message is expressed in K562 cells and other cancer cell lines. A Northern blot carrying mRNA from various cancer cell lines (Clontech) was probed with a radiolabeled probe from the coding sequence of FHL3. B, FHL3 message is expressed in adult skeletal muscle. A Northern blot carrying mRNA from various adult human tissues (Clontech) was probed with a radiolabeled probe from the coding sequence of FHL3. For A and B, both membranes were stripped and re-probed with a human ␤-actin probe (Clontech) (lower panels) to test for equal loading of mRNA on the membrane. adenocarcinoma), A599 (lung carcinoma), and MOLT-4 (melanoma) RNA. We also probed a Northern blot of mRNA from various adult tissues and observed high level expression in skeletal muscle as well as lower levels in heart (Fig. 2B). Thus, FHL3 mRNA is present not only in muscle tissue, as previously observed, but in erythroid cells of fetal type and other cancer cell lines. This result is consistent with other reports that FHL proteins function in several different cell types in addition to muscle cells.
Mapping the Interacting Domains of BKLF and FHL3-BKLF interacts with CtBP2 through a PVDLT motif centered around amino acid 72 (14). To determine whether FHL3 interacts with BKLF through this CtBP-contact region or through a distinct region, a series of BKLF deletions was generated. These constructs produce gal4DBD-BKLF fusion proteins and were tested for their ability to interact with gal4AD-FHL3 in the yeast two-hybrid system (Fig. 1C). This experiment indicated that FHL3 interacts strongly with BKLF in a region that lies between amino acids 160 and 224, but that amino acids 75-160 also contribute to binding. Importantly, this region is distinct from the domain in BKLF that binds CtBP2 (amino acids 51-75 (14)), and mutation of the PVDLT motif, which abrogates CtBP2 binding, has no effect on the interaction of BKLF with FHL3 (Fig. 1C). This raises the possibility that BKLF may be able to interact with both FHL3 and CtBP2 simultaneously, and the proximity of the FHL3 and CtBP2 binding sites within BKLF suggested that FHL3 might also make contacts with CtBP2 (see below).
FHL3 Also Interacts with CtBP2-To determine whether FHL3 could bind CtBP2, we first used the yeast two-hybrid assay. We tested whether a gal4DBD-FHL3 protein could interact with a gal4AD-CtBP2 protein. A strong interaction was observed between FHL3 and CtBP2, and no interaction was seen with the gal4AD alone (Fig. 3A). This interaction was also observed in a GST pull-down assay (Fig. 3B), as GST-FHL3 retained radiolabeled CtBP2 (lane 3), whereas GST alone did not (lane 2). The retention of radiolabeled CtBP2 by FHL3 in GST pull-down assays was generally low but was reproducible (see also Fig. 4C). Thus, FHL3 can also interact with CtBP2 in both a yeast two-hybrid assay and in GST pull-down assays. Interestingly, although the vast majority of CtBP-interacting proteins contain a PXDLS-like amino acid motif required for binding CtBP (13), FHL3 has no obvious PXDLS motif (see "Discussion").
Mapping the Interaction between FHL3 and BKLF and CtBP2-We used both GST pull-down assays and the yeast two-hybrid assay to determine which domains in FHL3 interacted with BKLF and which with CtBP2. A series of GST-FHL3 deletion constructs were generated and used in GST pull-down assays with radiolabeled BKLF and CtBP2 (Table I, Table I. Lane 1 in A contains molecular weight markers, and the Input lanes (lane 1) in B and C contain 20% of the radiolabeled BKLF and CtBP2, respectively, used in the assay. Equivalent amounts of GST-fusion proteins were used in each assay. FIG. 3. FHL3 binds the co-repressor CtBP2 in vivo (in yeast) and in vitro. A, Yeast two-hybrid assay shows that FHL3 binds CtBP2. Yeast carrying the plasmids shown were grown on medium lacking leucine and tryptophan (left panel) or histidine, leucine, and tryptophan (right panel). Growth of yeast carrying the gal4DBD-FHL3 and gal4AD-CtBP2 fusions on SD-H-L-T medium show that the bait and prey proteins interact. B, GST pull-down assay shows FHL3 and CtBP2 interact in vitro. 35 S-Labeled CtBP2 was retained by GST-FHL3 protein (lane 3) but not by GST alone (lane 2). The input (lane 1) contains 20% of the radiolabeled CtBP2 used in the assay. GST proteins used were the same as shown in Fig.  1B, lower panel. B and C). The same domains of FHL3 were also used in the yeast two-hybrid system, as gal4DBD fusions, and were tested against galAD fusions with both BKLF (amino acids 1-268) and CtBP2. A summary of the GST pull-down assays and yeast two-hybrid assays is shown in Table I. The results indicate that LIM domains half, 1, and 2 (H-1-2) are sufficient for interaction with BKLF, whereas, only LIM domains half and 1 of FHL3 are sufficient for the interaction with CtBP2. Other domains of FHL3, however, also appear to contact BKLF and CtBP2 (for instance LIM domains 3-4 contact both BKLF and CtBP2, and LIM domain 4 contacts CtBP2), and in some instances results of the yeast two-hybrid assay and the GST pull-down assays are not consistent. Overall, these experiments suggest that folding of the entire FHL3 protein may be important and that the complete binding surfaces may be more complex than just one discreet contiguous domain. Our results are similar to previous attempts to map the interaction between FHL2 and the AR (23) or PLZF (30). In these cases, deletion studies showed that the LIM domains H-1-2 and 3-4 of FHL2 both interact with the AR with similar affinities (22), and no single LIM domain of FHL2 was capable of interacting with PLZF (29). Taken together these experiments are consistent with the view that the four and a half LIM domains in these proteins fold into a complex protein interaction surface that may wrap around and make multiple specific contacts with its partner proteins.
FHL3 Can Homodimerize-It has previously been reported that FHL proteins can dimerize (22). We tested whether this was the case for FHL3. A strong interaction was observed between the proteins gal4DBD-FHL3 and gal4AD-FHL3 in the yeast two-hybrid assay. Deletion studies show that LIM domains H-1-2 of FHL3 are sufficient for contact with full-length FHL3 but that dimerization can also occur between LIM domains 3-4 and full-length FHL3 (Table I). Again this result demonstrates that the series of contact surfaces is likely to be complex. A full understanding of the contact surfaces will require structural analysis of the protein complexes.
BKLF, CtBP2, and FHL3 Form a Nuclear-specific Complex in Vivo-We next used co-immunoprecipitation assays to assess the in vivo interactions between BKLF, CtBP2, and FHL3 (Fig. 5). COS cells were co-transfected with pMT2 expression vectors for BKLF, plus HA-tagged CtBP2 and HA-tagged FHL3, or were mock transfected with equivalent amounts of empty vector. Nuclear extracts were prepared 48 h post-transfection and were immunoprecipitated with either anti-BKLF antibody or pre-immune anti-sera. Immunoprecipitates were probed by Western blotting with anti-HA antibody. Both CtBP2 and FHL3 are efficiently immunoprecipitated by anti-BKLF antibody (Fig. 5A, lane 6), whereas neither protein is immunoprecipitated with pre-immune sera (lane 4). Additionally, CtBP2 and FHL3 are not precipitated from mock transfected cells with either anti-BKLF or pre-immune sera (lanes 3 and 5). This same membrane was subsequently stripped and re-probed with anti-BKLF antibody (Fig. 5B). As expected, BKLF is efficiently immunoprecipitated with anti-BKLF antibody (Fig. 5B, lane 6) but not with pre-immune sera (lane 4), and no BKLF protein was precipitated from mock transfected cells (lanes 3 and 5). These results are consistent with previous in vitro experiments and confirm that BKLF can associate with both CtBP2 and FHL3 in vivo.
To further understand the interaction between BKLF, CtBP2, and FHL3, we used native gel-filtration chromatography to analyze whether these three proteins co-elute (Fig. 6). In   Fig. 4. Yeast were cotransformed with the gal4DBD-FHL3 constructs as shown, together with the gal4AD fusions with either BKLF-(1-268) or CtBP2 fusions. Interactions, as measured by growth of yeast carrying the plasmids on SD-H-L-T media, are shown in the right columns. this experiment we analyzed first nuclear and then equivalent amounts of cytoplasmic extract. Extracts were separated by gel-filtration chromatography, and the fractions were precipitated with trichloroacetic acid, separated by SDS-PAGE, and probed by Western blotting with anti-HA antibody (to detect HA-CtBP2 and FHL3-HA), then stripped and re-probed with anti-BKLF antibody. The monomer masses of BKLF, CtBP2, and FHL3 are 39, 49, and 31 kDa, respectively, but as described above FHL3 can self-associate, as can CtBP2 (14) and BKLF. 2 Analysis of nuclear extracts indicated that BKLF, CtBP2, and FHL3 first co-elute in fraction 17, with BKLF and CtBP2 peaking in this fraction (Fig. 6A). Examination of standards included in the experiment suggest that fraction 17 contains protein in excess of 232 kDa, so the most simple explanation is that this complex may consist of dimers of each protein (expected molecular mass, 239 kDa). All three proteins also co-elute in fraction 19, between the 67-and 158-kDa standards, with significant amounts of FHL3 found in this fraction. This fraction may contain all three proteins eluting independently as homodimers. Interestingly, CtBP2 and FHL3 also coelute independently of BKLF in fraction 14 and 15 between the 440-and 670-kDa standards, suggesting these two proteins may be found in additional higher molecular weight complexes within the nucleus. The elution profile from cytoplasmic extracts is significantly different (Fig. 6B). In particular, the majority of BKLF elutes in fraction 19 (i.e. between 67 and 158 kDa) and FHL3 is predominantly found in fractions 18 and 19. This observation suggests that a complex of around 240 kDa containing BKLF, CtBP2, and FHL3 may form in the nucleus but not in the cytoplasm.
To confirm this, co-immunoprecipitations were performed with cytoplasmic extract from COS cells transfected with BKLF, HA-CtBP2, and FHL3-HA using anti-BKLF antibodies, or pre-immune anti-sera (Fig. 7). Immunoprecipitates were probed with anti-HA antibody (Fig. 7A). No FHL3 protein is found to be associated with BKLF in the cytoplasmic extracts (Fig. 7A, lane 5), and only very small amounts of CtBP2 (lane 5). Additionally, neither CtBP2 nor FHL3 is immunoprecipi-  6. BKLF, FHL3, and CtBP2 are part of a nuclear-specific complex of ϳ240 kDa. COS cells were co-transfected with expression vectors for BKLF and HA-tagged CtBP2 and FHL3. Nuclear (A) and cytoplasmic (B) extracts were prepared and separated by native gel-filtration chromatography. Proteins in each fraction were precipitated, and probed by Western blotting, first with anti-HA antibody (upper panel in A and B), then stripped and re-probed with anti-BKLF antibody (lower panel in A and B). Nuclear BKLF, FHL3, and CtBP2 first coelute in fraction 17, at greater than 232 kDa, whereas the elution profile for cytoplasmic BKLF, FHL3, and CtBP2 is distinctly different, with the three proteins predominantly co-eluting in fraction 19, between 67 and 158 kDa. tated with pre-immune sera (lane 3), or from mock transfected cells (lanes 2 and 4). The blot was stripped and re-probed with anti-BKLF antibody, showing that BKLF is efficiently precipitated from cytoplasmic extract (Fig. 7B, lane 5). Thus, the in vivo interaction between BKLF and both FHL3 and CtBP2 seen in the nucleus (above) is not observed in the cytoplasm.
Co-expression of BKLF and CtBP2 Leads to the Nuclear Enrichment of FHL3-CtBP proteins are found distributed throughout the cell and have been hypothesized to have roles both in the nucleus and the cytoplasm (13). As shown above the DNA-binding protein BKLF is also detectable in both the nucleus and the cytoplasm, although it is more abundant in the nucleus (Fig. 6). Previous work on FHL proteins has also suggested that they exist in both the nucleus and the cytoplasm (20 -23, 31, 40 -42). Moreover, it has been shown that Rhosignaling can promote nuclear enrichment of FHL2 protein (31). Thus, it appears that the sub-cellular localization of FHL proteins may be subject to regulation. In the gel-filtration experiments discussed above, we noted that the elution profile for FHL3 is markedly different between nuclear and cytoplasmic fractions and that BKLF, CtBP2, and FHL3 only co-elute in high molecular weight fractions in nuclear extracts. We thus examined the levels of BKLF, CtBP2, and FHL3 in the nucleus and cytoplasm when different combinations of these three proteins are co-expressed in COS cells.
When FHL3 is expressed on its own, it is found predominantly in the cytoplasm (Fig. 8A, compare lanes 1 and 2). When co-expressed with CtBP2, again, the majority of FHL3 is found in the cytoplasm, as is CtBP2 (compare lanes 3 and 4). When FHL3 is co-expressed with BKLF, an increased fraction of FHL3 is found in the nucleus, although it is still primarily cytoplasmic, whereas BKLF is predominantly nuclear (Fig. 8A,  upper and lower panels, compare lanes 5 and 6). However, when FHL3 is co-expressed with both BKLF and CtBP2, FHL3 is significantly enriched in the nucleus, as is CtBP2 (Fig. 8A,  upper and lower panels, compare lane 7 with 1 and 3). Taken together with the co-immunoprecipitation assays and gel-filtration data above, it appears that a specific complex of ϳ240 kDa containing BKLF, CtBP2, and FHL3 may accumulate in the nucleus only when all three proteins are present and that this complex is not present in the cytoplasm.
To examine whether BKLF promotes nuclear accumulation of endogenous FHL3 and CtBP2 protein, we generated four stable cell lines expressing BKLF using K562 hematopoietic cells, which normally express very low levels of BKLF protein (Fig. 8B, lane 1, upper panel). K562 cells express FHL3 mRNA message (see above), and the original FHL3 prey plasmid, and full-length FHL3 were isolated from a K562 cell cDNA library. We generated an anti-FHL3 antibody and used this to test the effect of BKLF expression on nuclear localization of FHL3. The four K562 cell lines carrying the stable transgene for BKLF show high level BKLF expression in the nucleus, compared with the parental line (Fig. 8B, upper panel, compare lane 1  with lanes 2-5). Importantly, these same BKLF stable cell lines show a marked increase in FHL3 protein in nuclear extracts, compared with the parental cell line (Fig. 8B, lanes 1-5, middle  panel). Additionally, nuclear levels of CtBP2 are increased in the BKLF stable lines (Fig. 8B, lanes 1-5, lower panel). Thus, in a similar manner to that observed in transfected COS cells, BKLF can promote the nuclear accumulation of endogenous FHL3 and CtBP2 in erythroleukemia cells.
FHL3 Can Repress Transcription-Previous studies have shown that FHL proteins can function as transcriptional regulators. The FHL protein ACT serves as co-activator of CREM and CREB (21), and FHL2 as a co-regulator for the androgen receptor (22,23), WT1 (29), and PLZF (30). We thus examined the transcriptional activity of FHL3 (Fig. 9). Because FHL proteins are not known to bind DNA, we first fused FHL3 to the gal4DBD and tested its ability to regulate basal transcription of two artificial test promoters carrying gal4 binding sites. The gal4DBD-FHL3 fusion protein repressed the thymidine kinase (tk) promoter by greater than 12-fold in a dose-dependent manner (Fig. 9A, columns 5-7), whereas the gal4DBD alone has no effect on this promoter (Fig. 9A, columns 2-4). Additionally, the FHL3 fusion protein repressed the minimal promoter containing only a TATA box to about 15-fold (Fig. 9B, columns  5-7), again in a dose-dependent manner, whereas the gal4DBD only marginally affected its activity (Fig. 9A, columns 2-4).
We next tested the ability of FHL3 to regulate activated transcription in the context of a natural promoter. The erythropoietin promoter is strongly activated by the erythroid-specific transcription factor GATA-1 (43). The promoter contains  1 and 2) or co-expressed with either CtBP2 (lanes 3 and 4) or BKLF (lanes 5 and 6). When FHL3 is co-expressed with both BKLF and CtBP2, FHL3 is significantly enriched in the nucleus, as is CtBP2 (lanes 7 and 8, upper panel) and BKLF (lanes 7 and 8, lower panel), albeit to a lesser extent. B, stable expression of BKLF in K562 cells promotes nuclear accumulation of endogenous FHL3 and CtBP2. Nuclear extracts were prepared from four K562 cell lines carrying stable BKLF transgenes and parental K562 cells and were probed by Western blotting as shown. BKLF stable lines (lanes 2-5, upper panel) express high levels of BKLF protein in the nucleus, whereas K562 cells express little BKLF (lane 1, upper panel). FHL3 expression in the nucleus is significantly increased in the nucleus in the BKLF stable cell lines (lanes 1-5, middle panel), as is CtBP2 (lanes 1-5, lower panel). Lane 6 contains nuclear extract form COS cells transfected with expression vectors for BKLF, FHL3-HA, and HA-mCtBP2 (note that the small discrepancy in migration between HA-mCtBP2 and endogenous CtBP2 is due to addition of the HA tag). multiple CACCC elements and we have found that BKLF can potently repress GATA-1-mediated activation (Fig. 9C, columns 1-3). 2 The level of GATA-1 activation of the basal activity of this promoter is ϳ20-fold (Fig. 9C, column 1) and is assigned a -fold repression of 1. BKLF represses the promoter ϳ7-fold (Fig. 9C, columns 2-3). We tested the ability of FHL3 to repress GATA-1-mediated activation of this promoter. To direct FHL3 to the promoter, we generated a chimeric fusion protein between FHL3 and the zinc finger DBD of BKLF. As judged by electrophoretic mobility shift assays, this FHL3-zinc finger fusion protein can efficiently bind CACCC sites in DNA in a similar manner to BKLF (data not shown). As shown in Fig. 9C  (columns 4 and 5), the FHL3-zinc finger chimera represses GATA-1 activation as efficiently as BKLF, whereas FHL3 alone (columns 8 and 9) and the BKLF zinc finger DBD alone (col-umns 6 and 7) have no effect on GATA-1 activation.
To further test the activity of FHL3, we examined whether FHL3 (without a linked DBD) could potentiate BKLF repression (Fig. 9D). GATA-1 strongly activates the erythroid-specific A␥-globin promoter, a known GATA-1 target gene (44). Again, GATA-1 activation of this promoter, at ϳ20-fold, is set to a -fold repression of 1 (Fig. 9D, column 1). We have found that BKLF can potently repress this natural promoter, through binding to multiple CACCC sites in the promoter. 2 As shown in Fig. 9D (column 2), low levels of BKLF expression vector (100 ng) repress GATA-1 activity ϳ1.5-fold. Note that a low level of BKLF expression was chosen for minimal repression of the promoter, allowing further super-repression by FHL3 to be analyzed. Importantly, when FHL3 is co-expressed with BKLF (Fig. 9D, columns 3-5), BKLF repression is enhanced up to FIG. 9. FHL3 can repress both basal and activated transcription when tethered to a promoter with a heterologous DNA binding domain, and can potentiate BKLF repression. A, FHL3 represses the thymidine kinase (tk) promoter. NIH-3T3 cells were transfected with 5 g of the reporter plasmid p(gal4) 4 tkCAT (column 1), along with 0.5, 2, and 4 g of expression vectors for the gal4DBD (columns 2-4) or a fusion between the gal4DBD and FHL3 (columns 5-7). The gal4DBD-FHL3 fusion efficiently repressed the basal activity of this promoter, whereas the gal4DBD alone has no effect. B, FHL3 represses a basal promoter carrying the ␤-globin TATA box. NIH-3T3 cells were transfected with 5 g of the reporter plasmid p(gal4) 5 GH (column 1) along with 0.5, 2, and 4 g of expression vectors for the gal4DBD (columns 2-4) or a fusion between the gal4DBD and FHL3 (columns 5-7). The gal4DBD-FHL3 fusion efficiently repressed the basal activity of this promoter, whereas the gal4DBD alone has minimal effect on the promoter. C, FHL3 represses the erythropoietin promoter, when activated by GATA-1. NIH-3T3 cells were transfected with 1 g of the reporter plasmid p(EpoR)GH and 1 g of the GATA-1 expression vector pRcCMV.GF1 (columns 1-9) together with 0. 4-fold. Thus FHL3 can repress both basal and activated transcription when targeted to a promoter through heterologous DNA binding domains, and the native FHL3 protein can potentiate BKLF repression. This fits well with recent data, showing that FHL3 (and FHL2) can also act as co-repressors for the multi-zinc finger repressor PLZF (30). DISCUSSION Transcriptional regulation is mediated in part by the recruitment of multiprotein complexes to regulatory regions in DNA.
Revealing the composition and activities of these complexes will aid our understanding of how DNA-binding transcription factors regulate gene expression in a tissue-and temporalspecific manner. Here we show that the Krü ppel-like factor BKLF/KLF3 can bind to two co-regulatory molecules, FHL3 and CtBP2, both of which can repress transcription. This work expands our understanding of the activities of the FHL class of proteins: FHL proteins were previously known to behave primarily as co-activators; we show here that FHL3 can behave as a transcriptional repressor and forms a physical link between BKLF and its co-repressor CtBP2.
FHL proteins were initially identified as being highly expressed in muscle cells (24,25) and were originally proposed to play roles in cytoskeletal regulation. This proposal was primarily based on the observation that, like the cytoskeletal-associated LIM proteins zyxin and paxillin, the FHL proteins were also composed primarily of LIM domains, and were developmentally regulated during muscle hypertrophy (24). Recently, however, it has become clear that many FHL proteins participate in the regulation of gene expression within the nucleus where they serve as co-regulators of DNA-binding transcription factors. ACT binds the testis-specific factor CREM (21), and ACT, FHL2, and FHL3 all bind CREB (22). Additionally, FHL2 binds Wilms Tumor-1 (WT-1) (29), and the androgen receptor (in an agonist-dependent manner) (23). In these cases, the FHL proteins behaved as co-activators, but whether they act as conventional co-activators, recruiting components of the basal transcriptional machinery, or chromatin-modifying proteins or act in a different manner has not been established. There are also two reports of FHL proteins functioning to counter gene activation. KyoT2, a splice variant of FHL1, antagonizes the activity of RBP-J by displacing co-activators EBNA2 or Notch (20). Furthermore, FHL2 has been demonstrated to function as a co-repressor of the multi-zinc finger repressor PLZF (30), and in this same report, FHL3 was also shown to function as a PLZF co-repressor. These data support our observations that FHL3 binds the transcriptional repressor BKLF and its co-repressor CtBP2 and aid in the direct repression of transcription by BKLF. The finding that FHL proteins participate both in activation and repression fits best with the view that FHL proteins are neither primarily co-activators or co-repressors but, rather, serve as adaptor molecules that stabilize large transcriptional complexes with differing activities.
We have previously shown that BKLF recruits the protein CtBP2 as a co-repressor and that this interaction is dependent on the integrity of a PXDLS-CtBP recognition element in the repression domain of BKLF (14). CtBP proteins have been shown to function as co-repressors for a large number of regulatory molecules that contain PXDLS motifs in their repression domains (13). Having observed that CtBP2 binds FHL3, we searched the FHL3 sequence for PXDLS type sequences that might mediate the interaction. None were detected. This suggests that FHL3 binds a distinct region of CtBP proteins rather than occupying the putative PXDLS peptide pocket of CtBP (13,45). Deletion analysis of FHL3 is consistent with this view, because no single small peptide in FHL3 contacts CtBP2. Instead, it appears that several different LIM domains of FHL3 are required for contact, and it is likely that different LIM domains of FHL3 may make several contacts with different surfaces of CtBP2. BKLF binds CtBP2 using a PVDLT motif (amino acid residues 70 -75) (14) and binds FHL3 through a domain that may extend from around amino acids 80 -260. FHL3 also uses several different LIM domains to contact BKLF. The mode of contact utilized by FHL3 is reminiscent of that seen with other members of the FHL family that also seem to utilize multiple different LIM domains to contact their partner proteins (22,23,41,42,46). Detailed structural studies will be necessary to fully elucidate the configuration by which FHL proteins contact other proteins.
We have found that FHL3 behaves as a potent transcriptional repressor of both basal and activated transcription when tethered to a promoter region either by the gal4DBD or by the zinc finger DBD of BKLF. Furthermore, FHL3 expression (without a linked DBD) can potentiate BKLF repression of GATA-1 activation. Taken together with the in vitro and in vivo binding data between BKLF and FHL3 presented above, this argues strongly that FHL3 functions as a BKLF co-repressor. This is further supported by recent observations that FHL3 (and FHL2) can also act as a co-repressor for the multi-zinc finger transcription factor PLZF (30). Because FHL3 binds CtBP, the transcriptional repression observed may be due, at least in part, to its ability to recruit one or more endogenous CtBP proteins to the promoter. In a previous study, FHL3 behaved as transcriptional activator (23). Differences in the context of the promoter are most likely to explain these different findings. We have chosen promoters that BKLF and CtBP2 can potently repress in cellular assays, and these may be particularly sensitive to FHL3. Differences in cell lines, and hence other accessory proteins, may also be important. Furthermore, in the context of recruitment by DNA-binding factors, unique interactions with different FHL3 target proteins may alter the conformation of FHL3, allowing it to interact with either coactivators or co-repressors through the presentation of distinct protein interaction domains.
FHL proteins are not exclusively nuclear and have been shown to bind both nuclear and cytoplasmic partner proteins (20 -23, 41, 42, 46 -48). Furthermore, sub-cellular partitioning of FHL proteins has been implicated in the regulation of their function. For example, Rho-signaling promotes the nuclear localization of FHL2, and subsequent co-activation of the AR (31), and the kinesin KIF17b binds ACT and is thought to counter co-activation of CREM by ACT by shuttling it from the nucleus to the cytoplasm (32). We have found that the subcellular localization of FHL3 is influenced by the availability of its cofactors. When FHL3 is expressed in COS cells on its own, or with either BKLF or CtBP2, it is primarily cytoplasmic. However, co-expression with both BKLF and CtBP2 markedly increases the amount of FHL3 (and CtBP2) found in the nucleus. Additionally, stable transgene expression of BKLF in erythroleukemia cells promotes the nuclear accumulation of endogenous FHL3 and CtBP2. Indeed, gel-filtration chromatography shows that a large complex of ϳ240 kDa containing BKLF, CtBP2, and FHL3 is found in the nucleus and not in the cytoplasm. The full composition of this complex is not certain. However, given the molecular weights of the three proteins, BKLF (at 39 kDa), FHL3 (at 31 kDa), and CtBP2 (at 49 kDa), and the knowledge that each can self-associate, the simplest explanation is that the complex contains a dimer of each component (totaling 239 kDa). The presence of a nuclear-specific complex containing BKLF, CtBP2, and FHL3 is supported by co-immunoprecipitation data, showing that BKLF in the nucleus binds both CtBP2 and FHL3. Cytoplasmic BKLF behaves very differently in gel filtration and, as judged by co-immuno-precipitation studies, does not bind appreciable amounts of FHL3 or CtBP2. Post-translational modification of BKLF may be important in directing it to the nucleus and allowing it to associate with different partner proteins. Alternatively, BKLF, carrying a putative nuclear localization signal (49), may form a complex with CtBP2 and FHL3 in the cytoplasm, and this complex may then be shuttled to the nucleus. We have previously found that BKLF can behave as both a transcriptional activator (35) 2 and as a repressor (14). These intriguing results may be accounted for by the presence of different post-translational states and/or the availability of partners such as FHL3 and/or CtBP2.
Members of the Sp/KLF family, numbering greater than 20, share related DNA-binding domains composed of three Krü ppel-like zinc fingers, and all bind to similar sites in DNA (1-4). Many of these proteins are co-expressed in various cell types, and a current challenge is to understand whether this coexpression facilitates redundant regulation of similar genes or regulation of specific genes. So far there is strong evidence that some KLF members, such as EKLF/KLF1, have specific target genes, like ␤-globin, that are not activated by other members of the family. The observation that different family members share little homology outside their DNA-binding domains and hence interact with distinct cofactors may be of central importance. Potentially, each KLF may interact with a unique set of cofactors that allows it to regulate a specific subset of genes. On this note, it has been shown that FHL3 does not bind the related KLF Sp1 (22). Thus, the complex we observe between BKLF, CtBP2, and FHL3 may be unique to BKLF, allowing it to repress a specific set of target genes. Future work correlating target genes with cofactor interactions, for example using chromatin immunoprecipitation with antibodies to a specific KLF and its cofactors, may afford insight into this complex problem.