Identification of a Human T-cell Leukemia Virus Type I Tax Peptide in Contact with DNA*

The human T-cell leukemia virus Tax protein directs binding of a host factor, cAMP response element binding protein, to an extended recognition sequence in the proviral promoter. Prior cross-linking experiments have revealed that Tax makes restricted contact with this DNA at two symmetric positions, 14 nucleotides apart on opposite strands of the DNA. Tax lacks a conventional DNA binding domain, and the sequences in Tax that are in contact with DNA have not been previously identified. Analysis of cross-linked peptides now shows that the contact occurs between Tax residues 89 and 110, corresponding to a protease-sensitive linker joining two protein structural domains. The linker assumes a protease-resistant conformation in the cross-linked complex. Point mutations within the linker prevent cross-linking and interfere with Tax function. These data suggest that entry of Tax into the ternary complex may be coupled to folding of an unstructured protein domain, which then makes base-specific contacts with DNA.

Viruses are intracellular parasites that depend on the host cell biosynthetic apparatus to survive. In order to promote their own replication, many viruses have developed mechanisms to redirect the host cell transcription machinery to preferentially synthesize viral RNA. Human T-cell leukemia virus type I provides a well studied example. HTLV-I 1 is the causative agent of adult T-cell leukemia/lymphoma and HTLV-I associated myelopathy (reviewed in Refs. 1 and 2). As in all retroviruses, HTLV-I RNA synthesis is under the control of a promoter located in the proviral long terminal repeat. The HTLV-I encoded Tax protein recruits host cell transcription factors to the proviral promoter, leading to high levels of viral RNA synthesis (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13).
Biochemical studies suggest that Tax activates proviral transcription, in large part, through an interaction with the host cAMP response element binding protein (CREB). Tax alters the DNA binding specificity of CREB (6,(14)(15)(16)(17), which ordinarily binds a symmetric 8-base pair consensus sequence, TGACGTCA (18). CREB recognizes this site through a specific interaction of amino acids in the bZIP domain of the protein with functional groups in the major groove of DNA. In HTLV-I infected cells, Tax selectively stabilizes the binding of CREB to several extended sites in the proviral promoter known as Tax response elements. Each TxRE consists of an imperfect CREB site flanked by additional 5Ј G-rich and 3Ј C-rich elements (6,14,16,17,19). The effect of Tax on complex stability is attributable in part to an enhancement of CREB-CREB dimerization and in part to enhancement of protein-DNA interactions (6,13,14,(20)(21)(22). Cross-linking studies show that Tax itself makes two extremely restricted DNA contacts within the complex (23). These contacts are arranged symmetrically, 14 nucleotides apart, on opposite strands of the DNA. This places the contacts on the same face of the helix, 180°around the helical axis from the CREB dimerization domain.
The formation of a Tax-CREB-DNA ternary complex at the proviral site is made even more significant because of an accompanying change in the interaction of CREB with the transcriptional coactivator, CBP. CBP ordinarily interacts selectively with CREB that has been phosphorylated by cAMPdependent protein kinase, protein kinase A (8,18). The Tax-CREB-DNA ternary complex recruits CBP directly, bypassing the requirement for protein kinase A phosphorylation and allowing viral transcription to proceed constitutively, in the absence of cAMP-mediated signaling. (8,11,12,24). Thus, the interaction of Tax with the TxRE DNA is a key event in the viral life cycle.
Although Tax has been exhaustively mutagenized, it lacks an identifiable DNA binding domain, and the mechanism by which it redirects CREB binding to the extended proviral site remains fundamentally a mystery. In the present study, we have used photochemical cross-linking to identify the specific peptide within Tax that is in contact with DNA. Aryl azide groups were introduced into the backbone of DNA at two specific sites by reaction of 4-azidophenacyl bromide with monophosphorothioate-containing DNA (23,(25)(26)(27)(28). Analysis of cross-linked adducts has identified sites of Tax-DNA contact within an exposed loop of the Tax protein that connects two structural domains. When Tax enters into a ternary complex with CREB and DNA, this loop becomes inaccessible to proteases, and we suggest that it folds into a defined conformation in order to make base-specific contacts within the 5Ј and 3Ј regions of the extended recognition site. Amino acid residues in Tax that contact DNA are immediately adjacent, within the primary sequence, to residues that interact with CBP. Thus, the same small region of Tax seems to be involved both in redirecting CREB-DNA binding specificity and in conferring cAMP-independent transcriptional activity.

EXPERIMENTAL PROCEDURES
Protein Purification-TaxH 6 protein was expressed in Escherichia coli and purified by nickel-agarose chromatography as described in Ref. 23. CREB A protein was expressed in E. coli and purified by heparinagarose chromatography as described (14). The CREB bZIP protein (CREB 254 -327) was expressed in E. coli and purified identically to CREB A.
Binding Reactions-Binding reactions contained 0, 25, 50, or 100 nM CREB, 0 or 300 nM Tax, 520 pM [ 32 P]DNA, 5.2 g/ml poly(dI-dC), and binding buffer (12.5 mM HEPES, pH 7.9, 75 mM KCl, 0.4 mM EDTA, 2.5 mM MgCl 2 , 1 M ZnSO 4 , 6% glycerol, 0.6 mM 2-mercaptoethanol, and 0.2 g/ml phenylmethylsulfonyl fluoride) in a final volume of 20 l. The wild type DNA probe was a 24-base pair fragment containing the most promoter-proximal of the three TxREs present in the HTLV-I promoter, and the symmetric DNA probe was a 22-base pair fragment containing a consensus cAMP response element and flanking regions (for sequence, refer to Fig. 1). Reactions were incubated for 15 min at room temperature, and binding was analyzed by 5% PAGE using a Tris-glycine running buffer (14). Results were quantitated by PhosphorImager analysis.
Azido-modified Oligonucleotides-Oligonucleotides containing a single phosphorothioate modification were obtained from Cybersyn (Lenni, PA) and from the Medical College of Georgia Molecular Biology Core Facility. Azido-modified, double stranded oligonucleotides were constructed from the phosphorothioate-modified oligonucleotides and 4-azidophenacyl bromide by the methods described in Ref. 23.
Preparative Cross-linking and Isolation of Peptide Adducts-Tax (1700 nM) and CREB (1700 nM) were incubated with the azido-modified symmetric 32 P-DNA probe (850 nM) in the absence of poly(dI-dC) in binding buffer at room temperature for 15 min in a final volume of 6.5 ml. Protein-DNA complexes were cross-linked by exposure to 302-nm ultraviolet light (AlphaImager 2000, Alpha Innotech, San Leandro, CA) for 30 s. The cross-linked protein-DNA complexes were separated from the free protein and DNA by 10% SDS-PAGE. The cross-linked complexes were excised from the gel and extracted by incubation overnight at 4°C in 10 mM Tris-HCl, pH 7.0, 1 mM EDTA, and 0.1% SDS. The extracted material was subjected to complete digestion with trypsin under denaturing conditions (29). The sample was incubated for 0.5 h at 37°C, in 2 M urea, 5 mM CaCl 2 , 2.5 mM Tris, 0.25 mM EDTA, and 0.025% SDS at pH 8, and then trypsin (sequencing grade, Roche Molecular Biochemicals) was added to a final concentration of 5 g/ml, and the reaction was continued for 2 h at 37°C. The resulting peptide-DNA adducts were precipitated with ethanol, resuspended in formamide loading buffer (90% deionized formamide, 0.1% xylene cyanole, and 0.1% bromphenol blue), and separated by electrophoresis in a gel containing 8 M urea and 12% polyacrylamide. The peptide-DNA complexes were excised from the gel and extracted by overnight incubation in 20 mM NH 4 HCO 3 , pH 8.5, and 10% acetonitrile. Extracted complexes were submitted for peptide sequencing at the Emory University Microchemical Facility (Atlanta, GA).
Partial Trypsin Digestions-Tax (300 nM) and CREB bZIP (600 nM) were incubated with the azido-modified symmetric DNA probe (520 pM) in the presence of 7.3 g/ml poly(dI-dC) in binding buffer at room temperature for 15 min. The photocross-linked protein-DNA complexes were prepared as described previously (23). Trypsin (1.1 g/ml) (Sigma) was then added to the reaction and the reactions were incubated at 37°C for 0 -40 min. The reactions were terminated by the addition of phenylmethylsulfonyl fluoride to 1 mg/ml, and the products were analyzed by 10% SDS-PAGE.
For the N-terminal peptide sequencing of fragments 1 and 2, 100 g of Tax was incubated in 25 mM HEPES, pH 7.9, 150 mM KCl, 1 mM EDTA, 10% glycerol, and 4 mM 2-mercaptoethanol with 2.8 g/ml trypsin (N-tosyl-L-phenylalanine chloromethyl ketone-treated) (Sigma) for 20 min at 37°C in a total volume of 2 ml. The reaction was terminated with the addition of 200 l of 10 mg/ml phenylmethylsulfonyl fluoride. The reaction was dialyzed against 100 mM NH 4 OAc for 2 h and then against 10 mM NH 4 OAc for 2 h, lyophilized, resuspended in 1 ml of water, lyophilized again, and resuspended in 10 l of 1ϫ SDS loading buffer. The fragments were resolved by glycine-free SDS-PAGE and electroblotted onto polyvinylidene difluoride membrane (30). The electroblotted peptide fragments were submitted for peptide sequencing at Macromolecular Resources (Colorado State University, Fort Collins, CO). The sample for the sequencing of fragment 3 contained 15 g of Tax, 25 mM HEPES, pH 7.9, 150 mM KCl, 1 mM EDTA, 10% glycerol, and 17 g/ml trypsin in 75 l. The reaction was incubated at 37°C for 40 min and was terminated by the addition of 2 l of 10 mg/ml phenylmethylsulfonyl fluoride and 25 l of 4ϫ SDS loading buffer. The fragments were resolved by glycine free SDS-PAGE and electroblotted onto polyvinylidene difluoride membrane (30). The electroblotted peptide fragment was submitted for peptide sequencing at the Emory University Microchemical Facility.
Antibodies and Immunoblotting-Antibody 88003 is a polyclonal anti-C-terminal Tax antibody (3). The rabbit used to produce antibody 88003 was originally immunized with a Tax-␤-galactosidase fusion protein and was boosted with a C-terminal peptide conjugate. This resulted in a polyclonal serum that contains strong C-terminal reactivity and weak reactivity to the rest of the protein. Antibody 1314 (National Institutes of Health AIDS Reference Reagent Program) is a monoclonal anti-Tax antibody, which reacts with an epitope at the extreme C terminus of Tax. 2 Antibody 12701 is a polyclonal anti-Nterminal Tax antibody raised against peptide conjugate containing Tax residues 37-56 (SARLHRHALLATCPEHQITW) (Pocono Rabbit Farms, Canadensis, PA). Immunoblotting using antibodies 88003 and 1312 was performed using nitrocellulose membranes and 5-bromo-4-chloro-3-indolyl phosphate/nitroblue tetrazolium detection (Promega). Immunoblotting using antibody 12701 was performed using polyvinylidene difluoride membrane and enzyme-catalyzed fluorescence substrate (Vistra ECF substrate, Amersham Pharmacia Biotech).
Site-directed Mutagenesis of Tax-The plasmids containing the clustered point mutants of Tax (see Fig. 4A) were made using an adaptation of a commercial method (QuickChange Site-Directed Mutagenesis Kit, Stratagene, La Jolla CA). The wild type Tax plasmid (31) was subjected to polymerase chain reaction utilizing primers containing mutant sequences and a restriction site (SacI or HpaI). The mutant polymerase chain reaction products were purified, digested with the appropriate enzyme, purified again, ligated, subjected to DpnI digestion to remove any remaining wild type plasmid, and transformed into E. coli strain DH5␣. The Tax protein was purified in an manner identical to wild type.
In Vitro Transcription-In vitro transcription was performed as described (32) using 120 ng of HeLa nuclear extract protein and 750 ng of circular plasmid template. The template consisted of four tandem copies of the promoter-proximal HTLV-I TxRE (14), joined to position Ϫ52 of the HTLV-I core promoter, joined at position ϩ1 to a 380-nucleotide G-less cassette (gift of M. Anderson, Medical College of Georgia).

RESULTS
Design of Oligonucleotide Probes-A wild type oligonucleotide probe (Fig. 1A), which is the same as was used in our previous studies, consisted of a partially symmetric 24-nucleotide sequence representing the most promoter-proximal of the three TxREs in the HTLV-I long terminal repeat (14,23). A central imperfect CREB site was flanked by G-rich 5Ј flanking and C-rich 3Ј flanking sequences, as indicated. The binding of this DNA to CREB and Tax was measured in an electrophoretic mobility shift assay. Results were quantitated and are shown in Fig. 1B. The wild type oligonucleotide bound relatively weakly to CREB (K d Ͼ 100 nM), and the binding was stimulated 3-8-fold by Tax, consistent with previous results (23).
To increase the efficiency of binding and simplify the analysis of products generated in subsequent cross-linking experiments, a new, idealized, symmetric 22-nucleotide probe was designed (Fig. 1A). The self-annealed symmetric DNA, which contained a perfect match to the consensus CREB site (18), bound more strongly to CREB than the wild type probe (Fig.  1B). Tax further increased the level of this binding.
To generate probes for site-specific cross-linking, the symmetric DNA probe was synthesized with a phosphorothioate group substituted for a backbone phosphate at the positions indicated (Fig. 1A). Previous work has shown that these are the only positions on the DNA that strongly cross-link to Tax (23). The phosphorothioate-containing DNA was radiolabeled, selfannealed, and reacted with 4-azidophenacyl bromide to yield a photoreactive aryl azide, which is able to cross-link nonspecifically to protein functional groups within a 9-Å radius of its attachment point (25,33).
Use of Aryl Azide Probes to Map Protein-DNA Contacts-The goal of the present study was to isolate and characterize Tax peptides present in cross-linked adducts. Protein-DNA com-plexes were formed, cross-linked, and isolated on a preparative scale. Because only a small percentage of the aryl azide groups form productive cross-links with protein, the protein-DNA adducts needed to be separated from free protein by SDS-PAGE. Tax and CREB are similar in size (40 and 35 kDa, respectively), and the Tax-DNA and CREB-DNA adducts migrate at overlapping positions (23). These adducts are seen as the predominant radiolabeled band in Fig. 2A (arrow). This band was excised from the gel. The remainder of the gel was then stained with Coomassie Blue R250, which revealed that free proteins migrated at the position marked (arrow) and were well resolved from the adducts (Fig. 2A).
The excised protein-DNA complexes were extracted from the polyacrylamide matrix and digested with trypsin under denaturing conditions as described under "Experimental Procedures." The resulting peptide-DNA complexes were resolved by urea-PAGE and were detected by autoradiography (Fig. 2B). Two peptide-DNA complexes were seen. We presume that the upper complex is a Tax adduct because it was seen only when the binding reactions contained Tax protein (data not shown). The lower band is apparently a CREB adduct because it was seen in all reactions containing CREB, whether or not Tax was also present. The complexes have been labeled accordingly in Fig. 2. A small amount of free DNA was also present. This may reflect a breakdown of adducts during purification, as well as some non-cross-linked DNA that was retained in the original protein-DNA complexes by base pairing. The Tax-DNA and CREB-DNA adducts were individually excised from the gel, extracted, and subjected to Edman degradation to identify the peptides that were present.
The CREB-DNA adduct yielded protein sequence at the 2 pmol level. This sequence was (L)MXNX, where X denotes an unidentified amino acid, and the residue in parentheses is tentative. Although this sequence is short, it corresponds to a unique tryptic fragment within the CREB DNA binding domain, 276 LMKNR (Fig. 2C). Because there was no cleavage after Lys 278 , and because this residue was not recovered as a PTH amino acid, it may be the residue that participates in adduct formation. A three-dimensional model of the CREB-DNA complex, showing the position of this residue (presented under "Discussion") showed that the Lys 278 side chain lies in close proximity to the modified phosphate. These results serve as an important control because they show that data obtained from analysis of the cross-linked CREB peptide agrees with structural predictions (34) The Tax-DNA adduct yielded protein sequence at the 1 pmol level. This sequence was VLTPPITX(T)(T), which corresponds to a unique tryptic fragment of Tax, 89 VLTPPITHTTPNIPPS-FLQAR. Residue His 96 , which was not recovered as a PTH amino acid derivative, could be the actual site of cross-linking. Because this residue falls near the end of the readable sequence, however, this identification is tentative, and we cannot rule out the possibility that a more C-terminal residue in this peptide is the actual site of cross-linking. As described in the next section, the Tax peptide that we have identified is of particular interest because it appears to be part of a loop that connects two structural domains. It lies immediately adjacent to, but appears to be distinct from, a sequence that is involved in CBP binding (12).
Partial Proteolysis of Tax-Limited proteolytic digestion is an established method of determining protein domain structure. To apply this approach to the Tax-CREB system, ternary complexes were formed between Tax, the bZIP DNA binding domain of CREB, and DNA. The 9-kDa bZIP domain of CREB was used in place of full-length protein because it is small enough not to interfere with the visualization of proteolytic The residues that differ between the two probes are indicated in boldface type. Lowercase p indicates phosphate groups, which are shown explicitly because they are potential attachment points for aryl azide cross-linking groups. In the wild type probe, the vertical lines indicate attachment points for aryl azide groups that react strongly with Tax (23). In the symmetric probe, the vertical lines indicate the attachment points for aryl azide groups used for cross-linking in this study. B, binding of oligonucleotide probes to CREB and Tax. Binding reactions were performed as described under "Experimental Procedures" using 0, 25, 50,  fragments of Tax. Cross-linking was performed, followed by partial digestion with trypsin. Products were resolved by SDS-PAGE. The gel was silver-stained to identify the bulk Tax digestion pattern (Fig. 3A) and was subjected to PhosphorImager analysis to identify the digestion pattern of the crosslinked Tax (Fig. 3B).
Three Tax fragments that were resistant to limited proteolysis under native conditions were visualized by silver staining (Fig. 3A). Although the digestion was performed in the presence of CREB bZIP domain and DNA, free Tax was in large excess over bound Tax in these reactions and the bulk digestion pattern (as observed in Fig. 3A) was indistinguishable from that obtained with Tax alone (data not shown). The CREB bZIP domain migrated well ahead of these fragments.
We identified the proteolytic fragments of Tax by a combination of mass spectroscopy, immunoblotting and N-terminal sequence analysis. Fragment 1 contains the Tax N terminus but has lost a small segment at the C terminus. It was not reactive with an anti-C-terminal peptide antibody or with an anti-C-terminal monoclonal antibody but was reactive with an anti-N-terminal peptide antibody (Fig. 3C). Edman degradation of this fragment yielded sequence beginning with residue 2 of Tax. Mass spectroscopy indicated that the C-terminal cleavage occurred after Lys 346 or Arg 349 ; thus the missing C-terminal region was 10 or 13 amino acids long, including the histidine tag (data not shown). Fragment 2 contains sequence extending from Val 89 to the Tax C terminus, as determined by Edman degradation and the immunoblotting shown in panel C. Fragment 3 contains sequence starting at Val 89 , as determined by Edman degradation, and has lost a small segment at the C terminus as determined by immunoblotting. Presumably, the C-terminal cleavage sites are the same as in fragment 1.
A diagram of the fragment assignments is shown in Fig. 3D. Under these conditions, we can find no evidence for tryptic cleavage at Arg 110 or Arg 116 , as has been reported by another group (35). We have noted, however that additional degradation products appeared if tryptic fragments were subjected to prolonged dialysis at low ionic strength in preparation for sequencing, a procedure that we subsequently avoided (data not shown). It may be that differences in sample processing account for the discrepancy in results.
The pattern of protease susceptibility seen with the crosslinked fraction of the Tax differs dramatically from the bulk digestion pattern (compare Fig. 3, A and B). Fragments 2 and 3, resulting from cleavage between Lys 88 and Val 89 , are essentially absent from the radiolabeled, cross-linked products. Because chemical analysis of the adducts shows that cross-linking occurs distal to this cleavage site, both fragments 2 and 3 should have been visible had cleavage occurred. Quantitation of these results showed that the amount of cleavage between The CREB bZIP domain and its digestion products are Յ9 kDa and are not visible. Fragments 1, 2, and 3 have been assigned as described in text. B, phosphorimage showing digestion products derived from crosslinked Tax adducts. The gel is the same one as in panel A. Note the absence of fragments 2 and 3. C, immunoblotting of Tax partial digests. Duplicate reactions were run on the same gel. The gel was cut in half; one half was silver-stained, and the other half was used for Western blotting with antibodies as indicated. Antibody 88003 is directed primarily again a Tax peptide in the C terminus, antibody 1314 is a monoclonal antibody directed against a C-terminal epitope, and antibody 12701 is directed against a N-terminal Tax peptide (see under "Experimental Procedures"). D, diagram showing assignment of fragments 1, 2, and 3 within the Tax sequence.
Lys 88 and Val 89 in the cross-linked protein was reduced 10-fold or more relative to the bulk Tax. We conclude that cross-linking within the adjacent region, residues 89 -109, hinders tryptic cleavage. Cleavage at the C-terminal site (reflected in the appearance of fragment 1) is unaffected by cross-linking and provides an internal control in these reactions. Substantially the same results were obtained in reactions containing fulllength CREB as with the bZIP domain, although there was some background attributable to CREB cleavage products, particularly at early time points (data not shown).
By raising the DNA concentration to a level stoichiometric with Tax, we were able to obtain some protection of the bulk Tax against trypsin cleavage, but this protection was never complete (data not shown). The CREB bZIP domain is highly trypsin sensitive, and it is likely that by degrading CREB, trypsin treatment rapidly disassociates non-cross-linked ternary complexes.
Mutagenesis of the Cross-linked Region-Some previous mutational studies have addressed the function of Tax residues 89 -110, within the cross-linked Tax peptide. Notably, a double point mutation at 92 PP impairs CREB-dependent but not NFBdependent transcriptional activation (36). However, mutation of Ser 104 has little effect on Tax function (4).
To more directly investigate the requirement for amino acids 89 -110 for Tax function, we constructed four multiple point mutants, as diagrammed in Fig. 4A. Mutant m92 is a reconstruction of the 92 PP mutation (36) in the TaxH 6 background; the other mutants are novel. All of the mutant proteins were expressed in E. coli as soluble proteins and at similar levels (data not shown). Mutant proteins were purified and subjected to partial proteolysis to determine whether they retained the same overall domain structure as wild type Tax. Results showed that all but one of the mutants had the same pattern of trypsin cleavage as the wild type protein (Fig. 4B). The sole FIG. 5. Cross-linking and transactivation using mutant Tax proteins. A, Tax cross-linking. Aryl azide cross-linking reactions were performed as described under "Experimental Procedures" using 175 nM CREB bZIP domain. Reactions were performed using 175 nM wild type Tax (wt), in the absence of Tax (Ϫ), or in the presence of 175 nM each of the indicated Tax mutants. All lanes are from the same gel. The region containing the Tax-DNA adducts is shown. Cross-linking was quantitated by PhosphorImager analysis and is expressed relative to wild type Tax. B, bZIP cross-linking. Shown is the same experiment as in panel A. The region of the gel containing the bZIP-DNA adducts is shown. Cross-linking was quantitated and expressed as in panel A. C, TxRE-dependent transcription. In vitro transcription was performed as described under "Experimental Procedures" using HeLa cell nuclear extract and wild type or m95 mutant Tax as indicated. Reactions were performed in duplicate (with Tax) or quadruplicate (without Tax). RNA synthesis was quantitated by PhosphorImager analysis and expressed relative to synthesis in the absence of Tax. Graph shows mean transcription in two independent experiments. Symbols denote wild type Tax (q) or mutant m95 (f). Error bars denote S.D. Dashed horizontal line denotes relative transcription level of 1.0.

FIG. 4. Mutant Tax proteins.
A, mutagenesis of Tax. Clustered point mutants were generated as described under "Experimental Procedures" and are numbered according the position of the first mutated amino acid. Note that m92 is a reconstruction in the TaxH 6 background of the previously described M16 Tax mutant (36). B, partial proteolysis of Tax mutants. Purified wild type and Tax proteins were subjected to digestion with trypsin as described under "Experimental Procedures." Time of digestion is indicated in min. Proteolytic fragments 1, 2, and 3 are designated as in Fig. 3. exception, m92, became more trypsin-resistant. These data suggest that most of the mutations in the linker region (with the exception of m92) do not affect the overall pattern of Tax protein folding.
All of the mutants were then tested in functional assays. DNA binding and cross-linking experiments were performed in the presence of the CREB bZIP domain (Fig. 5, A and B). Cross-linking of m92 was reduced to about 50% of the wild type level (Fig. 5A), consistent with a previous finding that mutation of these amino acids resulted in 50% loss of the ability to stimulate HTLV-I transcription in vivo (36). Cross-linking of m95, which encompasses the histidine residue that is the putative site of adduct formation, was reduced even more severely, to 25% of the wild type level. Cross-linking of m99 and m107, which map to the C-terminal half of the peptide, was less severely affected.
Under the conditions used in these assays, wild type Tax, in addition to its own participation in the cross-linking reaction, also induces a modest increase in cross-linking of the CREB bZIP domain to DNA (Fig. 5B, compare lanes labeled wt and -). Three of the mutants, m92, m95, and m99, showed an impairment in the ability to induce bZIP cross-linking relative to the wild type Tax.
Purified wild type Tax protein is capable of activating transcription in vitro in a cell-free extract (3). Although the magnitude of the in vitro transactivation is not as great as in vivo, it nevertheless occurs by the same TxRE-dependent mechanism (3). We performed in vitro transcription experiments to compare the transactivation potential of wild type Tax and m95, the mutant that showed the most severely impaired function in the cross-linking assays. Reactions were performed using HeLa nuclear extract and a DNA template containing four tandem copies of an HTLV-I TxRE fused to the HTLV-I core promoter. Under these conditions, wild type Tax produced a 2-3-fold stimulation. By contrast, m95 produced little or no stimulation of RNA synthesis.
Taken together, the results of these studies show that mutations in the region of Tax residues 89 -110 impair Tax crosslinking and transactivation function without, in most cases, affecting the overall structure of the Tax protein. Results are consistent with the model that this region of Tax is part of a flexible domain that makes critical contacts with DNA.

DISCUSSION
The HTLV-I Tax protein redirects the binding of CREB by stabilizing its interaction with an extended sequence present in the HTLV-I proviral promoter. Although Tax and CREB together form a complex that has a different DNA binding specificity than CREB alone, previous studies have failed to identify a discrete DNA binding domain in Tax that could account for this change. In the present study, we begin to resolve the mystery by identifying specific peptides within CREB and Tax that form cross-links to DNA in the ternary complex.
The Tax peptide identified in the cross-linking experiments corresponds to residues 89 -110. Partial proteolysis of Tax shows that this peptide lies adjacent to a site that is highly sensitive to trypsin cleavage in the native structure. It appears that this site, which lies between Lys 88 and Val 89 , is part of a linker sequence that joins two Tax structural domains. Previous studies have shown that there is a Zn finger-like domain N-terminal to the cleavage site (4,37). The domain C-terminal to the cleavage site forms a protease-resistant core that mediates dimerization and other functions (7,12,35,38).
When Tax-DNA cross-linking occurs, the trypsin-sensitive site in the linker region becomes trypsin-resistant. The resistance is dependent on native protein structure. It was observed under native conditions (Fig. 3), but not under denaturing conditions used to prepare adducts for N-terminal sequence analysis (Fig.  2). One interpretation is that Tax contains a DNA binding loop that folds into a defined and protease-resistant conformation in the ternary complex. The coupling of protein folding to DNA binding is a common theme observed in many systems (39). An alternative interpretation for the protease resistance is that steric hindrance prevents trypsin from gaining access to the linker region when this region is cross-linked to DNA.
The proposed function of the Tax residues 89 -110 in mediating DNA interactions in the ternary complex is consistent with mutational analysis. Of note, a double point mutation at 92 PP has been shown previously to affect CREB-dependent but not NFBdependent transcriptional activation (36). Deletion of the entire N terminus has a similar but more pronounced effect (35). NFBdependent transcriptional activation reflects a cytosolic event that is not dependent on the DNA binding activity of Tax (35, 40 -44). Reconstruction of the 92 PP mutation in the TaxH 6 background (mutant m92) yielded a 50% reduction in cross-linking consistent with the previously reported phenotype (36). Mutation of a nearby cluster of polar amino acids (mutant m95) produced a more severe phenotype, whereas mutations in the distal part of the peptide were less affected. We suggest that m95 may define the actual sites of molecular contact between Tax and the flanking region of the TxRE.  6. Protein-DNA contacts. A, three-dimensional model of the location of CREB-DNA contacts. The structure is based on the GCN4-DNA crystal structure (34) (see text for details). CREB residue Lys 278 , which is believed to form a cross-link to the DNA, is indicated in yellow. The sites of cross-linker attachment to the DNA are indicated in orange. B, proposed trajectory of Tax. Guanosine 2-amino groups required for ternary complex formation are indicated in dark blue. Sites of cross-linker attachment are indicated in orange. Amino acid residues in CREB (Glu 299 , Met 291 , and Arg 284 ) that contribute to Tax binding are indicated in gray (47). The dashed line indicates the axis of rotational symmetry in the CREB-DNA complex. The solid outline indicates the trajectory that Tax may assume to approach all of these locations.
A structural model of the CREB-DNA complex is helpful in interpreting the cross-linking data. The model is based on the GCN4-DNA crystal structure (23,34). It differs from an earlier version (23) in that the CREB bZIP domain has been extended as a continuous ␣ helix beyond the region observed in the crystal structure, and the positions of the amino acid side chains have been optimized by molecular dynamics simulation. The CREB peptide identified in our experiments corresponds to a short tryptic fragment within the basic portion of the bZIP domain. Residue Lys 278 , which we tentatively identified as a residue that participates in adduct formation, is shown as highlighted in yellow in Fig. 6A and lies within a few angstroms of the site of cross-linker attachment, highlighted in orange.
The potential trajectory of Tax across the surface of CREB can be predicted based on our data and the results of previous studies. Four items of data are relevant. Previous binding experiments show that ternary complex formation is dependent on the 2-amino groups of guanosine residues in the flanking region of the TxRE (Fig. 6B, highlighted in dark blue) (45,46). Cross-linking experiments show that Tax contacts the single phosphate residue highlighted in orange and does not contact several other phosphates in the region (23). Extensive mutational analysis of the CREB bZIP domain has revealed several amino acids that appear to influence Tax binding (Fig. 6B, highlighted in gray). These lie on the outer surface of the DNA binding portion of the bZIP helix (47). Finally, biochemical and two-hybrid studies show that Tax is dimeric (12,35,38). Symmetry considerations suggest that the Tax dimerization interface should lie along the axis of rotational symmetry of the CREB-DNA complex (Fig. 6B, dashed line). Together, these data predict that Tax transits across the outer surface of the CREB bZIP domain in a position similar to that diagrammed in Fig. 6B. As noted previously, the binding of Tax does not appear to induce any change in CREB-DNA contacts (23).
Future studies involving the approaches described here may also help determine the geometry of CBP binding in the CBP-Tax-CREB-DNA quaternary complex. Our data predict that CBP will lie very close to the Tax-DNA interface, because the region of Tax that interacts with CBP (4,7,12,18,35,37,38) lies immediately N-terminal to the region that cross-links to DNA. We note that some of the mutations shown in Fig. 4B were identified by their affect on CBP-Tax-CREB-DNA quaternary complex formation, and although interpreted as affecting Tax-CREB interaction, the possibility of a direct interaction with CBP in this region has not been not excluded.
Tax has pleiotropic effects on the DNA binding properties of a number of transcription factors in addition to CREB, including other bZIP proteins, serum response factor, and certain GAL4 derivatives (5, 20 -22, 48 -54). It may be that the presence of a flexible DNA binding loop in Tax that mediates ternary complex formation is an adaptation that allows Tax to interact with and influence the DNA binding properties of transcription factors that vary widely in primary sequence.