QM/MM Study of the Absorption Spectra of

DsRed.M1 Chromophores
ELSA SANCHEZ-GARCIA, MARKUS DOERR, WALTER THIEL

Max-Planck-Institut fur Kohlenforschung, Kaiser-Wilhelm-Platz 1, 45470 Mulheim an der Ruhr, Germany

Received 7 August 2009; Revised 16 September 2009; Accepted 17 September 2009
DOI 10.1002/jcc.21443
Published online 14 December 2009 in Wiley InterScience (www.interscience.wiley.com).

Abstract: We report geometries and vertical excitation energies for the red and green chromophores of the
DsRed.M1 protein in the gas phase and in the solvated protein environment. Geometries are optimized using density
functional theory (DFT, B3LYP functional) for the isolated chromophores and combined quantum mechanical/molecular mechanical (QM/MM) methods for the protein (B3LYP/MM). Vertical excitation energies are computed
using DFT/MRCI, OM2/MRCI, and TDDFT as QM methods. In the case of the red chromophore, there is a general
blue shift in the excitation energies when going from the isolated chromophore to the protein, which is caused both
by structural changes and by electrostatic interactions with the environment. For the lowest pp* transition, these two
factors contribute to a similar extent to the overall DFT/MRCI shift of 0.4 eV. An enlargement of the QM region to
include active-site residues does not change the DFT/MRCI excitation energies much. The DFT/MRCI results are
closest to experiment for both chromophores. OM2/MRCI and TDDFT overestimate the rst vertical excitation
energy by 0.30.5 and 0.20.4 eV, respectively, relative to the experimental or DFT/MRCI values. The experimental
gap of 0.35 eV between the lowest pp* excitation energies of the red (cis-acylimine) and green (trans-peptide)
forms is well reproduced by DFT/MRCI and TDDFT (0.32 and 0.37 eV, respectively). A histogram spectrum for an
equal mixture of the two forms, generated by OM2/MRCI calculations on 450 snapshots along molecular dynamics
trajectories, matches the experimental spectrum quite well, with a gap of 0.23 eV and an overall blue shift of about
0.3 eV. DFT/MRCI appears as an attractive choice for calculating excitation energies in uorescent proteins, without
the shortcomings of TDDFT and computationally more affordable than CASSCF-based approaches.
q 2009 Wiley Periodicals, Inc.

J Comput Chem 31: 16031612, 2010

Key words: red uorescent protein; QM/MM; absorption spectra; excited states; DFT/MRCI; OM2; TDDFT

Introduction
There has been growing interest in the study and development
of red uorescent proteins (RFPs) in recent years.120 Many RFP
variants have been discovered or engineered such that the available uorescent proteins now cover most of the visible spectrum.
RFPs have found widespread applications in biochemistry and
biomedicine, for example, as marker for gene expression and for
tagging proteins within living cells.2126
The rst identied RFP is the DsRed protein from the Discosoma coral1,3,4 where the presence of a cis-acylimine group
between Phe65 and Gln66 extends the conjugation of the p system and thereby causes the observed red shift when compared
with the green uorescent proteins (GFPs).2,27 Wild-type DsRed
has shortcomings such as slow chromophore maturation, oligomerization into a stable tetramer, and association to even higher
aggregates,4 and therefore a number of other variants have been
engineered.5,6 The DsRed.M1 protein16 is a stable fast maturing
monomer, but with relatively low uorescence intensity. In
DsRed.M1, the chromophore exists in two forms: as an acylimine

with sp2 hybridization of the Ca of Gln66 (red form, responsible

for the red uorescence) and excitation maximum at 557 nm
(2.23 eV), and as a peptide form with sp3 hybridization of the Ca
of Gln66 (green form, not uorescent) and excitation maximum
at 480 nm (2.58 eV). According to the peptide nomenclature, the
acylimine form has a cis-conformation (syn-position of the C
O
bond with respect to the lone pair at the N atom), while the peptide chromophore adopts a trans-conformation (with the C
O
bond in anti-position to the N


H peptide bond).16,28
There have been several theoretical studies of the absorption
spectra of the isolated chromophores of uorescent proteins,27,2931
but the necessity of carrying out calculations of the chromophore

Additional Supporting Information may be found in the online version of

this article.
Correspondence to: W. Thiel; e-mail: thiel@mpi-muelheim.mpg.de
Contract/grant sponsor: The Deutsche Forschungsgemeinschaft; contract/
grant number: SFB 663, project C4

q 2009 Wiley Periodicals, Inc.

Sanchez-Garcia, Doerr, and Thiel Vol. 31, No. 8 Journal of Computational Chemistry

1604

within the protein environment is frequently emphasized.28,29,3134

The effect of the protein matrix varies from one uorescent protein
to the other. For example, in the E2GFP mutant, the optical properties are found to be more sensitive to the protein environment than
in related GFPs.33 The majority of the theoretical studies have
addressed the chromophores of green uorescent proteins including
mutants like asFP595. In most cases, time-dependent density functional theory (TDDFT) has been applied,29,30,3238 and to a lesser
extent also complete-active-space self-consistent-eld (CASSCF)
procedures.35,39 Recent CASSCF-based studies have cautioned
against using TDDFT for GFP-like chromophores, in spite of its
good performance for many organic molecules.40,41 One critical
case is the asFP595 anionic chromophore, which absorbs in the
long wavelength region of the visible spectra, where the TDDFT
errors tend to be higher.40,42 The known TDDFT shortcomings for
charge transfer states also need to be taken into account.43 In a
systematic computational study of a series of 10 chemically distinct GFP chromophore analogues, TDDFT (with several different
functionals) showed generally poor quantitative agreement with
experiment, particularly so in the case of the nonaromatic Y66S
and Y66L chromophores where it failed (probably due to the signicant contribution of double excitations to their excited state
wavefunctions).41 Another recent investigation44 reported comparisons of the computed rst vertical excitation energy for the anionic
and neutral forms of the GFP chromophore in the gas phase:
TDDFT, EOM-CC, CASPT2, and QMC calculations give quite
similar results for the anionic form, while TDDFT strongly underestimates the excitation energy for the neutral chromophore compared with the correlated ab initio methods.44 On the other hand,
in a quantum benchmark study of the electronic properties of the
GFP anionic chromophore in the gas phase, the lowest vertical excitation energy from TDDFT was signicantly higher than that
from correlated ab initio methods such as MRMP2.45
Combined quantum mechanical/molecular mechanical (QM/
MM) approaches46,47 have emerged as the method of choice for
capturing the effects of the protein environment on local electronic events such as chemical reactions or electronic excitations.
Some QM/MM work has been published for GFP and several of
its mutants.32,38,48 We have recently started QM/MM studies on
the structural and spectroscopic properties of red uorescent proteins.49,50 Here, we report on QM/MM results for the red and
green chromophores of DsRed.M1 within the solvated protein matrix. Using a DFT-based multireference conguration interaction
treatment (DFT/MRCI), the GUGA-CI method in combination
with the orthogonalization model 2 Hamiltonian (OM2/MRCI),
and time-dependent density functional theory (TDDFT) as QM
components, we investigate the effects of the environment and of
structural changes in the chromophore on the vertical excitation
energies and compare the merits of these methods.

Figure 1. cis-acylimine (left) and trans-peptide forms (right) of the

chromophore of DsRed.M1: models A, B, and C (see text).

basis set from the Turbomole basis set library.57 The geometries
of the isolated chromophores were optimized for two models
(see Fig. 1): Model A consists of the chromophore residue
CRQ66 (except atoms CG1, HG11, HG12, CD3, OE1, NE1,
HE11, and HE12) and atoms C, O, CA, and HA of residue
Phe65, while model B corresponds to the QM region QM1 (see
below) that includes residue CRQ66, atoms C, O, CA, and HA
of Phe65, and atoms N and HN of Ser69 (see Fig. 1 and Fig.
S1, Supporting Information).
A variant of model B is model C which contains the chromophore in the geometry adopted within the protein environment
(QM/MM optimized with QM region QM1, snapshot 1). This
geometry was taken as starting point for the optimizations of the
isolated chromophores (A and B). To evaluate the effects of geometry changes between the isolated chromophore and the chromophore embedded in the protein environment, the vertical excitation energies were calculated also for model C, without including the MM point charges.

QM/MM Structure Optimizations

Computational Details
Isolated Chromophores

The QM calculations on the isolated cis-acylimine and transpeptide chromophores were performed with Turbomole (version
5.9.1)51 using the B3LYP density functional5256 and the TZVP

For the QM/MM optimizations, we used randomly picked snapshots from QM/MM molecular dynamics (MD) simulations as
starting geometries. Because the details and results of the MD calculations have already been discussed in our previous work,49 we
only give a brief summary here. The initial set of coordinates for

Journal of Computational Chemistry

DOI 10.1002/jcc

QM/MM Study of the Absorption Spectra of DsRed.M1 Chromophores

the MD was taken from the X-ray structure available in the protein data bank,58 under the reference 2VAD.16 The protein was
described by the CHARMM22 force eld,59,60 the water molecules
were represented by the TIP3P model,61 and the QM region with
the chromophore was treated using the semi-empirical SCC-DFTB
method,62,63 as implemented in CHARMM33b1.6466 A 500 ps
constant-temperature production MD with a time step of 1 fs was
performed. The VMD program was used for visualizing the MD
trajectories and resulting geometries.67
QM/MM optimizations were performed with the program
package ChemShell.68 We used Turbomole51 for the QM part
and DL_POLY69 as driver of the CHARMM22 force eld for
the MM part. The QM part was treated with the B3LYP density
functional and the TZVP basis set from the Turbomole basis set
library. Geometry optimizations were performed using the
HDLC optimizer70 implemented in ChemShell. The optimizations were done in hybrid delocalized internal coordinates. The
from the center (atom
coordinates of all atoms beyond 15 A
active
CA2 of the chromophore) were xed. Within the 15 A
region, all atoms were allowed to move in each optimization
step. The optimization was nished when the maximum gradient
component was below 0.00135a.u. Two different QM regions
were employed: QM1 contains the same atoms as models B and
C (see above), and QM2 additionally includes the residues
His163, Arg95, Ser146, Glu215, W138, and W42.
Open valencies at the QM/MM border were saturated using
hydrogen link atoms. We applied an electrostatic embedding
scheme,71 that is, the xed MM charges were introduced into the
one-electron part of the QM Hamiltonian and the QM/MM electrostatic interactions were evaluated from the QM electrostatic
potential and the MM atomic charges. To avoid overpolarization
of the QM region at the boundary, the charge on the rst layer of
MM atoms (those connected to the QM region) was moved to the
second layer and a dipole (constructed from two point charges)
was placed on the recipient atom of this charge shift, so that the
charge and dipole of the MM system were conserved after the
shift.72,73 No electrostatic cutoffs were used.
Electronic Spectra

Calculations for the isolated chromophore were done at the QM

level with one of the QM methods specied below. The corresponding QM/MM computations for the protein include the effect
of the MM point charges on the vertical excitation energies.
TDDFT

Time-dependent density functional theory (TDDFT)7477 is

widely used for computing electronic spectra. One of the main
advantages of this method is its favorable computational cost,
which allows its application to rather large molecules. TDDFT
results are usually sensitive to the choice of the exchange-correlation functionals, in particular, when local or generalized-gradient functionals are compared with hybrid functionals. In this
study, we apply the B3LYP functional, which yields excitation
energies that are usually within 0.3 eV of high-level ab-initio
results.78 However, TDDFT has some well known limitations.
For example, it fails to correctly describe charge-transfer states

1605

and states with large double-excitation character.43 The TDDFT

calculations were performed with Turbomole (version 5.9.1),
using the TZVP basis from the Turbomole basis set library.
DFT/MRCI

We also calculated vertical excitation spectra with the combined

density functional multireference conguration interaction (DFT/
MRCI) method of Grimme and Waletzke,79 which is an alternative to TDDFT. Being more generally applicable than TDDFT,
it is still efcient enough for application to large systems. Taking experimental vertical excitation energies as the reference,
errors are usually below 0.2 eV.79 When compared with high
level ab-initio data, DFT/MRCI typically underestimates excitation energies by about 0.2 eV.78
The basic idea of the DFT/MRCI method is to take major
parts of the dynamic electron correlation into account by DFT,
while static correlation effects are included by a short CI expansion. The conguration state functions (CSFs) in the MRCI
expansion are built up from Kohn-Sham (KS) orbitals of a
closed-shell reference state. Diagonal elements of the effective
DFT/MRCI Hamiltonian are constructed from the corresponding
Hartree-Fock-based expression and a DFT-specic correction
term. The effective DFT/MRCI Hamiltonian contains ve empirical parameters, which depend only on the multiplicity of the
excited state, the number of open shells of a conguration, and
the density functional used. Currently, optimized parameter sets
are available for the DFT/MRCI Hamiltonian in combination
with the BH-LYP functional.5255,80 To speed up the calculations
(details see below), we did not use the original parameterization
from Grimme and Waletzke,79 but a modied set of parameters
also developed in the Grimme group81 (p1 5 0.629, p2 5
0.611, pJ 5 0.119, p[0] 5 8.000, a 5 0.503) which is currently
only available for singlet electronic states. The required DFT
wavefunctions were generated using Turbomole (version 5.71).51
As is common practice in the DFT/MRCI approach, the CI
space was kept moderate by only selecting congurations whose
estimated energy is beyond a threshold dEsel above the highest
reference space energy. Although the usual choice for this parameter is 1.0 Hartree, we used a value of 0.8 Hartree in our
calculations. This was possible due to the modied DFT/MRCI
parameter set applied, which improves convergence of the electronic energies with respect to dEsel. Using this combination of
dEsel and DFT/MRCI parameters, calculations are sped up by a
factor of about 20, while excitation energies are only slightly
changed (in our experience mostly higher by less than 0.1 eV).82
The MRCI reference space was determined iteratively by rst
performing a DFT/MRCI calculation with a conguration selection threshold dEsel 5 0.6 Hartree and a reference space that
included all congurations which can be generated by up to doubly exciting ten electrons within the ten frontier MOs. In the
next iteration, dEsel was increased to 0.8 Hartree and all congurations with a squared coefcient of at least 0.003 in the previous CI calculation were included in the reference space. Further
iterations were performed until the energies of all calculated
roots were converged to 1026 Hartree.
The same QM regions and boundary schemes were used as in
the preceding B3LYP/TZVP optimizations. Ten roots with singlet

Journal of Computational Chemistry

DOI 10.1002/jcc

Sanchez-Garcia, Doerr, and Thiel Vol. 31, No. 8 Journal of Computational Chemistry

1606

), Dihedral Angles (degree) and Vertical

Table 1. Selected Distances (A
Excitation Energies (eV) for the B3LYP/TZVP Optimized cis-acylimine
Chromophore Using Models A-C (see text).
cis-acylimine

Model A

Model B

Model C

OH


CZ
CG2


CB2
CB2


CA2
C2
O2
C1


CA1
CA1


N
N


C
C
O
CD2


CG2


CB2


CA2
N2


CA2


CB2


CG2
CD2


CG2


CA2


N2
CA1


N


C


O
DFT/MRCI excitation
energy
TDB3LYP excitation
energy

1.244
1.392
1.400
1.230
1.438
1.303
1.376
1.226
20.4
20.2
20.5
42.5
2.01
(1.191)
2.40
(0.935)

1.242
1.392
1.400
1.228
1.444
1.295
1.381
1.222
4.8
4.6
7.8
47.9
2.06
(1.084)
2.39
(0.808)

1.265
1.389
1.398
1.244
1.465
1.279
1.404
1.222
6.9
11.5
14.9
98.0
2.21
(1.205)
2.55
(0.953)

Table 1). However, in the acylimine moiety, the single bond

character of the C1


CA1 and N


C bonds and the double bond
character of the CA1


N bond are more accentuated with model
B than with model A. This tendency is even stronger with model
C, showing how the protein environment tends to favor the phenolate form of the chromophore, as discussed in our previous
work.49 The OH


CZ and C2
O2 bond lengths are larger in
model C than in models A and B since both the phenolate oxygen atom (OH) and the carbonyl oxygen of the imidazolinone
(O2) are involved in hydrogen bonds with other residues of the
protein environment (Fig. 2, Table 1).
In the trans-peptide form, the most notable change is the
elongation of the trans-peptide N


C bond in model C (Table
2), presumably due to steric strain within the protein environment (there are no obvious specic interactions). The OH


CZ
bond is again lengthened in model C because of hydrogen-bonding interactions with active-site residues. The O2 atom is
engaged in different hydrogen bonds in model B (with Arg95)
and in model C (with the terminal NH2 group, Fig. 1) so that
the C2
O2 bond is somewhat elongated in both cases (Table
2).

Oscillator strengths are given in parentheses.

multiplicity were calculated for each snapshot and the TZVP basis
set from the Turbomole basis set library was used.57 All valence
electrons (identied by their MO energies) were correlated.
OM2/MRCI

To obtain electronic spectra over the whole MD trajectory, the

transition energies to the lowest pp* state were computed by
OM2/MRCI using the GUGA-CI method in combination with
the semiempirical Hamiltonian from the orthogonalization model
2 (OM2).8385
The active space in these calculations consisted of 20 electrons in 18 p MOs for the cis-acylimine and 16 electrons in 14
p MOs for the trans-peptide, which has a smaller p system. The
SCF calculations were performed for the lowest singly excited
state in the restricted open-shell Hartree-Fock (ROHF) formalism. Three reference congurations were used in the MRCI procedure, corresponding to the closed shell ground state and to single and double HOMOLUMO excitations. With this setup, the
number of conguration state functions was 3745 for the peptide
and 11,133 for the acylimine.
The program package ChemShell68,86 was used for the OM2/
MRCI calculations of the electronic spectra of the MD snapshots. For the QM part and for the treatment of the isolated
chromophore, a development version of the MNDO program87
was used. The MM region was treated like in the other QM/MM
calculations. The OM2/MRCI calculations over the MD trajectory were performed for QM region QM1.

Results and Discussion

QM Calculations: Isolated Chromophore Geometries

For the cis-acylimine chromophore, the values of most relevant

bond distances are nearly the same for models A and B (Fig. 1,

Figure 2. Top: Overlay between the QM(B3LYP/TZVP)/MM optimized structures of the cis-acylimine form of DsRed.M1: snapshots
1 (green), 2 (blue), 3 (gray), and 4 (red), QM region QM2. Bottom:
Hydrogen bond interactions in snapshot 1 after optimization.

Journal of Computational Chemistry

DOI 10.1002/jcc

QM/MM Study of the Absorption Spectra of DsRed.M1 Chromophores

), Dihedral Angles (degree) and Vertical
Table 2. Selected Distances (A
Excitation Energies (eV) for the B3LYP/TZVP Optimized trans-peptide
Form Using Models AC (see text).
trans-peptide

Model A

Model B

Model C

OH


CZ
CG2


CB2
CB2


CA2
C2
O2
C1


CA1
CA1


N
N


C
C
O
N


H
CD2


CG2


CB2


CA2
N2


CA2


CB2


CG2
CD2


CG2


CA2


N2
CA1


N


C


O
DFT/MRCI excitation
energy
TDB3LYP excitation
energy

1.249
1.404
1.385
1.232
1.506
1.457
1.351
1.228
1.013
21.3
21.4
22.2
23.7
2.70
(1.190)
3.08
(0.929)

1.245
1.396
1.393
1.248
1.507
1.455
1.359
1.225
1.012
20.1
0.3
0.1
28.9
2.73
(1.250)
3.11
(1.047)a

1.266
1.400
1.386
1.243
1.511
1.456
1.383
1.229
1.013
5.6
5.9
9.4
2.7
2.66
(1.252)
2.99
(0.911)a

At the TDB3LYP level, the rst excited singlet state is of np* type (see
text): model B 3.05 (0.0000002), model C 2.97 (0.0084).
Oscillator strengths are given in parentheses.

QM Calculations: Vertical Excitation Energies

For the cis-acylimine form the inclusion of a larger number of

atoms in the chromophore model causes a slight deviation from
planarity (see the values of the dihedral angles between the phenolate and imidazolinone rings for models A and B in Table 1).
This slight out-of-plane distortion in model B can probably be
attributed to the repulsion between one of the hydrogen atoms
of the terminal NH2 group and the closest hydrogen atom in the
phenolate ring (Fig. 1). It does not affect the calculated vertical

1607

excitation energies much, which are quite similar for models A

and B both at the DFT/MRCI and the TDDFT level (Table 1).
In the protein (model C), the chromophore deviates somewhat
more from planarity, and the terminal C
O group adopts an
essentially perpendicular orientation, with the dihedral angle
CA1


N


C


O increasing from 488 in the gas phase to 988 in
the protein (Fig. 1, Table 1), such that it is less involved in the
conjugated p system. As a consequence, the computed vertical
excitation energies increase by 0.200.15 eV when compared
with models A and B.
For the trans-peptide form, the phenolate-imidazolinone part
of the chromophore is also slightly less planar in model C than
in models A and B, although the difference is less pronounced
here (Table 2). The calculated vertical excitation energies for
models A and B are again similar and only slightly larger than
those for model C. At the DFT/MRCI level, the variations are
very small (0.07 eV for B vs. C, 0.04 eV for A vs. C, and 0.03
eV for A vs. B) so that further analysis is not warranted.
QM Calculations: Electronic Structure

The electronic structure of the ve lowest excited singlet states

is characterized in Tables 3 and 4 on the basis of the DFT/
MRCI/TZVP wavefunctions for the cis-acylimine and trans-peptide forms of the DsRed.M1 chromophore (model A). The lowest transition in the cis-acylimine form is a high-intensity pp*
(HOMO?LUMO) transition, which dominates the absorption
spectrum. A low-intensity transition is found about 0.3 eV
higher, which mainly arises from the excitation of an electron
from the lone-pair MO at the phenolic oxygen (HOMO-2) to the
p* LUMO. Above these transitions, there are a number of relatively low-intensity transitions, which involve excitations within
p MOs and also from the lone-pair MO of the acyl-oxygen atom
to the p* LUMO. Although the S0?S1 transition is clearly
dominated by single excitations, the higher electronic transitions
have signicant contributions from double excitations (Table 3).

Table 3. DFT/MRCI Excitation Energies (eV), Oscillator Strengths (in parentheses), Dominant Congurations and Fractions of

Single- and Double-Excitation Contributions to the Lowest Excited States Corresponding to the cis-acylimine
Chromophore (model A).

Excitation energy
S1: 2.01 (1.191)
S2: 2.43 (0.000)
S3: 2.79 (0.073)

S4: 3.22 (0.003)

S5: 3.53 (0.342)

Contributions of single (S)

and double (D) excitations

Main contributionsa
(1)83.3%
(1)63.8%
(1)26.0%
(1)25.7%
(2)24.0%
(1)55.8%
(1)42.4%
(2)26.0%

H?L
H-2?L
H-1?L
H ?L11
H,H ?L, L
H-5?L
H?L11
H-1?L

S: 85.0%; D: 12.8%
S: 73.6%; D: 23.9%
S: 55.4%; D: 41.4%

S: 70.6%; D: 25.2%
S: 77.1%; D: 20.6%

H, HOMO; L, LUMO; signs of the contributions shown in parentheses.

Journal of Computational Chemistry

DOI 10.1002/jcc

Sanchez-Garcia, Doerr, and Thiel Vol. 31, No. 8 Journal of Computational Chemistry

1608

Table 4. DFT/MRCI Excitation Energies (eV), Oscillator Strengths (in Parentheses), Dominant Congurations and Fractions of

Single- and Double-Excitation Contributions to the Lowest Excited States Corresponding to the trans-peptide
Chromophore (model A).

Excitation energy
S1: 2.70 (1.190)
S2: 3.01 (0.000)
S3: 3.74 (0.005)
S4: 4.06 (0.071)
S5: 4.29 (0.051)

Contributions of single (S)

and double (D) excitations

Main contributionsa
(1)86.3 % H?L
(1)71.4 % H-1?L
(1)40.0 % H -2?L
(1)33.2 % H,H?L,L
(1)57.2 % H ?L14
(2)17.4 % H?L15
(1)68.7 % H-3?L

S: 88.4 %; D: 11.1 %
S: 80.5 %; D: 18.6 %
S: 52.4 %; D: 45.2 %
S: 87.1 %; D: 11.7 %
S: 76.0 %; D: 23.1 %

H, HOMO; L, LUMO; signs of the contributions shown in parentheses.

The lowest three excited states in the trans-peptide chromophore resemble those in the cis-acylimine chromophore. The rst
excited state corresponds to a bright pp* (HOMO?LUMO)
transition, followed by a dark np* and a relatively dark pp*
state. The p system is less extended than in the acylimine form
which generally results in higher excitation energies. The
absorption spectrum is again clearly dominated by the very
bright S0?S1 transition (Table 4).
The corresponding TDB3LYP/TZVP results for the rst ve
excited singlet states are given in the Supporting Information for
comparison (model A, Supporting Information Tables S1 and
S2). In agreement with the DFT/MRCI results, the lowest two
excited states are predicted to be of bright pp* and dark np*
type, but the energy gap between them is smaller at the
TDB3LYP level. For higher excitations, the computed states are
different for both methods; these states are not discussed further
because they are not in the focus of this work.
When models B and C are used for calculating the TDB3LYP
vertical excitation energies of the trans-peptide chromophore, the
pp* transition corresponds to the second excitation. In both cases,
the rst excitation involves a dark np* transition between HOMO1 and LUMO. The energy gap between the S1 and S2 states is
only 0.06 eV in model B and 0.02 eV in model C (Table 2). In
all cases, the TDDFT excitation energies are consistently 0.3 eV
higher than the DFT/MRCI values (Tables 1 and 2).

QM/MM Calculations: Chromophores Within the

Solvated Protein Matrix

The protein environment may change the character and the

sequence of the excited states of the chromophore. Therefore,
we rst briey summarize the QM/MM results for the ve lowest singlet excited states in the DsRed.M1 protein (snapshot 1)
using DFT/MRCI as QM component (see Tables S3 and S4 of
Supporting Information). As in the case of the isolated chromo-

phore (see Tables 3 and 4), the S1 state corresponds to the bright
pp* (HOMO?LUMO) transition, both in the cis-acylimine and
trans-peptide forms (QM region QM1). The S1 state remains
well separated from the S2 state, with computed S1-S2 gaps of
0.79 eV and 0.84 eV, respectively. These gaps are larger than in
the gas phase (0.42 eV and 0.31 eV, respectively), mainly
because of the signicant blue shift of the np* transition in the
protein (by about 0.8 eV) which even affects the order of the
excited states: In the gas phase, the S2 state is of np* type both
in the cis-acylimine and trans-peptide forms, whereas in the
DsRed.M1 protein, this remains true only for the trans-peptide.
By contrast, for the cis-acylimine form of DsRed.M1, the three
states above S1 are closely spaced and rather mixed: the S2 state
corresponds mostly to the dark pp* transition between HOMO
) and Dihedral Angles (degree) from
Table 5. Average Distances (in A
the QM/MM Optimizations of the cis-acylimine and trans-peptide
Snapshots and from the Crystal Structure.a
cis-acylimine
B3LYP/TZVP
R(O ser146 O crq)
R(O w42 O crq)
R(N his163 O crq)
R(N arg95 O crq)
R(O1 glu215 N2 crq)
R(O2 glu215 O w138)
R(O w138 N ser69)
CA1-N-C-O
CD2-CG2-CA2-N2

QM1
2.716
2.728
2.804
2.713
2.825
2.713
3.293b
99.6
14.1

QM2

trans-peptide

Crystal QM1

2.782
2.572
2.853
2.655
2.824
2.830
2.836
2.934
2.760
2.749
2.753
2.589
3.315b 2.793
98.4
171.2
17.2
1.8

2.740
2.702
2.787
2.718
2.836
2.735
2.887
1.1
10.8

QM2 Crystala
2.822
2.787
2.810
2.858
2.761
2.755
2.994
1.0
17.0

2.562
2.650
2.837
2.922
2.713
2.589
2.793
4.0
1.0

Reference 16.
Average values without including snapshot 4: 2.869 (QM1), 2.992
(QM2).
b

Journal of Computational Chemistry

DOI 10.1002/jcc

QM/MM Study of the Absorption Spectra of DsRed.M1 Chromophores

1609

Table 6. Computed DFT/MRCI (with and without Point Charges PC), OM2/MRCI, and TDDFT Vertical Excitation Energies (eV)

for the cis-acylimine and trans-peptide Forms of the DsRed.M1 Protein at Optimized QM(B3LYP/TZVP)/MM Structures Using
QM Region QM1: Results for Snapshots 1-4 and Average Values.
DFT/MRCI
cis-acylimine
Exp.
1
2
3
4
Average

2.23
2.44
2.45
2.37
2.35
2.40

2.58
2.71
2.71
2.73
2.72
2.72

OM2/MRCI(AS 10,18)

2.21 (1.205)
2.21 (1.220)
2.18 (1.208)
2.21 (1.238)
2.20

2.78 (1.301)
2.82 (1.314)
2.71 (1.309)
2.71 (1.303)
2.75

DFT/MRCI no PC

OM2/MRCI (AS 8,14)

(1.252)
(1.258)
(1.270)
(1.253)

DFT/MRCI
trans-peptide
Exp.
1
2
3
4
Average

DFT/MRCI no PC

(1.241)
(1.211)
(1.237)
(1.227)

2.66
2.65
2.66
2.65
2.65

(1.252)
(1.223)
(1.248)
(1.236)

2.97
2.89
2.89
2.89
2.91

(1.286)
(1.199)
(1.227)
(1.204)

TDB3LYP

2.75
2.76
2.71
2.69
2.73

(0.987)
(0.996)
(1.009)
(1.010)

TDB3LYP

3.10
3.09
3.12
3.12
3.11

(1.064)
(1.031)
(1.065)
(1.049)

Oscillator strengths are given in parentheses.

and LUMO 1 1, while the np* excitation from the lone pair of
the phenolic oxygen to the p* MO appears as S3 state (Table
S3, Supporting Information). In the following, we shall disregard
the higher excited states and focus on the bright pp* S1 state.
QM/MM Calculations: Active-Site Geometries

Four different snapshots from QM/MM MD trajectories of the

DsRed.M1 protein have been optimized at the DFT(B3LYP/
TZVP)/MM level. In the case of the cis-acylimine chromophore,
an overlay of the resulting structures for QM region QM2 shows
that they mainly differ in the position of W138 (Fig. 2, top).
W138 is the only residue in QM2 which does not directly interact with the chromophore. It is included because it interacts with
Glu215, which is known to play an important role for the properties of uorescent proteins.4,16 The interactions involving
W138 and Glu215 have been discussed in more detail in our
previous work.49
Table 5 presents average values of selected geometrical parameters from the QM/MM optimizations of the cis-acylimine
and trans-peptide forms of DsRed.M1. It lists the results for QM
regions QM1 (chromophore) and QM2 (chromophore plus
active-site residues) as well as the corresponding X-ray data.16
The structures are qualitatively similar, but there are also some
notable quantitative discrepancies. Better agreement with the Xray data is obtained with QM1 for interactions involving the
phenolate oxygen atom, and with QM2 for interactions involving
the imidazolinone ring. The calculated hydrogen bond distances
are generally larger when QM2 is used (except for O1glu215
N2crq).
The water molecule W138 is hydrogen-bonded to Glu215
(Ser69) in all (most) snapshots. In the case of the cis-acylimine
snapshot 4, this rather mobile water molecule (Fig. 2) has

moved far away from Ser69 in the MD simulation and remains

at a large distance after QM/MM optimization (Ow138 Nser69
with QM1 and 4.281 A
with QM2), contrary
values of 4.565 A

to the situation in snapshots 13 (average values of 2.869 A

with QM1 and 2.992 A with QM2).

The CA1


N


C


O dihedral angle in the acylimine part of
the chromophore is close to 1008 for both QM regions (Table 5)
and thus differs strongly from the crystallographic value. This
deviation has been thoroughly discussed in our previous work.49
The CD2


CG2


CA2


N2 dihedral angles indicate the planarity of the chromophore: in the computed structures, there are notable out-of-plane distortions (more so with QM2 than with
QM1, up to 178), whereas the chromophore appears to be essentially planar in the X-ray structure (Table 5).
QM/MM Calculations: Vertical Excitation Energies

The vertical excitation energies were calculated for both chromophores at the DFT/MRCI, TDB3LYP, and OM2/MRCI levels
of theory, with QM regions QM1 (Table 6) and QM2 (Table 7).
In all cases, the computed excitation energies are slightly lower
with the larger QM region (DFT/MRCI by 0.060.07 eV, OM2/
MRCI by 0.030.05 eV, TDB3LYP by 0.10 eV). The origin of
these small shifts was traced by DFT/MRCI test calculations
(using QM region QM2 in combination with QM1-based geometries) which indicate that these shifts are mostly due to structural
changes (and not to electronic effects).
DFT/MRCI calculations were also performed for QM region
QM1 without including MM point charges, in order to evaluate
the electrostatic effect of the protein environment on the excitation energies at the geometry adopted by the chromophore in the
protein (Table 6). For the cis-acylimine form, the excitation
energy of the isolated chromophore is 0.2 eV lower than that in

Journal of Computational Chemistry

DOI 10.1002/jcc

Sanchez-Garcia, Doerr, and Thiel Vol. 31, No. 8 Journal of Computational Chemistry

1610

Table 7. Computed DFT/MRCI (with and without Point Charges PC),

OM2/MRCI, and TDDFT Vertical Excitation Energies (eV) for the

cis-acylimine and trans-peptide Forms of the DsRed.M1 Protein at
Optimized QM(B3LYP/TZVP)/MM Structures Using QM Region
QM2: Results for Snapshots 1-4 and Average Values.
DFT/MRCI
cis-acylimine
Exp.
2.23
1
2.36 (1.250)
2
2.30 (1.248)
3
2.35 (1.273)
4
2.30 (1.251)
Average
2.33
DFT/MRCI
trans-peptide
Exp.
2.58
1
2.65
2
2.66
3
2.66
4
2.65
Average
2.66

(1.205)
(1.218)
(1.231)
(1.214)

OM2/MRCI (AS 10,18)

TDB3LYP

Therefore, we list in Table 7 the vertical excitation energies for

the second excitation (i.e., the bright pp* transition) for snapshots 1 and 4 (TDB3LYP, QM2, trans-peptide). In both cases,
the energy gap between S1 and S2 is small (0.21 eV and 0.04
eV, respectively). As already stated, the appearance of low lying
CT states is a typical artifact of TDDFT.
Spectra Over MD Trajectory

2.73
2.69
2.70
2.67
2.70

(1.282)
(1.280)
(1.293)
(1.289)

OM2/MRCI (AS 8,14)

2.90
2.87
2.91
2.85
2.88

(1.219)
(1.200)
(1.237)
(1.177)

2.65
2.63
2.65
2.61
2.63

(1.008)
(1.032)
(1.051)
(1.017)

TDB3LYP

2.99
3.00
3.00
3.01
3.00

(1.092)a
(1.073)
(1.093)
(0.706)a

At the TDB3LYP level, the rst excited singlet state is of np* type in
two of the snapshots (see text): snapshot 1, 2.77 (0.00278), snapshot 4,
2.97 (0.373).
Oscillator strengths are given in parentheses.

the protein. This difference is less pronounced for the trans-peptide form (0.07 eV), but in both cases the inclusion of the pointcharge environment leads to higher excitation energies.
Among the QM methods considered presently, DFT/MRCI
consistently gives vertical excitation energies that are closest to
experiment (typically 0.1 eV larger), and also reproduces the experimental gap of 0.35 eV between the lowest pp* excitation
energies of the cis-acylimine and the trans-peptide chromophores
very well (0.32 eV with QM1 and 0.33 eV with QM2).
TDB3LYP and OM2/MRCI yield similar excitation energies (0.3
0.5 eV above the experimental values), but TDB3LYP reproduces
the gap between the lowest pp* excitation energies of the cis-acylimine and trans-peptide forms (0.37 eV with QM1 and QM2)
better than OM2/MRCI where the calculated gap is too small
(0.16 eV with QM1 and 0.18 eV with QM2, Tables 6 and 7).
The vertical excitation energies discussed above refer to the
pp* HOMO?LUMO transition, that is, to the rst bright excitation. As mentioned before, the DFT/MRCI energy difference
between the rst excited pp* and np* states is larger in the protein than in the isolated chromophore. Although the p* MO is
not directly affected by the hydrogen bonding in the active site,
the n-MO is stabilized by such hydrogen bonds, so that more
energy is required to excite into the np* state. Therefore, the
S1?S2 energy gap becomes larger.
In the TDB3LYP calculations of snapshots 1 and 4 with
QM2 (Table 7), the rst excited state is a charge transfer (CT)
state dominated by a HOMO-1?LUMO excitation (Fig. 3).

Using the efcient OM2/MRCI method as QM component in

QM/MM calculations, we have calculated the vertical excitation
energies during a 500 ps MD trajectory in order to estimate the
effect of molecular motion on the excitation spectra. The calculations were done for 450 snapshots (one per ps starting after the
rst 50 ps). The histogram spectra generated from the OM2/
MRCI vertical excitation energies over the MD trajectory nicely
reproduce the experimental cis-acylimine and trans-peptide bands
with only small blue shifts (Fig. 4). For the cis-acylimine (transpeptide) chromophore, the excitation energies lie between 2.25
and 2.85 (2.53 and 3.12) eV during the whole trajectory. The experimental spectrum16 was digitalized for direct visual comparison
with the computed absorption bands (Fig. 4, bottom). The latter
were obtained as superpositions of Lorentzian functions with a
full width at half maximum of 0.05 eV and a height proportional
to the calculated oscillator strength. There is satisfactory agreement, with the computed excitation energy gap between the cisacylimine and trans-peptide chromophores of 0.23 eV being
slightly smaller than the experimental value of 0.35 eV. The comparison between the computed spectrum and the OM2/MRCI excitation energies of the individually optimized snapshots 14 for
the acylimine and peptide forms (Fig. 4, bottom) shows that molecular motion does not signicantly affect the vertical excitation
energies in the present case. This supports the validity of our procedure to determine vertical excitation energies in uorescent proteins from a small set of randomly selected snapshots.

Figure 3. HOMO, HOMO-1, and LUMO for the trans-peptide chromophore of DsRed.M1 at the optimized QM(B3LYP/TZVP)/MM
geometry of snapshot 4 (QM region QM2).

Journal of Computational Chemistry

DOI 10.1002/jcc

QM/MM Study of the Absorption Spectra of DsRed.M1 Chromophores

1611

forms of DsRed.M1. They are rather insensitive to the size of

the chosen QM region (changes of less than 0.1 eV when going
from QM1 to QM2). OM2/MRCI and TDDFT overestimate the
rst vertical excitation energies by 0.30.5 (0.20.4) eV relative
to experiment (DFT/MRCI). The experimental gap of 0.35 eV
between the lowest pp* excitation energies of the red and green
forms is well reproduced by DFT/MRCI and TDDFT (0.32 and
0.37 eV, respectively) whereas OM2/MRCI gives a smaller gap.
A histogram spectrum for an equal mixture of the red and green
form of DsRed.M1, generated by OM2/MRCI calculations on
450 snapshots along QM/MM MD trajectories, matches the experimental spectrum quite well, with a gap of 0.23 eV and an
overall blue shift of about 0.3 eV, indicating the usefulness of
OM2/MRCI for such purposes.
In an overall assessment, the presently applied QM methods
all perform reasonably well in the calculation of the electronic
spectra of the DsRed.M1 chromophores. DFT/MRCI is the computationally most demanding approach, but it also provides the
highest accuracy and is still practical for chromophores of this
size. Because it does not suffer from the well-known shortcomings of TDDFT and is more affordable computationally
than CASSCF-based approaches, it is an attractive choice for
QM/MM studies of electronically excited states in uorescent
proteins.

References

Figure 4. Top: Histograms obtained from OM2/MRCI excitation

energies calculated along 450 ps MD trajectories (450 snapshots).
The counts specify the number of snapshots with excitation energy
in the given range. Bottom: Comparison between the resulting
OM2/MRCI spectra (see text) and the experimental spectra.16

Conclusion
QM/MM geometry optimizations of the DsRed.M1 protein indicate some structural changes in the chromophore when compared with the QM gas-phase geometries. The active-site interactions induce slight deviations from planarity, and in the case
of the red cis-acylimine chromophore, the C
O moiety of
Phe65 twists further out of conjugation. These structural changes
lead to higher vertical excitation energies in the protein. The
electrostatic interactions between the chromophore and the protein environment generally also tend to increase the excitation
energies, especially so for the dark np* state. In the case of the
lowest singlet excited state of the red chromophore which is a
bright pp* state, the DFT/MRCI calculations give an overall
blue shift of 0.4 eV when going from the gas phase to the protein, with roughly equal contributions from the structural
changes and the electrostatic interactions.
The DFT/MRCI excitation energies are closest to experiment
both for the red (cis-acylimine) and the green (trans-peptide)

1. Matz, M. V.; Fradkov, A. F.; Labas, Y. A.; Savitsky, A. P.; Zaraisky,

A. G.; Markelov, M. L.; Lukyanov, S. A. Nat Biotechnol 1999, 17, 969.
2. Gross, L. A.; Baird, G. S.; Hoffman, R. C.; Baldridge, K. K.; Tsien, R. Y.
Proc Natl Acad Sci USA 2000, 97, 11990.
3. Wall, M. A.; Socolich, M.; Ranganathan, R. Nat Struct Biol 2000, 7, 1133.
4. Yarbrough, D.; Wachter, R. M.; Kallio, K.; Matz, M. V.; Remington, S. J.
Proc Natl Acad Sci USA 2001, 98, 462.
5. Bevis, B. J.; Glick, B. S. Nat Biotechnol 2002, 20, 83.
6. Gavin, P.; Devenish, R. J.; Prescott, M. Biochem Biophys Res Commun
2002, 298, 707.
7. Shaner, N. C.; Campbell, R. E.; Steinbach, P. A.; Giepmans, B. N. G.;
Palmer, A. E.; Tsien, R. Y. Nat Biotechnol 2004, 22, 1567.
8. Stepanenko, O.; Verkhusha, V.; Kazakov, V.; Shavlovsky, M.; Kuznetsova, I.; Uversky, V.; Turoverov, K. Biochemistry 2004, 43, 14913.
9. Shaner, N. C.; Steinbach, P. A.; Tsien, R. Y. Nat Methods 2005, 2, 905.
10. Shkrob, M. A.; Yanushevich, Y. G.; Chudakov, D. M.; Gurskaya, N. G.;
Labas, Y. A.; Poponov, S. Y.; Mudrik, N. N.; Lukyanov, S.; Lukyanov,
K. A. Biochem J 2005, 392, 649.
11. Loos, D. C.; Habuchi, S.; Flors, C.; Hotta, J.-I.; Wiedenmann, J.; Nienhaus,
G. U.; Hofkens, J. J Am Chem Soc 2006, 128, 6270.
12. Shu, X.; Shaner, N. C.; Yarbrough, C. A.; Tsien, R. Y.; Remington, S. J.
Biochemistry 2006, 45, 9639.
13. Ai, H.-W.; Shaner, N. C.; Cheng, Z.; Tsien, R. Y.; Campbell, R. E. Biochemistry 2007, 46, 5904.
14. Ip, D. T.-M.; Wong, K.-B.; Wan, D. C.-C. Mar Biotechnol 2007, 9, 469.
15. Shcherbo, D.; Merzlyak, E. M.; Chepurnykh, T. V.; Fradkov, A. F.; Ermakova, G. V.; Solovieva, E. A.; Lukyanov, K. A.; Bogdanova, E. A.; Zaraisky, A. G.; Lukyanov, S.; Chudakov, D. M. Nat Methods 2007, 4, 741.
16. Strongin, D. E.; Bevis, B.; Khuong, N.; Downing, M. E.; Strack, R. L.;
Sundaram, K.; Glick, B. S.; Keenan, R. J. Protein Eng Des Sel 2007, 20,
525.
17. Goulding, A.; Shrestha, S.; Dria, K.; Hunt, E.; Deo, S. K. Protein Eng Des
Sel 2008, 21, 101.

Journal of Computational Chemistry

DOI 10.1002/jcc

1612

Sanchez-Garcia, Doerr, and Thiel Vol. 31, No. 8 Journal of Computational Chemistry

18. Nienhaus, K.; Nar, H.; Heilker, R.; Wiedenmann, J.; Nienhaus, G. U. J Am
Chem Soc 2008, 130, 12578.
19. Schleifenbaum, F.; Blum, C.; Elgass, K.; Subramaniam, V.; Meixner, A. J.
J Phys Chem B 2008, 112, 7669.
20. Shaner, N. C.; Lin, M. Z.; Mckeown, M. R.; Steinbach, P. A.; Hazelwood,
K. L.; Davidson, M. W.; Tsien, R. Y. Nat Methods 2008, 5, 545.
21. Fischer, M.; Haase, I.; Simmeth, E.; Gerisch, G.; Muller-Taubenberger, A.
FEBS Lett 2004, 577, 227.
22. Forner, J.; Binder, S. BMC Plant Biol 2007, 7, 1471.
23. Hsu, Y.-Y.; Liu, Y.-N.; Wang, W.; Kao, F.-J.; Kung, S.-H. Biochem Biophys Res Commun 2007, 353, 939.
24. Merzlyak, E. M.; Goedhart, J.; Shcherbo, D.; Bulina, M. E.; Shcheglov, A.
S.; Fradkov, A. F.; Gaintzeva, A.; Lukyanov, K. A.; Lukyanov, S.; Gadella,
T. W. J.; Chudakov, D. M. Nat Methods 2007, 4, 555.
25. Shaner, N. C.; Patterson, G. H.; Davidson, M. W. J Cell Sci 2007, 120,
4247.
26. Turchin, I. V.; Savitsky, A. P.; Kamensky, V. A.; Plehanov, V. I.; Meerovich, I. G.; Arslanbaeva, L. R.; Jerdeva, V. V.; Orlova, A. G.; Kleshnin, M.
S.; Shirmanova, M. V.; Fiks, I. I. Proc SPIE-Int Soc Opt Eng 2007, 6449,
644915.
27. Boye, S.; Brondsted Nielsen, S.; Krogh, H.; Bloch Nielsen, I.; Pedersen, U.
V.; Bell, A. F.; He, X.; Tonge, P. J.; Andersen, L. H. Phys Chem Chem
Phys 2003, 5, 3021.
28. Olsen, S.; Smith, S. C. Chem Phys Lett 2006, 426, 159.
29. Olsen, S.; Prescott, M.; Wilmann, P.; Battad, J.; Rossjohn, J.; Smith, S. C.
Chem Phys Lett 2006, 420, 507.
30. Wan, S.; Liu, S.; Zhao, G.; Chen, M.; Han, K.; Sun, M. Biophys Chem
2007, 129, 218.
31. Voliani, V.; Bizzarri, R.; Nifosi, R.; Abbruzzetti, S.; Grandi, E.; Viappiani,
C.; Beltram, F. J Phys Chem B 2008, 112, 10714.
32. Marques, M. A. L.; Lopez, X.; Varsano, D.; Castro, A.; Rubio, A. Phys
Rev Lett 2003, 90, 258101.
33. Camilloni, C.; Provasi, D.; Tiana, G.; Broglia, R. A. J Phys Chem B 2007,
111, 10807.
34. Nifosi, R.; Luo, Y. J Phys Chem B 2007, 111, 505.
35. Amat, P.; Granucci, G.; Buda, F.; Persico, M.; Tozzini, V. J Phys Chem B
2006, 110, 9348.
36. Nemukhin, A. V.; Topol, I. A.; Burt, S. K. J Chem Theory Comput 2006,
2, 292.
37. Sun, M. Int J Quantum Chem 2006, 106, 1020.
38. Patnaik, S. S.; Trohalaki, S.; Naik, R. R.; Stone, M. O.; Pachter, R. Biopolymers 2007, 85, 253.
39. Schafer, L. V.; Groenhof, G.; Boggio-Pasqua, M.; Robb, M. A.; Grubmuller, H. PLoS Comput Biol 2008, 4, e1000034.
40. Bravaya, K. B.; Bochenkova, A. V.; Granovsky, A. A.; Savitsky, A. P.;
Nemukhin, A. V. J Phys Chem A 2008, 112, 8804.
41. Timerghazin, Q. K.; Carlson, H. J.; Liang, C.; Campbell, R. E.; Brown, A.
J Phys Chem B 2008, 112, 2533.
42. Jacquemin, D.; Perpete, E. A.; Scuseria, G. E.; Cioni, I.; Adamo, C.
J Chem Theory Comput 2008, 4, 123.
43. Dreuw, A.; Head-Gordon, M. Chem Rev 2005, 105, 4009.
44. Filippi, C.; Zaccheddu, M.; Buda, F. J Chem Theory Comput 2009, 5,
2074.
45. Epifanovsky, E.; Polyakov, I.; Grigorenko, B.; Nemukhin, A.; Krylov, A.
J Chem Theory Comput 2009, 5, 1895.
46. Senn, H. M.; Thiel, W. Top Curr Chem 2007, 268, 173.
47. Senn, H. M.; Thiel, W. Angew Chem Int Ed 2009, 48, 1198.
48. Grigorenko, B.; Savitsky, A.; Topol, I.; Burt, S.; Nemukhin, A. J Phys
Chem B 2006, 110, 18635.
49. Sanchez-Garcia, E.; Doerr, M.; Hsiao, Y.-W.; Thiel, W. J Phys Chem B
(accepted).
50. Sun, Q.; Doerr, M.; Li, Z.; Thiel, W.; Smith, S. 2009 (submitted).

51. Ahlrichs, R.; Bar, M.; Haser, M.; Horn, H.; Kolmel, C. Chem Phys Lett
1989, 162, 165.
52. Dirac, P. A. M. Proc R Soc London Ser A 1929, 123, 714.
53. Slater, J. C. Phys Rev A 1951, 81, 385.
54. Becke, A. D. Phys Rev A 1988, 38, 3098.
55. Lee, C.; Yang, W.; Parr, R. G. Phys Rev B 1988, 37, 785.
56. Becke, A. D. J Chem Phys 1993, 98, 5648.
57. Schafer, A.; Huber, C.; Ahlrichs, R. J Chem Phys 1994, 100, 5829.
58. www.pdb.org.
59. Mackerell, A. D., Jr.; Bashford, D.; Bellott, M.; Dunbrack, R. L.; Evanseck, J. D.; Field, M. J.; Fischer, S.; Gao, J.; Guo, H.; Ha, S.; Joseph-Mccarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F. T. K.; Mattos, C.; Michnick, S.;
Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher, W. E., III; Roux, B.;
Schlenkrich, M.; Smith, J. C.; Stote, R.; Straub, J.; Watanabe, M.; Wiorkiewicz-Kuczera, J.; Yin, D.; Karplus, M. J Phys Chem B 1998, 102, 3586.
60. Mackerell, A. D.; Feig, M.; Brooks, C. L. J Comput Chem 2004, 25, 1400.
61. Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein,
M. L. J Chem Phys 1983, 79, 926.
62. Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.; Frauenheim, T.; Suhai, S.; Seifert, G. Phys Rev B 1998, 58, 7260.
63. Frauenheim, T.; Seifert, G.; Elstner, M.; Niehaus, T.; Kohler, C.; Amkreutz, M.; Sternberg, M.; Hajnal, Z.; Di Carlo, A.; Suhai, S. J Phys 2002,
14, 3015.
64. Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Swaminathan, S.; Karplus, M. J Comput Chem 1983, 4, 187.
65. Mackerell, A. D.; Brooks, B.; Brooks, C. L., III; Nilsson, L.; Roux, B.;
Won, Y.; Karplus, M. Encyclopedia of Computational Chemistry;
Schleyer, P., Ed.; Wiley: Chicester, UK, 1998; p. 271.
66. Cui, Q.; Elstner, M.; Kaxiras, E.; Frauenheim, T.; Karplus, M. J Phys
Chem B 2001, 105, 569.
67. Humphrey, W.; Dalke, A.; Schulten, K. J Mol Graphics 1996, 14, 33.
68. Sherwood, P.; De Vries, A. H.; Guest, M. F.; Schreckenbach, G.; Catlow,
C. R. A.; French, S. A.; Sokol, A. A.; Bromley, S. T.; Thiel, W.; Turner, A.
J.; Billeter, S.; Terstegen, F.; Thiel, S.; Kendrick, J.; Rogers, S. C.; Casci,
J.; Watson, M.; King, F.; Karlsen, E.; Sjovoll, M.; Fahmi, A.; Schafer, A.;
Lennartz, C. THEOCHEM 2003, 632, 1.
69. Smith, W.; Forester, T. R. J Mol Graphics 1996, 14, 136.
70. Billeter, S. R.; Turner, A. J.; Thiel, W. Phys Chem Chem Phys 2000, 2,
2177.
71. Bakowies, D.; Thiel, W. J Phys Chem 1996, 100, 10580.
72. Antes, I.; Thiel, W. ACS Symp Ser 1998, 712, 50.
73. De Vries, A. H.; Sherwood, P.; Collins, S. J.; Rigby, A. M.; Rigutto, M.;
Kramer, G. J. J Phys Chem B 1999, 103, 6133.
74. Van Gisbergen, S. J. A. S.; Baerends, J. J Chem Phys 1995, 103, 9347.
75. Jamorski, C. C.; Salahub, M. E. J Chem Phys 1996, 104, 5134.
76. Petersilka, M. G. U. J.; Gross, E. K. U. Phys Rev Lett 1996, 76, 1212.
77. Bauernschmitt, R.; Haser, M.; Treutler, O.; Ahlrichs, R. Chem Phys Lett
1997, 264, 573.
78. Silva-Junior, M. R.; Schreiber, M.; Sauer, S. P. A.; Thiel, W. J Chem Phys
2008, 129, 104103.
79. Grimme, S.; Waletzke, M. J Chem Phys 1999, 111, 5645.
80. Becke, A. D. J Chem Phys 1993, 98, 1372.
81. Gerenkamp, M. PhD Thesis, Universitat Munster, 2005.
82. Parac, M.; Doerr, M.; Marian, C. M.; Thiel, W. J Comp Chem, DOI:
10.1002/jcc.21233.
83. Weber, W. PhD Thesis, Universitat Zurich, 1996.
84. Weber, W.; Thiel, W. Theory Chem Acc 2000, 495.
85. Koslowski, A.; Beck, M. E.; Thiel, W. J Comput Chem 2003, 24, 714.
86. www.chemshell.org.
87. Thiel, W. Max-Planck-Institut fur Kohlenforschung, Kaiser-WilhelmPlatz 1, D-45470 Mulheim, Germany, 2004, MNDO99 version 6.1.

Journal of Computational Chemistry

DOI 10.1002/jcc

Copyright of Journal of Computational Chemistry is the property of Wiley Periodicals, Inc., A Wiley Company
and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for individual use.