The novel protein KBP regulates mitochondria localization by interaction with a kinesin-like protein
Members of the Kinesin-3 family of kinesin-like proteins mediate transport of axonal vesicles (KIF1A, KIF1Bβ), distribution of mitochondria (KIF1Bα) and anterograde Golgi to ER vesicle transport (KIF1C). Until now, little is known about the regulation of kinesin-like proteins. Several proteins interact with members of this protein family. Here we report on a novel, K IF1 b inding p rotein (KBP) that was identified in yeast two-hybrid screens.
KBP was identified by using the yeast-two-hybrid system with an amino-terminal fragment of KIF1C as a bait that is strongly homologous to KIF1B. Here we investigated the interaction of KBP and KIF1B. The full length proteins coimmunoprecipitated after overexpression and in untransfected 293 cells. Immunofluorescence experiments revealed that KBP was mainly localized to mitochondria, as has been described for KIF1Bα. Overexpression of a deletion mutant or reduction of the KBP protein level using an anti-sense construct led to an aggregation of mitochondria. Such an effect is probably due to the lower activity of KIF1Bα in the absence of KBP, as was revealed in motility assays.
KBP is a new binding partner for KIF1Bα that is a regulator of its transport function and thus represents a new type of kinesin interacting protein.
KeywordsNIH3T3 Cell Motor Domain Motility Assay Untransfected Cell Perinuclear Localization
List of abbreviations
vesicular stomatitis virus
Intracellular transport in cells is mediated by three different types of motor proteins that travel along filamentous tracks. Actin filaments are used by the myosin family of proteins, while transport along microtubules (MT) is mediated by either kinesin-like proteins (KLPs) or dyneins. The functional difference between these two kinds of molecular motors lies in the polarity of movement. Dyneins transport their cargoes toward the minus end of MT, while most KLPs transport cargo towards the minus and plus end of MT, depending on the position of the motor domain at the amino- or carboxyl-terminus of the protein (for reviews see [1, 2]). The common feature of all KLPs is a high degree of homology in their motor domain, a 340 amino acids region that contains MT- and ATP-binding sites and classifies them into one of 14 families of the kinesin superfamily . Human and mouse genomes encode 45 KLPs  whose functions classically are transport of cargo, like protein rafts, lysosomes, chromosomes or various membrane vesicles. However, KLPs also can zipper, cross-link and influence the stability of MT for building and maintaining the mitotic and meiotic spindle apparatus (for reviews see [5, 6, 7].
Our previous research identified the protein KIF1C that localizes to the Golgi apparatus. Using a dominant-negative mutant we have shown that KIF1C is involved in transport of vesicles from the Golgi to the ER . Together with KIF1A and KIF1B, KIF1C belongs to the Kinesin-3 family. The three proteins show a high degree of sequence homology outside their amino-terminal motor domain and share a so called U104 domain. The biological function of this domain, also known as forkhead homology-associated domain  has not yet been described in KLPs but it may be involved in protein-protein interactions regulated by phosphorylation. Moreover, KIF1C has been implicated in the susceptibility of mouse macrophages to anthrax lethal toxin .
The other member of the Kinesin-3 family, KIF1B, is expressed as two main splicing variants that share 660 amino-terminal amino acids and have different roles in intracellular transport. The α-isoform associates with mitochondria , while the β-isoforms [12, 13]. are involved in transport of synaptic vesicles and lysosomes in non-neuronal cells . A missense mutation in the ATP binding domain of the motor may be causal for the development of Charcot-Marie-Tooth disease type 2A .
For the Kinesin-3 KLPs several binding proteins were identified. KIF1A binds to Liprin-α , and KIF1Bα interacts with the PDZ domain of the glucose transporter 1 binding protein, indicating a possible additional target of this KLP . In addition, PDZ domains of PSD-90, PSD-97 and S-SCAM are responsible for interaction with KIF1Bα as well . We reported previously on the binding of protein tyrosine phosphatase D1  and 14-3-3 proteins  to KIF1C. Here we describe the identification of a novel KIF1Bα interacting protein, KBP (K IF1 b inding p rotein). We show that KBP colocalizes with mitochondria and that it interacts with KIF1Bα. Moreover, we present evidence that this new protein plays a role for the regulation of mitochondrial distribution by regulating the KIF1Bα activity.
Identification and cloning of KBP
We have previously identified two proteins that bind to KIF1C, PTPD1 and members of the 14-3-3 protein family [8, 19]. To search for proteins that bind to KIF1C between the motor domain and the PTPD1 binding region, we employed the yeast two-hybrid system. Two screens were made, first with amino acids 435 – 622 as a bait, to identify possible U104 binding proteins, and second with amino acids 261 – 800 as a bait. Only the second screen with the larger bait resulted in possible interacting proteins. Out of 5 × 106 screened transformants, 10 independent isolates of the same cDNA were found to interact.
Expression of KBP
To estimate the transcript size and prove that the identified clone encodes the complete open reading frame, as well as to check the tissue distribution, Northern blot analysis using the whole cDNA as a probe was performed. A band of 2.6 kb was detected in all examined tissues, with the highest expression found in heart and skeletal muscle (Fig. 1B), which is similar to the expression pattern of KIF1B, as shown by Nangaku et al.  and Chen et al. . Next, we constructed a glutathione-S-transferase-KBP (GST-KBP) fusion protein and raised a rabbit polyclonal antiserum against it. When probing lysates from various cell lines with this antibody, a protein of approximately 72 kDa was detected (Fig. 1C). This is in agreement with the predicted molecular mass, and the size of the protein in human cells corresponded with the size of transiently overexpressed protein after transfection, indicating that we indeed cloned the complete open reading frame.
Localization of KBP in NIH3T3 cells
This interaction of KBP and KIF1Bα was confirmed by co-immunoprecipitation. Transfected 293 cells expressed either a VSV-epitope tagged form of KBP or KBP-VSV plus KIF1Bα. Cells were lysed, the proteins immunoprecipitated using anti-KIF1B serum, separated by SDS-PAGE, transferred to nitrocellulose and blotted with VSV antibody. As shown in Fig. 3C, KBP co-immunoprecipitated with KIF1B when both proteins were overexpressed. Finally, we tested the interaction of KBP and KIF1B in untransfected BHK cells (Fig. 3D). After immunoprecipitation of KBP, KIF1B was detected as a co-immunoprecipitated protein on a Western blot. In conclusion, KBP interacts and colocalizes with KIF1Bα.
Deregulation of KBP leads to aggregation of mitochondria
Mitochondria aggregation and the role of KBP
In addition to KIF1B, conventional kinesin (KIF5B) also was described to mediate transport of mitochondria . To confirm a specific role of KBP for KIF1Bα and not KIF5B, we tested the interaction between these proteins. Therefore, we constructed a plasmid encoding a GST fusion protein encoding amino acids 265–639 of KIF5B, corresponding to the previously used GST-KIF1Bα. Proteins were purified from bacteria and incubated with glutathione-sepharose and lysates from 293 cells transiently overexpressing KBP. As indicated in Fig. 6B, only GST-KIF1Bα associated with KBP. Neither GST-KIF5B nor GST-Grb2 (as a control) were able to bind KBP. The GST-fusion proteins are visible since the anti-KBP antiserum was generated using a GST-KBP fusion protein and indicate that similar amounts were used for the pull-down.
The above finding suggests that KBP does not bind the KIF1Bα's carboxyl-terminus, which was shown to mediate cargo interactions [16, 22]. To identify the KBP-binding site on KIF1Bα we generated a mutant with an amino-terminal truncation of 349 amino acids (KIF1Bα350-1135-VSV) which was not able to bind KBP (Fig. 6C). Together with the data presented in Figure 6B this confines the binding of KBP to KIF1Bα between amino acids 270 and 350 at the carboxyl-terminus of the motor domain. Thus, KBP plays an important role in mitochondria distribution in cells but it is neither a mitochondrial receptor nor a general adapter protein for mitochondria transporting KLPs.
KBP improves KIF1B motility
Summary of data from motility assays. Average values for velocities and distances traveled by beads are shown.
Distance traveled along MT
KIF1B /KBP (n = 61)
KIF1B/KBPΔ250–281 (n = 67)
In the present study we report on the identification of the new K IF1 b inding p rotein KBP. The association of KBP with KIF1Bα was verified in vitro, in 293 cells transiently overexpressing both proteins and in untransfected BHK cells. Anti-sense experiments revealed that KBP is required for a proper distribution of mitochondria in the cell. Finally, we showed that KBP increases KIF1Bα motility in vitro.
The bait used in the yeast two-hybrid screen that led to the identification of KBP contained the sequence from the motor domain to the PTPD1 binding region (amino acids 261–800). Additional data from binding experiments of KBP to KIF1C and KIF1Bα deletion mutants allowed us to narrow down the binding sequence of KIF1B/C for KBP to the carboxyl-terminal region of the motor domain. A high amino acid sequence homology of 90% between KIF1C and KIF1Bα in this region supports the conclusion that KBP can bind to both proteins, and a manuscript describing the role of KBP for KIF1C is in preparation. It is important to note that the binding region of KBP is located close to the K-loop, a region in KIF1 KLPs that increases the affinity of the motor domain for MT . For KIF1B, several splice variants involving or located next to the K-loop are known , and it will be interesting to check whether the presence of these variants has an effect on interaction with KBP . An interaction with the KIF1Bβ variant would open a role for KBP as a player in neurodegenerative diseases. Recently, the role of the K-loop in a 679 amino acid construct of the rat orthologue of KIF1C was analysed . In in vitro experiments this protein was not chemically processive, the K-loop increased the affinity for MT and stabilized a weak binding mode with ADP in the active site. The hypothesis was formed that rat KIF1C could work in teams of motor proteins, which could generate processive movement of attached organelles. This would be a similar mechanism as proposed for KIF1A . In our motility assays we observed that KBP increased the motility of beads in extracts from KIF1Bα overexpressing cells. When the motor protein was overexpressed with KBP, it traveled average distances of 2–3 μm and maximally 6 μm, while with KBPΔ251–281 we observed KIF1Bα moving an average of 1 μm. These experiments do not contradict the data of Rogers et al. and Klopfenstein et al. since it is possible that KBP facilitates grouping of motor proteins and thus increases the motility. However, it also indicates an alternative mechanism where associated proteins like KBP could increase the motility of single molecules of KIF1B.
Numerous proteins have been identified that interact with KLPs [26, 27]. These proteins may determine the localization of the KLP complex or which cargo is transported. Very little, however, is known about associating proteins that regulate either the cargo binding of the KLP or its activity. Such a KLP associating protein that is required for the KLP function has been described in yeast. The MT minus-end directed motor protein Kar3p is important for meiosis I in S. cerevisiae, and a knock out leads to an arrest in prophase I . The associating protein Cik1p targets Kar3p to MT, implicating this as a reason why loss of Cik1p also impairs meiosis, although the effect is less severe . Several studies describe a regulatory role of the carboxyl-terminus in conventional kinesin processivity. Kinesin light chains were reported to inhibit the binding of the heavy chain to MT . Further, the kinesin tail domain by itself has an ATPase inhibiting role that is relieved by cargo binding [31, 32, 33]. One function of the light chains and the carboxyl-terminus of the heavy chain may thus be to ensure that kinesin is active only in the presence of its cargo. However, most of these studies have been done in vitro and cannot account for effects of regulating proteins.
The drastic consequences of the loss of KBP expression for mitochondria localization could also indicate that KBP is targeting KIF1Bα to the mitochondria. However, KIF1Bα is attached to mitochondria also after depletion of KBP (Fig. 6A). In addition, KLPs normally bind their cargo with carboxyl-terminal sequences, whereas KBP interacts with KIF1Bα amino-terminal sequences. Therefore, we do not think that KBP is important for the cargo – kinesin connection. Alternatively and mentioned before, KBP may regulate the KIF1Bα activity, i.e., in the absence of KBP, KIF1B is not or less able to move mitochondria.
Only two widely expressed KLPs have been reported to transport mitochondria, KIF1Bα and KIF5B, while KIF5A and KIF5C are expressed predominantly in neuronal tissues . KIF5B did not interact with KBP, and it should not be affected by KBP depletion. Thus, even if there is some expression of KIF5B, a major role of KIF1Bα for mitochondria transport in NIH3T3 cells is likely. Recently, the Drosophila protein MILTON was described that associates with the KIF5 ortholog kinesin heavy chain and may play a role similar to KBP since its mutation impairs MT-plus end directed mitochondrial transport in axons and synapses . In addition to KLPs, dynein motor proteins probably can transport mitochondria in cells , and this transport would occur in a MT minus-end direction. When force moving mitochondria to the MT plus-ends is reduced, a perinuclear localization of the mitochondria would result. On the other hand, a complete depolymerization of MTs by nocodazole treatment led to a diffusion of mitochondria throughout the cytoplasm , demonstrating the specificity of the consequences of KBP depletion.
We have shown earlier that protein tyrosine phosphatase D1 and members of the 14-3-3 protein family interact with KIF1C, implying its regulation by phosphorylation. In this manuscript, we have identified the new protein KBP as a KLP interacting protein that is required for KLP activity. This suggests that vesicle transport in cells may be subject to regulation at the level of KLP – microtubule interaction and implicates KBP as a physiological mediator.
Yeast two-hybrid screen
Two fragments of the KIF1C cDNA encoding amino acids 435–622 (corresponding to the U104 domain) or amino acids 261–800 were cloned into the LexA fusion protein vector pBTM116 and transformed into the Saccharomyces cerevisiae strain L40 (MATa trp1 leu2 his3 LYS2::lexA-HIS3 URA3::lexA-lacZ ), generating the L40 LexA-KIF1CU104 and L40 LexA-KIF1C298–840 strains, respectively. A brain cDNA library fused to the GAL4 activation domain in the pACT2 vector (Clontech) was transformed into the L40 LexA-KIF1CU104 and L40 LexA-KIF1C298–840 strains and 5 × 106 transformants were screened for interaction as described . Yeast plasmid DNA was isolated from His+ β-Gal+ colonies, rescued into Escherichia coli HB101, retransformed into L40 LexA-KIF1C298–840, and yeasts assayed for β-galactosidase activity and growth on medium with complete supplement lacking amino acids Trp, Leu, Ura, Lys, and His (Bio101). The specificity of the interaction between KIF1C298–840 fragment and potential candidates was proven by transforming the candidate plasmid also into a L40 LexA and a L40 LexA-laminin strain.
DNA sequence analysis
Clones interacting specifically with KIF1C in the two-hybrid screen were subjected to sequencing by the dideoxynucleotide chain termination method. Non-redundant databases (GenBank, EMBL, DDBJ and PDB) were searched for homologous sequences using the BLAST program [39, 40].
Northern blot analysis
A multiple tissue Northern blot was obtained from Clontech. Hybridization was performed at 67°C with 5 × 106 cpm/ml of 32P-random primed DNA probe as described by the manufacturer.
cDNA cloning and protein purification
For transient expression in 293 and NIH3T3 cells, all cDNAs were cloned by standard procedures into the cytomegalovirus immediate early promoter-based expression plasmid pRK5 . To clone a KIF1Bα encoding cDNA, the 3'-part of the open reading frame was received from clones FH12320 (1941–3721 bp of this clone) and HJ02790 (1–685 bp of this clone) known in HUGE Protein Database as genes KIAA1448  and KIAA0591 , respectively. The 5'-part was amplified in a PCR reaction using a human brain cDNA library as a template. A cDNA fragment encoding the amino acids 265–639 of the human version of KIF5B was amplified with specific primers in a similar way. The corresponding region of KIF1Bα (amino acids 270–641) was cut out from the KIF1Bα cDNA using restriction enzymes, and both fragments were cloned into a pGEX vector. The corresponding fusion proteins were produced in Escherichia coli BL21 (Novagen) or 298 F' strains and purified as described .
The KIF1B and KBP rabbit polyclonal antisera were raised against glutathione-S-transferase (GST) fusion proteins of KIF1B (amino acids 743–1153) and KBP (full length), respectively. Monoclonal antibodies used were derived from the following hybridoma: kinesin heavy chain – H2; phosphotyrosine – 4G10; β-tubulin – KMX1; vesicular stomatitis virus (VSV) epitope tag – p5d4; KIF5 – SUK4; hemagglutinin (HA) epitope tag – 12CA5.
Cell lines and cell culture
293 cells were grown in F12/Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum and 2 mM glutamine. BHK, HeLa, NIH3T3 and C2C12 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum and 2 mM glutamine.
Transient expression, cell lysis, and immunoprecipitation
Transfection of 293 cells and NIH3T3 cells was performed using the method of Chen and Okayama . The cells were lysed in 1 ml of lysis buffer/10-cm plate (1% Triton X-100, 50 mM HEPES, pH 7.5, 10% glycerol, 150 mM NaCl, 1.5 mM EGTA, 10 mM sodium pyrophosphate, 100 mM NaF, 1 mM sodium orthovanadate, 10 μg/ml aprotinin, 1 mM phenylmethylsulfonylfluoride) and the lysates were precleared by centrifugation at 13,000 g for 15 minutes at 4°C. The lysates were adjusted for equal protein concentration, the appropriate antibody (3 μl of serum or 2 μg of purified antibody) and either Protein A- or Protein G-Sepharose were added, and the lysates were incubated for at least 3 hours at 4°C on a turning wheel. The immunoprecipitates were washed with HNTG buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 0.1% Triton X-100, 10% glycerol, 10 mM NaF, 1 mM sodium orthovanadate), separated on an 8 or 10% SDS-polyacrylamide gel, transferred to a nitrocellulose membrane, and incubated with the appropriate primary antibody. After three washes, the membranes were incubated with secondary horseradish peroxidase-conjugated goat anti-mouse or goat anti-rabbit antibodies (Sigma). After three washes they were visualized using the ECL system. Before reprobing, blots were incubated for 30 minutes in 62.5 mM Tris-HCl, pH 6.8, 2% SDS, and 0.1% β-mercaptoethanol at 55°C.
In vitro binding assays
For in vitro binding assays, 293 cells transiently overexpressing KBP were lysed and incubated for 4 hours at 4°C with GST or GST-KIF1Bα/KIF5B fusion proteins immobilized on glutathione-Sepharose. After washing with HNTG buffer, the proteins were separated by SDS-PAGE and analysed by Western blotting.
NIH3T3 cells were grown on uncoated glass coverslips, fixed for 20 minutes in ice-cold methanol at -20°C, washed with phosphate-buffered saline and incubated for 10 minutes with 0.1% NaBH4 and 0.1 M glycine in phosphate-buffered saline to block autofluorescence. Nonspecific antibody binding was blocked for 45 minutes with PBS with 0.045% fish gelatin containing 5% normal goat serum and 1% bovine serum albumine. Incubation with primary antibody was done for 1.5 hours at 37°C after dilution in PBS/gelatin containing 5% normal goat serum. The affinity purified KBP and KIF1Bα antibodies were used at concentrations of 10 ng/μl and 20 ng/μl, respectively. The β-tubulin antibodies were diluted as recommended by the manufacturer, and the monoclonal VSVantibodies at a concentrations of 20 ng/μl. After five washes with PBS/gelatin, primary antibody binding was detected with isotype-specific secondary antibody conjugated with either DTAF (goat anti-rabbit, Dianova) or Alexa (Alexa Fluor 546 goat anti-mouse and goat anti-rabbit and Alexa Fluor 488 goat anti-mouse, Molecular Probes). The coverslips were mounted in PermaFluor (Immunotech). To allow unbound protein to diffuse out, extraction of cells with 0.5% Triton X-100 was performed prior to fixation for 5–10 seconds [46, 47]. MitoTracker (Molecular Probes) staining was done according to manufacturer's recommendations.
Pictures were taken with a Leica TCS NT/SP Laser Scanning Confocal Microscope supplied with HCX PL APO immersion oil objective of magnification 63× and aperture 1.32–0.62. The pictures were prepared using TCS software supplied by the manufacturer.
Cell extracts and motility assays
Cell extracts were prepared in BRB80 buffer (80 mM Pipes, pH 6.8, 2 mM MgCl2, 1 mM EGTA) supplemented with protease inhibitors, through multiple passing through cell cracker (HGM, Heidelberg, Germany). The nuclei were spun down and the supernatant was spun again at 120,000 g to separate cytosol from membrane fraction The concentration of the cytosolic fraction was adjusted to 1 mg/ml and supplemented with 1 mM DTT, 20 μM taxol, 2 mg/ml BSA, 5 mg/ml casein and 1/20 volume energy mix (150 mM creatine phosphate, 20 mM ATP, 20 mM MgCl2, 2 mM EGTA, pH 7.7).
For motility assays with latex beads, 1 μl of beads (2.5% solid-latex, carboxylate 0.1 micron microspheres; Polysciences, Inc., Warrington) was left untreated or incubated for at least 30 minutes with 4 μl of KIF1B antibody solution (3 mg/ml). Then the antibody/beads suspension was diluted 1:6 in BRB80, 1 mM DTT, 20 μM taxol, 2 mg/ml BSA, 5 mg/ml casein and 1/20 volume energy mix. One μl of beads was mixed with 9 μl of cell extract and floated into a flow cell with coverslips coated with microtubules. The beads were left for 5 – 7 minutes in a humid chamber to sediment onto the microtubules. To inhibit unspecific KIF5B driven movements, SUK4 antibody was incubated with extracts for 10 minutes at a final concentration of 0.25 mg/ml, and then the extracts were used in the assay. Flow cells were prepared as described earlier , and microtubules were isolated from pig brain as described .
Motility was observed using video-enhanced differential interference contrast microscopy (VE-DIC) in real time (Olympus optics BX60 microscope, equipped with DIC optics; Japan, Tokyo). To determine the rates and distances, RETRAC software (Dr. N. Carter, Marie Curie Research Institute, Oxted, Surrey, UK) was used.
We are grateful to K. Kapp and K. Wojtachnio for critically reading the manuscript, to Viki Allan for helpful suggestions for the in vitro motility assays, and to the Kazusa DNA Research Institute for providing clones FH12320 and HJ02790. This work was supported by a grant from the Deutsche Forschungsgemeinschaft to R.L.
- 3.Lawrence CJ, Dawe RK, Christie KR, Cleveland DW, Dawson SC, Endow SA, Goldstein LS, Goodson HV, Hirokawa N, Howard J, Malmberg RL, McIntosh JR, Miki H, Mitchison TJ, Okada Y, Reddy AS, Saxton WM, Schliwa M, Scholey JM, Vale RD, Walczak CE, Wordeman L: A standardized kinesin nomenclature. J Cell Biol. 2004, 167: 19-22. 10.1083/jcb.200408113.PubMedCentralCrossRefPubMedGoogle Scholar
- 15.Zhao C, Takita J, Tanaka Y, Setou M, Nakagawa T, Takeda S, Yang HW, Terada S, Nakata T, Takei Y, Saito M, Tsuji S, Hayashi Y, Hirokawa N: Charcot-Marie-Tooth disease type 2A caused by mutation in a microtubule motor KIF1Bbeta. Cell. 2001, 105: 587-597. 10.1016/S0092-8674(01)00363-4.CrossRefPubMedGoogle Scholar
- 18.Mok H, Shin H, Kim S, Lee JR, Yoon J, Kim E: Association of the kinesin superfamily motor protein KIF1Balpha with postsynaptic density-95 (PSD-95), synapse-associated protein-97, and synaptic scaffolding molecule PSD-95/discs large/zona occludens-1 proteins. J Neurosci. 2002, 22: 5253-5258.PubMedGoogle Scholar
- 33.Stock MF, Guerrero J, Cobb B, Eggers CT, Huang TG, Li X, Hackney DD: Formation of the compact confomer of kinesin requires a COOH-terminal heavy chain domain and inhibits microtubule-stimulated ATPase activity. J Biol Chem. 1999, 274: 14617-14623. 10.1074/jbc.274.21.14617.CrossRefPubMedGoogle Scholar
- 43.Nagase T, Ishikawa K, Miyajima N, Tanaka A, Kotani H, Nomura N: Prediction of the coding sequences of unidentified human genes. IX. The complete sequences of 100 new cDNA clones from brain which can code for large proteins in vitro. DNA Res. 1998, 5: 31-39. 10.1093/dnares/5.1.31.CrossRefPubMedGoogle Scholar
- 46.Illenberger S, Zheng-Fischhofer Q, Preuss U, Stamer K, Baumann K, Trinczek B, Biernat J, Godemann R, Mandelkow EM, Mandelkow E: The endogenous and cell cycle-dependent phosphorylation of tau protein in living cells: implications for Alzheimer's disease. Mol Biol Cell. 1998, 9: 1495-1512.PubMedCentralCrossRefPubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.