FULL PAPER
Computationally Selected Epitopes
Integrated Approaches Toward High-Affinity Artificial
Protein Binders Obtained via Computationally Simulated
Epitopes for Protein Recognition
Zeynep Altintas,* Aref Takiden, Tillmann Utesch, Maria A. Mroginski,
Bianca Schmid, Frieder W. Scheller, and Roderich D. Süssmuth
Widely used diagnostic tools make use of antibodies recognizing targeted
molecules, but additional techniques are required in order to alleviate the
disadvantages of antibodies. Herein, molecular dynamic calculations are
performed for the design of high affinity artificial protein binding surfaces
for the recognition of neuron specific enolase (NSE), a known cancer
biomarker. Computational simulations are employed to identify particularly
stabile secondary structure elements. These epitopes are used for the
subsequent molecular imprinting, where surface imprinting approach is
applied. The molecular imprints generated with the calculated epitopes
of greater stability (Cys-Ep1) show better binding properties than those of
lower stability (Cys-Ep5). The average binding strength of imprints created
with stabile epitopes is found to be around twofold and fourfold higher for
the NSE derived peptide and NSE protein, respectively. The recognition of
NSE is investigated in a wide concentration range, where high sensitivity
(limit of detection (LOD) = 0.5 ng mL−1) and affinity (dissociation constant
(Kd) = 5.3 × 10−11 m) are achieved using Cys-Ep1 imprints reflecting the stable
structure of the template molecules. This integrated approach employing
stability calculations for the identification of stabile epitopes is expected
to have a major impact on the future development of high affinity protein
capturing binders.
1. Introduction
The recognition of biomolecules on surfaces or in solution
environments plays a key role in all processes of life and is
technologically utilized in the life science disciplines such as
Dr. Z. Altintas, A. Takiden, Dr. T. Utesch, Prof. M. A. Mroginski,
B. Schmid, Prof. R. D. Süssmuth
Institute of Chemistry
Technical University of Berlin
10623 Berlin, Germany
E-mail: zeynep.altintas@tu-berlin.de
Prof. F. W. Scheller
Institute of Biochemistry and Biology
University of Potsdam
14476 Potsdam, Germany
The ORCID identification number(s) for the author(s) of this article
can be found under https://doi.org/10.1002/adfm.201807332.
DOI: 10.1002/adfm.201807332
Adv. Funct. Mater. 2019, 1807332 1807332  (1 of 11)
www.afm-journal.de
biomedicine. Research in biomedicine
and biochemistry makes use of binding
molecules that specifically capture a pro-
tein from a complex medium for either
preparative or analytical use.[1–3] These
recognizing molecules, foremost anti-
bodies, generally are of proteinacious
origin and thus underlie the typical
limitations of biomolecules: tendency to
denaturation on surfaces, instability over
time, and high production costs.[4] An
alternative to protein-based recognition
molecules are imprinted polymers. These
artificial protein binders bear a consid-
erable potential to overcome the draw-
backs of proteins. Molecular imprinting
forms populations of specific binding
sites embedded in robust polymer net-
works by polymerization of functional
monomers and cross-linking agents in the
presence of an imprint molecule, which
is either the whole target molecule or
some fragment of it.[5,6] Recent advances
in molecular imprinting techniques have
introduced these artificial recognition
elements to in vivo imaging studies,[7,8]
target-induced recruiting of membrane
receptors,[9] engineered nanoparticles for medical applica-
tions,[10] diagnostics,[11–13] and bioselective membrane devel-
opment for selective removal of contaminants.[14] Common to
most of these studies surface-imprinting techniques or epitope
imprinting approach have been utilized to address problems
in protein imprinting.[15–17] Nevertheless, robust imprinting
methods that provide protein binders, which can compete
with established bioaffinity materials as antibodies in terms
of affinity, sensitivity, and selectivity, are still missing, mainly
due to the structural complexity and flexibility of protein
molecules.
Another key point that needs to be taken into considera-
tion is the common use of surface exposed linear peptides in
epitope imprinting.[4,17] The influence of secondary structure
elements, with an emphasis on α-helices, β-sheets, or turn
structures, on molecular imprinting has not been investigated
so far. The conformational stability of epitopes also depends
on the length and amino acid composition of the peptide as
well as its location in protein. In this proof-of-concept study,
© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.afm-journal.de

MD calculations and selection of epitopes

NSE

Ep1
Ep2
Ep3

RMSD (Å)
Ep4
Ep5
Ep6

Time (ns)
Protein recognition and capture

Whole protein
Protein or epitope
Imprinted polymer with
capture
corresponding cavities to
Epitope epitope

Gold sensor chip Protein or epitope

extraction

Figure 1.  Integrated research methodology illustrating computational modeling of NSE epitopes and the use of epitopes for molecular imprinting and
target recognition.

neuron specific enolase (NSE) is used as a model protein,
which is an important biomarker for small cell lung cancer
(Figure  1). From six a-helical epitopes of NSE, we employed
molecular dynamic (MD) calculations in water and buffered
environment, and determined differently stabile epitopes. We
combined our previously developed epitope surface imprinting
approach[17] with the most and least stable epitopes, where
significantly higher binding strength was obtained for target
molecules using the molecular imprints created with the stable
epitope. Our studies confirm that this integrated approach can
be potentially extended to the establishment of epitope libraries
for protein molecules to obtain highly efficient artificial protein
binders. Moreover, this approach is expected to decrease the
development time and research cost by eliminating unsuitable
epitope candidates.
2. Results and Discussion
2.1. Computational Analysis of NSE Biomarker and Its Epitopes
MD simulations are performed to investigate the stability of
six sequentially different (see Figure S1, Supporting Informa-
tion) epitopes in aqueous solution. All epitopes (10–12 mer)
are taken from surface regions of NSE as shown in Figure 2A.
Their location in the protein makes them potential target

Figure 2.  A) Representation of epitope locations in NSE protein. B) Root-mean-square deviation (RMSD) of the protein backbone during 300 ns for
the epitopes (Ep1: black, Ep2: blue, Ep3: cyan, Ep4: red, Ep5: violet, Ep6: green). The crystallographic structure served as reference.

Adv. Funct. Mater. 2019, 1807332 1807332  (2 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
molecules for molecular imprinting. Regions from inner parts
of the enzyme are not considered, since they are most likely
not involved in binding and recognition. The evaluation of the
root-mean-square deviation (RMSD) of the protein backbone in
300 ns long MD simulations indicates significant differences
in stability of the epitopes (Figure 2B). Immediately after water
contact, all peptides undergo structural changes, since their con-
formations are not any longer stabilized by the protein matrix
present in the crystal structure. After 100 ns, the differences
between the epitopes become more pronounced. Ep5 (average
RMSD: 3.73 ± 0.87 Å) and Ep2 (average RMSD: 3.72 ± 0.78 Å)
show strong fluctuations, indicating very unstable structure.
Ep4 (average RMSD: 3.43 ± 0.59 Å) and Ep6 (average RMSD:
3.32 ± 0.28 Å) reach a stable conformation showing RMSD
values of more than 3 Å with respect to the protein embedded
state. Compared to these values, Ep1 (average RMSD: 2.44 ±
0.54 Å) and Ep3 (average RMSD: 2.52 ± 0.68 Å) are much closer
to the crystallographic structure (average RMSD < 3 Å). The
higher structural fluctuations observed for Ep3 make Ep1 the
most suitable target for molecular imprinting according to
the RMSD analysis. The RMSD values are calculated during the
MD simulation trajectory. This means that each snapshot,
taken at a certain time, is aligned with the crystallographic
structure and the resulting RMSD difference between the
backbone atoms of both structures is recorded. The condi-
tions in all simulations are the same so that “external factors”
are reduced to the minimum. The noise levels indicate that all
epitopes have a certain degree of flexibility in aqueous solution.
This observation is expected, since even large biomolecules
exhibit internal motions at room temperature. The RMSD
results are also in overall agreement with the evolution of the
secondary structure (Figure  3). Due to its high flexibility, Ep5
does not form a well-defined secondary structure during the
300 ns long trajectory. Ep2, which also shows high fluctuations
Figure 3. Secondary structure evolution of the six epitopes in aqueous solution during 300 ns. Pink, blue, green, and white indicate α-helices,
3–10 helices, turns, and coils, respectively.
Adv. Funct. Mater. 2019, 1807332 1807332  (3 of 11)
www.afm-journal.de
in the RMSD value, loses the large parts of its α-helical fold and
adopts a partially disordered conformation. Ep1, Ep4, and Ep6
stay relatively close to their α-helical fold observed in the crys-
tallographic structure, while Ep3 transforms into a 3–10 helix
(Figure 3; Table S1, Supporting Information). Taking the RMSD
values and secondary structure plots together, the MD simula-
tions suggest Ep1 as best and Ep5 as worst target for molecular
imprinting. Therefore, these epitopes are selected to synthesize
the molecular imprints and perform comparative experimental
studies.
2.2. Synthesis of Epitope Imprints
The epitope imprints are prepared using scopoletin as func-
tional monomer. The polyscopoletin-based imprinted films
result in nonconductive and hydrophilic polymers, consenting
to noncovalent interactions with the epitope template.[18]
Voltammetry techniques allow following up entire molecular
imprinting process by using ferricyanide as a redox marker.
The oxidation and reduction peaks of the cyclic voltammetry
(CV) curves are clearly suppressed after the preadsorption of
Cys-Ep1 on the gold surface (Figure  4A, curves a and b). The
signal suppression drastically increases after the formation of
polyscopoletin polymer around the template (Figure 4A, curve c).
The removal of the template, as a result of anodic potential
application to the surface, relatively increases the oxidation and
reduction peaks when compared to the polymerization process
(Figure 4A, curve d). During the template removal process, the
applied potential (0.9 V) oxidizes the thiol groups of cysteine
peptide attached to the Au, and thus detached the template.
Preadsorption of the template molecule via the thiol group of
cysteine on the gold surface is favorable for the formation of
a stable molecularly imprinted polymer (MIP) film, as it can
© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.afm-journal.de

15 6
A) B) a
a
4
10
d

b 2
5
d 0 b
I/mA

I/mA
0 c
-2 c
Peptide adsorption (a)
Bare chip (a)
-5 Polymerization (b)
Peptide adsorbtion (b) -4
Polymerization (c)
TR 0.7 V (c)
TR 0.9 V (d) TR 0.9 V d)
-6
-10

-8
-15
-0.2 0.0 0.2 0.4 0.6 -0.2 0.0 0.2 0.4 0.6
E/V E/V
15
C) 15
D)
c
10 a 10

5 5

b
I/mA

0
I/mA

Bare chip
-5 Bare chip (a) -5 Peptide adsorbtion
TR 1.2 V (b) Polymerization
TR (0.7 V)
-10 TR 1.4 V (c) TR (0.9 V)
-10
TR (1.2 V)
TR (1.4 V)
-15
-15

-0.2 0.0 0.2 0.4 0.6 -0.2 0.0 0.2 0.4 0.6
E/V E/V

Figure 4.  A) Characterization of molecular imprinting process from bare chip surface to the end of template removal at optimum conditions using cyclic
voltammetry (CV). B) Comparison of template removal efficiency of 0.7 and 0.9 V anodic potential applications. C) Substantial or complete peeling
off polymer film due to high voltage (1.2 and 1.4 V) application for template removal. D) Overall comparison of molecular imprinting process along
with four different anodic potential applications for template removal step. Voltammograms recorded in K3[Fe(CN)6] (0.1 m)/potassium chloride (1 m).

avoid nonspecific interactions and form stabile cavities. The
oxidation potential plays key role for efficient template removal.
The potential application at lower voltage (i.e., 0.7 V) may not
remove the template molecules from the polymer network
(Figure 4B), whereas higher potentials (i.e., 1.2 and 1.4 V) may
strip off the entire polymer film (Figure 4C,D).
The thickness of the polymer films is another key point to
acquire successful imprints for target recognition.[13] Thinner
films are desirable over thicker films during MIP synthesis,
since the template removal is generally difficult in the latter
case. Herein, 30 polymerization cycles are found optimal, as it
delivers an appropriate film thickness that allows eradicating
the template easily (see Figure S2, Supporting Information).
2.3. Atomic Force Microscopy (AFM) Characterization
of Molecular Imprinting Process
To further analyze the imprinting process, the bare gold quartz
crystal microbalance (QCM) surface, template imprinted
surface, and template removed surface are imaged using
AFM technique. Additionally, we visualize the nonimprinted
polymer (NIP, produced under the optimal polymerization
conditions in the absence of the template mole­cules) in com-
parison. The 2D and 3D surface topography images (Figure S3,
left and middle, Supporting Information) along with cross-
section profiles (Figure S3, right, Supporting Information)
are recorded and analyzed. The cleaned QCM surface acted
as a reference point with a root mean square (RMS) value
of 0.33 ± 0.02 nm and a cross-sectional height of 1.45 nm
(Figure S3A, left and middle, Supporting Information). The
AFM images associated with electropolymerization of the mon-
omer on preadsorbed peptide on the gold chip give an RMS
roughness of 4.4 ± 0.6 nm with a cross-sectional height value of
19.3 nm (Figure S3B, Supporting Information). This is a signifi-
cant increase compared to the bare QCM gold chip and confirms
the formation of a thin polymer film in the nm range on the gold
electrode. Application of anodic potential at 0.9 V removes the
template molecules from the polymer network, thus, decreases
the RMS roughness (3.3 ± 0.3 nm) and cross-sectional height
(15.5 nm), as shown in Figure S3C (Supporting Information).
Considering the small size of the template molecule (>1 nm),[19]
the difference between the templated and template free polymers
is expected. The AFM results for NIP further confirm the size

Adv. Funct. Mater. 2019, 1807332 1807332  (4 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.afm-journal.de

and presence of template molecules in molecular imprinting
process, as the nonimprinted polymer shows 1.2 nm less RMS
roughness (3.2 ± 0.8 nm) (Figure S3D, Supporting Information).
The cross-section analysis and the topology images show an
overall smoother surface for the NIP, varying in large parts at
around 5 nm. Excluding the bare chip, which naturally has the
smallest roughness (0.33 nm), a comparison of the RMS rough-
ness of the samples results in two observations: 1) The rough-
ness decreased when changing from the templated surface to the
template-removed MIP, which is attributed to emptied cavities
in polymer network. 2) The nonimprinted polymer shows clearly
lower roughness, indicating that no peptide is caught inside the
polymer, resulting in an overall more homogeneous surface. To
visualize the details of the MIP and nonimprinted polymer sur-
faces on the gold chip, we perform further AFM measurements
in smaller scanning areas (i.e., 3 and 1 µm2). The RMS values
and cross-sectional profiles of both polymer surfaces provide
similar outcomes (Figures S4–S9, Supporting Information) as
obtained for the 10 µm2 area.
We also apply chemical mapping for template-free MIP and
NIP surfaces based on phase image analysis by AFM. The
phase image is mainly characterized by different chemical
components.[20] For example, there are two different morpho-
logical phases on template-free MIP surfaces, indicating empty
cavities and polymer network (Figure S10A–C, Supporting
Information), whereas only one phase is dominated on the NIP
surface (Figure S10D–F, Supporting Information). More impor-
tantly, the changes on the surfaces resulted in substantially
higher degree phase lag for MIPs (138° at 10 µm2) compared
to the nonimprinted polymer (27.2° at 10 µm2). To better
identify and visualize the surface morphologies, the analyses
are also performed at smaller scanning areas (3 and 1 µm2),
revealing the similar trends obtained at 10 µm2 for the target
(Figure S10B,C, Supporting Information) and nonimprinted
(Figure S10E,F, Supporting Information) polymer films.
2.4. Target Recognition and Capture Using Molecular Imprints
The detection of NSE derived peptide (Ep1) and NSE protein
using the Cys-Ep1 imprints is investigated over a wide concen-
tration range. The imprints recognized the peptide target in
the range of 2–500 × 10−6  m and this can be demonstrated by
square wave voltammograms (Figure  5A). Each concentration

3.5
A) 2 µM peptide
3.5
B)
3.0 4 µM peptide
8 µM peptide 3.0
2.5 16 µM peptide
32 µM peptide
Current (µA)

64 µM peptide 2.5
2.0
128 µM peptide
Current (µM)

256 µM peptide
1.5 500 µM peptide 2.0

1.0
1.5

0.5
1.0
0.0
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 0 100 200 300 400 500
Potential (V) Peptide concentration (µM)
70
D)
Relative signal / Peptide concentration (% µM)

60
C) 3.0
Relative signal suppression (%)

50 2.5

40
2.0

30
1.5
20
1.0
10

0 0.5

12 18 24 30 36 42
0 100 200 300 400 500
Relative signal (%)
Peptide concentration (µM)

Figure 5.  A) Square wave voltammograms recorded for the concentration dependent recognition of Ep1 in a range from 2 to 500 × 10−6 m. B) Raw data
of triplicate measurements obtained for an experimental series using square wave voltammetry technique. C) Concentration dependent Ep1 detection
using the Cys-Ep1 imprinted MIP electrodes (n = 6). D) The Scathard plot resulting in dissociation constant (Kd) of 12.2 × 10−6 m. Slope: −0.08037.

Adv. Funct. Mater. 2019, 1807332 1807332  (5 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.afm-journal.de

12
A) 14
B)
0.5 ng mL NSE
10
1 ng mL NSE 12
5 ng mL NSE
8 10 ng mL NSE
20 ng mL NSE 10

Current (µA)
Current (µA)

6 40 ng mL NSE
50 ng mL NSE
100 ng mL NSE 8
4

6
2

0 4

-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 0 20 40 60 80 100

Potential (V) Protein Concentration (ng mL )

80
C)
70
Relative signal suppression (%)

60

50 24

Relative Signal/Concentration (%/ng mL-1

20
40
16

12
30
8

20 4

0
10
10 20 30 40 50 60 70

0
0 20 40 60 80 100

Protein concentration (ng mL-1)

Figure 6.  A) Square wave voltammograms showing the NSE protein capture in a concentration range of 0.5–100 ng mL−1 (suppression of the current
signal is proportional to the concentration of NSE protein loaded to the MIP electrode surface). B) Raw data of triplicate measurements for one
experimental series with standard deviations. C) Overall results of six separate experiment series for NSE binding assays using the Cys-Ep1 imprinted
MIP electrode (n = 6). Inset: The Scathard plot, Kd = 5.3 × 10−11 m, Slope: −0.39673.

of experimental series is measured three times after 30 min
incubation on the MIP electrode to determine the variation of
current signal (Figure 5B). The raw data of six separate experi-
ment series are converted into the relative signal suppression
(Figure 5C), where small variations are observed on data with
a correlation coefficient of 0.993 based on the logarithmic
regression analysis (Figure S11, Supporting Information). The
affinity of the imprinted polymers toward the NSE derived pep-
tide is calculated according to these data by using the Scatchard
plot,[21] revealing a dissociation constant of 12.2 × 10−6  m
(Figure 5D).
The recognition of NSE is investigated under the same condi-
tions using the Cys-Ep1 imprinted MIP. NSE can be measured
in a concentration range from 0.5 to 100 ng mL−1 (Figure 6A,B).
A high affinity between the epitope imprints and the NSE pro-
tein is found based on the Scatchard plot, which results in a Kd
of 5.3 × 10−11  m (Figure 6C). This affinity is significantly higher
than those of many natural and synthetic receptors reported
so far. For example, Cys-modified cytochrome c-derived
peptide polyscopoletin imprints recognized and captured the
cytochrome c-derived peptide and cytochrome c protein with
affinity constants of 2.51 × 10−6 and 8.5 × 10−6 m, respectively.[17]
A photochemically polymerized epitope MIP, synthesized for
cytochrome c using a different monomer mixture, led to higher
affinity (7.23 × 10−9  m), where a fluorescence-based NanoO-
range protein assay was employed for target detection.[4]
The recognition and quantification of protein-based dis-
ease biomarkers with high sensitivity and selectivity can be
achieved using different molecular imprinting techniques.[21–23]
For example, the detection limits of 2 ng mL−1, 7.7 µg mL−1,
0.5 U mL−1, and 0.10 U mL−1 are reported for HIV-1
glycoprotein (gp41),[24] transferrin,[21] CA 125,[22] and CA15-3,[23]
respectively. The molecular imprints of gp41 are prepared
based on epitope imprinting using dopamine as a functional
monomer,[24] whereas transferrin imprints are synthesized with
scopoletin monomer by following whole protein imprinting
approach.[21] Phenol[22] and toluidine blue[23] are also suitable
functional monomers for the preparations of protein binders
in MIP format by employing the electropolymerization tech-
nique. Whole protein imprinting shows several drawbacks due

Adv. Funct. Mater. 2019, 1807332 1807332  (6 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
to the flexible nature and large size of proteins. The template
removal process may also be insufficient, directly influences
the efficiency of protein capture, thus resulting in a much
higher limit of quantification[21] than the current work.
The synthesis of molecular imprints in nanoparticle format
is among the most efficient methods for proteins in recent
years.[7,8] Nevertheless, the sensitivity and affinity of pro-
tein binders, which are developed using the computationally
selected stable epitopes in the current study, are higher than
those obtained with the previously reported nanoparticle-based
MIPs.[7,8] The use of computationally selected epitopes can be
combined with these techniques to improve the outcomes of
nanoMIP applications. Our approach is a new step to provide
a true rational design for MIPs, which can improve the current
status of clinical research as well as commercial applications of
MIPs by overcoming existing major problems.
We investigate the selectivity of molecular imprints by com-
paratively studying the NSE protein binding on nonimprinted
polymer. Due to the absence of specific cavities on the NIP, the
interaction of NSE with the nonspecific polymer produces sig-
nificantly lower signals (Figure S12, Supporting Information).
Thus, an imprinting factor (IF) of 4.6 is found by taking the
signal ratios between the target and nonimprinted polymer sur-
faces, indicating a good selectivity of the molecular imprints
toward the target biomarker.
As the next step, the recognition of Ep5 and NSE protein was
investigated using the Cys-Ep5 imprints. Binding of Ep5 to the
imprinted cavities led to gradual decrease on the current signal
in a concentration range of 2–64 × 10−6  m. Despite being least
stabile epitope, Ep5 could still be determined by its specific
cavities. However, capturing efficiency of Ep5-based imprints
is found to be much less than those of Ep-1 based imprints
(Figure  7A). The signal intensities of NSE protein binding to
the Cys-Ep5 MIP surfaces are substantially low when compared
to the Cys-Ep1 MIPs (Figure 7B). Obviously, the interaction
of NSE with the MIP surface via its less stable epitope Ep5 is
weaker than the binding via the more stable Ep1.
A) 50
Cys-Ep5 molecular imprints
Cys-Ep1 molecular imprints
40
Relative signal (%)
30
20
10
0 0
2 4 8 16
Peptide concentration (µM)
Figure 7. A) Binding of Ep1 and Ep5 on Cys-Ep1 (red columns) and Cys-Ep5 (black columns) molecular imprints, respectively. B) NSE protein
recognition using Cys-Ep1 (red columns) and Cys-Ep5 (black columns) molecular receptors. Black columns include standard deviations of three sets
of experiments. Red columns demonstrate six sets of experiments with standard errors.
Adv. Funct. Mater. 2019, 1807332
www.afm-journal.de
2.5. Cross-Reactivity Studies
Specificity of biorecognition elements is critically important
in disease detection. Hence, comprehensive cross-reactivity
studies are performed. The specific (Ep1) and nonspecific
(Ep4, Ep5) peptides are prepared at a fixed concentration of
32  × 10−6  m and then incubated with the Cys-Ep1 MIP elec-
trodes for 30 min. This moderate concentration is selected
based on the calibration curve of Ep1. The capturing of
bovine serum albumin (BSA) (control protein) and NSE
(target protein) by Cys-Ep1 molecular receptors is measured
at a fix concentration of 60 ng mL−1, chosen based on the cali-
bration curve of NSE protein. Furthermore, the mixture of all
analytes at same concentration (60 ng mL−1) is also measured
to identify the degree of signal decrease due to interfering
compounds in the sample solution. The results are expressed
as relative signal suppression data, where the binding of Ep1,
Ep4, and Ep5 to the Cys-Ep1 MIPs leads to ≈28.6%, ≈5.5%,
and ≈2.5% relative signal intensities, respectively (Figure 8A).
Despite Ep4 and Ep5 not having sequence similarity to the
Ep1, the cross-reaction of Ep4 is clearly higher than that of
Ep5. This may be linked to the unstable nature of Ep5. The
control protein, BSA shows ≈7.9% relative signal intensity
upon binding on the MIP electrode (Figure 8B). The binding
of NSE protein leads to ≈63% relative signal intensity,
whereas the mixture of all analytes (both target and control
molecules) at the same concentration decreases the signal
intensity (45%) due to the presence of interfering compounds
in the solution. However, it is worth noting that the selected
concentration in this study (60 ng mL−1) is quite high and
the obtained signal intensity is still significantly higher
than those of control molecules alone (Figure 8). Despite
the control peptides are also surface exposed NSE epitopes
with similar amino acid lengths (12 mer), their nonspecific
binding on the MIP surfaces is at low levels, indicating the
epitope specificity of the synthesized MIP via noncovalent
interactions (Figure 8).
B)
80
Cys-Ep5 molecular imprints
70 Cys-Ep1 molecular imprints
60
Relative signal (%)
50
40
30
20
10
32 64 1 5 10 20 50 100
Protein concentration (ng mL )
1807332  (7 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
A) 40
35
Relative signal suppression (%)
30
25
20
15
10
5 10
0 0
Ep4 Ep5 Ep1
Figure 8.  A) Cross-reactivity studies on Cys-Ep1 MIP electrodes using the target (Ep1) and nonspecific peptide molecules (Ep4 and Ep5) at a fix
concentration of 32 × 10−6 m. B) Binding of BSA (first column), mixture of all analytes (Ep4, Ep5, BSA, and NSE; second column), and NSE (third
column) at a fix concentration of 60 ng mL−1. The mixtures are prepared at equal concentrations (n = 4).
3. Conclusion
The conformation of biological molecules in aqueous environ-
ments plays an important role for their molecular interactions
and recognition. This behavior strongly effects the successful
implementation of biosensing platforms for the detection of
target molecules as possible conformational changes lead to
decreased sensing signals. To date, protein binders have been
developed using linear peptides with an unknown structure in
epitope imprinting process.[4,8,18] Despite successful outcomes
obtained to some extent, most of these works lack of providing
either high affinity, selectivity, or sensitivity. Herein, we aim to
investigate the role of highly stable epitope residues (namely
α-helices) extracted from proteins. We therefore predict the
conformational stability of six surface epitopes of NSE by MD
simulations. According to the results, we select a stable α-helical
(Ep1) and a less stable disordered peptide (Ep5) and use their
cysteine modified versions (Cys-Ep1 and Cys-Ep5) as tem-
plates for molecular receptor synthesis. The recognition of NSE
derived epitopes (Ep1 and Ep5) as wells as whole protein bio-
marker (NSE) is investigated in wide concentration ranges. The
detection of Ep1 and NSE can be achieved with high sensitivity
(LODpeptide = 2 × 10−6  m; LODprotein = 0.5 ng mL−1) and affinity
(Kd peptide = 12.2 × 10−6  m; Kd protein = 5.3 × 10−11  m). In contrast
to this finding, the recognition capability of Cys-Ep5 molecular
imprints for Ep5 and NSE is found to be poor, reflecting the
unstable structure of the template molecules. The cross-reac-
tivity studies confirm the high target specificity of Cys-Ep1
imprints toward Ep1 and NSE protein, which can be recognized
even in a complex medium including nonspecific molecules
(Ep4, Ep5, and BSA protein) at a high concentration. It is worth
noting that from the simulations, we cannot conclude that one
stable epitope will be better than another one in experiment.
But the simulations suggest that Ep1 (or another stable epitope)
will be a better candidate to synthesize MIPs than very unstable
epitopes, such as Ep5. Therefore, we select extreme cases to
use them as a proof-of-principle. The additional tests with Ep4
provide similar outcomes with Ep1 (see Figure S13, Supporting
Adv. Funct. Mater. 2019, 1807332 1807332  (8 of 11)
www.afm-journal.de
B) 70
60
Relative signal suppression (%)
50
40
30
20
BSA Mix NSE
Information), but this does not mean that all stable epitopes
would give similar results because the ideal conditions in the
MD simulation are not reachable in experiments.
Our fast and rational selection can be used for establishing
epitope libraries for protein molecules by eliminating unsuit-
able epitope candidates and ranking the best candidates based
on their stability analysis obtained from MD simulations. In
long term, such epitope libraries can significantly reduce the
application time and research cost. It is worth noting that some
proteins are rich in loop and β-sheet formations, but do not
possess α-helix. Our modelling studies can still be used for
such protein targets and the influence of epitope size on molec-
ular imprinting should also be researched. The novel integrated
approach has shown a good potential to contribute to some
limitations of medical diagnostic field. Research disciplines,
such as biomedicine, cell signalling, proteomics, diagnostic,
and biocatalysis, requiring recognition receptors can apply this
technique for designing stable and efficient receptors.
4. Experimental Section
Materials, Chemicals, and Reagents: Cys-Ep1 and Cys-Ep5 were used for
comparative molecular imprinting studies. NSE derived synthetic peptide
Ep1 (LKAVDHINST, 10 mer) and neuron-specific enolase from human
brain (≥95%, sodium dodecyl sulfate–polyacrylamide gel electrophoresis
(SDS-PAGE), Sigma Aldrich) were used for target recognition studies. Ep4
(AMRLGAEVYHTL, 12 mer) was synthesized for use in cross-reactivity
studies as another stable NSE epitope based on computational outcomes.
Ep5 (KSPTDPSRTITG, 12 mer) was synthesized for comparative peptide
binding experiments as the least stable epitope and it was also used for
cross-reactivity studies. Scopoletin (≥99%, C10H8O4, Mw: 192.17 g mol−1,
Sigma-Aldrich) was used as functional monomer to manufacture the
molecular imprints and nonimprinted polymers. Potassium chloride
(≥99.9%, KCl, Mw: 74.55 g mol−1, VWR Chemicals) and potassium
hexaferricyanide (≥99%, K3[Fe(CN)6], Mw: 329.26 g mol−1, Carl Roth)
were used to prepare the redox marker solution for electrochemical
measurements using CV and square wave voltammetry (SWV) techniques.
QCM gold chips (LOT Quantum Design GmbH, Darmstadt, Germany)
were used as working electrodes in electrochemical sensor for epitope
© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
imprinting. QCM chips (LOT Quantum Design GmbH, Darmstadt,
Germany) were used for AFM analyses. QCM chips were cleaned using
the mixture of hydrogen peroxide (35%, H2O2, Mw: 34.01 g mol−1, Carl
Roth), ammonia (25%, NH3, Mw: 17.03 g mol−1, Carl Roth), and double
distilled water. The phosphate buffer was prepared using monosodium
phosphate monohydrate (≥99%, NaH2PO4⋅H2O, Mw: 137.99 g mol−1,
Merck) and disodium phosphate dihydrate (≥99%, Na2HPO4, H2O,
Mw: 177.99 g mol−1, Merck). Double-distilled ultrapure water (Millipore,
Germany) was used for analyses.
Molecular Dynamic Simulations: The initial coordinates of all epitopes
were extracted from the crystallographic structure of NSE (pdb code: 1TE6,
chain A).[25] The peptides were described with the CHARMM 36 force
field[26] and protonated according to pH 7. The protonation configurations
of histidine residues were set by visual inspection. The epitopes were then
solvated in rectangular boxes containing ≈6000 TIP3P water molecules.[27] In
order to mimic the experimental buffer conditions and to neutralize the net
charge of the models, ≈40 × 10−3 m Na+Cl− were randomly added replacing
water molecules. The subsequent atomic MD simulations were performed
with the CUDA version of NAMD 2.9.[28] First, the energy of the models was
minimized. Second, they were heated to 300 K and thermally equilibrated.
To avoid abrupt structural changes in these preparing steps, position
restraints on all heavy atoms of the epitopes were stepwise released. Finally,
300 ns long productions runs were carried out under periodic boundary
conditions in an NPT (constant number of particles N, pressure P, and
temperature T) ensemble at 300 K and 1 atm. This was accomplished with
the Langevin piston algorithm[29] implemented in NAMD. A cut-off of 12 Å
was applied for van der Waals and short-ranged electrostatics. Long-range
electrostatics were treated with the Particle Mesh Ewald summation.[30] A
time step of 2 fs was used which was enabled by constraining all bonds
containing hydrogen atoms with the SHAKE algorithm.[31] The analysis was
done with VMD.[32] The RMSD of the peptide backbones and the secondary
structure evolution using STRIDE[33] were determined.
Peptide Synthesis: Synthesis protocol for automated solid-phase
peptide synthesis: Epitopes were synthesized according to a previously
reported recipe.[34] Automated solid-phase peptide synthesis was
performed in 50 µmol scale. Loading: To a 20 mL syringe reactor with frit
and cap were added 2 g of 2-chloro-tritylchloride resin (1.4 mmol g−1)
and 14 mL dry dichloromethane (DCM). The resin was preswollen
for 10 min and the solvent was removed by evaporation in vacuum.
A mixture of the amino acid (8.40 mmol, 3 eq.) and 5 equivalents of
N,N-diisopropylamine (DIPEA) dissolved in 10 mL dry DCM was added
to the resin. The syringe was agitated for 30 min at room temperature.
The solution was removed and the resin was washed (2 × 10 mL N,N-
dimethylformamide (DMF), 2 × 10 mL DCM). Capping of nonreacted
functional groups of the resin was performed with DCM, methanol,
and DIPEA 80:15:5 (2 × 15 mL, 10 min). After washing (5 × 10 mL
DMF), Fmoc-removal was achieved with DMF/piperidine (4:1, 10 mL,
1 × 2 min, 1 × 20 min). After final washing (2 × 10 mL DMF, 1 × 10 mL
methanol, 3 × 10 mL DCM), the resin was dried in vacuo.
Coupling of Fmoc/tBu-Protected Amino Acids: To 200 mg of the resin
(≈0.5 mmol g−1), a 0.25 m solution of the amino acid in DMF (2.5 eq.
relative to resin loading) was added. After addition of a 0.5 m solution
of DIPEA in DMF (2.5 eq.) and a 0.25 m solution of O-(benzotriazol-
1-yl)-N,N,N′,N′-tetramethyluronium tetrafluoroborate (TBTU) in DMF
(2.5 eq.), the reaction solution was mixed for 15 min. A second coupling
was performed for 15 min. For couplings subsequent to the 5th amino
acid, double couplings with 30 min coupling time were performed. For
couplings subsequent to the 10th amino acid, a third coupling with
45 min was performed. After each coupling cycle capping with 0.5 m
acetic anhydride in DMF (2 × 2.5mL, 10 min) was performed. Finally, the
resin was washed with DMF (6 × 2.5 mL).
Fmoc Removal: DMF/piperidine (4:1, 2.5 mL) was added to the resin
and mixed for 2.5 min. The procedure was repeated 4 times. The resin
was washed with DMF (6 × 2.5 mL). After the final coupling cycle, the
resin was washed with DCM (3 × 2 mL).
Global Deprotection: The resin was transferred to a 5 mL syringe
with frit and cap. After addition of the cleavage cocktail (trifluoroacetic
acid (TFA), H2O, triethylsilane (TES), 3,6-dioxa-1,8-octane-dithiole,
Adv. Funct. Mater. 2019, 1807332 1807332  (9 of 11)
www.afm-journal.de
92.5:2.5:2.5:2.5), the syringe was shaken for 3 h. The peptide was
precipitated in ice cold diethyl ether and centrifuged. The supernatant
was removed and the precipitate was washed with diethyl ether twice.
The peptide was resolved in MeCN/H2O (1:4) and lyophilized. Crude,
lyophilized peptide variants were dissolved in H2O/MeCN (1:1; v:v)
and were purified by C18 reversed-phase high performance liquid
chromotography (HPLC) (1260 Infinity C18, 212 × 250 mm, particle
size 10 µm, Agilent Technologies, Waldbronn, Germany), if not specified
otherwise. Table S2 (Supporting Information) shows the gradient system
that was used (flow-rate of 20 mL min−1; UV-detection at 210 nm) for the
different peptide variants. Elution started with a linear gradient from X%
to Y% buffer B for 15 min, followed by holding 100% buffer B for another
5 min (Buffer A, 0.1% TFA in H2O; buffer B, 0.1% TFA in methanol).
Corresponding retention times and m/z data for each peptide were
determined. Collected fractions containing the products were pooled
and lyophilized to afford the epitope variants as fluffy solids.
Electrochemical Measurements: A three-electrode system in an
electrochemical cell (2 mL) was used for measurements. QCM gold
chips were used as working electrodes, whereas a platinum wire and
a Ag/AgCl system were used as counter and reference electrodes,
respectively. CV and SWV measurements were performed using a
Reference 600 potentiostat (Gamry Instruments, Warminster, PA, USA),
in the presence of a redox marker (K3[Fe(CN)6] containing KCl. CV
measurements were applied at potential range of −0.2–0.8 V and scan
rate of 0.05 V s−1. The applied potentials for SWV measurements were
from −0.3 to 0.8 V at a frequency of 10 Hz and amplitude of 0.05 V.
Cleaning of QCM Electrodes: The QCM chips were thoroughly cleaned
prior to electrochemical measurements. For this, a mixture of NH3,
H2O2, and H2O (v:v:v: 1:1:5) was used. The gold chip was inserted into
a holder and then boiled at 90 °C for 8 min. After boiling, the solution
was cooled by distilled water and further rinsed several times. The gold
electrode was then dried under gentle flow of N2 gas.
Synthesis of Epitope Imprints: For the MIP preparation, the Cys-Ep1 or Cys-
Ep5 (50 × 10−6  m) in tris(2-carboxyethyl)phosphine (TCEP) buffer (pH 7.3)
was first immobilized on the QCM gold electrode based on self-assembled
monolayer formation for 3 h. Adsorption was facilitated by the thiol side
chain of cysteine which has high affinity toward gold, in the epitope
sequence. A 0.5 × 10−3 m scopoletin prepared in 0.1 m NaCl was deposited
onto the QCM gold surface by applying a multistep amperometry technique
for electropolymerization. The electrodeposition was executed by 30 cycles
starting with 0 V for 5 s and followed by 0.9 V for 1 s. The polymerization
mixture was then removed from the electrochemical cell and rinsed with
distilled water several times. A thin film was formed with these parameters.
The subsequent template removal was enabled by applying anodic
desorption with the multiple pulse amperometry technique at 0.9 V for
30 s. The MIP films were thoroughly washed with distilled water to remove
any remaining epitope templates from the surface. The nonimprinted
polymer was synthesized using the same procedure in the absence of the
template and the selectivity of the cavities was investigated in comparison
with the MIPs. The IF was calculated based on these comparative studies
by taking the signal ratio between the MIPs and NIPs.[35] The MIP cavities
were used for the recognition and detection of the NSE derived peptides
(Ep1 and Ep5) and the NSE biomarker.
AFM Characterization: AFM was used to characterize the polymer films
on the QCM electrode surfaces. Every step of the imprinting process
(bare gold surface, MIP synthesis, and template removal) was followed by
AFM (NanoWizard II, JPK Instruments AG., Germany). The cross-section
analysis, the RMS roughness, and surface topology values of each surface
were evaluated. The measurements were performed in dry state using
intermittent contact mode. Commercially available AFM probe (TAP300
AL-G) from Budget Sensors (Innovative Solutions Bulgaria Ltd., Bulgaria)
was used on cantilevers with a resonance frequency in the range of
300 ± 100 kHz. The scanning line rate was 0.2–0.5 Hz. The samples were
measured in different scanning areas (10, 3, and 1 µm2) to investigate
the surfaces in detail. Phase image analysis was also performed in
various scanning areas (10, 3, and 1 µm2) to visualize different chemical
components on the template removed MIPs in comparison with
nonimprinted polymer. The AFM measurements of the polymer films
© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
were carried out at room temperature and under dry conditions. The
images were processed by using JPKSPM Data Processing software.
Target Capturing: The imprints were initially used for recognition and
detection of the NSE derived synthetic peptide (Ep1). The samples were
prepared in phosphate buffer (pH 7.3) and the detection of the peptide was
investigated in a concentration range of 2–500 × 10−6  m using the Cys-Ep1
molecular imprints. Each sample was injected into the electrochemical cell
and incubated for 30 min prior to electrochemical measurements by SWV
using the redox marker solution (5 × 10−3  m K3[Fe(CN)6]/0.1 m potassium
chloride). On the other hand, the detection of NSE was investigated in
a concentration range of 0.5–100 ng mL−1. The binding of the protein
biomarker to the cavities was measured using the SWV technique. The MIP
sensor was washed with double distilled water prior to each electrochemical
measurement and square wave voltammograms recorded in K3[Fe(CN)6]
(5 × 10−3 m)/potassium chloride (0.1 m). The affinities between the imprints
and the target molecules were analyzed by determining the dissociation
constant Kd via the Scatchard plot. Accordingly, raw SWV data were
converted into signal suppression data as percentage. This standardization
of values made them comparable between the individual experiment series.
The data obtained in terms of signal suppression indicated the relative
signal of the sensor for each sample measurement. A plot of relative signal
of redox marker (%) versus relative signal of redox marker (%) divided by
epitope or protein concentration gave a straight line where the slope (m)
was used to calculate Kd using Equation (1)[35]
1
M=−
Kd  (1)
Cross-Reactivity Studies: The specificity and selectivity of the
synthesized MIPs were tested by investigating the cross-reaction of
the MIPs with nonspecific peptides (Ep4 and Ep5) and proteins (BSA).
The target molecules (Ep1 and NSE biomarker) were also tested as
part of this experiment under the same conditions. The mixture of all
nonspecific analytes (Ep4, Ep5, BSA) and NSE protein were also tested
to determine the interfering effect. All the samples were prepared
in phosphate buffer and incubated for 30 min on the Cys-Ep1 MIP
electrodes. Each MIP film was prepared independently for testing each
sample. The relative signal suppression was calculated for the samples
and the results were demonstrated in comparison.
Supporting Information
Supporting Information is available from the Wiley Online Library or
from the author.
Acknowledgements
Z.A. acknowledges financial support from the European Commission
(10041380), the Marie Curie Actions and the Technical University of
Berlin (60320037-12).
Conflict of Interest
The authors declare no conflict of interest.
Keywords
artificial protein binders, cancer markers, computationally simulated
epitopes, molecular imprinting, protein recognition
Received: October 17, 2018
Revised: December 31, 2018
Published online:
Adv. Funct. Mater. 2019, 1807332 1807332  (10 of 11)
www.afm-journal.de
[1] M. J. Taussig, O. Stoevesandt, C. A. K. Borrebaeck, A. R. Bradbury,
D. Cahill, C. Cambillau, A. de Daruvar, S. Dubel, J. Eichler, R. Frank,
T. J. Gibson, D. Gloriam, L. Gold, F. W. Herberg, H. Hermjakob,
J. D. Hoheisel, T. O. Joos, O. Kallioniemi, M. Koegll, Z. Konthur,
B. Korn, E. Kremmer, S. Krobitsch, U. Landegren, S. van der Maarel,
J. McCafferty, S. Muyldermans, P. A. Nygren, S. Palcy, A. Pluckthun,
B. Polic, M. Przybylski, P. Saviranta, A. Sawyer, D. J. Sherman,
A. Skerra, M. Templin, M. Ueffing, M. Uhlen, Nat. Methods 2007, 4,
13.
[2] A. Nematollahzadeh, W. Sun, C. S. A. Aureliano, D. Lutkemeyer,
J. Stute, M. J. Abdekhodaie, A. Shojaei, B. Sellergren, Angew. Chem.,
Int. Ed. 2011, 50, 495.
[3] A. Nematollahzadeh, P. Lindemann, W. Sun, J. Stute,
D. Luetkemeyer, B. Sellergren, J. Chromatogr. A 2014, 1345,
154.
[4] H. Nishino, C. S. Huang, K. J. Shea, Angew. Chem., Int. Ed. 2006,
45, 2392.
[5] G. Vlatakis, L. I. Andersson, R. Muller, K. Mosbach, Nature 1993,
361, 645.
[6] K. Mosbach, O. Ramstrom, Bio/Technology 1996, 14, 163.
[7] F. Canfarotta, A. Waters, R. Sadler, P. McGill, A. Guerreiro,
D. Papkovsky, K. Haupt, S. Piletsky, Nano Res. 2016, 9, 3463.
[8] A. Cecchini, V. Raffa, F. Canfarotta, G. Signore, S. Piletsky,
M. P. MacDonald, A. Cuschieri, Nano Lett. 2017, 17, 2307.
[9] B. Gruber, S. Balk, S. Stadlbauer, B. Konig, Angew. Chem., Int. Ed.
2012, 51, 10060.
[10] F. Canfarotta, S. A. Piletsky, Adv. Healthcare Mater. 2014, 3, 160.
[11] Z. Altintas, J. Pocock, K.-A. Thompson, I. E. Tothill, Biosens.
Bioelectron. 2015, 74, 996.
[12] Z. Altintas, M. Gittens, A. Guerreiro, K.-A. Thompson, J. Walker,
S. Piletsky, I. E. Tothill, Anal. Chem. 2015, 87, 6801.
[13] D. Dechtrirat, N. Gajovic-Eichelmann, F. F. Bier, F. W. Scheller, Adv.
Funct. Mater. 2014, 24, 2233.
[14] Z. Altintas, I. Chianella, G. Da Ponte, S. Paulussen, S. Gaeta,
I. E. Tothill, Chem. Eng. J. 2016, 300, 358.
[15] A. Cutivet, C. Schembri, J. Kovensky, K. Haupt, J. Am. Chem. Soc.
2009, 131, 14699.
[16] X. B. Hu, G. T. Li, J. Huang, D. Zhang, Y. Qiu, Adv. Mater. 2007, 19,
4327.
[17] D. Dechtrirat, K. J. Jetzschmann, W. F. M. Stocklein,
F. W. Scheller, N. Gajovic-Eichelmann, Adv. Funct. Mater. 2012, 22,
5231.
[18] M. Bosserdt, J. Erdossy, G. Lautner, J. Witt, K. Kohler,
N. Gajovic-Eichelmann, A. Yarman, G. Wittstock, F. W. Scheller,
R. E. Gyurcsanyi, Biosens. Bioelectron. 2015, 73, 123.
[19] H. P. Erickson, Biol. Proced. Online 2009, 11, 32.
[20] Z. Ye, X. Zhao, J. Microsc. (Oxford, U. K.) 2010, 238, 27.
[21] X. R. Zhang, A. Yarman, J. Erdossy, S. Katz, I. Zebger,
K. J. Jetzschmann, Z. Altintas, U. Wollenberger, R. E. Gyurcsanyi,
F. W. Scheller, Biosens. Bioelectron. 2018, 105, 29.
[22] S. Viswanathan, C. Rani, S. Ribeiro, C. Delerue-Matos, Biosens.
Bioelectron. 2012, 33, 179.
[23] J. A. Ribeiro, C. M. Pereira, A. F. Silva, M. G. F. Sales, Biosens.
Bioelectron. 2018, 109, 246.
[24] C. H. Lu, Y. Zhang, S. F. Tang, Z. B. Fang, H. H. Yang, X. Chen,
G. N. Chen, Biosens. Bioelectron. 2012, 31, 439.
[25] G. Q. Chai, J. M. Brewer, L. L. Lovelace, T. Aoki, W. Minor,
L. Lebioda, J. Mol. Biol. 2004, 341, 1015.
[26] R. B. Best, X. Zhu, J. Shim, P. E. M. Lopes, J. Mittal, M. Feig,
A. D. MacKerell, J. Chem. Theory Comput. 2012, 8, 3257.
[27] W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey,
M. L. Klein, J. Chem. Phys. 1983, 79, 926.
[28] J. C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid,
E. Villa, C. Chipot, R. D. Skeel, L. Kale, K. Schulten, J. Comput.
Chem. 2005, 26, 1781.
© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.afm-journal.de

[29] S. E. Feller, Y. H. Zhang, R. W. Pastor, B. R. Brooks, J. Chem. Phys.
1995, 103, 4613.
[30] T. Darden, D. York, L. Pedersen, J. Chem. Phys. 1993, 98, 10089.
[31] W. F. Vangunsteren, H. J. C. Berendsen, Mol. Phys. 1977, 34, 1311.
[32] W. Humphrey, A. Dalke, K. Schulten, J. Mol. Graphics Modell. 1996,
14, 33.
[33] D. Frishman, P. Argos, Proteins: Struct., Funct., Genet. 1995, 23, 566.
[34] E. Reyna-Gonzalez, B. Schmid, D. Petras, R. D. Sussmuth,
E. Dittmann, Angew. Chem., Int. Ed. 2016, 55, 9398.
[35] A. F. T. Waffo, C. Yesildag, G. Caserta, S. Katz, I. Zebger,
M. C. Lensen, U. Wollenberger, F. W. Scheller, Z. Altintas, Sens.
Actuators, B 2018, 275, 163.

Adv. Funct. Mater. 2019, 1807332 1807332  (11 of 11) © 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim