1. Predictive Structure-Based Models of Evolved Drug Resistance
Alissa Calderon a,b*, Carla Islas b, Robert P. Metzger a, Gary B. Fogel c, David Hecht a,b and B. Mikael Bergdahl a
a. Department of Chemistry and Biochemistry, San Diego State University, San Diego, CA 92182
b. Department of Chemistry, Southwestern College, Chula Vista CA 91910
c. Natural Selection, Inc., San Diego, CA 92121
Background
References
Methods
This research was supported by the National Institute of General Medical Sciences of the National Institutes
of Health under Award Number SC3GM100791. The content is solely the responsibility of the authors and
does not necessarily represent the official views of the National Institutes of Health.
Plasmodium falciparum (Pf), the causal agent of malaria,
provides an ideal system for modeling of the evolution of
drug resistance, and for developing predictive methods
that might assist in the forecasting of future adaptation.
In the most prevalent malaria strains, key amino acid
substitutions (N51I, C59R, S108N and I164L) confer
resistance to anti-folate compounds such as
pyrimethamine and cycloguanil that target Pf-
dihydrofolate reductase (DHFR). One method of better
understanding this process is by in silico evolution
modeling the effect of likely amino acid changes in Pf-
DHFR on anti-folate drug binding affinities. Towards this
goal we have recently demonstrated that in silico
evolution can correctly identify trajectories that will lead
to the development of pyrimethamine resistant variants
of Pf-DHFR in the malaria parasite P. falciparum [1-3].
Here we apply this methodology to predict the
development of resistance in wild-type Pf-DHFR to the
antibiotic trimethoprim.
ResultsResults
1). Hecht et al. (2012) “Modeling the evolution of drug resistance in
malaria”, JCAMD, 26:1343-1353.
2). Hecht et al. (2012) “Towards predictive structure-based models of
evolved drug resistance”, 2012 IEEE Symposium on
Computational Intelligence in Bioinformatics and Computational
Biology, San Diego, pp 120–126.
3). Fogel et al. (2013) “Modeling the Evolution of Drug Resistance in
Plasmodium falciparum”, 2013 IEEE Congress on Evolutionary
Computation, Cancun, Mexico.
4). Gasasira et al. (2010) “Effect of trimethoprim-sulphamethoxazole
on the risk of malaria in HIV-infected Ugandan children living in
an area of widespread antifolate resistance”, Malaria Journal,
9:177.
Variation
Each round of in silico evolution begins with a parent
DHFR sequence in FASTA format. Using a script, the
parent sequence is uploaded and a user inputs the
number of amino acid positions to vary for each round
of evolution, and the number of offspring sequences.
One of two amino acid replacement matrices is used
as a probability matrix for site specific amino acid
replacements. One matrix is based on a structure-
based alignment of two DHFR x-ray crystal structures
and homology [1-3]. A second amino acid substitution
matrix was generated from the PAM 250 matrix.
Homology Modeling
All homology models are generated using MOE
(www.chemcomp.com) with default settings.
Fitness & Selection
Docking experiments are performed using GOLD
(www.ccdc.cam.ac.uk). Each offspring sequence is
scored using a fitness function specifically designed
specifically to evaluate the ability of each offspring
sequence to maintain binding affinity for the co-factor
NADPH and the substrate dihydrofolate while
reducing affinity for the inhibitor:
i). Docked conformations of NADPH and 7,8 dihydrofolate
match x-ray conformations.
ii). Docked conformations of the inhibitor do not bind in the
active site pocket. Poses found to dock outside of the pocket
by visual inspection satisfy this constraint.
iii). Docking scores of NADPH and 7,8-dihydrofolate for each
offspring sequence should be roughly similar to that of the
parent sequence (as well as to that of the wild type DHFR
sequence). Lower docking scores imply loss of binding
affinity resulting in a selective disadvantage.
Methods
Several models of Pf-DHFR sequences predicted to be
resistant to trimethoprim resulting from three
independent runs of in silico evolution are presented
(Table 1). In all cases the NADPH docking scores were
comparable (or higher) to wt and the DHF docking scores
were lower. This is not surprising since the resistance
conveying amino acid substitutions occur in/around the
active site.
We tested our methodology against all combinations of
amino acid replacements known to confer resistance to
pyrimethamine (Figure 3). All substitutions, with several
possible exceptions (blue), resulted in no predicted
resistance to trimethoprim.
Higher levels of the I164L substitution are found in P.
falciparum from HIV co-infected patients (in Africa)
given trimethoprim on a prophylactic basis [4].
Interestingly, this substitution is more commonly found in
Asia.
Figure 3. Starting with the wild type amino acid sequence
for Pf-DHFR (green), all combinations of amino acid
replacements known to confer resistance to pyrimethamine
in the wild (red) were tested with trimethoprim. All
substitutions (with the possible exceptions of I164L,
N51I_I164L, and C59R_I164L) resulted in no predicted
resistance to trimethoprim (pathways with crossed circles).
Those sequences containing the I164L mutation (blue) gave
mixed results and need to be confirmed experimentally in
future studies.
Figure 1. Workflow for generating and evaluating variant
DHFR sequences. The initial input is the wt Pf-DHFR
sequence. The loop of variation, scoring, and, generation of
parent solutions for the next “generation” of evolution
continues until a termination criterion is satisfied [1-3].
Figure 2. Superposition of x-ray crystal conformations of
trimethoprim and NADPH bound to wt Pf-DHFR from
3FRB.pdb (colored in gold) vs. docked conformations (in
CPK), RMSD values <1.00Å.
Table 1. Summary of 3 independent runs of in silico evolution.
The winner for each run is presented along with the
corresponding docking scores. Two runs were performed
using the amino acid replacement matrix based on a
structure-based alignment (e.g. Position Matrix) and one run
was performed using the amino acid replacement matrix
based on PAM 250. Each round was terminated when a
predicted resistant sequence was identified.
Gen #
Substitution
Matrix
Mutation Ligand
PLP
Fitness
PLP
Score
wt wt NADPH 61.77 -42.58
DHF 73.12 -59.91
TMP 73.06 -66.18
Run 1 Position V45T NADPH 66.67 -66.6
Gen 1 Matrix DHF 30.4 -26.32
TMP N/A N/A
Run 2 Position V45T NADPH 84.42 -73.09
Gen 1 Matrix DHF 34.65 -26.44
TMP N/A N/A
Run 2 Position L53V NADPH 79.15 -63.55
Gen 1 Matrix DHF 37.95 -54.55
TMP N/A N/A
Run 2 Position S52L NADPH 77.26 -68.48
Gen 1 Matrix DHF 28.37 -27.73
TMP N/A N/A
Run 3 PAM 250 Y170F NADPH 88.44 -70.93
Gen 2 Matrix DHF 40.69 -32.13
TMP N/A N/A
Run 3 PAM 250 V103I NADPH 75.82 -64.54
Gen 2 Matrix DHF 25.41 -22.77
TMP N/A N/A
Conclusions/Future WorkDiscussion
These results imply that trimethoprim and
analogues could be effective vs. anti-folate drug
resistant strains of Malaria.
In future studies we plan to perform experimental
validation studies, expressing predicted resistant
sequences and performing inhibition assays.