Unraveling the Mechanism of a Potent Transcriptional Activator*

Despite their enormous potential as novel research tools and therapeutic agents, artificial transcription factors (ATFs) that up-regulate transcription robustly in vivo remain elusive. In investigating an ATF that does function exceptionally well in vivo, we uncovered an unexpected relationship between transcription function and a binding interaction between the activation domain and an adjacent region of the DNA binding domain. Disruption of this interaction leads to complete loss of function in vivo, even though the activation domain is still able to bind to its target in the transcriptional machinery. We propose that this interaction parallels those between natural activation domains and their regulatory proteins, concealing the activation domain from solvent and the cellular milieu until it binds to its transcriptional machinery target. Inclusion of this property in the future design of ATFs should enhance their efficacy in vivo.

The expression of specific genes is mediated by transcriptional regulators that bind to DNA and interact with various protein partners to exert transcriptional control. There is increasing evidence that malfunctioning transcriptional regulators play a central role in the etiology of human diseases as diverse as cancer and diabetes (1)(2)(3). The development of artificial transcription factors (ATFs), 1 molecules that target specific genes and control their transcriptional levels, has thus become a highly active area of research. ATFs will likely have a clear role in the study and eventual treatment of transcription-based disorders (2, 4 -7). A significant challenge has been the creation of fully functional artificial transcriptional activators that seek out and robustly up-regulate specific genes (4,5,7).
Endogenous activators are modular proteins minimally composed of two domains as follows: a DNA binding domain that dictates the genes to be targeted and an activation domain that governs the nature and the extent of the transcriptional response through interactions with the transcriptional machinery ( Fig. 1A) (6). The two domains typically operate in an independent fashion such that the DNA binding domain of one protein can be attached to the activation domain of another to generate a fully functional activator. The nature of the linker connecting the two domains is variable but most often consists of a dimerization domain that facilitates specific and high affinity DNA binding (6). Inspired by the architecture of natural activators, artificial activators have typically been generated through a modular replacement strategy in which the endogenous DNA binding domain is replaced with an engineered counterpart with desired DNA targeting properties (4,7,8). By using this strategy, protein, oligonucleotide, and small molecule DNA binding domains have been used in the construction of ATFs that function well in vitro and/or in cell culture (4 -11).
In contrast to the diversity of DNA binding domains used in ATF construction, it has proven much more challenging to identify non-natural activation domains that function robustly in vivo (4,5,7). In fact, most ATFs that elicit strong expression of a targeted gene in cells utilize activation domains that are derived from natural transcription factors, such as VP16 or Gal4 (4,10,(12)(13)(14)(15)(16). Attempts to develop non-natural activation domains have predominantly yielded peptides or, more recently, RNA molecules that function only moderately relative to natural systems (11,(17)(18)(19)(20)(21)(22)(23)(24)(25)(26). There are, however, two artificial activators that are exceptions to this trend and function strongly in vivo; these thus serve as excellent models for the design or discovery of potent artificial activation domains. One potent activation domain is P201, a hydrophobic peptide (YL-LPTCIP, see Fig. 1B) that when coupled to the DNA binding domain of Gal4 (residues 1-100) activates transcription to levels that are comparable to those obtained by some of the most potent natural activators ( Fig. 1) (19,20). In contrast to natural activators that typically interact with a number of transcriptional machinery proteins, the function of P201 is most likely mediated by targeting the Gal11 component of the transcriptional machinery. In support of this, a single point mutation in Gal11 (T322K) abolishes the ability of P201 to activate transcription in vivo and the ability to bind to that protein in vitro (20).
The second example of a potent artificial activation is observed when Gal4(1-100), a fragment of the transcriptional activator Gal4, is present in a yeast strain bearing a mutated version of Gal11. The first 50 residues of Gal4 compose the minimal DNA binding domain, and the next 50 residues form an extended dimerization domain (see Fig. 1B). In normal yeast strains, Gal4(1-100) is an inert DNA-binding protein. How-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ever, when a single point mutation (N342V) is introduced into Gal11, Gal4(1-100) stimulates high levels of transcription (27)(28)(29). Several studies indicate that this mutant version of Gal11, called "potentiator" or Gal11P, retains the normal function of Gal11 but is able to interact with the dimerization domain of Gal4 (Gal4 dd , residues 50 -100; see Fig. 1, B and D) (28). Structure-activity studies indicate that the primary interactions with Gal11P occur in helix 3 and loop 1 of Gal4 dd , both of which contain a number of hydrophobic residues (28).
Gal11, a primary target of two of the most potent artificial activators, is also one of the targets of natural yeast activators (20, 27, 29 -32). Gal11 resides in the mediator complex; this multiprotein complex interacts with RNA polymerase II and is thought to mediate the positive or negative signals of transcriptional regulators bound to their respective genes (33,34). In support of this model, a number of mediator subunits have been found to interact with transcriptional activators (35)(36)(37), and several of the targeted components, including Gal11, reside in the tail of this tri-lobed complex (Fig. 1C) (38,39). Because both of the potent artificial activators described above function through an interaction with Gal11, this suggests that Gal11 could be a privileged target for the development of potent activation domains.
Recent evidence has emerged indicating that exclusive targeting of Gal11 does not provide potent transcriptional activation domains. For example, peptides selected specifically for their ability to bind to Gal11 function only modestly as transcriptional activators despite affinities for Gal11 comparable with natural activation domains (23). This suggests that the binding interaction with Gal11 is unlikely to be the sole contributor to the unusually potent function of P201. To gain insight into additional properties of P201 that make it a robust artificial activator, we examined its ability to activate transcription in yeast strains bearing various mutants of Gal11 and to bind these Gal11 mutants in vitro. These experiments indicate the existence of a second binding interaction that is essential for P201-mediated activation. Remarkably, this interaction is not with a transcriptional machinery protein but with the hydrophobic dimerization domain of Gal4 (Gal4 dd ). We propose that this dynamic interaction between Gal4 dd and P201 plays a role analogous to the interactions between natural activation domains and their regulatory proteins, preserving the hydrophobic activation sequences from nonproductive binding events with other cellular proteins. This finding provides an important guiding principle for the future design of robust artificial activators.
Expression and Purification of GST-Gal11(186 -619)-The plasmid pGEXVsh1 was transformed into chemically competent BL21(DE3) pL-ysE Escherichia coli (Invitrogen), and cells were plated onto LB agar plates supplemented with ampicillin (0.1 mg/ml) and chloramphenicol (0.034 mg/ml). Cultures (50 ml) from single colonies were grown overnight at 37°C (275 rpm) in LB supplemented with ampicillin (0.1 mg/ml) and chloramphenicol (0.034 mg/ml) before addition to 1 liter of LB supplemented with ampicillin (0.1 mg/ml). After 3 h, the cultures were cooled to 16°C, and expression was induced with isopropyl 1-thio-␤-D-galactopyranoside (final concentration 0.5 mM) for 5 h. The cell pellet was lysed using sonication, and the GST-tagged protein was isolated from the cell lysate using glutathione-Sepharose beads (Amersham Biosciences). Elution from the beads was accomplished with 50 mM Tris⅐HCl buffer, pH 8.0, containing 15 mM glutathione and 0.1% Nonidet P-40. The protein solution thus obtained was concentrated to ϳ20 -25 M and the buffer exchanged to storage buffer (phosphatebuffered saline, pH 7.4, 1 mM dithiothreitol, 10% glycerol (v/v), and 0.01% Nonidet P-40) using a Millipore Ultrafree centrifugal filter device. The protein solution was stored in 50-l aliquots at Ϫ80°C until needed. The protein concentration was measured by using a Bradford  28) with key structural elements denoted (␣-helices, ␣1-3 and loops L1 and L2) is shown along with a scaled representation of FVQDϩYLLPTCIP (Gal4(97-100)ϩP201). C, the composite activator Gal4(1-100)ϩP201 (in green) was modeled by superimposing the crystal structure of Gal4 residues 1-64 (59) on the NMR-derived structure of Gal4 residues 53-97 (28). P201 and residues 98 -100 of Gal4 are shown as fully extended peptides (brown). RNA polymerase II holoenzyme structure (ϳ36 Å) is constructed from electron microscopy (39). The polymerase is colored in turquoise, and the head (H) and middle (M) lobes of the Mediator complex are colored in dark blue. The tail (T) domain of the Mediator is colored in purple; Gal11 the target of the artificial activator as well as several natural activators is a component of this module. The Gal4-binding sites can be separated from the promoter (TATA box) by variable distances, and this is represented by a gap in the DNA template. The interaction between the activator and the holoenzyme is sufficient to elicit the expression from the promoter. However, additional factors that are not shown in this illustration such as the TATA box-binding protein and other general transcription factors (TFIIB, TFIIF, TFIIE, TFIIH, and perhaps TFIIA) are required for transcription (27). D, the replacement of asparagine 342 in Gal11 (purple) with valine (shown as a green patch within Gal11) leads to a binding interaction with the dimerization domain of Gal4 (green). This interaction converts Gal4 dd into a potent artificial activator. assay (Bio-Rad) with bovine serum albumin as the standard. The identity and purity of the fusion protein was verified by reducing SDS-PAGE with appropriate molecular weight standards.
␤-Galactosidase Assays-For quantitative measurement of activity, freshly transformed colonies were used to inoculate 5-ml cultures of SC media containing 2% raffinose, 2% galactose but lacking uracil, histidine, and leucine. The cultures were incubated overnight at 30°C at 250 rpm. Following incubation, these cultures were used to inoculate 5-ml cultures of SC media, which were subsequently incubated overnight at 30°C with agitation to an A 600 of 0.6 -0.9. The yeast cells were harvested, and the culture was then lysed with glass beads in Z buffer (for 500 ml: 4.26 g of sodium phosphate, 2.76 g of sodium phosphate monobasic, 0.37 g of potassium chloride, 0.123 g of magnesium sulfate, pH 7.0) for 5 min in the cold. A portion of the cell extract was used to measure ␤-galactosidase activity via incubation with o-nitrophenyl-␤-D-galactopyranoside at 4 mg/ml in Z buffer. The reaction was stopped by adding 1 M Na 2 CO 3 , and the A 420 was measured. The activity reported was normalized to the total protein concentration of the extract, which was measured using a Bradford assay kit (Bio-Rad) with bovine serum albumin as the standard.
Peptide Synthesis and Fluorescent Labeling-XL Y (LTGLFVQDYL-LPTCIP) and XL R (LTGLFVQDRLLPTCIP) were synthesized on Rink amide resin from Fmoc (N-(9-fluorenyl)methoxycarbonyl)-protected amino acids, using standard protocols as in Ref. 61. At the conclusion of the synthesis, the peptides were fully deprotected, precipitated with cold ether, and purified to homogeneity using reversed-phase high pressure liquid chromatography (C18 column with a gradient solvent system: buffer A, 0.1% trifluoroacetic acid; buffer B, CH 3 CN). Each peptide was characterized using electrospray mass spectrometry. The peptides were labeled at the N terminus with fluorescein using 5/6carboxyfluorescein succinimidyl ester (Pierce) in accordance with the manufacturer's instructions, purified using reversed-phase high pressure liquid chromatography, and characterized using electrospray mass spectrometry. The labeled peptides were divided into 2-nmol aliquots and stored as dry pellets at Ϫ80°C.
Dissociation Constant Measurements-Dissociation constant measurements were carried out on a Spex Fluoromax-2 fluorimeter at room temperature. Prior to each experiment, a 2-nmol aliquot of fluoresceinlabeled XL Y or XL R was resuspended in 1 ml of Buffer S (phosphatebuffered saline, pH 7.4, 1 mM dithiothreitol, 10% glycerol (v/v), and 0.01% Nonidet P-40), and the concentration of the solution was determined by using the absorbance of fluorescein. The concentration of the solution was adjusted to 1.25 M with Buffer S, and 1 l of this solution was added to 49 l of GSTϩGal11, GSTϩGal11P, or GSTϩGal11DM (20 M final protein concentration for most measurements). After incubation for 10 min at room temperature, the first measurement was taken (excitation wavelength, 487 nm; emission wavelength, 516 nm). For each successive measurement, the solution was diluted by addition of Buffer S containing 25 nM fluorescein-labeled XL Y or XL R (to keep the concentration of the labeled component constant throughout the experiment), manually mixed, and incubated for 10 min at room temperature. At concentrations above 25 M, GSTϩGal11(186 -619) aggregates to a significant degree, and dissociation constants greater than 10 M thus were not measured. The same procedure was used for binding experiments with GSTϩGal4 dd or GSTϩGal4 dd ϩP201, except that a 10 nM concentration of XL Y or XL R was used. The data obtained were plotted in Origin 7.0 and fit to the following equation using the Levenberg-Marquardt least squares method: where a is the amplitude of the curve and b is the anisotropy of the free labeled peptide. All experiments were performed in triplicate, and errors were calculated using the mean Ϯ S.D. of the three experiments.
k off Measurements-All k off determinations were carried out at 25°C on a Beacon 2000 polarization system (Invitrogen). For the Gal11-XL Y interaction, fluorescein-labeled XL Y (25 nM in Buffer S) was incubated with 18 M GSTϩGal11(186 -619) until equilibrium was reached, as determined by anisotropy measurements. Following equilibration, a 1700-fold excess of unlabeled XL Y was added by mixing, and anisotropy measurements were taken every 11.5 s to monitor the release of the labeled peptide. An identical procedure was used for the XL Y ⅐Gal4 dd experiment, except that a 10 nM concentration of fluorescein-labeled XL Y was used and a 50 M concentration of GSTϩGal4 dd . A control experiment was carried out to determine the loss in anisotropy due solely to the dilution of the protein and accounted for in the data analysis. All data were plotted using Origin 7.0 and fit to a single phase exponential decay using the Levenberg-Marquardt least squares method.

RESULTS
In Vivo Activation-P201 was originally identified by a screen in Saccharomyces cerevisiae of random 8-residue peptides attached to Gal4(1-100). We initially postulated that the potent function of Gal4(1-100)ϩP201 could be due to some residual affinity of Gal4 dd (residues 50 -100 of Gal4) for wildtype Gal11. Thus the overall level of activation observed with Gal4(1-100)ϩP201 would be the result of two binding interactions, P201⅐Gal11 and Gal4 dd ⅐Gal11. To delineate the relative contributions of P201 and Gal4 dd , we first looked at the function of Gal4(1-100)ϩP201 and Gal4(1-100) in yeast strains bearing wild-type Gal11 or its mutants. The first of the key Gal11 mutations is the substitution of threonine 322 to lysine (T322K), a change that disrupts binding with P201. The second mutation results from replacement of asparagine 342 with a valine (N342V, also referred to as the Gal11P mutation), and this alteration promotes an interaction with Gal4 dd (20,28). We also tested the function of Gal4(1-100)ϩP201 and Gal4(1-100) in a yeast strain bearing the Gal11 double mutant (Gal11 DM) that has both the T322K and N342V mutations.
Gal4(1-100)ϩP201 functions well in the wild-type Gal11bearing strain, and as expected, the T322K substitution in Gal11 abolishes P201 activity (Fig. 2, A and B). In contrast, Gal4 dd does not activate transcription in yeast strains containing either wild-type Gal11 or Gal11 with the T322K mutation ( Fig. 2, C and D). Introduction of a second mutation in Gal11 (N342V, Gal11DM) restores activation function to Gal4 dd to levels nearly identical to those observed with Gal11P (Fig. 2, E and F). The latter result demonstrates that the T322K mutation has little effect on the ability of Gal4 dd to activate transcription. The Gal11P mutant contains binding sites for both Gal4 dd and for P201, but as illustrated in Fig. 2G, the levels of transcription elicited by Gal4(1-100)ϩP201 in this strain are similar to those obtained in the presence of wild-type Gal11 ( Fig. 2A).
In the final experiment, we examined the function of Gal4(1-100)ϩP201 in yeast bearing Gal11DM to decouple the effect of P201 on activation mediated by Gal4 dd (Fig. 2H). The T322K mutation in Gal11 abrogates P201-mediated activation ( Fig.  2B) but has little effect on the function of Gal4 dd (Fig. 2F). It would be expected, then, that Gal4(1-100)ϩP201 would activate transcription in Gal11DM yeast primarily because of the interaction of Gal4 dd with the binding surface generated by the N342V mutation (Fig. 2H). However, activation was significantly diminished in this strain (Fig. 2H). In other words, P201 somehow blocks the function of Gal4 dd . Identical results were obtained in yeast strains bearing a single binding site for Gal4 (see supplemental Fig. 1 for details). Taken together, these results suggest that although Gal4 dd and P201 do not appear to function cooperatively, there is a functional interplay between the two peptides. Subsequent experiments were therefore designed to first more clearly define the P201⅐Gal11 interaction, and second to investigate a possible Gal4 dd ⅐P201 interaction.
Defining the Minimal Gal11-binding Peptide-The results from the in vivo activation studies above indicate that Gal4 dd does not interact with wild-type Gal11 in the context of Gal4(1-100)ϩP201. However, previous mutagenesis studies have shown that residue 97 in helix 3 of Gal4 dd is important for P201 function, and there thus remained the possibility that some portion of Gal4 dd is necessary for P201 to interact effectively with Gal11 (19). To examine the role of helix 3 of Gal4 dd in P201-mediated activation, we carried out alanine-scanning mutagenesis of this region (Fig. 3A). To avoid any effect on DNA binding, we replaced the DNA binding domain of Gal4 with the DNA binding domain of the bacterial protein LexA. As illustrated in Fig. 3A, the replacement of residues 96 -98 in Gal4 dd with alanine eliminated P201-mediated transcription, whereas substitution of residues 92-94 had modest effects. The in vivo results were correlated in vitro by examining the interaction of the peptides P201, Gal4(96 -100)ϩP201, and Gal4(93-100)ϩP201 with Gal11. These experiments revealed that only Gal4(93-100)ϩP201 binds to Gal11 with any appreciable affinity ( Fig. 3B and data not shown). Thus, both the in vivo functional data and the in vitro binding experiments are consistent with a requirement for both the P201 sequence and residues 93-100 of Gal4 dd for interaction with Gal11 and for transcriptional activation. All subsequent binding experiments were therefore carried out with this minimal peptide (Gal4(93-100)ϩP201, LTGLFVQDYLLPTCIP), referred to as XL Y .
Interaction of XL Y with Gal4 dd and Gal11-The data of Fig.  2H suggest that XL Y binds directly to Gal4 dd and interferes with its ability to interact with Gal11DM. To probe a possible interaction between XL Y and Gal4 dd and to measure the affinity of XL Y for Gal11, we designed a series of fluorescence polarization assays. For this purpose, XL Y was synthesized by standard methods and labeled with carboxyfluorescein (61). Gal4 dd and the central region of Gal11, residues 186 -619, were expressed and purified as GST fusion proteins. The central region of Gal11 was used in these experiments because it is targeted by XL Y as well as Gal4 dd ; it can be readily generated using a bacterial expression system, and it is the largest fragment of Gal11 that is well behaved at micromolar concentrations in vitro.
As shown in Fig. 3, B and C, XL Y binds to both Gal11 and Gal4 dd with dissociation constants in the micromolar range and interacts with Gal11 ϳ2-fold more tightly (2.2 versus 5 M, respectively). We also measured the affinity of XL Y for Gal4 dd ϩP201. This interaction is roughly 3.5-fold weaker than the interaction with Gal4 dd lacking the fused P201 sequence (compare Fig. 3, C with D). This result indicates that the fused P201 peptide does not inhibit the XL Y ⅐Gal4 dd interaction. As a control, we also examined the binding properties of a P201 sequence containing an arginine rather than a tyrosine residue (XL R , LTGLFVQDRLLPTCIP); this sequence is inactive as a transcriptional regulator in vivo (19). In our experiments XL R did not bind to Gal11, Gal4 dd , or to Gal4 dd ϩP201 under any conditions examined (data not shown).
Next, we tested to see if XL Y would interact with Gal11 proteins bearing either the 11P (N342V) mutation singly or in combination with the T322K mutation (Gal11DM). Consistent with the in vivo data, we found that introduction of the 11P mutation causes a 3-fold reduction in affinity (Fig. 3E). Also consistent with the lack of activation in vivo (Fig. 2, B and H), XL Y does not bind to Gal11DM within the limits of our assay. Furthermore, the inactive version of Gal4(93-100)ϩP201, XL R , shows no measurable binding interactions with either Gal11P or Gal11DM (data not shown).
NMR studies of Gal4 dd suggest that it has several solventexposed hydrophobic residues. XL Y is also a very hydrophobic peptide. To probe the specificity of the Gal4 dd ⅐XL Y interaction, we examined the interaction of Gal4 dd with another hydrophobic Gal11-binding peptide, AHYYYPSE (23). This peptide has a similar affinity for Gal11 (4.8 M) relative to XL Y (2.2 M) but is far less robust than XL Y as a transcriptional activation domain (23). Moreover, as illustrated in Fig. 3F, this peptide exhibits no detectable binding to Gal4 dd . Consistent with the inability to bind Gal4 dd in vitro, we find that this peptide, unlike XL Y , does not interfere with the ability of Gal4 dd to stimulate transcription in strains bearing the Gal11DM mutant (see supplemental Fig. 2 for details). In further experiments we found that XL Y does not bind nonspecifically to unrelated proteins such as bovine gamma globulin that display affinity for hydrophobic surfaces (62) (Fig. 3G). Thus, the Gal4 dd ⅐XL Y binding event cannot simply be ascribed to nonspecific hydrophobic interactions.
Interaction of Gal4 dd with Gal11-We next determined the strength of the interaction between Gal4 dd and the Gal11 variants by electrophoretic mobility shift assays. Fluorescence polarization was not used for these experiments because the large size of the two components (Gal4 dd bound to DNA ϩ Gal11 and its mutants) leads to a minimal change in anisotropy upon complex formation. The Gal11 proteins and the assay conditions (buffer, temperature, and incubation times) employed were identical to those used in the fluorescence polarization assays outlined above to allow for meaningful comparisons of

FIG. 2. A-H, levels of transcription elicited by the two artificial activators in a strain bearing different Gal11 alleles.
The reporter bears two cognate 17-bp Gal4-binding sites 191 bp upstream of the Gal1 promoter (TATA element) fused to the lacZ reporter gene. This position is analogous to the location of the natural upstream activating sequence (UASg) at this promoter. The transcriptional activation potential of the activators is determined by measuring the activity of ␤-galactosidase (␤-gal) in strains transformed with single copy plasmids expressing different alleles of Gal11 at endogenous levels. Only the hydrophobic loop (68 -72), the hydrophilic loop, and the second and third helices of Gal4 are shown. The P201 peptide (along with residues 97-100 of Gal4) is shown as an extended peptide. The hydrophobic loop and the third helix (marked by brackets) interact with Gal11P or Gal11DM. Consistent with previous results, Gal4 dd is inert in the absence of the 11P mutation. Various alleles of Gal11 are shown. The N342V mutation in Gal11P and Gal11DM is shown as a gray patch and the T342K mutation as a white patch. The ␤-galactosidase activity is reported based on quadruplicate independent measurements (see "Materials and Methods"). In parallel control experiments, activation by full-length natural Gal4 (residues 1-881) was not affected in the four strains bearing different Gal11 mutants, indicating that these alleles do not alter the natural Gal11 functions (data not shown, see also Ref. 20) .  FIG. 3. Binding behavior of the minimal activator peptide, XL Y . A, results from alanine-scanning mutagenesis studies on helix 3 of Gal4 dd . The LexA DNA binding domain (1-87) was fused inframe to Gal4(40 -100)ϩP201, and transcriptional activation was monitored by a reporter bearing two LexA operators 191 bp upstream of the Gal1 promoter driving the expression of the lacZ gene. The use of the LexA DNA binding domain ensures that any loss of activity upon mutagenesis is not due to loss of DNA binding of Gal4ϩP201 (for example see Ref. 28). The ␤-galactosidase (␤-gal) activity averaged from quadruplicate measurements is reported, and the S.D. did not exceed 15%. As shown, residues 93-100 of this region are essential for P201-mediated transcriptional activation. This, in combination with the binding data of 3B, indicates that the minimal functional activator is Gal4(93-100)ϩP201, named XL Y . B, determination of the dissociation constant for the XL Y ⅐Gal11 interaction by fluorescence polarization. Fluorescein-labeled XL Y (Gal4(93-100)ϩ P201) at a constant concentration was incubated at 25°C with increasing concentrations of Gal11. The fluorescence polarization at each concentration was measured, and the resulting data fit using the Levenberg-Marquardt least squares method to obtain the dissociation constants. Each experiment was performed in triplicate (R 2 Ͼ 0.98) with the error indicated. C, determination of the dissociation constant for the XL Y ⅐Gal4 dd interaction by fluorescence polarization. Conditions were identical to those used for B. D, determination of the dissociation constant for the XL Y ⅐Gal dd ϩP201 interaction by fluorescence polarization. Conditions were identical to those used for B. E, determination of the dissociation constant for the XL Y ⅐Gal11P interaction by fluorescence polarization. Conditions were identical to those used for B. F, the hydrophobic activator peptide AHYYYPSE does not interact with Gal4 dd as monitored by fluorescence polarization. Conditions were identical to those used for B. G, XL Y does not interact with bovine gamma globulin as monitored by fluorescence polarization. Conditions were identical to those used for B. the binding affinities measured by the two techniques. A concentration of 30 nM Gal4(1-100) was sufficient to bind saturably to its cognate DNA site, and this Gal4-DNA complex was incubated with increasing concentrations of wild-type Gal11, Gal11P, or Gal11DM. As demonstrated in Fig. 4, Gal4 dd does not interact with wild-type Gal11, consistent with the lack of activation observed in vivo. Introduction of the 11P mutation, however, leads to half of the Gal4-DNA complex being further shifted to a ternary DNA-Gal4-Gal11P complex at a Gal11P concentration of ϳ18 M. This binding is virtually unchanged by the introduction of the second (T322K) mutation (compare unnumbered lanes 10 and 16 in Fig. 4). Because Gal11 and its variants begin to aggregate significantly and precipitate at concentrations above 24 M, the concentration of Gal11 could not be further increased to determine the complete binding isotherm; however, the half-maximal binding at 18 M provides a reasonable indication of the K D . Although it is difficult to make a direct quantitative comparison because different techniques were used to assess binding, the results suggest that the affinity of Gal4 dd for Gal11P is somewhat weaker than for XL Y . It is thus not surprising that the Gal4 dd ⅐XL Y interaction inhibits Gal4 dd -mediated transcription but does not negatively impact XL Y -mediated activation.
Site of the Gal4 dd -XL y Interaction-Both the in vivo activation data of Fig. 2 and the in vitro binding data of Fig. 3 are consistent with an interaction between Gal4 dd and XL Y , but two key questions regarding this interaction remained. 1) Does this interaction contribute to XL Y -mediated activation? 2) Where within Gal4 dd is the site of this interaction? We addressed these questions by a combination of deletion analysis and mutagenesis experiments (Fig. 5).
In the first set of experiments, the P201 sequence was at-tached to the LexA DNA binding domain along with fragments of Gal4 dd of increasing length. As depicted in Fig. 5A, XL Ymediated transcriptional activation was not observed until the entire Gal4 dd region (residues 50 -100) was incorporated into the protein construct. The last result of this panel points to the importance of helix 1 and loop 1 for this process, because the construct containing only helices 2 and 3 (Gal4(73-100)) showed no activity.
To refine this picture further, additional alanine-substitution mutagenesis experiments of Gal4 dd were carried out. For this purpose, we again employed the chimeric LexA-Gal4 protein described in Fig. 3A to avoid any impact on DNA binding. Using the structural model of Gal4 dd to guide these efforts, we mutated residues in loop1 (L1), helix 2 (␣2), loop2 (L2), and helix 3 (␣3) that are important for Gal4 dd ⅐Gal11P interactions (see Fig. 1B and Ref. 28). As shown in Fig. 5B, changing residues 68, 69, 72, and 75 to alanine has a profound effect on XL Y -mediated activation, with little or no activity observed, suggesting that loop 1 is the likely site of interaction with XL Y . To probe this further, we examined the binding of XL Y to Gal4 dd containing one of those mutations, 69A, again using fluorescence anisotropy, and no detectable binding was observed (see supplemental Fig. 3 for details). As summarized in Fig. 5C, several of these residues are also essential for the Gal4 dd -Gal11P-binding interaction (28). Most interestingly, not all substitutions affected XL Y function; L86A and I89A mutants were functional (Ͼ30% active). Both of these residues are essential for the Gal11P⅐Gal4 dd interaction in vitro and for Gal4 dd transcriptional activity in vivo (28). In contrast, residues Phe-68 and Leu-77 flanking loop 1 and Asp-84 at the C-terminal end of ␣2 are essential for XL Y function but not for the Gal4 dd ⅐Gal11P interaction.
The residues of Gal4 dd that are important for XL Y function and for Gal11P interaction are also summarized in Fig. 5C. There are several residues in loop 1 that are essential both for XL Y function and for the ability of Gal4 dd to interact with Gal11P. However, residues that are essential for XL Y function are not entirely coincident with those required for the Gal4 dd ⅐Gal11P interaction. Taken together, the results suggest that Gal4 dd uses distinct but overlapping surfaces along its length to interact with Gal11P and with XL Y .
Dynamics of Binding-To further characterize the XL Y ⅐Gal4 dd and the XL Y ⅐Gal11 interactions, we measured the off-rates again by using fluorescence polarization methods (40). Fluorescein-labeled XL Y was incubated with either Gal11 or Gal4(52-100) under conditions of saturated binding (10 -25 nM XL Y , 18 M Gal11, or 50 M Gal4(52-100)). Following equilibration at room temperature, a 1700-fold excess of unlabeled XL Y peptide was added as a single aliquot. This leads to a measurable decrease in polarization as the bound fluorescent peptide is exchanged with the unlabeled peptide. As shown in Fig. 6, we find the dissociation of XL Y from Gal4(52-100) is faster than the dissociation from Gal11. In fact, virtually all of the exchange occurred prior to the first measurement (11.5 s). Thus, the k off value of 0.24 s Ϫ1 for XL Y dissociating from Gal4(52-100) calculated from these data represent the lower limit of this rate. In contrast, the dissociation rate for the Gal11⅐XL Y interaction is significantly slower, with a k off of 0.06 s Ϫ1 . In contrast, the k off for Gal4 dd dissociating from Gal11P has been measured previously as Ͼ5 s Ϫ1 by NMR titration (28). Thus, although the measurements between methods are not directly comparable, this off-rate is significantly faster than the XL Y ⅐Gal4 dd off-rate reported in Fig. 6.

DISCUSSION
The hydrophobic octapeptide P201 functions as an unusually robust transcriptional activator when fused to the Gal4 DNA binding and dimerization domains (residues 1-100) (19,20). Our study demonstrates that in addition to the 8 amino acids that make up P201, residues 93-100 of Gal4 are essential for transcriptional activation and also for interaction with Gal11 (Fig. 3A). The minimal functional activator is thus the 16residue sequence Gal4(93-100)ϩP201, named XL Y . The potency of XL Y as an activation domain is particularly remarkable when one considers that other artificial activator peptides of greater size and surface area up-regulate only moderately (17,22,23). Moreover, the potency of XL Y is not simply a function of targeting Gal11, because artificial activator peptides of similar size and even similar affinity for Gal11 do not activate nearly as well as XL Y (23).
The first indication that an interaction with the dimerization domain of Gal4 might contribute to functional potency of XL Y came from studies in yeast strains bearing different mutants of Gal11 (Fig. 2). The most revealing result was obtained with Gal4(1-100)ϩP201 in a yeast strain with Gal11 containing two point mutations (Gal11DM) (Fig. 2H). One of the mutations (N342V) generates a surface that is targeted by Gal4 dd , whereas the second mutation (T322K) abrogates binding to XL Y . If in the context of Gal4(1-100)ϩP201, Gal4 dd and XL Y each target Gal11 independent of one another, then in the Gal11DM strain one would expect to observe activation levels comparable with those obtained with Gal4 dd alone in either the  Fig. 3A experiments. The activity reported is an average of quadruplicate independent measurements, and in all cases the S.D. was less than 15%. B, alanine substitution mutagenesis was performed on key structural elements of the Gal4 dd . The LexA-Gal4(50 -100)ϩP201 chimera shown in Fig. 3A was used for this study, and transcriptional activation was measured using the same reporter setup as the Fig. 3A experiments. The activity reported is an average of quadruplicate independent measurements, and in all cases the S.D. was less than 15%. Several mutations dramatically affected the activation properties of XL Y (Gal4(93-100)ϩP201). C, alanine mutagenesis data from Figs. 3A and 5A are summarized and compared with data from Ref. 28. The symbols L1 and L2 are the loops, and ␣1-3 represents the three helices as shown in Fig. 1A and derived by Hidalgo et al. (28). The residues that were replaced with alanine are in boldface type, and their effect on activation by XL Y or Gal4 dd in the respective strains (wild type Gal11 and Gal11P) are marked by a plus or minus symbol. Any mutation that negatively impacted functional activity more than 20%, implicating that residue in the XL Y ⅐Gal4 dd -binding interaction, is scored with a minus sign. The mutants that had an effect on one interaction but not the other are numbered. ␤-gal, ␤-galactosidase. Fluorescein-labeled XL Y was incubated with Gal11 (A) or Gal4 dd (B) such that saturated binding was observed at 25°C. The complexes thus obtained were rapidly mixed with a 1700-fold excess of unlabeled XL Y peptide, and the change in polarization due to displacement of the fluorescently labeled XL Y over time was monitored. The data thus obtained were fit to a single exponential decay to determine the respective off rates. The data displayed are the average of three individual experiments.
Gal11P or Gal11DM strains. Instead, we observed a significant decrease in activation, suggesting that XL Y blocks the interaction between the N342V surface of Gal11DM and Gal4 dd . Consistent with this interpretation, the XL Y ⅐Gal4 dd -binding interaction (5 M) is qualitatively stronger than that between Gal4 dd and Gal11P (ϳ18 M by electrophoretic mobility shift assays).
Additional mutagenesis experiments shown in Fig. 5 reveal that the solvent-exposed hydrophobic loop 1 of Gal4 dd is the most likely site of the XL Y ⅐Gal4 dd interaction and, furthermore, that this interaction is essential for the transcriptional activation function of XL Y . The mutation of several loop 1 residues abrogates XL Y -mediated transcription. Moreover, Gal4 dd , in which residue 69 of loop 1 is replaced by an alanine, shows no detectable binding interaction with XL Y (see supplemental Fig.  3). Thus, the XL Y ⅐Gal4 dd interaction is just as essential as the XL Y ⅐Gal11-binding interaction for XL Y to function as an activator. Consistent with this picture, residues 97-100 of Gal4 dd appear conformationally mobile in the NMR structure of Gal4 (50 -103), and if fully extended, these residues along with the eight residues of P201 would span ϳ36 Å, more than sufficient to reach the solvent-exposed hydrophobic loop 1 (Fig. 1B).
These data also provide a clear explanation for how XL Y inhibits Gal4 dd -mediated activation in yeast strains bearing Gal11DM. Structure-activity studies of Gal4 dd employing alanine-scanning mutagenesis as well as binding studies indicate that the hydrophobic residues in loop 1 and the third helix are required for interaction with Gal11P (28). The mutagenesis results of Fig. 5 show that the function of XL Y is dependent on several residues of Gal4 dd that are also necessary for interaction with Gal11P. The data also demonstrate that although there is significant overlap, not all residues important for XL Y function are important for the Gal4 dd ⅐Gal11P interaction and vice versa. Thus, Gal4 dd interacts with XL Y and with Gal11P using two partially overlapping surfaces. More importantly, the results show that interaction with Gal4 dd is essential for the ability of XL Y to function as a robust activator in vivo. The functional importance of this XL Y ⅐Gal4 dd interaction is further supported by the observation that a similar peptide (AHYYYPSE) that interacts with Gal11 nearly as well as XL Y (23) but does not bind to Gal4 dd is not able to activate transcription to the same extent as XL Y (Fig. 3F and supplemental Fig. 2).
The ϳ2-fold weaker interaction between XL Y and Gal4 dd relative to Gal11⅐XL Y (Fig. 3) provides some explanation as to why the XL Y ⅐Gal4 dd interaction does not inhibit XL Y -mediated transcription. In addition, the kinetic data of Fig. 6 are suggestive of a mechanism where rapid interconversion between closed and open states of XL Y correlates with inactive and active states of the activator (discussed below). In future experiments, the bound and unbound states of XL Y will be characterized in more detail through stopped-flow experiments of the constructs bound to DNA to probe the effects of cooperative DNA binding and interactions between adjacent DNA-bound activators on this process.
By having identified this unexpected XL Y ⅐Gal4 dd interaction, we are faced with the question as to why this interaction is necessary for XL Y to function as a potent activator. One possible explanation is that the XL Y ⅐Gal4 dd complex presents XL Y in an appropriate conformation for the interaction with Gal11. However, several observations lead us to disfavor this explanation (although further experimentation will be required to rigorously exclude it). Perhaps most important is the transferability of this phenomenon. For example, XL Y also functions as a potent transcriptional activator when fused to the DNA binding domain of Pho4 (a helix-loop-helix transcription factor), suggesting that the XL Y ⅐Gal4 dd complex is not the active component of XL Y -mediated transcriptional activation (19). Similarly, we have recently identified several other peptidic activation domains unrelated to XL Y that function more strongly (10 -100-fold enhancement) when attached to Gal4(1-100) relative to other DNA binding domains; each of these activation peptides interacts with Gal4 dd , and some do not require Gal11 to function as transcriptional activators. 2 Furthermore, the hydrophobic residues of XL Y that are required for interaction with Gal11 (tyrosine, for example) are also required for XL Y to bind to Gal4 dd . The simultaneous participation of these residues in interactions with Gal4 dd and Gal11, which would be required if the closed complex is the transcriptionally active form, is unlikely.
Taken together, the in vivo activation data and the kinetic and thermodynamic measurements are more consistent with a model in which XL Y when bound to Gal4 dd is inactive and unable to interact with Gal11 (Fig. 7). We propose that it is only in the open or exposed state that the peptide is functionally active, interacting with Gal11, and facilitating the cooperative assembly of the transcriptional machinery at the promoter (Fig. 7).
The concealment and exposure of XL Y are similar to a frequently observed feature of natural activation domains in which the hydrophobic residues that are vital for activator function are masked by either intramolecular or intermolecular interactions with other regulatory proteins (41)(42)(43)(44)(45)(46)(47)(48)(49)(50)(51)(52)(53). For example, in the structure of the p53 activation domain complexed with the inhibitor protein MDM-2, the hydrophobic residues that are known to play a critical role in activation are found buried in a hydrophobic cleft on the surface of MDM-2 (41,42). In response to physiological signals, the residues are released, and the activation domain is capable of stimulating the expression of target genes. There are also several examples of intramolecular concealment of activation domains characterized in yeast such as Leu3 and Put3. In both of these cases, the full-length activator protein is nonfunctional when tested in vivo or in vitro (49 -53). In the presence of a specific metabolite or small molecule, however, the activating region is revealed, and the protein stimulates the expression of genes (49 -53). FIG. 7. A model depicting the role of the XL Y ⅐Gal4 dd interaction in XL Y -mediated transcriptional activation. We propose that in the "closed" conformation, the XL Y peptide is inactive, whereas in the open conformation it can interact with Gal11 and likely other proteins that may lead to nonproductive complexation or even degradation. An assumption inherent to this model is that the affinity of XL Y for Gal4 dd is higher than its affinity for chaperones, the proteolytic machinery, or other undefined factors that target exposed hydrophobic peptides (depicted by the size of the arrows). The assumed difference in affinity would potentially shield the activator to a sufficient degree from nonproductive interactions in the cytoplasm. In the nucleus, the interaction between DNA-bound XL Y and mediator-associated Gal11 would recruit the transcriptional machinery to the promoter. The cooperative assembly of the transcriptional machinery at the promoter would further stabilize the XL Y ⅐Gal11 interaction.
Thus, in these examples another surface within the activator protein has evolutionarily co-evolved to bind and conceal the activating region in the absence of the relevant signal and to expose it when instructed to do so.
One purpose of the concealment or masking of natural activation domains is regulatory, preventing uncontrolled stimulation of transcription. In addition, the masking prevents unproductive binding as well as premature degradation of the activator, thus contributing to the potency of activator function. We propose that it is the latter role that the XL Y ⅐Gal4 dd interaction plays. The closed conformation renders XL Y inaccessible not just to Gal11 but also to other cellular proteins that lead to nonproductive complexation or even degradation of activators. In other words, the XL Y ⅐Gal4 dd interaction enables XL Y to escape cellular surveillance by protein-folding chaperones that bind to unstructured hydrophobic peptides and/or by the machinery that degrades them (54 -56). This model assumes that the affinity of XL Y for Gal4 dd is higher than its affinity for chaperones, the proteolytic machinery, or other undefined cellular factors that interact with exposed hydrophobic peptides in vivo. Thus, in the absence of the interaction with Gal4 dd , interactions with a variety of cellular proteins lead either to premature destruction of XL Y or to nonproductive interactions, both of which would negatively impact the overall effectiveness of XL Y as a transcriptional activation domain. Gal11 has a significant advantage over the other possible XL Ybinding partners because the XL Y ⅐Gal11 interaction initiates cooperative assembly of the transcriptional machinery complex at the promoter, likely a significant driving force. A similar cooperative assembly with chaperones or the proteolytic machinery is, in contrast, highly unlikely.
The transient exposure of hydrophobic residues may also play a role in the second example of a potent artificial activator, Gal4 dd . The NMR structure of Gal4 dd indicates that the key hydrophobic residues in the third helix, critical for interaction with Gal11P, are protected from solvent by dimerization of two Gal4 dd monomers (28). These studies also indicate that the inter-monomer interaction displayed by the third helix of the dimerization domain is weak, and if it is stabilized by the introduction of additional residues of Gal4, the ability of Gal4 dd to function as an activator in vivo and to bind Gal11P in vitro is severely diminished (28,57,58). Thus, in this example of potent artificial activation as well, the hydrophobic residues that interact with a target are likely transiently occluded/ exposed from bulk solvent and most cellular proteins.
In closing, this property of concealment and exposure is unlikely to be restricted to XL Y and Gal4 dd . We propose that any hydrophobic surface that permits a balance of exposure and occlusion of hydrophobic activation domains, comprised of biopolymers, unnatural polymers, or small molecules, would enhance the potency of the artificial activator. In support of this, preliminary results suggest that the ability to transiently interact with such a surface increases the potency of unrelated short peptides that are otherwise weak activators up to 2 orders of magnitude. 2 Inclusion of this property may greatly aid in the design of synthetic transcriptional activators that function robustly in a cellular context.