Structure prediction of LDLR-HNP1 complex based on docking enhanced by LDLR binding 3D motif
Reyhaneh Esmaielbeiki1*, Declan P. Naughton1 and Jean-Christophe Nebel1.
1 Faculty of Science, Engineering and Computing, Kingston University, Kingston-upon-Thames, KT1 2EE, UK
*Correspondence to: Reyhaneh Esmaielbeiki, Faculty of Science, Engineering and Computing, Kingston
University, Kingston-upon-Thames, KT1 2EE, UK. T: +44 (0)20 8417 7159. F:+44 (0)20 8417 2972. E-mail:
Date of initial submission: 14-07-2011
Date of revised submission: -
Date of final acceptance: 26-08-2011
Human antimicrobial peptides (AMPs), including defensins, have come under intense scrutiny owing to their
key multiple roles as antimicrobial agents. Not only do they display direct action on microbes, but also recently
they have shown to interact with the immune system to increase antimicrobial activity. Unfortunately, since
mechanisms involved in the binding of AMPs to mammalian cells are largely unknown, their potential as novel
anti-infective agents cannot be exploited yet. Following the reported interaction of Human Neutrophil Peptide 1
dimer (HNP1) with a low density lipoprotein receptor (LDLR), a computational study was conducted to
discover their putative mode of interaction.
State-of-the-art docking software produced a set of LDLR-HNP1 complex 3D models. Creation of a 3D motif
capturing atomic interactions of LDLR binding interface allowed selection of the most plausible configurations.
Eventually, only two models were in agreement with the literature.
Binding energy estimations revealed that not only one of them is particularly stable, but also interaction with
LDLR weakens significantly bonds within the HNP1 dimer. This may be significant since it suggests a
mechanism for internalisation of HNP1 in mammalian cells.
In addition to a novel approach for complex structure prediction, this study proposes a 3D model of the LDLR-
HNP1 complex which highlights the key residues which are involved in the interactions. The putative
identification of the receptor binding mechanism should inform the future design of synthetic HNPs to afford
maximum internalisation, which could lead to novel anti-infective drugs.
Keywords: Low density lipoprotein Receptor; 3D motif; protein-protein interaction; docking; human alpha
defensin; human immune system.
Human antimicrobial peptides (AMPs), including mammalian defensins, have come under intense scrutiny
owing to their key multiple roles as antimicrobial agents against a range of bacteria, fungi and viruses. These
roles have been reported to involve immunostimulation via chemotaxis, direct action on viral particles, and
binding to, followed by internalisation, into mammalian cells where antimicrobial activity is manifested through
inhibition of viral replication, via inhibition of protein kinase C signalling [1, 2, 3].
These molecules provide enormous scope for the investigation of mechanisms involved in infection, along with
immune response events, and represent a reservoir of potential novel anti-infective agents. In this vein, the use
of synthetic AMPs to treat HIV was reported as early as 1993 . Given the appearance and growth in numbers
of drug resistant infections, and the relative paucity of new clinically effective antimicrobial agents, further
studies are warranted to optimise the activities of natural and synthetic AMPs.
One key step, which requires further study, is to optimise the binding of (synthetic) AMPs to mammalian cells
to afford internalisation for intracellular defences to operate. Following the reported interaction of human α-
defensins with a low density lipoprotein receptor (LDLR) [1, 5], a plausible approach is to study potential
interactions between AMPs and the LDLR.
The LDLR family contains seven homologous members and is responsible for mediating different types of
ligands especially cholesterol into the cell . Their structure is composed of several domains which include a
ligand binding (LB) domain  composed of ligand binding modules (LAs), also named complement-type
repeats (CRs), a beta-propeller domain and transmembrane and cytoplasmic sections (Figure 1, top row).
Figure 1. Modular structure of LDLR receptor family: general domain pattern (top) and schematic
representation of the LDLR-ligand binding modes of known complexes (bottom).
The Low-density lipoprotein receptor family interacts with a wide variety of human and virion proteins 
through their homologous LA modules which are between 40-50 residues long [9,10] (Figure 2). LA’s structure
is stabilised by three disulfide bonds and a calcium ion [11, 12]. This ion is an essential element of the ligand
binding domain conformation since it is required to establish interaction between LDLR and the ligand [8, 13].
Figure 2. Multiple sequence and structure alignment of ligand binding domains of LDLR family complexes. In
the sequence alignment, residues involved in calcium interaction are denoted by dots. The three conserved
acidic residues and conserved tryptophan/phenylalanine are highlighted with black arrows. Sequence numbering
is based on 2KRI:B. In the structure alignment the three conserved acidic residues and conserved tryptophan are
shown on 2KRI structure. Residues numbering is based on 2KRI:B. The ligand binding domains associated with
each colour are: 2KRI-A4: red, 2FCW-A3:green, 2FCW-A4:dark blue, 2FYL-CR5:purple, 2FYL-CR6:yellow,
2KNY-CR17:orange, 1N7D-A4:deep teal, 1N7D-A5: gray, 1V9U-V3: pink.
High-resolution crystal structures of the available LDLR complexes have revealed that electrostatic forces play
an important role in interactions . This key function is captured by the minimal interaction motif described by
Jensen et al.  (Figure 3), which also highlights a hydrophobic element in the interaction. The receptor’s
conserved acidic residues (ASP/GLU) interact with a ligand’s lysine through a salt bridge creating a
hydrophobic environment for the side chain of a receptor’s tryptophan (TRP). In addition, a hydrophobic side
chain, ψ, (usually Leucine or isolecine) from the ligand sits next to TRP.
Figure 3. Minimal binding motif defined by Jensen et al. .
This paper investigates the mode of interaction between a class of α-defensins and LDLR by the mean of
predicted structural models [1, 5]. First, we produce a novel 3D motif which describes the binding
characteristics of LDLR-ligand interactions. Then, the motif is used as constraint to evaluate LDLR-α-defensin
complex models generated by state of the art docking software.
MATERIALS AND METHODS
Protein Data sets
Our study relies on the investigation of all 3D complexes involving a ligand binding domain of the LDLR
family. Query of the RCSB Protein Data Bank  using Blast  on March 2011 revealed that the structures
of six complexes have been resolved (Table 1). They all involve human proteins belonging to three members of
the LDLR family, i.e. Low-density Lipoprotein receptor (LDLR), lipoprotein receptor-related protein 1 (LRP1)
and Very low-density lipoprotein receptor (VLDLR). The sequences of their ligand binding domain were
extracted from Uniprot , where their accession numbers are P01130: LDLR, P98155: VLDLR and Q07954:
LRP1, respectively. Although their ligand binding modules are named LA, CR and V for LDLR, LRP and
VLDR, respectively, in this paper we use LA when referring to any of them.
Table 1. Known 3D structures of complexes involving members of the LDLR family.
PDB Code Receptor and Domains Ligand Ligand Complete Name
LDLR LA3,LA4 MRAP D3
associated protein Domain 3
associated protein Domain 1
LRP CR5,CR6 MRAP D1
LDLR LA4 Apo(H) Beta-2-glycoprotein 1
LRP CR17 Apo(E) Apolipoprotein E
LDLR LA4,LA5 LDLR beta propeller -
VLDLR V3 HRV2 VP1 Human rhinovirus 2 Viral Protein 1
In agreement with the existing 2D motif , sequence alignment of the LA modules (Figure 2) using ClustalW
 shows highly conserved acidic residues and a tryptophan/phenylalanine (TRP/PHE) - pairwise E-values
were calculated using Blast  (Table 2). Structural conservation of the ligand binding domains of the
receptors, i.e. LA3, LA4, LA5, CR5, CR6, CR17 and V3, are illustrated (Figure 2) and quantified (Table 2)
using the 3D alignment tool Pymol .
Since LA4 is the domain which is the most common in this set - in three cases out of six – it is used as
representative for the purpose of α-defensin docking. Similarly, among these AMPs, defensin Human
Neutrophil Peptide-1 (HNP1), which has been specifically shown to interact with LDLR [1, 5], is selected as
representative. Sequences and structures of HNP1 and LA4 were extracted from the PDB : 3GNY  and
2KRI  codes respectively.
Table 2. E-value between sequences of the Ligand binding domains and RMSD between their 3D structures.
Creation of 3D motif
Extending the existing 2D motif  using the approach suggested by Nebel et al. [21, 22], we produce a 3D
motif which describes the conserved 3D positions of the key atoms involved in LDLR-ligand interaction (Figure
4.A). From the receptor, the conserved acidic residues and TRP/PHE are represented by their alpha carbon
atoms. In addition, in order to add a constraint regarding interaction with the calcium ion, we include the oxygen
2FCWA3 2FCWA4 2FYLCR5 2FYLCR6 2KRIA4 2KNYCR17 1N7DA4 1N7DA5 1V9UVD3
RMSD Between Structures (Å)
2FCWA3 2FCWA4 2FYLCR5 2FYLCR6 2KRIA4 2KNYCR17 1N7DA4 1N7DA5 1V9UVD3
atom of the TRP/PHE carboxyl group with whom it interacts. On the ligand side, the basic residue interaction is
expressed by the side-chain nitrogen atom(s) which form(s) hydrogen bond(s) with acidic residues of the
Figure 4. (A) The 3D motif is represented by spheres. The blue ones show positions of N atoms from the ligand.
The black ones are the C-alpha atoms of the ASP and TRP and the red sphere is the O atom of TRP. Location of
the calcium is marked by a grey sphere. Image produced using Pymol. (B) Number of ranks to achieve 100%
recall of the top predictions (or recall of top predictions with top positions). In the legend the complexes names
are followed by C and M for curves based on cluster size and 3D motif method, respectively.
The actual coordinates of the consensus atoms forming the 3D motif are calculated by the multiple structure
alignments of these atoms using all available receptor-protein complexes. Here, only atoms from the receptor
side are used as superimposition constraints. Since their 3D structures are very well conserved - their average
RMSD is 0.28 Å - positions of all receptor atoms in the 3D motif are approximated by the average coordinates
of the aligned atoms. On the other hand, given that every ligand displays a very different receptor binding sites,
there is no consensus 3D position regarding the location of the nitrogen atom(s) of the basic residue(s).
However, there must be specific constraints in terms of their distance and orientation from the receptor. In our
3D motif, we express implicitly these constraints by storing all the actual nitrogen positions available in our
Note that among the 9 binding sites of the available structures, we excluded that of 2FYLCR6 in the
construction of our motif since its LA module is structurally different from the others as measured by an average
RMSD of 2.97 Å (Table 2).
Docking predictions are performed using the ClusPro 2.0 docking program , which, in addition to be freely
available for academic research, has demonstrated to be the state of the art at CAPRI 2009 (Critical Assessment
of Predicted Interactions) [24, 25, 26]. Cluspro works by initially calculating 70,000 docking models. Then, the
1000 models with the best energy conformation are selected and clustered using PIPER . Models with the
most neighbours within a 9 Å C-alpha RMSD cut-off are chosen as cluster representatives and are qualified by
the size of their associated cluster.
The ClusPro docking results are generated according to different constraints. For each category, software
produces a set of predicted models ranked according to their cluster size. Since previous studies have
highlighted the important role of electrostatic and hydrophobic interactions in LDLR complexes [8, 9, 14], we
only consider predictions generated under ‘electrostatic favoured’ and ‘Van der Waals + Electrostatic forces
(VDW/elec)’ modes. In this work default software parameters are used.
3D motif evaluation
Our 3D motif was evaluated in docking prediction task using a leave-one-complex-out cross validation. First, a
resolved 3D complex involving LDLR is selected. Secondly, a 3D motif is produced using all the other available
LDLR complexes. Thirdly, the two chains involved in the complex are submitted to ClusPro which generates a
set of putative complex models. Then, the fitting of the 3D motif to each model is used to score predicted
complexes. Finally, the produced ranked list is compared with the list of models ranked according to their
quality as expressed by their RMSD with respect to the actual resolved structure.
3D motif fitting is performed by superimposition on the binding site of the predicted LDLR-ligand complex
using receptor atoms as constraints. We define the quality of a prediction as the shortest distance between the
nitrogen of the basic residue of the ligand and those present in the 3D motif.
LDLR-HNP1 model prediction
Using the procedure previously described, LDLR-HNP1 complex estimates are generated by ClusPro and
ranked using our 3D motif fitting measure. Then, the best models according to that score are further analysed in
order to establish which ones are in agreement with the literature.
Finally, the stability of the remaining modelled complexes is quantified by both calculating the number of
intermolecular contacts and estimating pair wise interaction energies between the different chains involved in
those complexes. Detailed information on residue-residue and atom-atom contacts is provided by the Contact
Map Analysis server which is part of the software suite SPACE . In addition, since previous studies [29,
30] have shown good correlation between experimental measurements and energy calculations produced by the
FoldX software [31, 32], its latest version, v3.0 beta5.1 (http://foldx.embl.de/) has been selected to evaluate
binding energy between the two HNP1 monomers and between LDLR and each of the HNP1 chains.
Modes of interactions of LDLR-ligand complexes
Within the LDLR-ligand complexes, two modes of interaction between the LB module and the ligand have been
identified (Figure 1, bottom rows). In the first mode, two ligand binding modules of LDLR are required to
establish an interaction with the ligand. In 2FCW  the third and fourth modules of the ligand binding domain
(LA3,4) bind to MRAP domain 3 (MRAPD3). In 2FYL , two modules of complement-type ligand binding
repeats (CR5,6) interact with two different sections of MRAP domain 1 (MRAPD1). Similarly, LA4,5 of 1N7D
 bind to two different sites of LDLR beta propeller.
In the second mode, only one ligand binding module of LDLR binds to the ligand. Apo(H) and Apo(E) bind to
A4 in 2KRI and CR17 in 2KNY, respectively. In 1V9U , the third LB module of VLDLR (V3) interacts
with Human rhinovirus 2 (HRV2) viral proteins VP1.
As a whole, the available six structures describe 9 different binding sites, since three complexes operate in the
first mode of interaction.
3D motif validation
Our 3D motif, displayed in Figure 4.A, is evaluated against predictions of 9 binding sites. Results are reported
in Figure 4.B, where the number of ranks required to achieve 100% recall, r100%recall, is expressed as a function
of the number of top quality predictions, t. A perfect prediction evaluation scheme would place the t best
predictions on the t top-most positions of the ranked list, whereas the worst evaluation scheme would require the
whole list to recall the t best predictions.
Although Cluspro developers do not recommend judging the produced models according to their associated
cluster size, software output shows models ranked according to that score which obviously influence user’s
usage of these models. Therefore, we also show on Figure 4.B how cluster size would perform if used to rank
In every case, ranking based on 3D motif fitting produces curves closer to the perfect prediction than those
generated from cluster size ranking. As a consequence fewer models are needed to recall the top quality
predictions when the 3D motif is used to access LDLR interaction predictions. If the LDLR-Apo(E) (2KNY)
complex is excluded, our 3D motif allows the discovery of the 4 best quality models within a shortlist of 15.
Usage of the cluster score would require listing 53 models to achieve the same outcome. The different behaviour
displayed by 2KNY could be explained by the fact that this model is not a true complex since the fragment of
Apo(E) has been fused with a linker to CR17 to ensure interactions between both domains .
This experiment validates the usage of the LDLR 3D motif as a good indicator of model quality.
Literature study of LDLR-HNP1 complex
Since HNP1 has a hydrophobic and cationic face that resembles the binding patch of ligands which interact with
LDLR [5, 35, 36] (Figure 5.A), its mode of interaction may be similar to those previously studied. In addition,
this area belongs to a pocket detected by both Fpocket  and CastP software  (Figure 5.B).
Regarding the hydrophobic aspect, Ala-scanning mutational study of HNP1 revealed tryptophan26 (W26) is a
key residue in direct interaction with target proteins and enables the peptide to form dimmers . In addition,
either W26 or phenylalanine28 (F28) mutation decreases HNP1 antibacterial activity. The importance of W26 is
further highlighted by the fact it is either conserved or replaced by an amino acid displaying an aromatic side
chain in other human α-defensins.
Figure 5. (A) HNP1 sequence and 3D structure of the HNP1 dimer. The secondary structure of HNP1 is shown
above the sequence. W26 and F28 are highlighted using arrows in the sequence and orange sticks in the 3D
structure. R24 is also marked in red. (B) The pocket detected for HNP1 dimer using CastP software . Images
produced using Pymol.
As for the cationic face, HNP1 sequence comprises four basic residues, i.e. arginines, which could play a role
similar to the lysines present in the studied LDLR-ligand complexes. Among these basic residues, arginine24
(R24) has been reported as an important residue for interacting with bacterial lipids .
Although beta sheets are dominant in HNP1 and LDLR structures, the study of known LDLR-ligand complexes
does not support the involvement of beta sheets in their interactions. Actually, this study suggests formation of a
salt bridge between HNP1’s R24 and LDLR acidic residues and that either W26 or F28 plays the role of ψ in the
minimal motif (Figure 3).
Docking prediction of LDLR-HNP1 complex
Cluspro produced a total of 43 predicted models using both the electrostatic and VWD/elec categories. Those
models were ranked using our 3D motif and, as suggested by our previous experiment, only the top 15 are
considered for further analysis (Table 3). Since R24 and either W26 or F28 are expected to be involved in the
interaction, only Model.002.01, Model.006.02 and Model.006.18 are in agreement with literature findings.
Pairwise structural alignment reveals high similarity between Model.006.18 and Model.002.01 (1.61 Å RMSD).
This shows that Cluspro converged towards a specific docking configuration from two different sets of
constraints. Model.002.01 is chosen as representative of this configuration. In addition, as required by the
minimal motif (Figure 3), Model.002.01 and Model.006.02 have candidates for the role of ψ since the TRP 144
of LA4 interacts with both W26:B and F28:A of HNP1 (Figure 6). Both models position their R24 N atoms at
similar locations (RMSD < 0.2Å). However, there is approximately a 90-degree angle between the positions of
the ligands which leads to a 13.13 Å RMSD between those two putative complex configurations.
Figure 6. Proposed LDLR-HNP1 interaction models for (A) model.002.01 and (B) model.006.02. Structures of
HNP1 and A4 are shown in cyan and red, respectively. Calcium ion is represented as a grey sphere. R24 creates
salt bridge with the aspartic residues which are shown as black dashed lines. W144 of A4 and F28 and W26 of
HNP1 provide the hydrophobic interactions. Images produced using Pymol.
Table 3. Residues involved in interaction between LDLR and HNP1 according to docking results.
Model IDs starting by 002 and 006 are produced according to electrostatic and VDW+elec constraints,
respectively. Contacts between residues are identified by SPACE .
Model ID ASP Residue(s)
ARG residue: Chain
D147,149,151 R24:B W26:B, F28:A
D149,151 R15:A W26:A,I6:A,L25:A
D147,149,151 R24:B W26:B,F28:A
D147,151 R14:B I10:B
D147,151 R14:B -
D147,149,151 R24:B I6:B
D147,151 R15:A -
D149 R15:B I20:A
D147,149 R14:A W26 :B,F28:A
D149,151 R14:B -
D149,151 R24:B I6:B
D149,151 R24:B W26:B,F28:A
D147,149 R14:B -
D149,151 R14:A -
D147,149,151 R15:A -
Complex stability analysis based on FoldX binding energy calculations (Figure 7) reveals that Model.002.01 is a
much more stable LDLR-HNP1 complex than Model.006.02. Although Cluspro energy values (-712.5 and -
143.3 Kcal/mol for Model.002.01 and Model.006.02 respectively) are not particularly accurate , they are in
agreement with FoldX conclusions. In addition, the fact that Cluspro simulations based on two different sets of
constraints led to the configuration exemplified by Model.002.01 supports the presumption of its higher
stability. It is interesting to notice that, for this model, the strength of the LDLR-HNP1 bonds weakens the bond
between the two HNP1 monomers (Figure 7).
Figure 7. Complex stability expressed by interaction energy estimated by FoldX for the structures. HNP1 dimer
and A4 are shown by rectangle and circle, respectively. (A) HNP1 dimer (PDB Code: 3GNY), (B)
Model.002.01, (C) Model.006.02. Energies are in Kcal/mol.
The major objective of this investigation was to establish whether a LDLR-HNP1 interaction can occur based
upon computational models. Previous reports of i) the versatility of ligand recognition exhibited by the LDLR
family , ii) an interaction between human α-defensins with LDLR [1,5], and iii) its role in internalising
ligands (such as cholesterol and amyloid-beta [1,3,43] led to examination of the putative interaction.
The study relating to HNP1 dimer formation, conducted using SPACE , revealed that the structure contains
33 intermolecular contacts including 3 hydrogen bonds (see Table S1 and Table S2). The dimer binding energy
calculation of -8.45 Kcal/mol (Figure 7.A) is commensurate with several models where stable interactions occur
The major observation from the modelling is that interactions between the different chains of Model002.01 are
very strong, -10.59 Kcal/mol as a whole (Figure 7.B). For model.006.02, a very different scenario is depicted
where the energy saving for interaction with the dimer is greatly diminished where the interaction with one
monomer requires 4.75 Kcal/mol (Figure 7.A). This thermodynamically unfavourable scenario points to
Model.002.01 as preferential.
The strength of binding seen in Model.002.01 is reflected in the levels of intra-molecular interactions. In
addition to the contacts present in the dimer, binding to the receptor generates a further 48 contacts including 8
as hydrogen bonds and 3 as electrostatic interactions, (see Table S3 and Table S4).
Within the complex, an intriguing feature is the modulation of dimer interaction energies depending on which
model is studied. A considerably weaker level of dimer binding strength is observed for Model.002.01 which
may have ramifications for internalisation should this step proceed through the monomer form. In contrast, for
Model.006.02, the binding interaction for the dimer remains strong.
One aim of this study involved identification of the receptor binding mechanism for the purposes of informing
the future design of synthetic HNPs to afford maximum internalisation. This report highlights the key putative
contacts between HNP1 and the LDLR, and moreover, emphasises the potential importance of maintaining the
HNP1 dimer form for binding and potentially for internalisation. Further computational studies are required to
clarify the mechanism of internalisation and interaction with membrane [42, 43].
These insights, from computation study based drug design, provide a number of avenues towards novel
synthetic antimicrobial peptides which can be synthesised and tested through conventional assays. Strengthening
or weakening LDLR-HNP interactions may have synergistic or dysergistic effects on the two key aspects,
namely docking and internalisation. In this vein, strengthening the links that make the HNP1 dimer, even to the
extent of forming fixed permanent bond to anchor the dimer link, may be an avenue to greater efficacy in some
forms of antimicrobial activity.
LIST OF ABBREVIATIONS
LDLR: Low Density Lipoprotein Receptor
HNP1: Human Neutrophil Peptide 1
HIV: Human Immunodeficiency Virus
3D motif and LDLR-HNP1 models are available upon request from the authors.
Table S1: Contacts between HNP1’s monomers.
Table S2: Statistics of contacts in HNP’s monomers.
Table S3: Contacts between A4 and HNP1’s dimer in Model.006.02 and Model 002.01.
15 Download full-text
Table S4: Statistics of contacts in Model.006.02 and Model 002.01.
This work was in part supported by grant 6435/B/T02/2011/40 of the Polish National Centre for Science.