Leukemia inhibitory factor (LIF) is a member of the IL-6 cytokine family, having pleiotropic actions such as maintaining stem cell pluripotency and enabling blastocyst implantation. Because the action of LIF is mediated by a ligand-receptor interaction with the LIF receptor (LIF-R), an antagonist for LIF-R has been developed to inhibit LIF-induced signaling. In this study, we present a novel method for the production and purification of an antagonist to human LIF-R (hLA). His-tagged hLA was expressed in
, and simple purification methods without any endopeptidase cleavage were designed. In addition, we determined the optimal temperature conditions for enhancing the production of soluble hLA. Finally, the bioactivity of His-tagged hLA was examined using STAT3 phosphorylation and receptivity of human endometrial ECC-1 cells. Our strategy provides a rapid and efficient method to produce biologically active recombinant hLA.
Leukemia inhibitory factor (LIF), a member of the IL-6 family of cytokines, consists of 180 amino acids in the mature form and plays multiple roles in a broad range of cell types
. It has been reported that LIF induces the differentiation of the murine myeloid leukemia cell line M1 into macrophages, while inhibiting proliferation of the cells
. LIF is also known as differentiation-stimulating factor, hepatocyte-stimulating factor III, differentiation inhibitory factor, human interleukin for DA cells, melanoma-derived lipoprotein lipase inhibitor, and differentiation retarding factor
biological effects of LIF include promotion of the survival of neurons, regulation of hematopoiesis, stimulation of osteoblasts, maintenance of stem cell pluripotency, and facilitation of blastocyst implantation
. Blocking the action of LIF, and therefore all the pleiotropic activities of the molecule, could be useful for treating LIF-related diseases.
LIF acts through the receptor complex of LIF receptor (LIF-R) and glycoprotein 130 (gp130), a common receptor for IL-6 family cytokines, followed by downstream signaling molecule pathways such as STAT3, PI3K, and MEK/ERK pathways
. In light of this information, several strategies for inhibiting LIF activity have been developed, including the use of a small molecule inhibitor for downstream signaling molecules
, antibodies directed against LIF or LIF-R
, recombinant soluble LIF-R
, an antagonist for gp130
, and an antagonist for LIF-R
Among them, LIF-05, the result of nine site mutations in human LIF, showed a selective and powerful binding affinity for LIF-R, with no detectable binding to gp130
. LIF-05 and PEGLA, the polyethylene glycol-conjugated form, demonstrate inhibitory activity on cartilage degradation
and embryo implantation
. The contraceptive potential of LIF-05 and PEGLA were confirmed
. Furthermore, PEGLA administered vaginally had no effect on other tissues such as bone and on the central nervous system
. Thus, LIF-05 is a promising contraceptive agent with strong activity and low side effects.
Previously, LIF-05 was expressed as a glutathione Stransferase (GST) fusion protein
. The expression of GST-fused LIF-05 has the advantage of solubility, although two steps of purification, GST-affinity chromatography, followed by removal of the carrier or uncleaved fusion protein are required. In this study, we designed a Histagged LIF-05 expressed in
, which could be purified by a simple method without the need for any endopeptidase cleavage. In addition, we defined the optimum temperature conditions for enhancing the production of soluble LIF-05. Finally, the bioactivity of His-tagged LIF-05 was investigated using STAT3 phosphorylation and the receptivity of human endometrial ECC-1 cells.
Materials and Methods
- Construction of the Expression Vector
The human LIF cDNA image clone was purchased from Open Biosystems. A DNA sequence encoding the mature 180-aminoacid polypeptide
was amplified by polymerase chain reaction (PCR) using SolGent Pfu DNA Polymerase (SolGent Co., Daejeon, Korea). The amplified mature form of human LIF cDNA was subcloned into the pET-53 vector in order to express a Histagged mature human LIF protein in bacterial cells. The sequence was verified by a sequencing service of Macrogen Co. (Seoul, Korea), and the resulting construct was named pET-hLIF. In order to construct a His-tagged hLIF antagonist (hLA)-expressing vector, nine site mutations were introduced into hLIF by using a QuikChange II XL site-directed mutagenesis kit (Stratagene, La Jolla, CA, USA), according to a previous study
. The mutations were verified by sequence analysis, and the resulting construct was named pET-hLA. Specific oligonucleotides used for cloning pET-hLIF and pET-hLA are listed in
Information of primers used for cloning of human LIF, and site-directed mutation.
Information of primers used for cloning of human LIF, and site-directed mutation.
- Prediction of the Secondary and Three-Dimensional Structures
A structural model of the hLIF signal peptide (amino acids (aa) 1 –22) was constructed using the Modeller, which is a relative three-dimensional (3D) protein modeling system. The 3D models of the hLIF signal peptide (PDB IDs: 2Q7N, 1 PVH, 1 LKI, and 1A7M) were used. Sequence alignment of hLIF for the prediction of the secondary structure was visualized using Clustal W.
- Expression of Soluble hLA
The constructed pET-hLA vector was transformed into E. coli BL21(DE3)pLysS and selected with antibiotics (100 µg/ml of ampicillin). For soluble hLA protein expression, the selected clones were grown overnight at 37℃ in 3 ml of LB broth containing ampicillin. The overnight cultures were then diluted 1:100 in 10 ml of fresh LB broth containing ampicillin and grown at 37℃ until the OD
reached approximately 0.5. Recombinant hLA protein expression was induced by adding isopropyl-β- D -1-thiogalactopyranoside (IPTG; Sigma-Aldrich, MO, USA) to a final concentration of 0.5 mM. The cells were incubated at three different conditions, i.e. 37℃ for 3 h, 20℃ overnight, or 4℃ overnight with shaking at 100 rpm. Cells were harvested by centrifugation at 5,000 rpm for 30 min. The pellets were resuspended in 1ml of phosphate-buffered saline (PBS) buffer and lysed by sonication on ice. The lysates were fractioned into soluble and insoluble parts by centrifugation at 15,000 rpm for 30 min at 4℃. To examine the expression level of soluble hLA, the total cell lysates and fractions were analyzed on a 10% SDS-polyacrylamide gel and detected with Coomassie brilliant blue R350 (GE Healthcare, Uppsala, Sweden).
- Purification of Recombinant hLA
For the production of large amounts of recombinant hLA, 10 ml of overnight culture was added to 1 L of fresh LB broth containing ampicillin at 37℃ and incubated until the OD
was 0.5–0.7. Recombinant protein expression was induced by adding 0.5 mM IPTG, and then the cells were incubated at 4℃ overnight with shaking at 100 rpm. The cells were collected by centrifugation at 5,000 rpm for 30 min, and the pellets were resuspended in 6 ml of 20 mM imidazole in PBS buffer and lysed by sonication on ice. The cell lysates were resolved by centrifugation and loaded onto a Ni-NTA column (Thermo Scientific) in 20 mM imidazole (SigmaAldrich). After three washes, the protein was eluted with 250 mM imidazole. Proteins were desalted on a PD-10 column (GE Healthcare) and eluted with PBS. The purification efficiency was examined by Coomassie staining and western blotting.
- Cell Culture
The human endometrial ECC-1 cell line was supplied from American Type Culture Collection (Manassas, VA, USA). The cells were cultured as monolayers at 37℃ in an atmosphere containing 5% CO
/air in Dulbecco’s modified Eagle’s medium (DMEM; Welgene, Daegu, Korea) containing 10% heat-inactivated fetal bovine serum (Sigma-Aldrich) and 1% penicillin/streptomycin (Life Technologies). The human trophoblastic JAR cells were purchased from the Korean Cell Line Bank (Seoul, Korea) and maintained in RPMI-1640 (Welgene) containing 10% heatinactivated fetal bovine serum and 1% penicillin/streptomycin.
- Western Blot Analysis
ECC-1 cells were treated with 50 ng/ml of human recombinant LIF (R&D systems, Minneapolis, MN, USA) and/or 50 ng/ml of hLA. Total proteins were extracted and equal amounts of total proteins (20 µg) were separated by SDS-PAGE and electrotransferred to nitrocellulose membranes. The membranes were blocked with 5% blocking solution and were incubated with antibodies against phospho-STAT3, STAT3 (Cell Signaling, Danvers, MA, USA), LIF, phospho-p65, and p65 (Santa Cruz Biothechnology Inc., Santa Cruz, CA, USA). Next, the membranes were incubated with the appropriate horseradish peroxide-linked secondary antibody. The signal was visualized using the ECL chemiluminescence system (GE Healthcare).
- Adhesion Assay
ECC-1 cells were seeded onto 6-well plates and cultured to reach 100% confluence, in order to avoid nonspecific binding to the culture plate. Media were replaced, and the cells were incubated in serum-free DMEM containing recombinant human LIF and/or hLA (50 ng/mL, respectively) for 48 h. The JAR cells were labeled with 5-chloromethylfluorescein diacetate (CMFDA) fluorescence dye (CellTracker Green; Life Technologies) for 10 min at 4℃. Next, the labeled JAR cells (1 × 10
cells) were washed with 1× PBS, gently added onto the ECC-1 cell monolayer, and incubated in JAR growth medium. After gently shaking at 40 rpm for 30 min at 37℃, the cells were vigorously washed to exclude non-binding JAR cells. The attached JAR cells were visualized and four different microscopic focal fields were pictured by using a fluorescent microscope (Axio Imager 2; Carl Zeiss, Oberkochen, Germany). The number of attached cell numbers in the adhesion assays was quantified with ImageJ software (NIH, Bethesda, MA, USA). The experiments were independently performed three times.
- Statistical Analysis
Data from the adhesion assays are presented as experimental number of cells attached divided by the control number and are expressed as the mean ± SD. Differences between the mean values of two groups were determined by Student t-test, using GraphPad Prism (GraphPad Software, San Diego, CA, USA). The minimal level of significance was set at a
value of 0.05.
- Prediction of the Secondary and Three-Dimensional Structures of hLIF
The domain structure of full-length human LIF (aa 1–202) is shown in
A. Human LIF consists of a signal peptide (aa 1–22) and mature leukemia inhibitory factor (aa 23–202), harboring three disulfide bonds (Cys12-Cys134, Cys18-Cys131, and Cys60-Cys163). Although a great deal of progress has been made in research regarding hLIF, the 3D structure of the human LIF signal peptide domain has not been solved to date. In this study, we determined a molecular model of the signal peptide by homology modeling based on the four structures of the PDB IDs: 2Q7N, 1PVH, 1LKI, and 1A7M. The secondary structure of human LIF (1–22) was predicted, and the signal peptide is shown by a loop region. Interestingly, we found that the secondary structure of full-length human LIF consisted of only α-helices and loop regions without β-sheets (
B). The human LIF was composed of four long α-helices and two short α-helices, which were connected
loop regions (
C) with a hydrophobic core in the four long α-helices. The residues Ala, Val, Phe, Ile, and Leu in the hydrophobic core are structurally important and are situated in a completely hydrophobic environment in the interior of the protein. The α-helices form an intermolecular hydrophobic core together with the opposite α-helices. This hydrophobic core is surprisingly densely packed with side chains in the interior of the protein. Hydrophobic effects have been shown to be a result of hydrophobic residues in the interior of a protein and to contribute to its conformational stability. Thus, the flexible side chains could be interaction motifs with the human LIF receptor. The residues Q25, S28, Q32, S36, A112, D120, I121, G124, and S127 are located in helices α1 and α5, which are a part of the receptor-binding site. As shown in
, many of the packing interactions involve charged residues and putative binding residues in the helices α2 and α3 and the long loop, forming polar interactions with the side chains of amino acids or water molecules. Helix α4 is closed at the window for substrate binding by the α2 and α3 helices. In addition, helix α6 is largely closed at the window for substrate binding by the long loop region of the N-terminus. The two helices α2 and α3 and the N-terminal long loop disrupt the binding to the LIF receptor. The hydrophilic residues on the ends of the N-terminus or C-terminus are dense (
D). Six residues in these regions had positive and negative charges in a globular fold. These charges may also promote the formation of the target partner complex, LIF receptor. Based on these results, we can estimate the structural and functional roles of human LIF.
Structure and modeling of hLIF. (A) Schematic diagram showing the domain of full-length hLIF (amino acids (aa) 1–202). (B) Secondary structure and sequence alignment of hLIF. The unsolved structure of the hLIF signal peptide (1–22) is shown in a green box. α-Helices are represented as cyan ellipses and loops are represented as light magenta lines. The nine mutated residues are indicated by yellow boxes. (C) The structure of the full-length hLIF (1–202) is shown as a ribbon representation. The interaction sites between hLIF and hLIF receptor are shown in yellow in the α-helices. The modeled structure of the hLIF signal peptide (1–22) is outlined by a green circle. (D) The surface representation of hLIF (1–202) is shown. The relative distribution of the surface charge is shown, with the acidic regions in red, basic regions in blue, and neutral regions in white. Predictions of the binding sites between hLIF and receptor are shown in black. The modeled structure of the hLIF signal peptide is outlined by a green circle.
- Expression and Purification of Soluble hLA Protein
The pET-hLA vector was transformed into BL21(DE3)pLysS to express recombinant hLA proteins that include an Nterminal 6× His-tag. In order to acquire soluble hLA proteins, transformants were cultured at various temperatures after IPTG induction. The results showed that soluble recombinant hLA proteins were rarely expressed at 20℃ and 37℃. However, the
cultured at 4℃ after IPTG induction successfully produced soluble recombinant hLA proteins (
A, indicated by asterisk).
Expression and purification of recombinant hLA protein. (A) pET-hLA-transformed E. coli BL21(DE3)pLysS cells were cultured at various temperatures. Expression of the recombinant protein was induced by addition of IPTG (0.5 mM) when the OD600 of the culture reached 0.5, and the cells were incubated at the indicated temperature (4℃ and 20℃) for 15 h and at 37℃ for 3 h in LB medium. M: protein size marker; T, total cell protein; S, soluble protein; IS, insoluble protein; the asterisk indicates the expressed soluble hLA protein. (B) The cells were lysed by sonication, and soluble fractions were applied to a Ni-NTA affinity column. The fractions 2, 3, and 4 were collected. (C) The collected fractions were desalted on a PD-10 column.
As shown in
, the density of
grown at 4℃ was much lower than that of cells grown at 20℃ and 37℃; however, the expression of the recombinant hLA protein was about 4.2% of the total protein in the soluble fraction of cell lysates (
). The recombinant hLA proteins were obtained in the imidazole fraction and desalted over a PD-10 column (
C). The final amount and yield of recombinant hLA were 0.29 mg and 11.8%, respectively, from 1 L of culture. The identities of the purified hLA proteins were confirmed by western blot analysis with anti-LIF antibody (data not shown).
Final cell densities ofE. coligrown at indicated temperature.
Final cell densities of E. coli grown at indicated temperature.
Purification of recombinant hLA protein expressed in BL21(DE3)pLysS ofE. coli.
aResults are derived from 2.1 g wet cell weight equally with 1 L culture volume. Protein concentrations were determined by BCA protein assay using BSA as a standard. bThe relative amount of proteins was estimated by densitometry of SDS-PAGE. cThe yield was calculated by the amount of target protein at that step divided by the amount of target protein, hLA, in the first step (defined as 100%).
- Biological Activity of Purified hLA Protein
The biological activity of the purified recombinant hLA protein was examined by assessing the phosphorylation of STAT3 and receptivity toward JAR cells by using human endometrial ECC-1 cells. Recombinant human LIF was treated to stimulate the phosphorylation of STAT3, a major signaling molecule induced by LIF-R activation
. As shown in
A, hLA successfully inhibited the STAT3 phosphorylation induced by LIF treatment. In addition, recombinant hLA did not activate phosphorylation of the p65 subunit of NF-κB, a major signaling pathway stimulated by LPS (
B), indicating that the level of bacterial endotoxin in the purified hLA is below the limit required to activate an LPS-elicited inflammatory signal.
Effect of hLA on STAT3 phosphorylation in LIFstimulated ECC-1 cells and p65 phosphorylation. (A) ECC-1 cells were pretreated with or without 50 ng/ml of hLA for 1 h and then treated with or without 50 ng/ml of LIF. Total proteins were extracted and subjected to western blot analysis using antiphospho STAT3 antibody (upper panel) and anti-STAT3 antibody (lower panel). (B) ECC-1 cells were treated with 50 ng/ml of hLA for 1 h. Total proteins were extracted and subjected to western blot analysis using anti-phospho p65 antibody (upper panel) and anti-p65 antibody (lower panel).
Next, in order to stimulate the expression of adhesive molecules involved in embryo implantation, we cultured ECC-1 cells with 50 ng/ml of recombinant human LIF and/or hLA for 48 h. The CMFDA-labeled JAR cells (1 × 10
cells) were added to ECC-1 cells in a monolayer and incubated with gentle shaking, which was followed by removal of the unbound cells. The results showed that hLA significantly inhibited LIF-induced adhesion between ECC-1 and JAR cells (
). Taken together, the results of these two assays suggest that purified hLA can inhibit LIF/LIF-R signaling and equivalently counteract commercially available recombinant human LIF.
Effect of hLA on LIF-enhanced adhesion of JAR cells to ECC-1 cells. (A) ECC-1 cells were treated with 50 ng/ml of LIF or hLA for 48 h, and fluorescently labeled JAR cells were added for attachment. The unbound cells were washed off, and the attached cells were visualized by fluorescence microscopy. (B) The attached JAR cells were counted, and the ratio of attached cells to the experimental or to untreated cells was presented as the mean ± SD. *** denotes p < 0.001 in a comparison between two groups.
In this study, we describe a novel method for the production of soluble recombinant hLA in
. We cloned cDNA encoding human LIF into a pET-53 bacterial expression vector, and nine sites in the mature human LIF gene were mutated by site-directed mutagenesis to produce an antagonist for human LIF-R, according to a previous report
and our simulation study on the structural and functional roles of hLIF (
). Various affinity-tagging methods have been used to increase the yield and solubility of recombinant proteins. To enhance the solubility of recombinant protein in
, most researchers use proteintags such as glutathione S-transferase, maltose-binding protein, and N-utilization substance A
. Previously, the expressed recombinant hLA in
also used the GSTfusion system and required enzymatic cleavage to remove potent interfering activity
. His-tagged protein expression has also been widely used to produce recombinant proteins. Although the method has advantages with respect to cost and convenience, the solubility of the recombinant protein can remain a problem
. Recently, the bioactive human LIF protein was successfully expressed in a His-tagged form
. Thus, in this study, we designed an expression vector where hLA is expressed as a fusion protein with an N-terminal 6× His-tag.
Initially, the pET-hLA-transformed
cells were cultured at 37℃ for 15 h, after which expression of the recombinant protein was induced with IPTG. However, our results showed that recombinant hLA was barely detected in the soluble fraction (
, lane 10). The most common method used for improving the solubility of recombinant proteins in
is to decrease the temperature at which the target protein is expressed
. Therefore, the expression of recombinant hLA was investigated at 4℃ and 20℃. The results showed that the expression of soluble recombinant hLA protein was enhanced in
cultured at 4℃, after IPTG induction (
A, lane 4).
As predicted by DISULFIND, the recombinant hLA has three disulfide bonds at the same sites as those in wild-type human LIF
. The disulfide bonds play an important role in protein folding, stability, solubility, and bioactivity. Thus, the insoluble proteins expressed at 20℃ and 37℃ may be expressed as insoluble inclusion bodies owing to the inability to form disulfide bonds in the reducing environment of the cytoplasm in
. Previously, methods were developed to enhance the proper formation of multidisulfide bonded proteins to obtain a high yield
; for example, wild-type human LIF used thioredoxin
and protein disulfide isomerase
. In this study, with the modification of the reducing environment of the cytoplasm and the adjustment of various expression conditions, the target molecule was overexpressed as soluble protein in
The recombinant hLA expressed in the soluble fraction was purified on a Ni-NTA affinity column and further desalted over a PD-10 column (
C). Although there was only a low-to-medium expression level of recombinant hLA (
), the current protocol for purification is a simple method, not requiring enzymatic cleavage or a refolding process. All tags, whether on proteins or peptides, potentially interfere with the biological activity of recombinant proteins
. To exclude this possibility, we confirmed that the biological activity of recombinant hLA remained intact. Purified hLA successfully inhibited LIF-induced phosphorylation of STAT3 (
A). In addition, the purified and desalted recombinant hLA did not increase phosphorylation of the p65 subunit of NFκB, a major signaling molecule activated by bacterial endotoxins, including LPS (
B). These results suggest that the purified His-tagged hLA is correctly folded and can be used for antagonizing LIF-R without concern of endotoxin-induced inflammation. Next, we investigated the contraceptive action of hLA by using
models of embryo implantation
. The attachment of human trophoblastic JAR cells to endometrial ECC-1 cells was induced by treatment with commercially purchased human LIF. However, co-treatment of purified hLA with LIF significantly abrogated the LIF-elicited adhesion between JAR cells and ECC-1 cells (
In conclusion, our strategy provides an efficient and rapid method to produce a biologically active recombinant human LIF-R antagonist. Given the pleiotropic actions of LIF, this His-tag g ed recombinant hLA may be useful for intervention in LIF-related physiological or pathological conditions.
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIP) (No. 2014R1A5A2009936).
The LIF receptor antagonist PEGLA is effectively delivered to the uterine endometrium and blocks LIF activity in cynomolgus monkeys.
DOI : 10.1016/j.contraception.2012.09.032
DISULFIND: a disulfide bonding state and cysteine connectivity prediction server.
Nucleic Acids Res.
DOI : 10.1093/nar/gkl266
Leukemia inhibitory factor can substitute for nidatory estrogen and is essential to inducing a receptive uterus for implantation but is not essential for subsequent embryogenesis.
Affinity maturation of leukemia inhibitory factor and conversion to potent antagonists of signaling.
J. Biol. Chem.
DOI : 10.1074/jbc.M310103200
The effect of leukaemia inhibitory factor (LIF) on embryogenesis.
Reprod. Fertil. Dev.
DOI : 10.1071/RD9920449
The pleiotropic actions of leukemia inhibitory factor.
Models for study of human embryo implantation: choice of cell lines?
DOI : 10.1095/biolreprod.109.077800
Characterization of the receptor binding sites of human leukemia inhibitory factor and creation of antagonists.
J. Biol. Chem.
DOI : 10.1074/jbc.271.20.11971
Highlevel expression and efficient one-step chromatographic purification of a soluble human leukemia inhibitory factor (LIF) inEscherichia coli.
Protein Expr. Purif.
DOI : 10.1016/j.pep.2013.04.006
Birth control vaccine targeting leukemia inhibitory factor.
Mol. Reprod. Dev.
DOI : 10.1002/mrd.22002
n-Butylidenephthalide (BP) maintains stem cell pluripotency by activating Jak2/Stat3 pathway and increases the efficiency of iPS cells generation.
DOI : 10.1371/journal.pone.0044024
Vaginally administered PEGylated LIF antagonist blocked embryo implantation and eliminated non-target effects on bone in mice.
DOI : 10.1371/journal.pone.0019665
Novel inhibitors for murine and human leukemia inhibitory factor based on fused soluble receptors.
J. Biol. Chem.
DOI : 10.1074/jbc.M706610200
Determining the LIF-sensitive period for implantation using a LIF-receptor antagonist.
DOI : 10.1530/REP-09-0113
Cloning, expression, and functional characterization of in-house prepared human leukemia inhibitory factor.
A new type of cytokine receptor antagonist directly targeting gp130.
J. Biol. Chem.
DOI : 10.1074/jbc.273.42.27213
The crystal structure and biological function of leukemia inhibitory factor: implications for receptor binding.
DOI : 10.1016/0092-8674(94)90449-9
Monoclonal anti-leukemia inhibitory factor antibody inhibits blastocyst implantation in the rhesus monkey.
DOI : 10.1016/j.contraception.2006.05.070
Thi Vu TT
Soluble expression of human leukemia inhibitory factor with protein disulfide isomerase inEscherichia coliand its simple purification.
DOI : 10.1371/journal.pone.0083781
LIF-STAT signaling and trophoblast biology.
Leukemia inhibitory factor: part of a large ingathering family.
Int. Rev. Immunol.
DOI : 10.3109/08830189809043003
Overview of tag protein fusions: from molecular and biochemical fundamentals to commercial systems.
Appl. Microbiol. Biotechnol.
DOI : 10.1007/s00253-002-1158-6
Role of a LIF antagonist in LIF and OSM induced MMP-1, MMP-3, and TIMP-1 expression by primary articular chondrocytes.
DOI : 10.1016/j.cyto.2009.03.001
Post Uiterweer ED
In-vitromodel systems for the study of human embryo-endometrium interactions.
Reprod. Biomed. Online
DOI : 10.1016/j.rbmo.2013.08.002
Blocking LIF action in the uterus by using a PEGylated antagonist prevents implantation: a nonhormonal contraceptive strategy.
Proc. Natl. Acad. Sci. USA
DOI : 10.1073/pnas.0710110104