Adding Specificity to Artificial Transcription Activators

Paramjit S. Aroraa
aDepartment of Chemistry, New York University, New York, New York 10012

Available online 25 March 2005.

In this issue, Mapp and colleagues describe a significant advance in the design of artificial
transcription activators that function in a cell-type-specific manner. [1] The authors show that
peptides selected for binding a component of the yeast transcription complex require its presence
for effective transcriptional activation.

Article Outline
References

An overriding goal for chemical biologists is to develop tools for decoding the complicated
networks of protein-protein interactions that execute the genetic program of an organism. One
intricate and incompletely understood cellular process involves the initiation of transcription that
ultimately leads to the transfer of information from DNA into RNA. Transcriptional activators that
govern the expression of a specified gene are minimally composed of two modules: a sequence-
specific DNA binding domain (DBD) that finds the promoter region of interest, and an activation
domain (AD) that recruits the appropriate cellular machinery to the promoter via protein-protein
interactions (Figure 1) [2].

Full-size image (24K)

High-quality image (96K)
Figure 1. Activation of Gene Transcription
A minimal artificial activator composed of two separate functional domains: a DNA binding domain
(DBD) and an activation domain (AD).

View Within Article

Chemists are aiming to develop synthetic transcription activators (and repressors) that can
selectively modulate the expression of any gene of interest [3]. Recent research efforts have
afforded several ligands for sequence-specific binding of target DNA. These designed DBDs
include pyrrole-imidazole polyamides [4], peptide nucleic acids [5], triplex-forming
oligonucleotides [6], and zinc finger proteins [7]. While the DNA binding properties of these
molecules have been characterized in detail, the precise role of the ADs in transcription has not
been as clearly defined. The ADs are involved in the recruitment of coactivators, chromatin-
modifying enzymes, and other components of the transcriptional machinery. Although several ADs
may bind to a common target in transcriptional machinery, unlike DBDs they do not always share
defined structural motifs [2].
Activation domains comprised of multiple acidic and hydrophobic residues (“acidic activators”)
function effectively in eukaryotes. An important feature of these strong ADs may be their ability to
interact with multiple targets in the transcription complex [2]. But if this promiscuity is a necessary
qualification for a strong AD, it may be difficult to build ADs that are cell-type- and species-
specific. Thus, a key question for the development of next generation of transcription activators is
whether potent activation can be achieved by specifically targeting individual components of the
transcriptional machinery. Several research groups have now begun addressing this fundamental
issue by creating ligands for distinct proteins found in the transcription complex. Montminy and
Kodadek utilized phage display to isolate peptide ligands for p300/CREB binding protein (a histone
acetyltransferase) and yeast repressor (Gal80), respectively, and have shown that these peptides are
strong activators of transcription [[8] and [9]]. Schepartz and coworkers have explored miniature
protein scaffolds that display α-helical motifs to isolate high-affinity ligands for the CREB binding
protein (CBP) [10]. The isolated miniprotein afforded a potent activator of p300/CBP-dependent
transcription when fused to a Gal4 DBD [11]. Uesugi and coworkers used a small-molecule library
to isolate ligands that mimic the AD of ESX (an epithelial-specific transcription factor) [[12] and
[13]]. Subsequently, these small-molecule ADs were fused to a sequence-specific DNA binding
domain derived from pyrrole-imidazole polyamides to create synthetic transcription factors [14].
The Uesugi approach shows that it is possible to generate small-molecule activators by identifying
inhibitors of the transcription factor-target protein interactions. Mapp and coworkers successfully
showed that small-molecule transcription factors may also be constructed by incorporating key
functional groups from acidic activators into isoxazolidine-based scaffolds [15]. This strategy
elegantly translates the amphipathic nature of the acidic activators into the small-molecule regime.
Mapp’s group is also simultaneously pursuing an alternative method for the development of potent
activators [16]. This approach involves screening of synthetic peptide libraries to isolate short ADs
that target specific components of the transcription machinery. In the reported case, peptides that
target Gal11 (Med15), a component in yeast mediator complex, were isolated. A key finding in this
paper was that the binding affinities of the peptides for Gal11 may not be the sole determinants of
their activities, but that binding sites may also play significant roles. Schepartz and coworkers
arrived at a similar conclusion through their studies on the miniprotein activators [11]. Together
these efforts pave the way to the development of potent activators by systematically analyzing and
targeting protein surfaces found in the transcriptional complex.
In this issue of Chemistry & Biology, Mapp and coworkers show that the activity of the isolated
peptide-ADs is sensitive to the nature and location of the DBD and to the presentation of the ADs
on the DBD. These experiments raise a number of intriguing questions regarding the location of the
DBD relative to the initiation site and its effect on transcription activation. Significantly, the authors
show that the artificial activators are only functional in the cells that contain the target protein. This
level of specificity is uncommon in natural activators and constitutes a promising advance in the
field [17].
The work summarized above describes continuing fundamental advances at the interface of
chemistry and biology toward artificial control of gene expression. The latest addition by Mapp and
colleagues provides a concrete foundation for designing a new generation of cell-type-specific and
species-specific artificial activators. Certainly, much remains to be elucidated, because the ultimate
goal is to generate cell-permeable transcription factors that not only turn gene transcription on or off
but respond to extracellular signals as part of signal transduction cascades [3]. However, we can
anticipate that chemists will continue to bring fresh perspectives and a zest for understanding
biology at the molecular level to this highly fertile ground for exciting research.

Summary
Misregulated transcription is linked to many human diseases, and thus artificial transcriptional
activators are highly desirable as mechanistic tools and as replacements for their malfunctioning
natural counterparts. We previously reported two artificial transcriptional activation domains
obtained from synthetic peptide libraries screened for binding to the yeast transcription protein
Med15(Gal11). Here we demonstrate that the transcriptional potency of the Med15 ligands is
increased through straightforward structural alterations. These artificial activation domains
upregulate transcription via specific Med15 binding interactions and do not function in mammalian
cells, which lack Med15. This functional specificity stands in contrast to most natural or artificial
activation domains that function across all eukaryotic cell types. The results indicate that the
screening strategy holds excellent promise for identifying peptide and small molecule
transcriptional activators that function by unique mechanisms with advantageous specificity
properties.

Article Outline
Introduction
Results and Discussion
Exchanging the DNA Binding Domain
DNA Binding Site Location
Ligands 17 and 32
Activation Domain Multimers
Specificity of Function
Significance
Experimental Procedures
General Methods
Plasmid Construction
17+LexA, 28+LexA, and 32+LexA
Gal4(1-147)+17, Gal4(1-147)+28, and Gal4(1-147)+32
pM+28
Gal4(1-147)+28x2, Gal4(1-147)+32x2, Gal4(1-147)+28+32, and Gal4(1-147)+32+28
Gal4(1-147)+28x3 and Gal4(1-147)+32x3
Med15Δ(186-619)
β-Galactosidase Assays
Secreted Alkaline Phosphatase Assay
Immunofluorescence Staining
Acknowledgements
Supplemental Data
References

Introduction
Transcriptional activators play an essential role in gene regulation by recruiting the RNA
polymerase II holoenzyme to the genes with which they are associated (Figure 1) [1]. Many human
diseases are characterized by aberrant gene transcription patterns linked to malfunctioning
transcriptional regulators [[2] and [3]]. For example, in the case of medulloblastoma, one of the
most malignant pediatric cancers, the concentration of the transcriptional repressor REST/NRSF is
abnormally high, resulting in the suppression of genes critical for proper differentiation of neuronal
cells [[4] and [5]]; recent evidence suggests that upregulating the transcription of REST-regulated
genes can mitigate the tumorigenic potential of medulloblastoma cells [[6] and [7]]. Thus, artificial
transcriptional activators composed of small molecules or proteins have emerged as important tools
to better characterize the relationship between aberrant transcription patterns and disease and in the
longer term, to define key characteristics of transcription-based therapeutics [[8], [9] and [10]].

Full-size image (28K)

High-quality image (149K)
Figure 1. Transcriptional Activation in Eukaryotes
(A) Schematic of transcriptional activation.
(B) The architecture of a transcriptional activator.

View Within Article

Natural activators are exquisitely specific in their function, upregulating cognate genes in particular
cell types to predetermined levels upon demand [1]. Part of that specificity is derived from one of
the two essential components of an activator, a DNA binding domain (DBD) that recognizes
cognate DNA sequences (Figure 1B). The other key component is an activation domain (AD) that
interacts with a variety of proteins that constitute the transcriptional machinery and dictates the
level of gene upregulation [1]. The AD-transcriptional machinery interactions are tightly regulated
by signaling pathways because natural ADs exhibit promiscuous binding behavior [11] that leads to
uncontrolled transcriptional stimulation. An additional layer of functional specificity is imposed by
transcriptional regulatory networks that dictate when and where a given activator is expressed [[12]
and [13]], leading to cell-type specific function.
The functional specificity profile of artificial transcriptional activators is much less sophisticated,
and the most success has been attained in gene targeting specificity [[9], [10] and [14]]. For
example, artificial activators that upregulate predetermined genes have been constructed by the
replacement of endogenous DBDs with novel protein DBDs [15] or with synthetic variants such as
peptide nucleic acids [16], triplex-forming oligonucleotides [[17] and [18]], and hairpin polyamides
[19]. In addition, artificial activators that function only in the presence of a small molecule have
been developed and offer some control over the timing of gene activation, thus serving as a
substitute for the signaling pathways that regulate natural AD function [[20] and [21]]. As in natural
activators, the AD of artificial activators contacts the transcriptional machinery to upregulate
transcription. However, it is typically difficult to predict the level of transcriptional stimulation that
will be elicited by a given artificial activator due to many additional factors that impact AD
function. These factors include the DBD to which the AD is attached, the position of the DNA
binding site relative to the gene, the concentration of the AD present at the gene, and the affinity of
the AD for the transcriptional machinery [[11], [18], [19], [22], [23], [24], [25], [26] and [27]].
Finally, artificial activators that target particular cell types or organisms remain elusive. This lack of
specificity can largely be attributed to the activation domains employed in artificial activator
construction. These are typically ADs derived from or closely related to natural activators that in the
context of an artificial activator operate outside of the endogenous regulatory pathways [[9], [15],
[25], [28], [29] and [30]]. Thus, the ADs interact with a wide range of protein targets and, as a
result, function in all eukaryotic systems [[16] and [24]].
We previously described a strategy for identifying artificial ADs that employs a screen for ligands
of an individual transcriptional machinery protein (Figure 2) [31]. The focal protein of that study
was Med15(Gal11) [32], a common target of natural ADs that resides in the mediator complex of
the yeast transcriptional machinery [[33] and [34]]. Among the ligands identified from two
synthetic peptide combinatorial libraries were two 8 residue peptides that function as activation
domains when attached to a DBD and have sequence compositions distinct from any known ADs
(Figure 2). In this article, we show that the activity of ADs discovered via the screen can be
increased by simple modifications such as altering the DNA binding domain to which the AD is
attached, consistent with the function of most ADs. However, we also find that the ADs are
functionally dependent on the presence of a single protein, the original target Med15. This
functional specificity stands in contrast to typical natural or artificial ADs, and, as demonstrated
with one of the Med15-dependent ADs, leads to cell-type-specific transcriptional activation.
Further, although the experiments described here were carried out using peptide-based activators,
the screening strategy is readily extendable to small molecule combinatorial libraries, leading in the
future to new classes of small-molecule-based transcriptional activators [[35] and [36]] that
function in a cell-type-specific or organism-specific manner.

Z.Q. Wu, G. Belanger, B.B. Brennan, J.K. Lum, A.R. Minter, S.P. Rowe, A. Plachetka, C.Y.
Majmudar and A.K. Mapp, Targeting the transcriptional machinery with unique artificial
transcriptional activators, J. Am. Chem. Soc. 125 (2003), pp. 12390–12391. Full Text via CrossRef |
View Record in Scopus | Cited By in Scopus (15)
31]
(A) Two libraries of synthetic peptides were screened for their ability to bind to the transcription
factor Med15.
(B) Two of the resulting ligands showed good activity as transcriptional activation domains in
Saccharomyces cerevisiae compared to a natural AD sequence, VP2, when they were attached to a
protein DNA binding domain (LexA).
(C) A schematic of the integrated reporter gene used for the in vivo activation studies. As indicated,
the two DNA binding sites for LexA were positioned 50 bp from the TATA box. The transcriptional
activity of each LexA fusion protein was measured by assaying β-galactosidase activity.

View Within Article

Results and Discussion

The Med15 ligands were initially evaluated in several functional contexts in order to increase
transcriptional potency and facilitate subsequent experiments testing the requirement for Med15.
Among the factors that affect the potency of ADs are the DNA binding domain to which the AD is
attached and the position of the DNA binding site relative to the gene being upregulated. Both of
these factors were first investigated using the more potent of the Med15 ligands, 28 (Figure 2). In
addition, artificial transcriptional activators of greater potency were constructed using dimers and
trimers of the Med15 ligands as the activation domains.

Exchanging the DNA Binding Domain

Natural transcriptional activators are modular proteins, and activation domains typically function
independently of the identity of the DNA binding domain. ADs derived from the potent viral protein
VP16, for example, activate transcription efficiently when fused to a wide range of DBDs,
including entirely nonnatural versions such as polyamides [[22] and [25]] or triplex forming
oligonucleotides [18], although the level of transcription varies. The ability of natural ADs to
interact with multiple protein targets may account for this functional flexibility. Although changing
the DBD and hence presentation of the AD might affect interactions with a subset of transcription
protein targets, it is unlikely to affect all. However, given that our ligands were identified based
upon their ability to bind to a single transcription protein, we initially chose a DBD that most
closely mimicked the presentation of the ligands in the binding screen. Thus, because the peptides
were attached to solid support with the amino terminus free, the selected ligands were fused to the
LexA DBD at the amino terminus of the protein (Figure 2).
To test if the Med15 ligands require a particular presentation to function as ADs, we first
investigated the function of 28 fused to a commonly used DBD derived from the S. cerevisiae
protein Gal4, Gal4(1-147). The first 50 residues of Gal4 form a Cys6-Zn cluster that binds to DNA
and the subsequent 50 mediate dimerization through helix-helix interactions that enhance sequence-
specific DNA binding [[37] and [38]]. The remaining 47 amino acid residues serve as a linker
region of undefined structure between the AD and the DBD. A plasmid was constructed encoding
ligand 28 fused to the carboxyl terminus of Gal4(1-147). Thus, ligand 28 is displayed in the
opposite orientation (carboxy terminus free) relative to the original LexA fusion. This plasmid and
the original 28+LexA construct were each transformed into yeast strains containing two cognate
binding sites for either LexA or Gal4 positioned 191 bp upstream of a lacZ reporter gene. Activation
levels mediated by each of the DBD+28 fusions were then quantitated by measuring β-galactosidase
activity [39].
As shown in Figure 3A, the Gal4(1-147)-based activator was more potent than the 28+LexA fusion
protein, with 14.5-fold activation compared to the DBD alone. Thus, the Med15 ligand does not
require a free amino terminus to function as an activator despite the original screening conditions.
The change in orientation from amino terminus to carboxy terminus is not the only factor
contributing to the increase in function. In preliminary experiments, attachment of 28 to the carboxy
terminus of the DBD Gal4(1-100) produced an activator approximately 4-fold less active than
Gal4(1-147)+28. Thus, the additional linker residues present in the Gal4(1-147)+28 fusion protein
likely play a role in the activity increase, projecting the ligand from the DNA more effectively and
providing more favorable conditions for the Med15 interaction. A similar reliance upon linkers has
been noted with natural transcriptional activation domains such as ATF14 and VP2, both acid-rich
ADs derived from the potent viral coactivator VP16 [[22] and [25]], as well as nonnatural activation
domains [24]. Future studies of a wider range of DNA binding domains and linkers of varying
composition will be employed to quantify this effect.

Figure 3. Increasing the Activity of the Med15 Ligands

The transcriptional potency of each artificial transcriptional activator was measured by quantitating
β-galactosidase activity using standard methods [39] and fold activation levels were obtained by
dividing that activity by that obtained with DNA binding domain alone. Each activity bar represents
the average of at least 3 independent experiments with the indicated errors (SDOM). See Table 1 for
yeast strain details.
(A) The activity of ligand 28 fused to different DNA binding domains. Note that the position of the
DNA binding sites in this yeast strain is different than in the strain in which the results shown in
Figure 2 were obtained.
(B) Promoter position-dependence of ligand 28 fused either to LexA (left two bars) or Gal4(1-147)
(right two bars).
(C) The activity of 28, 17, and 32 fused to Gal4(1-147).

View Within Article

Table 1.
Yeast Strains Used in This Study

Strain Genotype Source

JPY9::pZZ4
1 Matα his3Δ200 leu2Δ1 trp1Δ63 ura3-52 lys2Δ385 gal4Δ11::pZZ41a [23]

JPY52::JP18 Matα his3Δ200 leu2Δ1 trp1Δ63 ura3-52 lys2Δ385 gal4Δ11

[23]
5 med15::LYS2, URA::pJP185b

JPY52::JP18 Matα his3Δ200 leu2Δ1 trp1Δ63 ura3-52 lys2Δ385 gal4Δ11 med15:LYS2,

[23]
8 URA::pJP188c
Matα his3Δ200 leu2Δ1 trp1Δ63 ura3-52 lys2Δ385 gal4Δ11 med15:TRP1,
ZL2 [34]
URA::pJP169d
Full-size table
a pZZ41 contains two LexA binding sites 50bp upstream of the TATA box.
b pJP185 contains two Gal4 binding sites 50bp upstream of the TATA box.
c pJP188 contains two Gal4 binding sites 191bp upstream of the TATA box.
d pJP169 contains two LexA binding sites 191bp upstream of the TATA box.

View Within Article

DNA Binding Site Location

The position of the DNA binding site relative to the transcription start site can also affect how much
an activator upregulates a gene [[23], [40], [41] and [42]]. Gal4(1-147)+28 was tested in yeast
strains bearing 2 binding sites either 50 bp or 191 bp upstream of the TATA box, again using
quantitative β-galactosidase assays. As illustrated in Figure 3B, the relationship between potency
and binding site position varied with the DNA binding domain employed. For 28+LexA, function
was better at a binding site distance of 50 bp from the TATA box, with fold activity dropping by half
when moved to 191 bp. The opposite trend was observed with Gal4(1-147) as the DBD. In that
case, lower levels of activation were obtained when the binding sites were closer to the TATA box.
These latter results parallel the activity of the Med15 protein itself when it is employed as an
artificial activator. The fusion protein Gal4(1-147)+Med15 is a potent activator at the 191 bp
distance; at 50 bp, however, activity drops by half [23]. The results obtained with Gal4(1-147)+28
are suggestive of a mechanism by which 28 specifically recruits Med15 to DNA in an orientation
analogous to the positioning in the Gal4(1-147)+Med15 fusion protein.

Ligands 17 and 32
For ligand 28, the best activity was obtained when Gal4(1-147) was used as a DBD and the DNA
binding sites were positioned 191 bp upstream relative to the TATA box. Two additional activating
ligands were then tested in this functional context (Figure 3C). One of these is ligand 17, shown in
Figure 2. The second is ligand 32 (sequence AYFEVPSE), the next most active ligand identified in
the original screen that interacts with Med15 with an affinity similar to 28 and 17 (KD 1.3 μM
versus 4.8 and 2.2 μM, respectively). In addition, the binding site of ligand 28 is distinct from those
of 32 and 17 ([31] and Figure S3). Although the sequence of 32 is different from the other two
artificial ADs, it bears the most resemblance to the largest class of natural activation domains, the
so-called acid-rich ADs that typically contain polar residues interspersed with hydrophobic amino
acids [1]. An example of this class of ADs is VP2, a positive control used in the original binding
screen and functional assays (Figure 2B). Plasmids encoding either 17 or 32 attached to the
carboxyl terminus of the Gal4(1-147) DBD were prepared by standard methods and then
transformed into the yeast strains used in earlier experiments. As shown in Figure 3C, the two
ligands showed quite different effects. In the case of 17, only 2-fold activity relative to the DBD
alone was observed, comparable to the results obtained with the LexA DBD (Figure 2B). Similar
results were obtained when 17 was fused to Gal4(1-100); in addition, moving the binding sites
closer to the transcriptional start site did not provide an increase in activity (see Figure S1 for
details). In contrast, ligand 32 exhibited quite modest activity when fused to the amino terminus of
LexA (1.4-fold) but, as shown in Figure 3C, the activity increased to 11.5-fold when it was attached
to Gal4(1-147), comparable to the activity of 28. Overall, these experiments provided two artificial
activators with improved functional profiles, Gal4(1-147)+28 and Gal4(1-147)+32, and these two
activators were used for all further investigations. The results further indicate that the transcriptional
activity of ligands obtained from future screening experiments (small molecules or peptides) can be
readily improved by straightforward optimization experiments, analogous to typical artificial
activation domains.

Activation Domain Multimers

One of the principle mechanisms by which levels of gene transcription can be increased is through
augmenting the local concentration of ADs present at a given gene, although the origin of this effect
is poorly understood [[43], [44], [45] and [46]]. This can be accomplished by increasing the number
of activators bound to DNA, leading to transcriptional levels greater than the sum of those observed
with the individual activators [[44], [46] and [47]]. Alternatively, the AD itself can be oligomerized
within a single transcriptional activator to produce a similar effect [[27], [47] and [48]]. In one
example, an artificial activator containing two copies of a well-studied AD derived from VP16
upregulated transcription 5-fold more effectively than an activator containing a single copy of the
AD [48]. From the standpoint of artificial activator design, the latter approach is particularly
attractive because it requires a change only in the structure of the artificial transcriptional activator
rather than alteration of the regulatory DNA sequences associated with the targeted gene in order to
produce greater levels of transcription.
To test if the potency of artificial activators constructed with our Med15 ligands could be further
increased by oligomerization, plasmids encoding the Gal(1-147) DBD fused to one, two, or three
copies of either peptide 28 (Figure 4A) or peptide 32 (Figure 4B) were constructed such that each
peptide sequence was separated from the next by a flexible linker sequence (see Experimental
Procedures). In addition, constructs containing a single copy of 28 and 32 separated by the linker
sequence were also prepared (Figure 4C). The resulting plasmids were then transformed into the
yeast strain used for the experiments shown in Figure 3, and, as before, the activity of each
construct was measured as a function of β-galactosidase activity. The results of the experiments are
displayed as activation relative to the level obtained with an artificial activator containing a single
copy of the ligand (Figure 4). For all of the dimeric constructs, an approximately linear increase in
function was observed upon expanding the number of peptide modules from 1 to 2, indicating that
this is an effective strategy for tuning potency. Similarly, the trimer of 28 was 3-fold more active
than Gal4(1-147)+28. Among all of the constructs, only the trimer of 32 (Figure 4B) showed
activity beyond a simple additive increase with greater than 95% confidence (more than three times
the level of 32 alone); however, this is still well below the cooperative levels observed with
multimers of acid-rich ADs [48]. This was somewhat unexpected because of the three Med15
ligands, 32 bears the greatest sequence resemblance to natural ADs with a balance of polar/acidic
and hydrophobic amino acids. Comparable results were obtained when the number of DNA binding
sites was increased (data not shown). As illustrated more fully in the following section, the overall
lack of cooperative function observed in these experiments may be indicative of a distinct
mechanistic difference between the artificial ADs isolated from our binding screen and most natural
ADs.
 Full-size image (41K)
High-quality image (287K)
Figure 4. Tuning the Potency of the Artificial Activation Domains
The transcriptional activity of each construct was measured by quantitating β-galactosidase activity
in the yeast strain JPY52::JP188 and is reported as a ratio to the activity of Gal4(1-147)+28 or
Gal4(1-147)+32 (defined as 1). The activity shown is the average of 3 separate experiments with the
indicated error (SDOM). See the Experimental Procedures section for additional details. Multimers
containing 4 copies of each peptide were also constructed; however, yeast bearing these plasmids
grew poorly and Western blots of the extracts indicated that the tetramer constructs were expressed
at very low levels.
(A) The relative fold activation of multimers of 28 on Gal4(1-147).
(B) The relative fold activation of multimers of 32 on Gal4(1-147) DBD.
(C) The relative fold activation of mixed dimeric constructs.

View Within Article

Specificity of Function
Natural activation domains typically interact with a number of transcriptional machinery proteins.
The well characterized yeast transcriptional activator Gal4, for example, has more than 10 identified
target proteins, although the physiological relevance of all the interactions has yet to be established
[[11], [26], [49], [50], [51], [52], [53] and [54]]. One consequence is that deletion or mutation of a
single transcription protein target rarely leads to complete loss of activator function [[34], [53] and
[55]]. Our activator peptides were identified based upon their ability to interact with a single
transcriptional machinery protein, however, and perhaps the most compelling question surrounding
their function is if the protein target Med15 is required for them to activate transcription. To
evaluate this possibility, β-galactosidase assays were carried out in a yeast strain in which Med15
had been deleted from the genome. This experiment is possible because Med15 is not an essential
protein, although yeast bearing this alteration exhibit a slow growth phenotype [56]. As shown in
Figure 5, we compared the function of Gal4(1-147)+28 and Gal4(1-147)+32 in yeast strains either
bearing Med15 (dark bars) or bearing no Med15 (white bars) and noted a nearly complete loss of
function. This was in contrast to the positive control, Gal4(1-147)+ATF14, a sequence taken from
the potent viral coactivator VP16, that showed only a 2-fold decrease in activation levels. Although
ATF14 is known to interact with Med15, it has several additional putative targets in the
transcriptional machinery, and thus its function is attenuated rather than abrogated in the absence of
Med15 [57].
 Full-size image (27K)
High-quality image (172K)
Figure 5. Specificity of Med15 Ligands
(A) The fold activity of 28 and 32 in yeast strains with Med15, Med15Δ(186-619), or with no
Med15 present (Med15 delete). Each activity value is the average of three individual experiments
with the indicated error (SDOM) See the Experimental Procedures section for additional
information.
(B) Fold activation in human embryonic kidney 293 cells. HEK293 cells were transiently
transfected with plasmids coding for each construct and SEAP activity was measured using standard
methods [59]. See Experimental Procedures for details.

View Within Article

This point was further investigated by carrying out the same set of experiments using a yeast strain
in which the central region of Med15 (residues 186–619) had been deleted [58]. This mutation
minimized the deleterious phenotype of the Med15 delete strain and enabled us to test if the binding
sites for ligands 28 and 32 were in this region because the original binding screen was carried out
with this fragment. Gratifyingly, nearly identical results were obtained, with 28 and 32 showing
little or no activity in this strain while the fold activation of ATF14 was similar to the strain with
Med15 present (light bars). Taken together, these data suggest that both 28 and 32 are dependent
upon a binding interaction with Med15 for transcriptional activation to occur.
One interesting feature of acid-rich ADs such as Gal4 or VP16 is that they function in all eukaryotes
tested, from yeast through humans [[60], [61] and [62]]. Despite differences in RNA polymerase II
holoenzyme composition, there is evidently significant conservation across species with regard to
activator targets. It has been challenging, however, to identify metazoan homologs of Med 15 [33].
Recently, compelling evidence for homology between the amino terminus of Med15 and the
mammalian protein ARC105 was reported; both proteins contain a so-called GACKIX domain
often found in targets of activators [63]. The significance of this similarity has yet to be determined,
however, because the amino terminus of Med15 can be removed and overall function is maintained
[34]. In addition, the two proteins have sequence similarities in the carboxy terminal region,
including a glutamine-rich stretch of amino acids. However, the region of Med15 used in our
original binding screen (residues 186–619) shares little sequence similarity with ARC105 or any
other identified metazoan protein, and we thus anticipated that activators that function through
interaction with this region would not be able to function in mammalian cells. To test this idea, a
plasmid encoding the most active of the peptides (28) was transiently transfected into human
embryonic kidney cells (HEK293 cells) along with a reporter plasmid bearing five Gal4 binding
sites within an E1b promoter upstream of a SEAP reporter gene following standard protocols. As a
positive control, we also examined the activity of a VP16-derived activation domain fused to
Gal4(1-147), known to function well in this system. As indicated in Figure 5B, no activation by
ligand 28 was observed while the VP16-derived AD functioned well in this context. This data
reinforces the earlier results indicating that 28 is dependent upon Med15 for function, and further,
indicates that the ligand screening strategy can be used to identify artificial ADs that are specific for
a particular cell type, depending on the target protein. Given the emerging role of cell-type-specific
transcription factors and factors expressed only at certain points in development (for example, see
[[64] and [65]]), ligands for those proteins will be particularly valuable for functionally specific
artificial activator construction.
Significance
The results presented here indicate that artificial activation domains discovered through a
binding screen differ from typical natural or artificial ADs in several key respects. Similar to
natural ADs, the potency of artificial activators constructed from the Med15 ligands can be
readily increased by simply reiterating the AD sequences within the construct. In contrast, a
strong synergistic increase in transcriptional levels is not observed. This is most likely related
to the functional specificity of the ADs, as subsequent experiments revealed that for at least
two of the artificial ADs, Med15 is required for transcription function. In the future, artificial
transcriptional activators constructed with the Med15-specific ligands used in combination
with ligands targeting other individual transcriptional machinery proteins will thus be
outstanding tools for probing the mechanistic origins of transcriptional synergy. Further,
since the screening strategy provides activation domains that function through binding
interactions with individual transcriptional machinery proteins, targeting other cell-type-
specific or organism-specific proteins provides a mechanism for the creation of artificial
activators whose functional specificity extends beyond that imposed by the DNA binding
domain. Finally, as the screening strategy is equally applicable to combinatorial libraries of
small molecules, these results provide a framework for building tunable, uniquely specific
small molecule transcriptional regulators.

Experimental Procedures

General Methods
Restriction enzymes, T4 polynucleotide kinase, and T4 DNA ligase were purchased from New
England Biolabs and used as recommended. Oligonucleotides were obtained from Invitrogen. The
plasmid YCplac111 and all of the yeast strains used to test the activity of our activator constructs
were generously provided by Dr. A. Ansari (University of Wisconsin) [66]. The plasmids pGBKT7,
pG5SEAP, pM, and pM3-VP16 were obtained from BD Biosciences. The human embryonic kidney
293 cells used for testing the activity of 28 were purchased from the American Type Culture
Collection (ATCC) and maintained as recommended. The QuikChange Site-Directed Mutagenesis
Kit used to generate the YCplac111 Med15Δ(186-619) plasmid was purchased from Stratagene. All
other chemicals and supplies were purchased from Fisher unless otherwise noted. All techniques
used for yeast manipulations were carried out in accordance with standard protocols [39]. All other
general molecular biology techniques were carried out as described [67].

Plasmid Construction

17+LexA, 28+LexA, and 32+LexA

Plasmids encoding 17+LexA, 28+LexA, and 32+LexA were generated from pNLexA (Origene) as
previously described [31].

Gal4(1-147)+17, Gal4(1-147)+28, and Gal4(1-147)+32

Plasmids encoding Gal4(1-147)+17, Gal4(1-147)+28, and Gal4(1-147)+32 were generated from
pGBKT7 by first annealing oligonucleotides encoding each peptide ligand, (5′-AA TTC GGT TCT
GGT GGT TCT GGT NNN NNN NNN NNN NNN NNN NNN NNN TAA-3′ and 5′-TCGA NNN
NNN NNN NNN NNN NNN NNN NNN ACC AGA ACC ACC AGA ACCG-3′ or 5′-AA TTC TGT
GGT TCT GAT GCT TTG GAT GAT TTT GAT TTG GAT ATG TTG TAA-3′ and 5′-TC GA TTA
CAA CAT ATC CAA ATC AAA ATC ATC CAA AGC ATC AGA ACC ACA G-3′) resulting in
sticky ends corresponding to the restriction sites EcoRI/SalI. The duplex oligonucleotides were
phosphorylated with T4 polynucleotide kinase and ligated with T4 DNA ligase into pGBKT7
predigested with EcoRI/SalI. The resulting plasmids were amplified in DH5α E. coli (Invitrogen),
selected on LB-agar plates containing 50 μg/ml kanamycin, and isolated from cultures using a
QIAprep Spin Miniprep Kit (Qiagen). The sequences of the isolated plasmids were verified by
sequencing at the University of Michigan Core Facility.

pM+28
For use in the human cell experiments, a plasmid encoding Gal4(1-147)+28 was generated from pM
by ligation of an oligonucleotide pair encoding peptide 28 (5′-AA TTC GGT TCT GGT GGT TCT
GGT GCT CAT TAT TAT TAT CCA TCT GAA TAA-3′ and 5′-TCGA TTA TTC AGA TGG ATA
ATA ATA ATG AGC AGA ACC ACC AGA ACCG-3′) into pM that had been predigested with
EcoRI/SalI. The resulting plasmid was amplified in DH5α E. coli (Invitrogen), selected on LB-agar
plates containing 0.1 mg/ml ampicillin and isolated from cultures using a QIAprep Spin Miniprep
Kit (Qiagen). The sequence of the isolated plasmids was verified by sequencing at the University of
Michigan Core Facility.

Gal4(1-147)+28x2, Gal4(1-147)+32x2, Gal4(1-147)+28+32, and Gal4(1-147)+32+28

To construct plasmids encoding the Gal4(1-147)+28x2, Gal4(1-147)+32x2, Gal4(1-147)+28+32,
and Gal4(1-147)+32+28 fusions, two pairs of oligonucleotides coding for each of the peptides and
linkers were designed and separately annealed. The annealed oligonucleotides for either peptide 28
or peptide 32 (5′-AATTT GGT TCT GGT GGT TCT GGT NNN NNN NNN NNN NNN NNN
NNN NNN GGT ACC G-3′ and 5′-GATCC GGTACC NNN NNN NNN NNN NNN NNN NNN
NNN ACC AGA ACC ACC AGA ACC A-3′) contained sticky ends complementary to
EcoRI/BamHI sites and an additional KpnI site upstream of the BamHI sticky end. The second set
of annealed oligonucleotides for peptide 28 or peptide 32 (5′-GATCC GGT TCT GGT GGT TCT
GGT NNN NNN NNN NNN NNN NNN NNN NNN TAA G-3′ and 5′-TCGAC TTA NNN NNN
NNN NNN NNN NNN NNN NNN ACC AGA ACC ACC AGA ACCG-3′) contained sticky ends
complementary to BamHI/SalI sites. The two sets of annealed oligonucleotides were 5′
phosphorylated using T4 polynucleotide kinase. Both pairs of annealed oligonucleotides were then
ligated simultaneously with T4 DNA ligase into EcoRI/SalI digested pGBKT7. The resulting
plasmids were amplified in SMART E. coli cells (Gene Therapy Systems), selected on LB-agar
plates containing 50 μg/ml kanamycin, and isolated from cultures using a QIAprep Spin Miniprep
Kit (Qiagen). The sequences of the isolated plasmids were verified by sequencing at the University
of Michigan Core Facility.

Gal4(1-147)+28x3 and Gal4(1-147)+32x3

To construct plasmids encoding Gal4(1-147)+28x3 and Gal4(1-147)+32x3, the respective Gal4(1-
147) dimer construct, either Gal4(1-147)+28x2 or Gal4(1-147)+32x2 was sequentially digested
with KpnI and BamHI. Oligonucleotides corresponding to 28 or 32 (5′-T GGT TCT GGT GGT
TCT GGT NNN NNN NNN NNN NNN NNN NNN NNN TTC G-3′ and 5′-GATCC GAA NNN
NNN NNN NNN NNN NNN NNN NNN ACC AGA ACC ACC AGA ACC AGTAC-3′) were
annealed, resulting in sticky ends complementary to KpnI/BamHI sites and forming an EcoRI site
upstream of the BamHI site. These annealed oligonucleotides were 5′ phosphorylated using T4
polynucleotide kinase and then ligated with T4 DNA ligase into KpnI/BamHI digested Gal4(1-
147)+28x2 or Gal4(1-147)+32x2, amplified in SMART E. coli cells (Gene Therapy Systems),
isolated from cultures using a QIAprep Spin, Miniprep Kit (Qiagen), and sequenced at the
University of Michigan Core Facility.

Med15Δ(186-619)
The YCplac111+Med15Δ(186-619) plasmid was generated from the parent YCplac111 full-length
Med15 plasmid using site-directed mutagenesis. Briefly, two sets of oligonucleotides were designed
to insert XhoI restriction sites either after nucleotide 558 (last nucleotide in codon for amino acid
186) or before nucleotide 1858 (first nucleotide in amino acid 620 of Med15). The first set, (5′-CAA
TTA CTG CAA AGA ATT CTC GAG CCT AAC ATT CCA CCC-3′ and 5′-GGG TGG AAT GTT
AGG CTC GAG AAT TCT TTG CAG TAA TTG-3′) with homology to the region surrounding
nucleotide 558, was used to amplify the parent YCplac111 plasmid that encodes full-length Med15.
The methylated parent plasmid was then digested with DpnI and the nicked mutagenized plasmid
was amplified in SMART E. coli cells (Gene Therapy Systems), selected on LB-agar plates
containing 0.1 mg/ml ampicillin and isolated from cultures using a QIAprep Spin Miniprep Kit
(Qiagen). This modified plasmid was then subjected to the same mutagenesis procedure using the
second set of oligonucleotides, (5′-GGG AAA GTA TGA GAA TTC TCG AGC AAA TTT TAA
GAA GAC-3′ and 5′-GTC TTC TTA AAA TTT GCT CGA GAA TTC TCA TAC TTT CCC-3′)
designed to insert an XhoI restriction site upstream of nucleotide 1858 of Med15. After insertion of
both of the XhoI restriction sites, the amplified plasmid was digested with XhoI then gel purified
and the resulting YCplac111+Med15Δ(186-619) plasmid was religated with T4 DNA ligase,
amplified in SMART E. coli cells and selected on LB-agar plates containing 0.1 mg/ml ampicillin.
The new YCplac111+Med15Δ(186-619) plasmid was subsequently isolated using a QIAprep Spin
Miniprep Kit (Qiagen) and the sequence was verified at the University of Michigan Core Facility.

β-Galactosidase Assays
The function of the ligand-DBD fusions were examined in yeast by a quantitative liquid β-
galactosidase assay in accordance with established methods [39]. Briefly, the plasmids encoding the
ligand+DBD fusions and the DBD plasmid (negative control) were transformed into yeast using the
LiOAc method or by electroporation, and transformed colonies were selected by growth on
synthetic complete (SC) media containing 2% raffinose and lacking the appropriate amino acid(s)
for selection. Freshly transformed colonies were used to inoculate 5 ml cultures of SC media
containing 2% raffinose and lacking the appropriate amino acids. The cultures were incubated
overnight at 30°C with agitation. Following incubation, these cultures were used to inoculate 5 ml
cultures of SC media containing 2% raffinose, 2% galactose and lacking the appropriate amino
acids that were subsequently incubated overnight at 30°C with agitation to an OD660 of 0.6–0.9.
The yeast cells were harvested and resuspended in breaking buffer (100 mM Tris-HCl (pH 8.0),
20% glycerol) containing the Complete Protease Inhibitors cocktail (Roche). The cells were lysed
by vortexing with glass beads. A portion of the cell extract was used to measure β-galactosidase
activity via incubation with o-nitrophenyl-β-D-galactopyranoside (1 mg/ml) in Z buffer (60 mM
Na2HPO4, 40 mM NaH2PO4, 10 mM KCl, 1 mM MgSO4•7H2O, and 50 mM 2-mercaptoethanol
[pH 7]). The reaction was stopped by adding 1 M Na2CO3 and the OD420 was measured on a Varian
Cary 300 UV-vis spectrometer. The activity reported was normalized to the total protein
concentration of the extract, measured using a Bradford assay kit (Bio-Rad) with BSA as the
standard. Western blot analysis was conducted on each reaction to confirm appropriate expression
of each of the constructs.

Secreted Alkaline Phosphatase Assay

The function of Gal4(1-147)+28 was examined in human cells using a quantitative secreted alkaline
phosphatase (SEAP) assay in accordance with standard protocols [59]. For this purpose, 4 μg of
plasmid encoding Gal4(1-147)+28, the Gal4(1-147) plasmid (pM, negative control), or the Gal4(1-
147)+ VP16(411-455) fusion (pM3-VP16, positive control) were transiently transfected into an
equal number of human embryonic kidney 293 cells (ATCC) using the PolyFect reagent (Qiagen)
according to the manufacturer’s protocol. Each transfection reaction also contained 2 μg of the
SEAP reporter plasmid pG5SEAP. The transfected cells were cultured at 37°C in a 5% CO2
incubator in DMEM (Mediatech) supplemented with penicillin (100 units/ml), streptomycin (100
μg/ml), L-glutamine (290 μg/ml) and heat-inactivated FBS (Hyclone). After 72 hr, supernatant from
the culture was removed and assayed for SEAP activity. Briefly, 250 µl of the supernatant was
heated to 65°C to inactivate any endogenous phosphates, after which it was added to an equal
volume of 2× SEAP buffer (1 mM MgCl2, 20 mM L-homoarginine, and 2 M diethanolamine [pH
9.8]) and incubated at 37°C for 10 min. Finally, 20 µl of 20 mM p-nitrophenol phosphate in 1×
SEAP buffer was added and the OD405 was measured at 5 min intervals for 1.5 hr using a plate
reader (Molecular Devices). The activity was calculated as the change in light absorbance per
minute per sample. The fold activity of Gal4(1-147)+28 and of the positive control were derived by
comparison to activity obtained for Gal4(1-147). Each value represents the average of three
individual experiments and the error is reported as the standard deviation of the mean.

Immunofluorescence Staining
To verify that the Gal4(1-147) peptide ligand fusions were being expressed and transfected in
approximately equal amounts in HEK293 cells, immunofluorescence staining was performed.
Briefly, the transiently transfected cells were fixed on glass slides using 2% paraformaldehyde.
After multiple washes using blocking buffer (0.05% saponin, 5% BSA, and PBS [pH 7.2]) anti-
Gal4 antibody (Covance) was added (1:2000 dilution) and incubated for 2 hr at room temperature.
After six 5 min washes, an FITC-conjugated anti-rabbit secondary antibody (Santa Cruz Biotech)
was added (1:150 dilution) and incubated for 45 min at room temperature. The slides were then
washed with blocking buffer 6 times for 5 min each and Hoechst (Chemicon), a nuclear stain that
enables visualization of all cells, was added to the slides. The cells were visualized under a
microscope (Leica DM LB connected to Spot RT slider camera, Diagnostic Instruments). The
Supplemental Data (Figure S2) shows images of cells transfected with Gal4(1-147), Gal4(1-
147)+28, or Gal4(1-147)+VP16(411-455). The green signal in the images is due to FITC, indicating
the cells that express Gal4, while the blue signal is due to the Hoechst stain, showing all the cells.
An overlay of both these images shows that >95% of the cells expressed the Gal4 fusion proteins.