Defining the RNA internal loops preferred by benzimidazole derivatives via 2D combinatorial screening and computational analysis

Article abstract of DOI:10.1021/ja200212b

RNA is an important therapeutic target; however, RNA targets are generally underexploited due to a lack of understanding of the small molecules that bind RNA and the RNA motifs that bind small molecules. Herein, we describe the identification of the RNA internal loops derived from a 4096 member 3 × 3 nucleotide loop library that are the most specific and highest affinity binders to a series of four designer, druglike benzimidazoles. These studies establish a potentially general protocol to define the highest affinity and most specific RNA motif targets for heterocyclic small molecules. Such information could be used to target functionally important RNAs in genomic sequence.

Full text of DOI:10.1021/ja200212b

ARTICLE
pubs.acs.org/JACS
Defining the RNA Internal Loops Preferred by Benzimidazole
Derivatives via 2D Combinatorial Screening and Computational
Analysis
Sai Pradeep Velagapudi,†,‡ Steven J. Seedhouse,‡ Jonathan French,‡ and Matthew D. Disney*,†
‡Department of Chemistry, The University at Buffalo, The State University of New York, Buffalo, New York 14260, United States
†The Kellogg School of Science and Engineering, The Department of Chemistry, The Scripps Research Institute, Scripps Florida,
130 Scripps Way #3A1, Jupiter, Florida 33458, United States
S Supporting Information
b
ABSTRACT: RNA is an important therapeutic target; however, RNA
targets are generally underexploited due to a lack of understanding of the
small molecules that bind RNA and the RNA motifs that bind small
molecules. Herein, we describe the identification of the RNA internal loops
derived from a 4096 member 3 ꢁ 3 nucleotide loop library that are the most
specific and highest affinity binders to a series of four designer, druglike
benzimidazoles. These studies establish a potentially general protocol to
define the highest affinity and most specific RNA motif targets for
heterocyclic small molecules. Such information could be used to target
functionally important RNAs in genomic sequence.
’ INTRODUCTION
studies, small molecule microarrays are hybridized with RNA
libraries in which the randomized regions in the RNA are
restricted to small motifs (internal loops or hairpins, for example)
that have a high probability of being found in genomic RNAs.
The members of the RNA motif library that bind to a small
molecule displayed on the array surface are excised and se-
quenced. The features that the bound RNAs share, and hence
those that contribute to binding affinity, are then determined
computationally using statistical analysis.11,12 The results of the
statistical analysis are confirmed by in solution binding measure-
ments. Interestingly, statistical significance for particular features
in RNA structures that bind ligands can be used to predict affinity
and specificity.11,12 2DCS and statistical analysis have been
previously used to determine the RNA motifs preferred by
aminoglycosides. The results have shown that there is unique
RNA motif and sequence space for many of the aminoglycosides
tested.9ꢀ14
Although these studies have helped to establish 2DCS as a
potentially general technique to identify and annotate RNA
motifꢀligands interactions, the exclusive use of aminoglycosides
as small molecule binders could limit its scope and hence its use
in in the development of chemical genetics probes targeting
cellular RNAs. For example, aminoglycosides display less than
favorable profiles for use in cell culture and animals.15,16 There-
fore, an improved set of ligands would include compounds that
are cell permeable, nontoxic, and more charge neutral.17,18
Herein, the 2DCS approach is utilized to identify the preferred
RNA plays a number of diverse roles in cellular biology
including encoding and translating protein, regulating the
amount of protein expressed under different cellular conditions,
and many other functions.1ꢀ4 Because of this, RNA is an
attractive target for chemical genetics probes or therapeutics.
Many studies have identified small molecules that bind RNA;5ꢀ7
however, few RNAs outside of the bacterial ribosome have been
exploited for small molecule intervention. One of the major
reasons that there are few therapeutics that elicit their effects by
modulating RNA function is the limited amount of informa-
tion that is gathered from traditional small molecule screening.
Moreover, the information is sparse compared to the structural
diversity of RNA present in genomic sequence, and it is not
general in that it does not define the preferred RNA mofits for a
small molecule.
Because the motifs that comprise an RNA structure can be
accurately predicted from sequence via free energy minimization
or phylogenic comparison,8 strategies to design small molecules
that bind RNA could be developed if structure prediction
algorithms were used in conjunction with databases of small
moleculeꢀRNA interactions. To enable this approach, however,
much more information is required than what is currently
available on the RNA motifs that bind small molecules and the
small molecules that bind RNA motifs.
In an effort to obtain RNA motifꢀligand partners, multi-
dimensional combinatorial screens, for example 2D combinato-
rial screening (2DCS), have been developed to identify the
optimal RNA motifs from a library of discrete secondary
structures (1, Figure 1) that bind small molecules.9,10 In 2DCS
Received: January 9, 2011
Published: May 23, 2011
r
2011 American Chemical Society
10111
dx.doi.org/10.1021/ja200212b J. Am. Chem. Soc. 2011, 133, 10111–10118
|

Journal of the American Chemical Society
ARTICLE
concentrated under vacuum. The residue was dissolved in water and
purified by reverse phase HPLC to afford 14.5 mg of a yellow powder
(75% yield); tR = 16.5 min. MSꢀESI(þ) LRMS: Calcd.: 476; observed:
477 [M þ Hþ]; HRMS: Calcd.: 477.2721 [M þ Hþ]; observed
477.2720 [M þ Hþ]. 1H NMR (CD3OD, 500 MHz) δ 1.61 (2H, m,
J = 7 Hz), 2.01 (2H, m, J = 7 Hz), 2.28 (2H, t, J = 7.5 Hz), 3.06 (3H, bs),
3.14 (2H, t, J = 6.5 Hz), 3.2 (2H, with solvent), 3.54ꢀ79 (4H, bd), 4.02
(2H, t, J = 6.5 Hz), 7.09 (2H, d, J = 9 Hz), 7.14 (1H, s), 7.23 (1H, d, J =
6.5 Hz), 7.54 (1H, d, J = 9 Hz), 7.91 (2H, d, J = 8.5 Hz). 13C NMR
(CD3OD, 75 MHz) δ 26.26, 29.69, 33.27, 37.77, 43.57, 46.00 (with
solvent), 50.10, 54.62, 68.93, 101.04, 115.29, 115.99, 117.06, 119.32,
127.63, 130.65, 134.15, 150.46, 150.70, 164.77, 175.35.
Synthesis of N-(3-Azidopropyl)-4-(2,6-di-tert-butyl-4-(6-(4-methyl-
piperazin-1-yl)-1H,-2,50-bibenzo[d]imidazol-20-yl)phenoxy) butanamide
(11). An 82 mg (0.16 mmol) portion of 4-(2,6-di-tert-butyl-4-(6-(4-
methylpiperazin-1-yl)-1H-benzo[d]imidazole-2-yl)phenoxy)butanoic acid,
125 mg (0.24 mmol) of PyBOP, and 31 mg (0.24 mmol) of DIPEA were
added to dry DMF under argon, and the solution was stirred for 30 min.
To this solution, 25 mg (0.24 mmol) of 3-azidopropylamine were added.
The solution was stirred for 36 h under argon and then concentrated
under vacuum. The residue was dissolved in water and purified by reverse
phase HPLC to afford 73 mg of a yellow powder (76% yield); tR = 27 min.
MSꢀESI(þ) LRMS: Calcd.: 588; observed: 589 [M þ Hþ], 1177 [2M þ
Hþ]; HRMS: Calcd.: 589.3973 [M þ Hþ]; observed: 589.3996 [M þ
Hþ]. 1H NMR (CD3OD, 500 MHz) δ 1.48 (18H, s), 1.71 (2H, t, J =
6.5 Hz), 2.20 (2H, m, J = 7 Hz), 2.32 (2H, t, J = 7.5 Hz), 3.00 (3H, s),
3.17 (2H, bd), 3.21 (2H, m, J = 6.5), 3.3 (2H, with solvent), 3.64 (2H, d,
J = 11.5 Hz), 3.77 (2H, t, J = 7 Hz) 3.90 (2H, d, 10.5), 7.37 (1H, s), 7.44
(1H, d, J = 8.5 Hz), 7.75 (1H, d, J = 8.5 Hz), 8.13 (2H, s). 13C NMR
(CD3OD, 75 MHz) δ 25.82, 28.74, 31.35, 31.97, 36.36, 36.84, 42.58,
47.52, 49.15, 53.64, 76.50, 100.12, 114.38, 117.55, 118.45, 126.74,
133.31, 146.60, 149.76, 163.14, 174.04.
Synthesis of N-(3-azidopropyl)-4-(2,6-di-tert-butyl-4-(6-(4-methyl-
piperazin-1-yl)-1H,-2,50-bibenzo[d]imidazol-20-yl)phenoxy) butanamide
(9). A 9 mg (0.016 mmol) sample of 4-(2,6-di-tert-butyl-4-(6-(4-methyl-
piperazin-1-yl)-1H,-2,50-bibenzo[d]imidazol-20-yl)phenoxy) butanoic acid,
11 mg (0.023 mmol) of PyBOP, and 3 mg (0.023 mmol) of DIPEA were
dissolved in DMF under argon. The solution was stirred for 30 min
followed by addition of 3 mg (0.023 mmol) of 3-azidopropylamine. The
resulting solution was stirred for 36 h under argon and then concen-
trated under vacuum. The residue was dissolved in water and purified by
reverse phase HPLC to afford 7.2 mg of a yellow powder (72% yield);
tR = 29 min. MSꢀESI(þ) LRMS: Calcd.: 704; observed: 705 [M þ Hþ],
359 [(M/2)þHþ]; HRMS: Calcd.: 705.4275 [M þ Hþ]; observed:
705.4645 [M þ Hþ]. 1H NMR (DMSO-d6, 500 MHz) δ 1.47 (18H, s),
1.63 (2H, t, J = 7 Hz), 2.08 (2H, t, J = 7 Hz), 2.23 (2H, t, J = 7 Hz), 2.91
(3H, s), 3.10 (4H, m, J = 6.5 Hz), 3.25 (2H, bs), 3.34 (2H, t, J = 7 Hz),
3.59 (2H, d, J = 11.5 Hz), 3.72 (2H, t, J = 7.5), 3.92 (2H, bd, J = 11.5 Hz)
7.27 (1H, s), 7.33 (1H, d, J = 8.5 Hz), 7.73 (1H, d, J = 8.5), 7.98 (2H, d,
J = 8.5), 8.10 (1H, d, J = 8 Hz), 8.19 (2H, s), 8.49 (1H, s), 10.19 (1H, bs).
13C NMR (CD3OD, 75 MHz) δ 26.72, 29.73, 32.43, 33.01, 37.29, 37.81,
43.56, 50.10, 54.62, 77.36, 101.10, 115.64, 115.51, 116.59, 119.52, 120.39,
124.74, 127.93, 128.40, 134.80, 147.01, 150.74, 155.87, 163.53, 175.10.
Synthesis of 9-Fl. A 200 nmol sample of 5-(N-(2-propyne)-for-
mamide)-fluorescein (Fl) was added in methanol to a solution containing
150 nmol of 9, 200 nmol CuSO4, and 400 nmol freshly dissolved ascorbic
acid. The final volume was brought to 700 μL with methanol. The reac-
tion mixture was placed in an Emrys Creator monomode microwave sys-
tem from Biotage operating at 2.45 GHz frequency with a pulsed micro-
wave irradiation power of 0ꢀ300W. The temperature was controlled by
using an infrared thermometer perpendicular to the sample vessel. The
reaction was maintained at 110 °C with stirring for 4 h. The product was
purified by preparative TLC on a Whatman 20 ꢁ 20 cm Silica gel TLC
plate; a mobile phase of 16:8:1 ethyl acetate/methanol/triethylamine
Figure 1. Secondary structures of the RNAs used in this study.
Oligonucleotide 1 is the 3 ꢁ 3 nucleotide internal loop library.
Oligonucleotides 2ꢀ6 are competitor oligonucleotides used to con-
strain selected interactions to the randomized region in 1. Oligonucleo-
tide 7 is the cassette into which the 3 ꢁ 3 nucleotide library was inserted.
RNA targets for a series of benzimidazole ligands that are based
on the cell permeable Hoechst scaffold. The results show that the
combination of 2DCS and computational analysis of the selected
sequences can allow for the accurate prediction of small molecule
affinity and specificity for each RNA structure contained in an
RNA internal loop library.
’ EXPERIMENTAL SECTION
Instrumentation. Mass spectra were collected on a LCQ Advan-
tage Ion Trap LC/MS equipped with a Surveyor HPLC system. HPLC
was completed on a Waters 1525 Binary HPLC Pump equipped with a
Waters 2487 Dual Absorbance Detector system. High-resolution mass
spectra were collected at the Scripps Florida Proteomics Facility or at
UC Riverside.
Chemicals. Hexanes and N-N-dimethylformamide (DMF) were
from EMD; 3-bromopropylamine hydrobromide was from TCI; sodium
azide was from Fisher Scientific; 4-hydroxybenzaldehyde, diisopropy-
lethylamine (DIPEA), and ethyl 4-bromobutanoate were from Alfa
Aesar; 3,5-di-tert-butyl-4-hydroxybenzyaldehyde was from Acros Or-
ganics. Benzotriazol-1-yl-oxytripyrrolidinophosphonium hexafluoro-
phosphate (PyBOP) was from Advanced ChemTech. All were used
without further purification. The HPLC solvents used were HPLC grade
acetonitrile from EMD and water obtained from a Barnstead NANO-
pure Diamond Water Purification System operating at 18.2 mΩ-cm.
Preparative HPLC. All HPLC purifications were performed at
room temperature monitoring at 218 and 254 nm. A SymmetryPrep C18
(7 μm, 19 ꢁ 150 mm column) was used with a flow rate of 10 mL/min
and a linear gradient of 0% to 90% B in A over 45 min. (A is water
þ0.1% trifluoroacetic acid (TFA) (v/v) and B is acetonitrile þ0.1%
TFA (v/v).)
Analytical HPLC. The purity of all compounds was determined by
analytical HPLC. A Waters Symmetry C18 (5 μm, 4.6 ꢁ 150 mm)
column was used with a flow rate of 1 mL/min and a linear gradient of
0% to 100% B in A over 50 min (A is 0.1% TFA in water, B is 0.1% TFA
in methanol). Absorbance was monitored at 218 and 254 nm.
Synthesis of N-(3-Azidopropyl)-4-(4-(6-(4-methylpiperazin-1-yl)-
1H-benzo[d]imidazole-2-yl)phenoxy)butanamide (10). A 16.2 mg
(0.041 mmol) portion of 4-(4-(6-(4-methylpiperazin-1-yl)-1H-benzo-
[d]imidazol-2-yl)phenoxy)butanoic acid, 31.2 mg (0.06 mmol) of PyBOP,
and 15.5 mg (0.12 mmol) of DIPEA were dissolved in dry DMF. The
mixture was stirred under argon for 30 min followed by addition of 7 mg
(0.06 mmol) of 3-azidopropylamine.19 After 36 h, the reaction was
10112
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
was applied for approximately 1 h. After 1 h, methanol was added such
that the mobile phase was 16:11:1 ethyl acetate/methanol/triethylamine.
The product (9-Fl) was scraped from plate, extracted into methanol by
vortexing for 30 s, and characterized by MS-ESI(þ) LRMS: Calcd.:
1117; observed: 1118 [M þ Hþ], 560 [Mþ2Hþ], 373 [Mþ3Hþ];
MALDI HRMS: Calcd.: 1118.5247 [M þ Hþ]; observed: 1118.5293
[M þ Hþ]. The purity of the product was determined by analytical
HPLC (tR = 34 min); 11% yield as determined by absorbance at 496 nm
and 40 μg/mL BSA; AB1) at 60 °C for 5 min and allowed to slowly cool
to room temperature. 9-Fl or 11-Fl was then added to final concentra-
tion of 50 nM. Serial dilutions (1:2) were then completed in 1X AB1
supplemented with 50 nM 9-Fl or 11-Fl. The solutions were incubated
for 30 min at room temperature and then transferred to a well of 96 well
black plate. Fluorescence intensity was measured on a Bio-Tek FLX-800
plate reader. The change in fluorescence intensity as a function of RNA
concentration was fit to eq 1:22
in 1X PBS, pH 7.4 using an extinction coefficient of 45 000 Mꢀ1cmꢀ1
.
2
0:5
I ¼ I0 þ 0:5Δεð½FLꢂ0 ꢀ ðð½FLꢂ0 þ ½RNAꢂ0KtÞ ꢀ 4½FLꢂ0½RNAꢂ0Þ
Þ
Synthesis of 11-Fl. A 150 nmol sample of 5-(N-(2-propyne)-
formamide)-fluorescein in methanol was added to a solution containing
750 nmol of 11, 200 nmol CuSO4, and 400 nmol freshly prepared
ascorbic acid. The final volume was brought to 700 μL with methanol.
The reaction mixture was microwaved as described for 9-Fl. The
product, 11-Fl was purified by TLC as described for 9-Fl and character-
ized by MS-ESI(þ) LRMS: Calcd.: 1001; observed: 1002 [M þ Hþ];
HRMS: Calcd.: 1002.4872 [M þ Hþ]; observed: 1002.4843 [M þ Hþ].
The purity of 11-Fl was determined by analytical HPLC (tR = 35 min);
70% yield as determined by absorbance at 496 nm in 1ꢁ PBS, pH 7.4
ð1Þ
where I and I0 are the observed fluorescence intensity in the presence
and absence of RNA respectively, Δε is the difference between the
fluorescence intensity in the absence of RNA and in the presence of
infinite RNA concentration, [FL]0 and [RNA]0 are the concentrations
of 9-Fl or 11-FL and RNA respectively, and Kt is the dissociation
constant. Dissociation constants for a DNA containing the preferred
Hoechst 33258 binding site were determined analogously.
Determination of the Binding Affinity of 8 for Nucleic
Acids. Dissociation constants for RNAs selected to bind 8 were
determined using an in solution assay. A selected RNA was annealed
in AB1 at 60 °C for 5 min and allowed to slowly cool to room
temperature. The RNA was then titrated into a well of a black 96 well
plated containing 500 nM of 8 in AB1. After each addition of RNA, the
sample was allowed to equilibrate for 5 min before measuring fluores-
cence intensity on a SpectralMax M5 plate reader (excitation/emission:
345/460 nm). RNA was added until the fluorescence intensity signal
saturated. The change in fluorescence intensity as a function of RNA
concentration was fit to eq 1. The dissociation constant for a DNA
containing the preferred Hoechst 33258 binding site was determined
analogously.
using an extinction coefficient of 45 000 Mꢀ1cmꢀ1 20
.
Microarray Construction. Alkyne-functionalized microarrays were
constructed as previously described.10 The azido-functionalized Hoechst-
like library (8ꢀ11) was immobilized onto alkyne-functionalized arrays
via 1,3 Huisgen dipolar cycloaddition reaction. Five different concentra-
tions (5.0 mM, 3.0 mM, 1.8 mM, 1.1 mM, and 0.65 mM) of 8ꢀ11 were
applied to the array surface in 300 nL of spotting solution [10 mM Tris-
HCl (pH 8.5), 500 μM CuSO4, 1 mM ascorbic acid, 100 μM tris-
(benzyltriazolylmethyl)amine (TBTA) (dissolved in 4:1 butanol/di-
methyl sulfoxide, and 10% glycerol]. The slides were placed in a
humidity chamber for 3 h. They were then washed in water for 5 min,
followed by washing with 0.5ꢁ PBS for 5 min, and then water for 25 min.
The slides were allowed to dry at room temperature.
Computational Determination of Features in Selected
RNAs that Lead to High Affinity and Selective Binding. The
RNAs that were selected to bind to a ligand were analyzed via the RNA
Privileged Space Predictor Program (RNA-PSP, v 2.0).12 Briefly, RNA-PSP
extracts the nucleotides derived from the randomized positions in each
selected RNA. The extracted sequences are then analyzed for features
(a guanine across from an adenine, for example) that appear more often
in the selected sequences than in the entire library. By comparing the
percentage of selected RNAs that have a feature of interest to the
percentage of RNAs in 1 with the same feature, the statistical significance
can be calculated as a Z-score:
RNA Selection (2DCS). Radioactively labeled internal loop library
(1, 100 pmol) and competitor oligonucleotides (2ꢀ6, 27.7 nmol each)
(Figure 1) were folded separately in 1ꢁ Hybridization Buffer (HB1;
20 mM (4-(2-hydroxyethyl)-1-piperazine ethanesulfonic acid (Hepes),
pH 7.5, 150 mM NaCl, and 5 mM KCl) at 60 °C for 5 min and allowed to
cool slowly to room temperature. The folded oligonucleotides were then
mixed together in a final volume of 400 μL. They were carefully pipetted
onto a slide pre-equilibrated with HB1 supplemented with 40 μg/mL
bovine serum albumin (BSA) (HB2). The oligonucleotides were evenly
spread across the slide surface with a piece of parafilm. The arrays were
hybridized at room temperature for 25 min, and the RNA solution was
removed. The slides were then washed by submersion in HB2 for 2 ꢁ 5
min with gentle agitation and then with NANOpure water.
After drying on the benchtop, the arrays were exposed to a phos-
phorimager screen and imaged using a BioRad FX phosphorimager. The
selected RNAs were mechanically removed by excising the agarose from
the surface and RT-PCR amplified as described previously.9
Cloning and Sequencing. RT-PCR products were cloned into
pUC-19 vector using BamH1 and EcoR1 restriction enzyme sites and
transformed into DH5-R Escherichia coli. Blue-white screening revealed
that the majority of colonies appeared light blue, despite the presence of
the RT-PCR products due to a small, in-frame insert. Thus, light-blue
colonies were used to inoculate 1 mL of TB medium containing 50 mg/L
ampicillin. Plasmids were sequenced by Functional Biosciences, Inc.
(Madison, WI).
n1p1 þ n2p2
Φ ¼
ð2Þ
ð3Þ
n1 þ n2
ðp1 ꢀ p2Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Zobs
¼
Φð1 ꢀ ΦÞð1=n1 þ 1=n2Þ
where n1 is the size of Population 1 (the selected mixture), n2 is the size
of Population 2 (1, 4096), p1 is the observed proportion of Population 1
(selected mixture) displaying the feature, and p2 is the observed
proportion for Population 2 (entire library) displaying the feature.
The Z-scores are then manually converted to the corresponding two-
tailed p-value, which represents the confidence that a feature is preferred
by the ligand and did not occur randomly.
Determination of the Binding Affinities of 9-Fl and 11-Fl
for Nucleic Acids. Dissociation constants for the binding of RNAs to
9-Fl and 11-Fl were determined using an in solution, fluorescence-based
assay.21 To prevent nonspecific binding of the ligands to 96 well plates,
Corning nonbinding surface 96 well plates were used and exposure of
ligands to the plates was always less than 40 min. A selected RNA was
folded in Assay Buffer (8 mM Na2HPO4, 190 mM NaCl, 1 mM EDTA,
A single selected RNA motif has multiple features that are statistically
significant (occur with g95% confidence). Therefore, the Z-scores for
each feature are summed to afford the sum Z-score. The sum Z-score is
strongly correlated with the relative affinity of an RNA for a ligand.12
Thus, a plot of the sum Z-score as a function of affinity allows for a
scoring function to be derived to predict the binding affinity of each
member of 1 to a ligand.
10113
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
Figure 2. Top, structures of the bis-benzimidazoles and benzimidazoles that were conjugated onto alkyne-agarose microarray surfaces via HDCR.
Bottom, microarray surface after hybridization with internally labeled library 1 and competitor oligonucleotides 2ꢀ6. Spots of ligand-bound RNAs that
were excised are indicated with a circle. The moles of ligand spotted are 1.5, 0.9, 0.5, 0.3, and 0.1 nmoles. A, array prior to excision of bound RNA. B, array
after excision of bound RNA.
’ RESULTS AND DISCUSSION
Huisgen dipolar cycloaddition reaction (HDCR).31,32 Arrays
were then hybridized with radioactively labeled 1 in the presence
of excess competitor oligonucleotides 2ꢀ6 (Figure 1). A dose
response for the ligands binding to 1 was observed (Figure 2).
The RNAs captured by the ligands at various concentrations
were excised and amplified. The RNAs bound to ligand at the
lowest loading that could be amplified over background were
cloned and sequenced because they are the highest affinity
binders.9 For the cases of 8 and 11, bound RNAs where
1.5 nmoles of material were delivered to the array surface were
analyzed, whereas the bound RNAs where 0.5 nmoles of 9 were
delivered to the array surface were analyzed.
As can be observed from Figure 2, compounds 8, 9, and 11 gave
signals that were much higher than background, whereas 10 did
not yield appreciable signal above background. RNAs bound to 10
could not be amplified from the array due to the low signal. Most
interestingly, compounds that had greater steric bulk (9 and 11)
gave higher signals on the array than the corresponding parent
structures, 8 and 10, respectively (Figure 2). This is interesting
because 9 and 11 do not bind DNA (Kd . 50 000 μM).
The sequences of the RNA loops selected to bind 8, 9, and 11
were determined by RT-PCR amplification, cloning, and sequen-
cing. Statistically significant features (a guanine across from an
adenine, for example) within the selected RNAs were then
determined by using RNA-PSP.11,12 RNA-PSP statistically ana-
lyzes the selected sequences and structures to identify RNA motif
space that is privileged to bind a particular ligand, represented as
a Z-score. That is, the proportion of the selected RNAs that
display a feature of interest is compared to the population of the
entire library (1) that displays the same feature (eqs 2 and 3).
RNA-PSP reports the features that are over-represented with at
least a 95% confidence interval. Importantly, it also scores each
We previously reported a series of investigations in which the
RNAs preferred by aminoglycoside derivatives were determined
by screening a small library of aminoglycosides against an
RNA library displaying discrete secondary structural elements
(Figure 1).9,10,23 This approach, 2D combinatorial screening
(2DCS), coupled with identifying statistically significant, unique
features in the selected RNAs,11,12 afforded high affinity and
selective interactions. This information was then used in drug
discovery efforts to target triplet-repeating and tetra-repeating
RNA transcripts that cause disease.24ꢀ27 The clinical use of
aminoglycosides, however, is associated with side effects includ-
ing ototoxicity and renal disease.15,16,28,29 Therefore, the 2DCS
approach was applied to more druglike ligands (top of Figure 2).
For these experiments, four benzimidazoles were synthesized
(top of Figure 2) including 8, which is an azide-functionalized
derivative of Hoechst 33258,26 and 10, which is a benzimidazole
version of 8. Compounds 9 and 11 were synthesized to contain
steric bulk at the phenolic side chain of Hoechst 33258. Previous
modeling studies have shown that the addition of steric bulk to
Hoechst derivatives decreases their affinities for AT-rich DNA
structures.30 Binding assays were completed and confirm this
finding, showing that there was no detectable binding of 9 and 11
(Kd’s . 50 000 nM) to a DNA hairpin with the preferred
Hoechst 33258 binding site (a 50AATT sequence). Therefore,
9 and 11 could be specific for RNA. Binding assays with 10 also
show that it is a weak DNA binder to that same sequence (Kd .
50 000 nM). In contrast, the binding of 8 to this DNA is higher
affinity with a Kd of 250 nM.26
Identifying Unique RNA Motif Space for Each Ligand.
Ligands 8ꢀ11 were arrayed onto alkyne-functionalized agarose
microarrays and conjugated to the surface using a Cu-catalyzed
10114
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
Figure 3. Venn diagram that illustrates the types of unique and overlapping RNA sequence spaces selected to bind 8, 9, and 11. The Venn diagram
analysis was enabled by using the RNA-PSP v 2.0 program.12 Data for 8 are shown at the 95% confidence interval and data for 9 and 11 are shown at the
99% confidence interval. Structures that are listed as pairs refer to two nucleotides that are across from each other in the selected structures.
library member in 1 for its fitness to bind to a specific ligand
(relative to the entire library) by summing the Z-score for each
privileged feature in that RNA, or a sum Z-score.12
The highest affinity interactions were observed for the RNA
structures selected to bind 8 with Kd’s that range from about 100
to 1200 nM. Lower affinity interactions are observed with RNAs
selected to bind 9 and 11. The highest affinity interactions are in
the low micromolar regime, with 9 having Kd’s between 1000 and
3000 nM and 11 having Kd’s between 1000 and 20 000 nM.
Analysis of the microarray signals, the affinity data, and the
data from the Venn diagram allows one to determine if higher
signal for related ligands at a given loading indicates interactions
are higher affinity or that more types of RNAs are bound. In the
case of this selection, it appears that that the higher signals on the
array for compound 9 relative to 8 are due to 9 binding more
types of RNA structures.
Prediction of RNAꢀLigand Affinity Using Statistical Anal-
ysis. A quantitative relationship between statistical analysis and
affinity can be derived using an approach we call structure-
activity relationship through sequencing (StARTS).12 In this
approach, the statistical score (Z-score at 95% confidence) for
each feature in a selected RNA calculated by RNA-PSP is
summed to afford a sum Z-score. Measured affinities are then
plotted as a function of the sum Z-score. These data can be fit to a
pseudofirst-order equation (Figure 5), which can be used to
predict the affinities of all members of 1.
A Venn diagram was constructed by using the RNA-PSP
analysis to compare the RNA motif spaces that bind each ligand
and to identify overlap (Figure 3). Previous analyses have shown
that unique space indicates a specific RNA motifꢀsmall molecule
interaction, whereas overlapping space (a feature that belongs to
more than one ligand) indicates that an RNA motif is not unique
to one small molecule.11 Qualitatively, a Venn diagram gives
insights into specificity. A large amount of overlapping space is
observed between ligands 9 and 11, and these two ligands
capture more types of RNA motif space than 8. In general, the
features unique to 8 contains purines on both sides of the loop,
for example (50NNA)/(30GNN) and (50NAN)/(30NGN). In
contrast, 9 generally prefers pyrimidines on both sides of the
loop such as (50UNN)/(30NUN) and (50NUN)/(30NYN). Much
of the RNA space unique to 11 contains a purine on one side of
the loop and a pyrimidine on the other (50NNU)/(30RNN) and
(50UNN)/(30NRN).
The affinities of a subset of the selected RNAs were deter-
mined for the corresponding ligands using in solution binding
measurements (Figure 4). Compound 8 was used directly to
determine dissociation constants as it is fluorescent, and a large
change in fluorescence intensity is observed in the presence of
RNA. In contrast, the dynamic range in fluorescence intensity
upon binding of 9 and 11 is in general too small to accurately
determine dissociation constants as previously observed for
DNA aptamers.30 Thus, these compounds were conjugated to
fluorescein to afford 9-Fl and 11-Fl.
StARTS plots for 8, 9, and 11 show that there is a strong
correlation between affinity and the sum Z-score, in good
agreement with a previous study (Figure 5).12 For all three
ligands, the RNA motifs with the highest sum Z-scores have the
highest affinities. Likewise, selected loops with lower sum
Z-scores have lower affinities. Thus, this computational approach
is apparently general and allows the accurate prediction of the
10115
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
Figure 4. Secondary structures of a subset of RNA loops selected to bind 8, 9, and 11 via 2DCS. Secondary structures were predicted by the RNA
structure program.8 The nomenclature for the loops refers to the ligand that the loops were selected to bind followed by an internal loop (IL) number.
Values below the loop identifier are the Kd’s for the RNAꢀligand complex (nM). None of the compounds binds to cassette 7 or library 1, indicating that
binding is specific to the selected randomized regions.
highest affinity RNA binders from a population of structures via
statistical analysis.
with RNA and the decreased number of hydrogen bond accep-
tors and donors. The statistical analysis and StARTS of the 2DCS
selection identifies RNAs that bind selectively to 11 over 8 and 9
despite its less favorable chemical properties. The selectivity data
are graphically represented in Figure 6, which contains a plot for
each ligand and the sum, Z-scores of all RNAs. Selected RNAs for
each ligand are represented as different colors, and RNAs whose
affinities were measured are labeled. Thus, the relative sum
Z-score for selected internal loops can allow for predictions of
the selectivity of small moleculeꢀRNA motif interactions.
Comparison of the Binding of Benzimidazoles to RNA and
to DNA. The RNA motifs selected to bind 9 and 11 are much
higher affinity than binding of the ligands to AT-rich DNA. This
is likely due to different binding modes. Hoechst 33258 interacts
with the DNA minor groove.33 Introduction of steric bulk onto
the phenolic side chain, affording 9 and 11, would preclude DNA
binding because the bulky tert-butyl groups would sterically clash
with the DNA minor groove.30 There are many potential binding
modes of 9 and 11 to the selected RNA structures in which the
bulky side chains do not occlude binding. For example, the
ligands could interact directly with the bases of selected internal
loop nucleotides or with the RNA major or minor grooves, which
have very different shapes than DNA. Moreover, RNA grooves
Instead of the qualitative prediction of selectivity afforded by
the Venn diagram, the StARTS approach can provide more
quantitative predictions. Because affinity scales with sum Z-score
values (via a pseudofirst-order equation), the affinity of a loop for
a different ligand can be estimated from the corresponding
StARTS plot. For example, the affinity of 8 IL4 (a loop selected
to bind 8) for11 can be estimated by calculating 8IL4’ssumZ-score
for 11 and using the pseudofirst-order equation that describes the
relationship between sum Z-score and affinity for 11.
The ability of the StARTS analysis to predict selectivity was
validated by studying the affinities of 8 IL4, 9 IL5, and 11 IL5 for
the other arrayed ligands. The sum Z-scores for the loops and the
ligands they were selected to bind are in top 0.4% for all members
of library 1. In contrast, their sum Z-scores for the other ligands
range from the top 68% to the top 10% indicating that they
should bind more weakly (Figure 6). Indeed, no saturable
binding was observed at 30 μM RNA when 11 IL5 or 8 IL4
were tested for binding 9 and 11, and no saturable binding was
observed when up to 4.5 μM 9 IL5 or 11 IL5 was added to 8. One
would expect that 11 would bind more weakly to RNA relative to
8 and 9 due to its dimished hydrophobic surface area to stack
10116
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
Figure 6. StARTS plots containing all library members for each ligand.
Each plot depicts the sum Z-scores for each member of 1 as a function of
affinity. Top, selection for 8; middle, selection for 9; and bottom,
selection for 11. The open or colored circles are the RNA motifꢀligand
partners shown in Figure 5. A library member refers to the calculated
sum Z-score for the entire 4096 member library 1 based on the analysis
of selected structures for binding a given ligand. As noted in the text,
compounds that have lower relative sum Z-scores are weaker binders.
Thus, by comparing the position of an RNA in each StARTS plots, the
specificity of the RNAꢀligand interactions can be estimated.
Figure 5. StARTS12 plots for 8, 9, and 11. Filled circles indicate
structures to which affinities were measured (Figure 4). Open circles
indicate predicted affinities for members of 1 with sum Z-score values in
the top ∼10%.
could expand or contract around loop nucleotides,34ꢀ39 a degree
of flexibility that a fully paired DNA likely does not have.
Comparisons to Previous Studies. Previous studies using
2DCS have exclusively investigated the binding of aminoglyco-
side ligands to RNA structures.9,10,13,14 These studies found that
selected interactions can occur with nanomolar Kd’s to internal
loops9,10,13 and low micromolar Kd’s to hairpin loops.11,14 Amino-
glycosides have a large number of amines and are therefore highly
cationic ligands. Herein, we identified RNAꢀligand interactions
with affinities that are comparable or somewhat weaker than amino-
glycosides but with more druglike ligands that have significantly
diminished charge.
Other studies have disclosed RNAꢀligand interactions using
ligands related to 8ꢀ11. For example, Hoechst 33258, a deriva-
tive of 8, has low nanomolar26,40 to micromolar41 affinities for a
variety of RNA targets. Compounds similar to 10 and 11 were
found as hits from an 180 000 member chemical library that bind
a loop in the Hepatitis C virus internal ribosomal entry site. The
initial hit bound with a Kd of ∼100 000 nM. It was optimized to
afford ligands with Kd’s ranging from 700 to 17 000 nM.42 Thus,
our ability to identify and characterize RNA structures that bind
small molecules with Kd’s from the low nanomolar to the
10117
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118

Journal of the American Chemical Society
ARTICLE
micromolar range and predict ligand selectivity via 2DCS bodes
well for this approach identifying RNA motifꢀdruglike ligand
interactions using larger and more diverse chemical libraries.
(25) Lee, M. M.; Pushechnikov, A.; Disney, M. D. ACS Chem. Biol.
2009, 4, 345.
(26) Pushechnikov, A.; Lee, M. M.; Childs-Disney, J. L.; Sobczak, K.;
French, J. M.; Thornton, C. A.; Disney, M. D. J. Am. Chem. Soc. 2009.
(27) Disney, M. D.; Lee, M. M.; Pushechnikov, A.; Childs-Disney,
J. L. Chembiochem 2010.
’ ASSOCIATED CONTENT
(28) Wright, G. D.; Berghuis, A. M.; Mobashery, S. Adv. Exp. Med.
S
Biol. 1998, 456, 27.
Supporting Information. Details of chemical synthesis,
b
(29) Hutchin, T.; Haworth, I.; Higashi, K.; Fischel-Ghodsian, N.;
Stoneking, M.; Saha, N.; Arnos, C.; Cortopassi, G. Nucleic Acids Res.
1993, 21, 4174.
(30) Sando, S.; Narita, A.; Aoyama, Y. Chembiochem 2007, 8, 1795.
(31) Chan, T. R.; Hilgraf, R.; Sharpless, K. B.; Fokin, V. V. Org. Lett.
2004, 6, 2853.
RNA binding assays, and computational analysis of the selected
structures for binding each arrayed ligand. This material is
available free of charge via the Internet at http://pubs.acs.org.
’ AUTHOR INFORMATION
(32) Kolb, H. C.; Finn, M. G.; Sharpless, K. B. Angew. Chem., Int. Ed.
Engl. 2001, 40, 2004.
Corresponding Author
Disney@scripps.edu
(33) Dervan, P. B. Bioorg. Med. Chem. 2001, 9, 2215.
(34) SantaLucia, J., Jr.; Turner, D. H. Biochemistry 1993, 32, 12612.
(35) Wu, M.; Turner, D. H. Biochemistry 1996, 35, 9677.
(36) Burkard, M. E.; Turner, D. H. Biochemistry 2000, 39, 11748.
(37) Batey, R. T.; Rambo, R. P.; Doudna, J. A. Angew. Chem., Int. Ed.
Engl. 1999, 38, 2326.
(38) Doudna, J. A. Nat. Struct. Biol. 2000, 7 (Suppl), 954.
(39) Weeks, K. M.; Crothers, D. M. Science 1993, 261, 1574.
(40) Cho, J.; Rando, R. R. Nucleic Acids Res. 2000, 28, 2158.
(41) Dassonneville, L.; Hamy, F.; Colson, P.; Houssier, C.; Bailly, C.
Nucleic Acids Res. 1997, 25, 4487.
’ ACKNOWLEDGMENT
This work was funded by the National Institutes of Health
(1R01GM079235-01A2) and the Research Corporation. M.D.D. is
a Dreyfus New Faculty Awardee, a Dreyfus Teacher-Scholar, and a
Research Corporation Cottrell Scholar.
’ REFERENCES
(42) Seth, P. P.; Miyaji, A.; Jefferson, E. A.; Sannes-Lowery, K. A.;
Osgood, S. A.; Propp, S. S.; Ranken, R.; Massire, C.; Sampath, R.; Ecker,
D. J.; Swayze, E. E.; Griffey, R. H. J. Med. Chem. 2005, 48, 7099.
(1) Blount, K. F.; Breaker, R. R. Nat. Biotechnol. 2006, 24, 1558.
(2) Calin, G. A.; Croce, C. M. Oncogene 2006, 25, 6202.
(3) Poehlsgaard, J.; Douthwaite, S. Nat. Rev. Microbiol. 2005, 3, 870.
(4) Fedor, M. J.; Williamson, J. R. Nat. Rev. Mol. Cell Biol. 2005,
6, 399.
(5) Thomas, J. R.; Hergenrother, P. J. Chem. Rev. 2008, 108, 1171.
(6) Chow, C. S.; Bogdan, F. M. Chem. Rev. 1997, 97, 1489.
(7) Gallego, J.; Varani, G. Acc. Chem. Res. 2001, 34, 836.
(8) Mathews, D. H.; Disney, M. D.; Childs, J. L.; Schroeder, S. J.;
Zuker, M.; Turner, D. H. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 7287.
(9) Childs-Disney, J. L.; Wu, M.; Pushechnikov, A.; Aminova, O.;
Disney, M. D. ACS Chem. Biol. 2007, 2, 745.
(10) Disney, M. D.; Labuda, L. P.; Paul, D. J.; Poplawski, S. G.;
Pushechnikov, A.; Tran, T.; Velagapudi, S. P.; Wu, M.; Childs-Disney,
J. L. J. Am. Chem. Soc. 2008, 130, 11185.
(11) Paul, D. J.; Seedhouse, S. J.; Disney, M. D. Nucleic Acids Res.
2009, 37, 5894.
(12) Velagapudi, S. P.; Seedhouse, S. J.; Disney, M. D. Angew. Chem.,
Int. Ed. Engl. 2010, 49, 3816.
(13) Tran, T.; Disney, M. D. Biochemistry 2010, 49, 1833.
(14) Aminova, O.; Paul, D. J.; Childs-Disney, J. L.; Disney, M. D.
Biochemistry 2008, 47, 12670.
(15) Silva, J. G.; Carvalho, I. Curr. Med. Chem. 2007, 14, 1101.
(16) Mingeot-Leclercq, M. P.; Brasseur, R.; Schanck, A. J. Toxicol.
Environ. Health 1995, 44, 263.
(17) Herzog, A.; Schutze, H. R. Dtsch. Tierarztl. Wochenschr. 1968,
75, 476.
(18) Disney, M. D.; Stephenson, R.; Wright, T. W.; Haidaris, C. G.;
Turner, D. H.; Gigliotti, F. Antimicrob. Agents Chemother. 2005, 49, 1326.
(19) Lampkins, A. J.; O’Neil, E. J.; Smith, B. D. J. Org. Chem. 2008,
73, 6053.
(20) Durroux, T.; Peter, M.; Turcatti, G.; Chollet, A.; Balestre,
M.-N.; Barberis, C.; Seyer, R. J. Med. Chem. 1999, 42, 1312.
(21) Childs-Disney, J. L.; Wu, M.; Pushechnikov, A.; Aminova, O.;
Disney, M. D. ACS Chem. Biol. 2007, 2, 745.
(22) Wang, Y.; Rando, R. R. Chem. Biol. 1995, 2, 281.
(23) Disney, M. D.; Childs-Disney, J. L. Chembiochem 2007, 8, 649.
(24) Lee, M. M.; Childs-Disney, J. L.; Pushechnikov, A.; French,
J. M.; Sobczak, K.; Thornton, C. A.; Disney, M. D. J. Am. Chem. Soc.
2009, 131, 17464.
10118
dx.doi.org/10.1021/ja200212b |J. Am. Chem. Soc. 2011, 133, 10111–10118