Computational docking simulations of a DNA-aptamer for argininamide and related ligands

Computational docking simulations of a DNA-aptamerfor argininamide and related ligands

H. Bauke Albada1 • Eyal Golub1 • Itamar Willner1

Received: 14 January 2015 / Accepted: 9 April 2015 / Published online: 16 April 2015

� Springer International Publishing Switzerland 2015

Abstract The binding properties of sequence-specific

nucleic acids (aptamers) to low-molecular-weight ligands,

macromolecules and even cells attract substantial scientific

interest. These ligand-DNA complexes found different

applications for sensing, nanomedicine, and DNA nan-

otechnology. Structural information on the aptamer-ligand

complexes is, however, scarce, even though it would open-

up the possibilities to design novel features in the com-

plexes. In the present study we apply molecular docking

simulations to probe the features of an experimentally

documented L-argininamide aptamer complex. The dock-

ing simulations were performed using AutoDock 4.0 and

YASARA Structure software, a well-suited program for

following intermolecular interactions and structures of

biomolecules, including DNA. We explored the binding

features of a DNA aptamer to L-argininamide and to a

series of arginine derivatives or arginine-like ligands. We

find that the best docking results are obtained after an en-

ergy-minimization of the parent ligand-aptamer complexes.

The calculated binding energies of all mono-substituted

guanidine-containing ligands show a good correlation with

the experimentally determined binding constants. The re-

sults provide valuable guidelines for the application of

docking simulations for the prediction of aptamer-ligand

structures, and for the design of novel features of ligand-

aptamer complexes.

Keywords Molecular dynamics � Dissociation constant �Binding energy � YASARA � AutoDock

Introduction

Aptamers are single-stranded nucleic acids sequences ex-

hibiting selective binding properties toward low-molecular

weight ligands, macromolecules and even cells [1, 2], and

the resulting aptamer-ligand complexes yield various

topologies. The unique binding properties of aptamers were

implemented in the recent years in nanomedicine [3–7], to

develop numerous sensors [8–10], to assemble pro-

grammed DNA structures [11–13], and to trigger DNA

machines [14–17] and logic gate operations [18, 19]. The

aptamers are elicited by the ‘systematic evolution of li-

gands by exponential amplification’, SELEX [20–22],

process that involves the selection and amplification of

nucleic acids exhibiting binding affinities toward the li-

gand. In order to understand the nature of aptamer-ligand

interactions and their possible implications for future ap-

plications, a comprehensive characterization of the com-

plexes is required. To date, this includes sequencing of the

aptamer, generation of a reliable three-dimensional (3D)

model and the elucidation of the binding modes [23].

Specifically, the identification of the interactions of the

nucleotide bases with the guest ligand involves NMR

studies [24], mutations of the base in the aptamer [25], and

in rare examples, with proteins, the crystallographic de-

termination of the 3D structures of the aptamer-ligand

complexes [26, 27]. In 2012, roughly one-third of the

*1000 oligonucleotide aptamers were DNA-based [28].

Electronic supplementary material The online version of thisarticle (doi:10.1007/s10822-015-9844-5) contains supplementarymaterial, which is available to authorized users.

& Itamar Willner

[email protected]

1 The Center for Nanoscience and Nanotechnology, Institute of

Chemistry, The Hebrew University of Jerusalem,

91904 Jerusalem, Israel

123

J Comput Aided Mol Des (2015) 29:643–654

DOI 10.1007/s10822-015-9844-5

http://dx.doi.org/10.1007/s10822-015-9844-5

http://crossmark.crossref.org/dialog/?doi=10.1007/s10822-015-9844-5&domain=pdf

http://crossmark.crossref.org/dialog/?doi=10.1007/s10822-015-9844-5&domain=pdf

Of those *1000 oligonucleotide aptamers, about 20 %

binds small molecules. In view of this limited number of

DNA aptamers for small molecules, it is not surprising that

very little is known about the chemical details of the in-

teractions between the aptamers and ligands. In fact, the

Brookhaven Protein Data Bank (PDB, http://www.pdb.org/

pdb/home/home.do) currently contains just over 1500

DNA and over 1000 RNA structures in general (canonical

DNA/RNA structures and aptamer-ligand complexes

alike); in the partially overlapping Nucleic Acid Database

(NDB, http://ndbserver.rutgers.edu/), this number is higher

with just over 4600 DNA and 2600 RNA structures. This

number is rather small compared to the *100,000 struc-

tures of proteins and mixed species that are deposited in the

PDB. Furthermore, within the low number of oligonu-

cleotide structures, only a few ligand-aptamer structures

are found. Although the predictive structural information

on the aptamer-ligand complexes provided by NMR

spectroscopy is a significant tool to elucidate the structures

of these complexes, the approach is limited to relatively

short oligonucleotide sequences. With long nucleic acid

sequences, intramolecular base interactions and chain

folding, lead to non-resolved NMR parameters. Accord-

ingly, reliable computational simulations of the structures

of aptamer-ligand complexes could be an important ad-

vance in the chemistry of aptamers, and perhaps allow the

prediction of structural features of aptamer-ligand com-

plexes that could provide means to improve or even modify

aptamer functionalities. While several molecular dynamics

computational programs were applied to examine structural

features of nucleic acids, mostly for RNA [29–32], the use

of such simulations to monitor and characterize aptamer-

ligand complexes are scarce [33]. Additionally, it is im-

portant to emphasize that currently applied force fields are

underdeveloped for oligonucleotides [34], especially when

compared to sophisticate force fields that are available for

proteins; only the AMBER force fields contain parameters

that describe both nucleotides and proteins. Furthermore,

computational analysis of oligonucleotide molecules is

complicated by the highly charged nature of the backbone,

the flexibility of the oligomer [35], and the large confor-

mational changes that can occur as a result of external

triggers like metal-ions, temperature, and ligands [36]. On

top of this, the profile of the energy-landscape of an

oligonucleotide can be profoundly different in the presence

or absence of the ligand [37]. Among the many confor-

mations of an aptamer existing at room temperature [38]

the ligand can bind to only a single or a very few confor-

mations, thereby lowering the minimal energy of that

structure [39]. For this induced-fit binding, the energy of

the interaction has to be sufficient to induce a conforma-

tional change, and that the timeframe of the interaction is

sufficiently long to allow this change to occur [40]. In light

of this discussion, it is not surprising that the scarce number

of docking studies of ligands to RNA-aptamers yields

success rates of 40–60 % of a correct positioning of the

ligand in the experimentally determined binding site [41].

This falls short of the 70–90 % that is usually obtained for

protein–ligand docking studies [39].

In the present study we seek to establish a computational

strategy that will aid in the understanding of the structural

features of DNA-aptamer ligand complexes. We selected

one ligand-aptamer complex (the argininamide-aptamer

complex, 1OLD) as a test example that could validate the

correlation between the computational model and the ex-

perimentally resolved structure, with the hope that such

computational simulations could be extended to other li-

gand-aptamer complexes, and potentially allow the future

prediction of the structures of ligand-aptamer complexes.

Furthermore, this approach would allow the introduction of

mutations in the binding site of the aptamer and permit the

rapid computational evaluation of the affinities of the

mutated binding site to the ligand and eventually allow the

‘‘computational’’ assessment of association affinities of

structurally related ligands to the binding site. For this, we

use the YASARA Structure software package [42]. At this

point, a few papers have been published in which

YASARA Structure has been used for molecular modeling

and/or docking studies of oligonucleotides [43]. We se-

lected this software since it offers a complete set of force

fields and tools to perform computational modeling of

DNA, RNA, proteins, and carbohydrates, which would

help us to later-on design unique features in the ligand-

aptamer complexes. Accordingly, we performed docking

simulation studies using the known structures of the com-

plexes between argininamide to its corresponding aptamer

(PDB-code: 1OLD; NDB-code: 1OLD). We compare the

computed binding energies of the aptamer complexes with

different ligand derivatives to the experimental values [44].

We find a good correlation between the computational

simulations and the experimental results and we support

that, in principle, such simulation could be considered as a

means to study structural features of aptamer-ligand

complexes.

Results of the docking simulations

Docking simulation of LARM and on the various

NMR-structures of the aptamer

Docking simulations of LARM (1) on each of the seven

deposited NMR-based aptamer-structures (see Fig. S1), i.e.

the so-called ‘seed structures’, were performed in order to

probe the suitability of each structure with respect to the

docking study. This was done using the AutoDock

644 J Comput Aided Mol Des (2015) 29:643–654

123

http://www.pdb.org/pdb/home/home.do

http://www.pdb.org/pdb/home/home.do

http://ndbserver.rutgers.edu/

Lamarckian Genetic Algorithm (LGA) approach and the

AMBER03 force field (without parmbsc0 correction [47]).

The family of AMBER force fields has been successfully

used in previous docking simulation studies that involved

oligonucleotides [48, 49], also in combination with the

YASARA software package [42] (see also the ‘‘Materials

and methods’’ section). YASARA itself has proven to be

very suitable to perform a variety of molecular dynamics

and docking simulations [42, 50–52]. The AutoDock LGA-

based docking procedure performs a requested number of

docking simulations and produces the docked ligands as

final structure, each docking associated with a calculated

binding energy. Here it should be noted that we report the

binding energy, which is calculated by subtracting the

energy of the soup (i.e. the bound state) from the energy of

the unbound state, resulting in a positive value.

All docking simulations predicted that LARM (1) binds in

the determined binding site, although the number of runs that

reproduce the expected binding was very lowwith 1–7 out of

100. Analysis of the overlay of the predicted and determined

bindingmode showed that a docking simulation of the ligand

on the NMR-structures was not able to reproduce the deter-

mined binding mode very well; the RMSD value of the

docked and experimentally determined binding mode of the

ligand ranged from 1.49 to 4.04 A (Fig. 1, see also Fig. S1).

In addition, the relatively high number of clusters that were

predicted by the docking simulation, ranging from 7 to 22

clusters, indicated the presence of multiple hotspots for

binding. The bis-cationic nature of the ligand is the likely

underlying cause for this large number of clusters. Since a

cluster represents a hotspotwhere a ligand can bind in several

conformations, the number of identified clusters increases

when the cluster RMSD is lowered; the docking simulation

itself is not affected, only the analysis of the docking result

afterwards is different when the cluster RMSD value is

changed (Table 1). Therefore, we maintained a cluster

RMSD of 5 A for the rest of our docking simulation studies.

Docking LARM using AutoDock LGA predicts that dock-

ing of the ligand on the 3rd deposited structure of the aptamer

produces the tightest binding, even though only 4 % of the

docked ligands are predicted to bind in the expected binding

pocket (entry 3,Table 1). It seems thatmismatches in the force

field that is used for the NMR-based structure calculation and

the one used for these docking simulations result in poor

docking simulation results. Application of the older

AMBER99 force field did not result in improved docking

Fig. 1 Overlay of the structures of LARM (1) when bound to the

aptamer as determined by NMR (red), and as predicted by the

docking simulation (yellow). The structures of the aptamer were taken

from the deposited NMR structures and are shown in grey; the

molecular surface of the aptamer is shown to emphasize the binding

pocket. a–g Correspond to structures 1–7, respectively, as deposited

in the PDB and NDB

J Comput Aided Mol Des (2015) 29:643–654 645

123

results (Table S1, in the Supporting Information). For the

docking simulation of the other seven ligands that were ex-

perimentally studied for their binding affinity for this ap-

tamer—i.e. DARM, LARG,DARG, agmatine, ethyl-guanidine,

L-lysine, and NG-methyl L-arginine (see Fig. 3 for their struc-

tures) [44] – on theNMRstructures of the aptamer the reader is

referred to the Supporting Information (Tables S2–S8).

With these results in hand, we explored if an energy

minimization (EM) of the ligand-aptamer complex would

lead to docking results that correlate with the experimental

binding findings.

Docking simulation on the aptamer-structures

that were obtained by energy-minimization

of the LARM ligand-aptamer complexes

Straightforward energy-minimization simulations of the

seven NMR-structures of the aptamer-LARM complexes

were carried out, followed by a docking simulation of theLARM ligand to the seven aptamer structures that were

obtained by this straightforward energy-minimization

(Table 2).

Comparison of these results with those obtained for

the docking simulation of LARM on the aptamer struc-

ture as deposited shows that the average binding energies

of LARM docked on the aptamer structure obtained by

EM studies are very comparable. Importantly, the RMSD

value of the docked and experimentally determined

binding mode of the ligand was in a narrow range of

2.37–2.92 A (the overlay of the ligand-aptamer complex

as predicted by the docking simulation with the parent

complex of the NMR study was performed by aligning

the aptamer-structure only), indicating that the variation

in the predicted binding mode is smaller after a

straightforward EM of the NMR-structures of ligand-ap-

tamer complexes (Fig. 2).

Table 1 Docking results for LARM (1) on seven NMR-structures of its DNA aptamer (1OLD)

NMR

structure

Best binding free energy

(kcal/mol)

Average binding free energy

(kcal/mol)

KD

(pM)

NBS nclusters (cluster RMSD,

in A)aRMSD ligand

(A)

1 11.19 n.a. 6320 1 7 (5), 12 (4), 18 (3) 3.85

2 10.66 10.62 ± 0.06 15,470 2 15 (5), 20 (4), 25 (3) 1.87

3 11.65 11.30 ± 0.35 2900 4 15 (5), 19 (4), 24 (3) 2.75

4 10.62 10.57 ± 0.04 16,460 5 11 (5), 17 (4), 25 (3) 3.55

5 10.57 10.22 ± 0.21 17,960 5 12 (5), 16 (4), 25 (3) 1.49

6 10.80 10.70 ± 0.15 12,090 2 12 (5), 13 (4), 20 (3) 2.06

7 11.03 10.64 ± 0.32 8160 7 22 (5), 27 (4), 34 (3) 4.04

KD = the calculated dissociation constant of the strongest binding ligand based on the formula DG = RT�ln(KD). NBS = number of ligands that

bind in the binding site. nclusters = number of clusters that have been calculated based on the specified cluster RMSD. RMSD ligand = the heavy

atom RMSD value that has been calculated for the docked ligand when compared to the experimentally determined binding modea The docking results were repeated three times, each time with three different cluster RMSD values, i.e. 3, 4, and 5 A

Table 2 Docking results for LARM (1) on seven 1OLD-aptamer structures that were obtained by an EM simulation of the seven published

structures of the LARM-aptamer complex

Structure Best binding

free energy (kcal/mol)

Average binding


KD (pM) NBS nclusters RMSD ligand (A)

1 12.31 11.42 ± 0.31 942 23 9 2.77

2 11.85 11.49 ± 0.25 2050 29 11 2.38

3 12.54 11.58 ± 0.34 643 21 13 2.79

4 12.28 11.61 ± 0.32 998 7 9 2.40

5 12.31 11.51 ± 0.30 940 20 14 2.37

6 11.83 11.42 ± 0.34 2130 4 12 2.42

7 11.58 11.37 ± 0.09 3250 22 13 2.92


bind in the binding site. nclusters = number of clusters that have been calculated using a cluster RMSD of 5 A


123

Docking of other ligands than LARM

on the aptamer-structures that were obtained

by an energy minimization of the deposited ligand-

aptamer complex

To assess if the aptamer structures, as obtained by the EM

of the seven deposited ligand-aptamer structures, are also

better suitable to bind arginine-like ligands, we performed

a docking simulation of the other seven ligands that were

experimentally studied for their binding affinity for this

aptamer—i.e. DARM (Table 3), LARG and DARG

(Table 4), agmatine and ethyl-guanidine (Table 5), L-

lysine, and NG-methyl L-arginine (Table S9 in the Sup-

porting Information) (see Fig. 3 for the structures) [44].

Docking of DARM (2)

As was the case for the docking simulation of LARM (1)

on this aptamer, docking of DARM (2) on the energy

minimized structure of the LARM-aptamer complex also

results in higher calculated binding energies and a higher

number of ligands that are docked in the expected

binding site, and the range of the number of clusters is

lower with 8–13 for the current simulation. Even though

the aptamer was energy minimized when bound to theLARM ligand, the procedure produced an aptamer

structure that was more suitable for binding of DARM as

well. For seed-structure 7, two clusters of ligands were

predicted to bind in the binding site with the guanidine

inserted into the binding pocket, resulting in a total

number of 73 ligands that bind to the binding site. It

appears that the pocket in this structure is wider than in

the other seed-structures, facilitating two potential bind-

ing modes; this is also reflected in the lower binding

energies that were calculated for both ligands on this

structure (vide infra).

Experiments have revealed that DARM (2) binds slightly

stronger to the aptamer than LARM (1), i.e. KD = 135 lMfor LARM and KD = 98 lM for DARM [45]. This is only

marginally reflected in the docking simulations. Consid-

ering the strongest binding ligand to one of the seven seed

structures, one could argue that the calculated KD-value of

Fig. 2 Overlay of the structures of LARM when bound to the aptamer

as determined by NMR (red), and as predicted by the docking

simulation on the aptamer-structure that was obtained after an EM

simulation of the LARM-aptamer complex (yellow). The structures of

the aptamers are shown in grey, the molecular surfaces are depicted in

order to emphasize the binding pockets. a–g correspond to struc-

tures 1–7, respectively, as deposited in the PDB and NDB


123

343 pM for DARM and of 643 pM for LARM reflects this

slightly preferred binding of DARM, and the average

strongest binding energies show a similar trend, with

12.12 ± 0.44 kcal/mol for DARM and 11.61 ± 0.32 k-

cal/mol for LARM. However, these differences in calcu-

lated affinities for LARM and DARM are not significant, as

can be clearly seen in Fig. 4.

Docking of LARG (3) and DARG (4)

After these docking simulations of the amino amide ligandsLARM (1) and DARM (2), docking of the amino acid li-

gands LARG (3) and DARG (4) was performed. Binding

experiments revealed that these two ligands have sig-

nificantly lower affinities for the aptamer, with KD-values

Table 3 Docking results for DARM (2) on seven 1OLD-aptamer structures that were obtained by an EM simulation of the seven ligand-aptamer

complexes



Average binding


KD (pM) NBS nclusters

1 12.39 11.25 ± 0.47 825 52 9

2 12.47 11.94 ± 0.36 720 32 11

3 12.91 12.12 ± 0.44 343 5 13

4 11.84 11.34 ± 0.22 2080 18 8

5 11.77 11.27 ± 0.36 2350 38 8

6 11.41 11.05 ± 0.30 4300 6 13

7a 11.63 11.46 ± 0.16 2970 11 11

11.13 10.65 ± 0.34 6980 62


bind in the binding site. nclusters = number of clusters that have been calculated using a cluster RMSD of 5 Aa Docking of DARM (2) on this aptamer structure resulted in two binding modes in which the guanidine-group was inserted in the binding pocket

Table 4 Docking results for LARG (3) and DARG (4) on seven 1OLD-aptamer structures that were obtained by an EM simulation of the seven

ligand-aptamer complexes



Average binding


KD (pM) NBS nclusters

LARG (3)

1 9.17 8.71 ± 0.31 188,880 17 11

2 9.91 8.99 ± 0.51 54,700 44 14

3 9.42 9.10 ± 0.29 124,470 9 11

4 9.55 8.73 ± 0.46 99,890 49 12

5 9.65 8.82 ± 0.47 84,290 34 12

6 9.35 8.96 ± 0.24 139,780 15 13

7a 8.75 8.73 ± 0.03 388,630 3 19

8.61 8.14 ± 0.32 489,610 41DARG (4)

1 9.17 8.53 ± 0.28 188,250 23 16

2 9.78 8.90 ± 0.56 67,520 56 14

3 9.88 8.89 ± 0.38 57,030 28 11

4 9.55 8.73 ± 0.46 99,890 49 12

5 9.18 8.61 ± 0.34 186,930 38 13

6 9.52 8.97 ± 0.29 105,670 9 14

7 9.07 8.43 ± 0.25 224,770 26 16


bind in the binding site. nclusters = number of clusters that have been calculated using a cluster RMSD of 5 Aa Docking of LARG (3) on this aptamer structure resulted in two binding modes in which the guanidine-group was inserted in the binding pocket


123

of approximately 2.5 mM (for comparison, KD * 100 lMfor LARM and DARM).

In line with the experimentally determined weaker bind-

ing of the amino acid ligands to the aptamer, the docking

simulations predict a similar weaker interaction of both of

these ligands with the aptamer (Table 4). Whereas amino

amide ligands have binding energies that range from

11.37 ± 0.09 to 11.61 ± 0.32 kcal/mol (Table 2) and

10.65 ± 0.34–12.12 ± 0.44 kcal/mol (Table 3) for LARM

(1) and DARM (2), respectively, the amino acid ligands have

binding energies that range from 8.14 ± 0.32–9.10 ± 0.29

to 8.43 ± 0.25–8.97 ± 0.29 kcal/mol, for LARG (3) and

Table 5 Docking results for AGM (5) and EtG (6) on seven 1OLD-aptamer structures that were obtained by an EM simulation of the sevenLARM ligand-aptamer complexes

Structure Best binding free energy (kcal/mol) Average binding free energy (kcal/mol) KD (pM) NBS nclusters

Agmatine, AGM (5)

1 11.16 11.02 ± 0.08 6560 16 5

2 11.85 11.10 ± 0.55 2070 86 3

3 11.84 11.47 ± 0.26 2080 11 5

4 11.75 10.89 ± 0.36 2450 83 3

5 11.41 10.61 ± 0.41 4320 63 8

6 11.37 10.95 ± 0.29 4640 27 7

7 10.92 10.23 ± 0.24 9910 90 6

Ethyl-guanidine, EtG (6)

1 6.84 6.80 ± 0.04 9,770,000 100 1

2 7.45 7.42 ± 0.06 3,470,000 97 2

3 7.35 7.29 ± 0.15 4,100,000 92 3

4 7.43 7.38 ± 0.05 3,610,000 86 2

5 7.46 7.45 ± 0.01 3,400,000 97 3

6 7.18 7.12 ± 0.08 5,480,000 83 3

7 6.52 6.48 ± 0.04 16,760,000 99 2


bind in the binding site. nclusters = number of clusters that have been calculated using a cluster RMSD of 5 A

Fig. 3 Ligands used in the

docking simulation on the

DNA-aptamer for argininamide.

The ionized forms of the ligands

at neutral pH are shown, the

circles highlight the position if

the charges. LARM or DARM

L- or D-argininamide, LARG orDARG L- or D-arginine, AGM

agmatine, EtG ethyl-guanidine,LLys L-lysine, L(NGMe)Arg

NG-methyl L-arginine


123

DARG (4), respectively. These calculated KD-values corre-

spond quitewell to themeasuredKD-values of the amino acid

and amino amide ligands. This lower affinity of the amino

acid ligandwhen compared to their amino amide counterpart

can be attributed to the presence of a negatively charged

carboxylate moiety (pKa * 3.5). This weakens the interac-

tion of the ligand with the negatively charged aptamer

structure, not only in the binding experiments but also in the

docking simulation.

Whereas an energy-minimization of the LARM-aptamer

complex greatly improved the docking simulation of both

ARM ligands, when compared to the simulation performed

on the deposited aptamer-structures (Fig. 4), this difference

is smaller for both ARG ligands. The calculated binding

energy increased for each seed-structure, but the difference

between the two simulations was not significant (Fig. 5).

Docking of AGM (5) and EtG (6)

Besides binding to both enantiomers of ARM and ARG,

the 1OLD-aptamer also binds to two non-amino acid li-

gands that contain a guanidine-moiety, i.e. agmatine, AGM

(5), and ethyl-guanidine, EtG (6) (see Fig. 3 for the

structures). These two achiral ligands had experimentally

determined KD-values of 100 lM for agmatine and

2.5 mM for ethyl-guanidine. To test if the energy-mini-

mized structures were suitable for a docking simulation

using these two ligands, all seven aptamer structures that

Fig. 4 Comparison of the

calculated binding energies ofLARM (1) and DARM (2) whendocked on the NMR-structure of

the aptamer (lighter colored

bars), or on the structure of the

aptamer that was obtained after

an EM simulation of the LARM-

aptamer complexes (darker

colored bars)


calculated binding energies ofLARG (3) and DARG (4) whendocked on the NMR-structure of

the aptamer (light colored bars),

or on the structure of the




colored bars)


123

were obtained were applied in a docking simulation using

these two ligands (Table 5).

Concerning agmatine (AGM, 5), the docking simulation

predicts that this ligand indeed binds in the binding site that

was determined for LARM. The binding energies tend to be

slightly lower than those that were calculated for LARM andDARM, but the differences are not significant (Table 5). Since

the determined KD-values for these three ligands are com-

parable, it can be concluded that the docking simulation is

able to produce a result that is very comparable to the ex-

perimentally determined affinities. When it comes to ethyl-

guanidine (EtG, 6), of which the binding affinity was ex-

perimentally determined to be similar to those of LARG andDARG, the calculated affinities for the aptamer structures are

significantly lower than expected (Table 5). Whereas LARG

and DARG have a binding energy that ranges from

8.14 ± 0.32–9.10 ± 0.29 and 8.43 ± 0.25–8.97 ± 0.29 k-

cal/mol (Table 4), for LARG and DARG, respectively, the

range of 6.48 ± 0.04–7.45 ± 0.01 kcal/mol that was calcu-

lated for EtG (6) is significantly lower over the entire range.

Regarding this, it should be noted that these two ligands

not only lack a chiral center, they are also notably smaller

than both ARM and both ARG ligands: agmatine lacks a

C(O)NH2 or C(O)O- moiety, resulting in a 25 % reduction

in the number of heavy atoms (see Fig. 3 for the structures

of the ligands). Ethyl-guanidine has only half of the

number of heavy atoms that are present in ARM or ARG.

The significant reduction of the size of the ligand reduces

the potential number of interactions between ligand and

aptamer, resulting in the lower calculated binding energy

(vide infra).

Comparison of the docking results performed on the

NMR structures of the aptamer and on the structures of the

aptamer that were obtained by EM simulation of theLARM-aptamer complex shows that in the case of these

two ligands, a significant but small improvement is again

achieved in most of the cases (Fig. 6). Only with struc-

ture 5 and 7 for agmatine (5) and structure 7 for ethyl-

guanidine (6) does the EM not lead to stronger binding

ligands.

Docking of LLYS (7) and NG-methyl L-arginine (8)

Lastly, we studied the importance of the mono-substituted

guanidine moiety for the interaction between a ligand and

the aptamer by docking L-lysine (LLYS, 7) and NG-methyl

L-arginine on the aptamer structure. It should be noted that

a KD-value could not be determined experimentally for

either of these two ligands [44]. The inability of L-lysine to

bind to the aptamer is reflected in the low calculated

binding affinity and the low number of ligands that are

placed in the expected binding pocket (Table S9). The

average binding affinities range from 7.77 ± 0.20 to

8.91 ± 0.26 kcal/mol for LLYS. This is slightly lower than

that for LARG, but the difference is small, indicating that

the docking simulation is too optimistic when it comes to

the docking of LLYS. A similar outcome is obtained for

NG-methyl L-arginine (L(NGMe)Arg, 8, Table S9). Whereas

the docking simulation predicts that the methylated LARG

ligand (8) is a stronger binder for the aptamer than LARG

(3), a binding constant for ligand 8 could not be determined

experimentally. It appears that the computational analysis

of the binding energy of the docked ligands of this NG-

methylated derivative of LARG is rather optimistic.

Discussion

A docking simulation is an approximation of actual binding

events, and the binding energy is calculated based on

several terms, as shown in Eq. 1 (see ‘‘Materials and

methods section’’) [46]. Of these components, DGsol is the


calculated binding energies of

AGM (5) and EtG (6) whendocked on the NMR-structure of

the aptamer (light colored bars),

or on the structure of the




colored bars)


123

most challenging term as it models desolvation of the

residues at the interface between the ligand and the ap-

tamer, and the hydrophobic effect. Since the procedure by

which affinities are calculated has been calibrated using the

binding-constants of known protein–ligand complexes, it

was unclear if reliable results could also be obtained for

DNA-aptamer ligand complexes.

Although the absolute values of the calculated and ex-

perimentally determined binding affinities are significantly

different, a very good correlation between the computed and

experimental values exists (Fig. 7). Onlywhen the size of the

ligand is substantially altered, especially with the removal of

potential H-bond partners as in EtG, and to a lesser extent

AGM, the predicted affinity deviates from the determined

values. For the moment, we attribute this deviation to an

improper calculation of the entropic factor that is caused by

the replacement of water molecules at the ligand-aptamer

binding interface. Ligands LLYS (7) and L(NGMe)Arg (8) are

omitted in this graph since their binding affinity for the ap-

tamer could not be determined experimentally.

In this study, we assessed if docking simulations can

reproduce the trend in the binding affinities of several

known ligand-aptamer complexes, and which would be the

most suitable approach by which the best results could be

obtained. Whereas many docking simulation have been

performed on protein–ligand complexes, only a few have

studied the interactions of small molecules with DNA [53].

However, most of these DNA-oriented studies were only

directed to groove-binding molecules [54–57] or to DNA-

intercalators [58], and no study has compared binding

events using several known ligand-aptamer complexes [33,

41, 59]. Even though the docking simulation on the

structure of the aptamer as deposited in the databank pro-

duces acceptable results, a straightforward EM of the li-

gand-aptamer complex before the docking simulation is

performed results in better docking outcomes, as was in-

ferred from the improved binding energies and a larger

number of ligands that were docked in the binding-site.

Conclusions

We demonstrated that the implementation of AutoDock in

the YASARA Structure software package is well suited to

perform docking simulations of ligand-aptamer complexes

routinely. We show that, although docking the ligands on

the NMR-structure of the aptamer is feasible, the docking

results can be significantly improved by performing an EM

simulation of the ligand-aptamer complexes before the

docking simulation is performed.

In case of the LARM-aptamer complex, that was de-

posited in the PDB and NDB with code 1OLD, the best

docking results were obtained using aptamer structures

that were obtained by a straightforward EM simulation of

the LARM-aptamer complexes. The binding energies im-

proved, and the binding mode of the docked ligand re-

sembled that of the experimentally characterized structure

more closely; in each docking simulation, the docked li-

gand with the strongest binding energy was placed in the

expected binding site of the aptamer.

K DK Dµ

µ

Fig. 7 Comparison of the KD-values that were: experimentally

determined KD-values (in lM, wide grey bars behind the green bars,

the KD-values are given above the bars), the average KD-values as

predicted by a docking simulation on each structure of the aptamers as

determined by NMR (in pM, light bars), and the average KD-values as

predicted by a docking simulation on each of the seven structures of

the aptamers after EM of the LARM-aptamer complexes (in pM, dark

green bars). See Fig. 3 for the structures of the ligands, the average

KD-values and their error bars were calculated using the values of the

strongest binding ligand


123

This study shows that the YASARA Structure software

package is useful to evaluate binding events between li-

gands and a DNA-aptamer. Using these results, we can

now better understand the binding interactions that dom-

inate ligand-aptamer complexes. We realize that our results

present the first step towards the implementation of mole-

cular dynamics to predict and understand the ligand-ap-

tamer interactions at the molecular level. The validation of

the approach to other ligand-aptamer complexes is a future

challenge. It is our hope that such molecular dynamics

computations will provide a powerful predictive tool to

identify the structures of ligand-aptamer complexes, and to

discover new ligand-mutated aptamer structures.

Materials and methods

For the validation of the YASARA Structure software

package with respect to the docking of ligands to DNA

aptamers, we studied the DNA-aptamers for argininamide

(PDB-code: 1OLD1) [60] using AutoDock [61] that is

embedded in YASARA Structure. Molecular graphics were

created using YASARA (www.yasara.org) and POVRay

(www.povray.org) (Figs. 1, 2, and Figs. S1 and S2).

Structures of the ligands (Fig. 3) were drawn using

ChemDraw Ultra (version 12.0.3.1216). Figs. 4, 5, 6 and 7

have been prepared using Microsoft Excel.

Computations were performed in YASARA Structure

(version 14.8.17) [41] using the AMBER03 force field [64]

(which shares the nucleic acid parameters with AMBER99

[62]) and point charges derived with the AM1-BCC

method [63]. EMs were performed on the content of the

neutralized cell using the AMBER03 (with PME for

longrange electrostatics [65], and a charge cut off at

7.86 A) force field and the ‘em_run.mcr’ macro. This

macro fills the simulation cell with water, changes the

oxygen atoms that are closest to the negatively charged

phosphate backbone to positively charged potassium cations

in order to neutralize the content of the simulation cell, runs

an EM of the water molecules that were added, and then

runs the main EM using the steepest descent temperature

control; the simulation converges as soon as the energy

improves by less than 0.05 kJ/mol per atom during 200 steps

[66]. Since only short minimizations were run, we did not

apply the parmbsc0 correction [47], which improves the

stability of nucleic acids during long-term molecular dynamics

simulations.

Docking studies on the aptamer were performed using

the built-in docking simulation macro ‘dock_run.mrc’ with

0.375 A grid resolution, 100 runs, and a cluster RMSD of

5 A for the docking conformations. The macro applies

AutoDock 4.0 [46], which uses an empirical scoring system

based on the free energy of binding [67], with the LGA

[68]. The docking simulation was performed on the entire

aptamer structure, i.e. a global docking simulation. In this,

a grid-map of the entire rigid aptamer structure is calcu-

lated in which information on electrostatics, H-bond

forming potential, and steric constraints are stored. Then,

the flexible ligand is fitted to this rigid grid-map. In this

process, the ligand is initially positioned outside the ap-

tamer, at random positions, and explores translations, ori-

entations, and conformations until a binding site is found.

After the ligand has been docked to the aptamer, the affi-

nity of the ligand and the predicted binding site is calcu-

lated by Eq. 1,

DGbind ¼ DGVdW þ DGH�bond þ DGelec þ DGconform

þ DGtor þ DGsol ð1Þ

where DGVdW models dispersion/repulsion (i.e. Van der

Waals) interactions, DGH-bond models hydrogen bonding

interactions, DGelec models electrostatics interactions,

DGconform models the deviations from covalent geometry,

DGtor models the restriction internal rotor and global ro-

tation and translation, and DGsol models desolvation upon

binding [46]. Here it should be recalled that the binding

energies that are reported are identical to the DGbind, which

is the free energy of binding (which has a negative value

for a binding event), but that the sign is flipped from - to

?. The average binding energies and their standard de-

viations were determined using Excel. The numbers of li-

gands that bind in the binding site were determined by

visual inspection of all docked ligands. The numbers of

clusters were calculated by the YASARA Structure soft-

ware package using a cluster RMSD of 5 A.

Acknowledgments This research is supported by the Israel Science

Foundation.

Conflict of Interest The authors declare that they have no conflict

of interest.

References

1. Mayer G (2009) Angew Chem Int Ed 48:2672–2689

2. Patel DJ, Suri AK (2000) Rev Mol Biotechnol 74:39–60

3. Campolongo MJ, Tan SJ, Xu JF, Luo D (2010) Adv Drug Deliv

Rev 62:606–616

4. Gallas A, Alexander C, Davies MC, Purib S, Allen S (2013)

Chem Soc Rev 42:7983–7997

5. Liu X, Xu Y, Yu T, Clifford C, Liu Y, Yan H, Chang Y (2012)

Nano Lett 12:4254–4259

1 It should be noted that upon retrieval of the structure, the

C-terminal moiety of the LARM ligands was not the expected

carboxamide, but an enol (i.e. Ca(OH)=CH2). This was first corrected

into the carboxamide before the rest of the computations were

performed.


123

http://www.yasara.org

http://www.povray.org

6. Afonin KA, Grabow WW, Walker FM, Bindewald E, Dobro-

volskaia MA, Shapiro BA, Jaeger L (2011) Nat Protoc

6:2022–2034

7. Willner I, Zayats M (2007) Angew Chem Int Ed 46:6408–6418

8. Kolpashchikov DM (2010) Chem Rev 110:4709–4723

9. Wang F, Lu CH, Willner I (2014) Chem Rev 114:2881–2941

10. Famulok M, Mayer G (2011) Acc Chem Res 44:1349–1358

11. Wilner OI, Willner I (2012) Chem Rev 112:2528–2556

12. Liu D, Park SH, Reif JH, La Bean TH (2004) Proc Natl Acad Sci

USA 101:717–722

13. Krishnan Y, Simmel FC (2011) Angew Chem Int Ed

50:3124–3156

14. Teller C, Willner I (2010) Curr Opin Biotechnol 21:376–391

15. Yan H, Zhang X, Shen Z, Seeman NC (2002) Nature 415:62–65

16. Bath J, Turberfield AJ (2007) Nat Nanotechnol 2:275–284

17. Beissenhirtz MK, Willner I (2006) Org Biomol Chem

4:3392–3401

18. Okamoto A, Tanaka K, Saito I (2004) J Am Chem Soc

126:9458–9463

19. Stojanovic MN, Stefanovic D (2003) Nat Biotechnol

21:1069–1074

20. Tuerk C, Gold L (1990) Science 249:505–510

21. Ellington AD, Szostak JW (1990) Nature 346:818–822

22. Sefah K, Shangguan D, Xiong X, O’Donoghue MB, Tan W

(2010) Nat Protoc 5:1169–1185

23. Feigon J, Dieckmann T, Smith FW (1996) Chem Biol 3:611–617

24. Thomas JR, Hergenrother PJ (2008) Chem Rev 108:1171–1224

25. Mannironi C, DiNardo A, Fruscoloni P, Tocchini-Valentini GP

(1997) Biochemistry 36:9726–9734

26. Padlan CS, Malashkevich VN, Almo SC, Levy M, Brenowitz M,

Girvin ME (2014) RNA 20:447–461

27. Krauss IR, Pica A, Merlino A, Mazzarella L, Sica F (2013) Acta

Cryst D69:2403–2411

28. McKeague M, DeRosa MC (2012) J Nucl Acids Article ID

748913, 20 pages. See also: the Aptamer Base (http://apta

merbase.semanticscience.org/)

29. Stelzer AC, Frank AT, Kratz JD, Swanson MD, Gonzalez-Her-

nandez MJ, Lee J, Andricioaei I, Markovitz DM, Al-Hashimi HM

(2011) Nat Chem Biol 7:553–559

30. Fulle S, Christ NA, Knestner E, Gohlke H (2010) J Chem Inf

Model 50:1489–1501

31. Haller A, Souliere MF, Micura R (2011) Acc Chem Res

44:1339–1348

32. Fulle S, Gohlke H (2010) J Mol Recognit 23:220–231

33. Pfeffer P, Gohlke H (2007) J Chem Inf Model 47:1868–1876

34. Fadrna E, Spackova N, Sarzynska J, Koca J, Orozco M, Chea-

tham TE III, Kulinski T, Sponer J (2009) J Chem Theory Comput

5:2514–2530

35. Westhof E, Cruz JA (2009) Cell 136:604–609

36. Phan AT, Kuryavyi V, Patel DJ (2006) Curr Opin Struct Biol

16:288–298

37. Boehr DD, Nussinov R, Wright PE (2009) Nat Chem Biol

5:789–796

38. Beckers MLM, Buydens LMC (1998) J Comput Chem

19:695–715

39. Zhang Q, Sun X, Watt ED, Al-Hashimi HM (2006) Science

311:653–656

40. Bosshard HR (2001) News Physiol Sci 16:171–173

41. Guilbert C, James TL (2008) J Chem Inf Model 48:1257–1268

42. Krieger E, Darden T, Nabuurs S, Finkelstein A, Vriend G (2004)

Proteins 57:678–683

43. Caulfield T, Devkota B (2012) Proteins 80:2489–2500

44. Harada K, Frankel AD (1995) EMBO J 14:5798–5811

45. Lin PO, Tong SJ, Louis SR, Chang Y, Chen WY (2009) Phys

Chem Chem Phys 11:9744–9750

46. Morris GM, Goodsell DS, Halliday RS, Huey R, Hart WE, Belew

RK, Olson AJ (1998) J Comput Chem 19:1639–1662

47. Perez A, Marchan I, Svozil D, Sponer J, Cheatham TE III,

Laughton CA, Orozco M (2007) Biophys J 92:3817–3829

48. Holt PA, Chaires JB, Trent JO (2008) J Chem Inf Model

48:1602–1615

49. Lin PO, Tsai C-W, Wu JW, Ruaan R-C, Chen W-Y (2012)

Biotechnol J 7:1367–1375

50. Bisignano P, Moran O (2010) Biochimie 92:51–57

51. Krieger E, Dunbrack RL, Hooft RW, Krieger B (2012) Methods

Mol Biol 819:405–421

52. Albada HB, Rosati F, Coquiere D, Roelfes G, Liskamp RMJ

(2011) Eur J Org Chem 2011:1714–1720

53. Snyder RD, Holt PA, Maguire JM, Trent JO (2013) Environ Mol

Mutagen 54:668–681

54. Ricci CG, Netz PA (2009) J Chem Inf Model 49:1925–1935

55. Netz PA (2012) Int J Quant Chem 122:3296–3302

56. Tooth YY, Lipkowitz KB, Long EC (2006) J Chem Theory

Comput 2:1453–1463

57. Rohs R, Bloch I, Sklenar H, Shakked Z (2005) Nucl Acids Res

33:7048–7057

58. Gilad Y, Senderowitz H (2014) J Chem Inf Model 54:96–107

59. Reshetnikov R, Golovin A, Spiridonova V, Kopylov A, Sponer J

(2010) J Chem Theory Comput 6:3003–3014

60. Lin CH, Patel DJ (1996) Nat Struct Biol 3:1046–1050

61. Morris GM, Huey R, Lindstrom W, Sanner MF, Belew RK,

Goodsell DS, Olson AJ (2009) J Comput Chem 30:2785–2791

62. Wang J, Cieplak P, Kollman PA (2000) J Comput Chem

21:1049–1074

63. Jakalian A, Jack DB, Dayly CI (2002) J Comput Chem

23:1623–1641

64. Duan Y, Wu C, Chowdhury S, Lee MC, Xiong G, Zhang W,

Yang R, Cieplak P, Luo R, Lee T (2003) J Comput Chem

24:1999–2012

65. Kawata M, Nagashima U (2001) Chem Phys Lett 340:165–172

66. Krieger E, Nielsen JE, Spronk CA, Vriend G (2006) J Mol Graph

Model 25:481–48667. Huey R, Morris GM, Olson AJ, Goodsell DS (2007) J Comput

Chem 28:1145–1152

68. Solis FJ, Wets RJB (1981) Math Oper Res 6:19–30


123

http://aptamerbase.semanticscience.org/

http://aptamerbase.semanticscience.org/

Documents

Computational docking simulations of a DNA-aptamer for argininamide and related ligands