Magnetic capture from blood rescues molecular motor function in diagnostic nanodevices
© Kumar et al.; licensee BioMed Central Ltd. 2013
Received: 16 January 2013
Accepted: 29 April 2013
Published: 3 May 2013
Skip to main content
© Kumar et al.; licensee BioMed Central Ltd. 2013
Received: 16 January 2013
Accepted: 29 April 2013
Published: 3 May 2013
Introduction of effective point-of-care devices for use in medical diagnostics is part of strategies to combat accelerating health-care costs. Molecular motor driven nanodevices have unique potentials in this regard due to unprecedented level of miniaturization and independence of external pumps. However motor function has been found to be inhibited by body fluids.
We report here that a unique procedure, combining separation steps that rely on antibody-antigen interactions, magnetic forces applied to magnetic nanoparticles (MPs) and the specificity of the actomyosin bond, can circumvent the deleterious effects of body fluids (e.g. blood serum). The procedure encompasses the following steps: (i) capture of analyte molecules from serum by MP-antibody conjugates, (ii) pelleting of MP-antibody-analyte complexes, using a magnetic field, followed by exchange of serum for optimized biological buffer, (iii) mixing of MP-antibody-analyte complexes with actin filaments conjugated with same polyclonal antibodies as the magnetic nanoparticles. This causes complex formation: MP-antibody-analyte-antibody-actin, and magnetic separation is used to enrich the complexes. Finally (iv) the complexes are introduced into a nanodevice for specific binding via actin filaments to surface adsorbed molecular motors (heavy meromyosin). The number of actin filaments bound to the motors in the latter step was significantly increased above the control value if protein analyte (50–60 nM) was present in serum (in step i) suggesting appreciable formation and enrichment of the MP-antibody-analyte-antibody-actin complexes. Furthermore, addition of ATP demonstrated maintained heavy meromyosin driven propulsion of actin filaments showing that the serum induced inhibition was alleviated. Detailed analysis of the procedure i-iv, using fluorescence microscopy and spectroscopy identified main targets for future optimization.
The results demonstrate a promising approach for capturing analytes from serum for subsequent motor driven separation/detection. Indeed, the observed increase in actin filament number, in itself, signals the presence of analyte at clinically relevant nM concentration without the need for further motor driven concentration. Our analysis suggests that exchange of polyclonal for monoclonal antibodies would be a critical improvement, opening for a first clinically useful molecular motor driven lab-on-a-chip device.
In the recent decades, healthcare costs have soared throughout the industrialized world and this development is predicted to continue, e.g. with the costs reaching 30% of the gross domestic product in the US 2035 (compared to 15% in 2007 and 5% in 1960 ) Together with environmental and climate issues this is one of the biggest challenges facing industrialized nations. The efforts to develop new biosensing devices that are cheaper, faster, and more accurate should be viewed in this context. Such devices would allow detection of diseases and environmental changes at an early stage with increasing chances for interventions at a low cost.
Of interest in this connection are lab-on-a-chip devices [2–4] where miniaturized chips, can perform a series of analyses and be used at the point of care or in the field rather than in a centralized laboratory. However, while appreciable progress has been made towards such devices  they often require expensive and bulky accessory equipment [3, 4]. For instance, pumps that drive microfluidics flow demand increasingly more power the greater the miniaturization  and the manufacturing of the chip components becomes increasingly challenging and expensive. To overcome these problems it has been proposed that biological molecular motors, with their inherent extensive miniaturization, biodegradability and self-propelling features, may be used to transport analytes e.g. from recognition to detection chambers, achieving separation, concentration as well as certain forms of detection [6–8]. Several important steps towards a functional molecular motor driven diagnostic device have also been realized (reviewed in [8–13]) such as: (i) attachment of antibodies to cytoskeletal microtubule  and actin filament  shuttles, followed by molecular motor-driven transportation of analytes (viruses, protein antigens etc.) bound to the antibodies, (ii) nano/microfabrication of devices for guided transportation of the motor propelled shuttles to concentrate analytes at a detector site [6, 7, 16–18] and (iii) long-term storage of ready-to-use devices without loss of activity [19–21].
Despite the above developments, challenges remain before a commercially viable molecular motor driven device is realized. Particularly, we showed recently  that complex fluid environments, such as blood plasma, blood serum and cell lysates may have deleterious effects on molecular motor driven propulsion of both actin filaments and microtubules, unless the samples are diluted > 100 times. These problems are essential to overcome since extensive sample dilution is highly undesirable in high-sensitivity detection. Separation of targeted molecules from biological fluids, using magnetic microparticles with antibodies immobilized on their surface, may be useful in this connection. Such separation is a technique of growing importance in biosensing [23, 24] and more recently, also magnetic nanoparticles have been used for similar purposes .
Here we utilize magnetic nanoparticles to investigate whether an approach with a magnetic pre-separation step is a way forward in alleviating the deleterious effects of complex fluid environments on motor driven diagnostics devices. Our results are promising, showing effective exchange of the deleterious fluid components for optimized biological buffers thus enabling subsequent motor driven transportation. We also found evidence that the presence of antigen (analyte) cross-links magnetic particles and actin filaments to an appreciable degree if both particles and filaments are conjugated with polyclonal antibodies against the analyte in question. Whereas the method is potentially useful already at the present stage of development, our microscopy and spectroscopy based analysis suggests that there is appreciable room for further optimization as discussed in some detail.
Effect of antigen (Ag) on number of actin filaments, fluorescence intensity and actomyosin motility quality in final stage of the magnetic separation procedure in Figure 3
Velocity (μm s-1)
# of filaments
Buffer B with Ag (n = 3)
6.44 ± 0.48
0.51 ± 0.03
628 ± 38
0.89 ± 0.11
Serum (100%) with Ag (n = 3)
5.86 ± 0.44
0.38 ± 0.05
523 ± 164
0.81 ± 0.11
Serum(100%) without Ag (n = 3)
6.26 ± 0.71
0.27 ± 0.08
27 ± 11*1
0.22 ± 0.07*2
The rather low fraction of motile filaments in the experiments in Figure 1, 4 and Table 1 is due to (i) omission of the incubation step with blocking actin (non-fluorescent actin at 1 μM concentration) often used to block ATP-insensitive myosin heads, and (ii) omission of a step with actin affinity separation by ultracentrifugation (to remove the ATP-insensitive heads) prior to the motility assay. The first of these omissions was important in order to leave as many myosin heads as possible free to bind actin filaments.
There was a tendency for a particularly low fraction of motile filaments in step IV (Figure 3) in absence of antigen in the sample solution (lower row, Table 1). The basis for this finding can be sought in the very low number of short actin filaments (Figure 4b) under these conditions. Whereas, the low number is due to lack of cross-linking of actin filaments to MPs in the absence of antigen, we attribute the fragmentation of the filaments to shearing forces e.g. related to repeated mixing and pipetting. The short filaments detach from the surface with higher probability than long filaments but only if they are motile and not if they are bound to rigor-like ATP-insensitive heads. This leads to an appreciable effect on the fraction of motile filaments particularly when the total number of filaments is low as was the case in the initial absence of antigen.
For increased specific cross-linking efficiency between MPs and actin filaments we attempted to use magnetic microparticles (average size 1 μm, according to manufacturer) instead of nanoparticles. However, due to the high binding capacity of the larger particles there was an appreciable tendency for formation of large aggregates composed of both filaments and microparticles (Additional file 1: Figure S2). These large aggregates may be potentially useful for magnetic concentration of actin-MP aggregates but are clearly not useful in separation schemes where actomyosin driven transport is required, e.g. for motor driven concentration [7, 29], subsequent to the magnetic pre-separation. In this case, nanoparticles must be used. Their considerably smaller surface area (although not smaller surface/volume ratio) make them less sticky consistent with the findings that actomyosin driven motility is less affected by large than small nanoscale cargoes . Furthermore, the small size of nanoparticles prevents clogging of micro and nanoscale channels [31, 32] of the type that we have generally used to guide HMM propelled actin filaments. The increase in the number of actin filaments on the surface (Figure 4, Table 1) at the end of the procedure in Figure 3, in itself, reports the presence of analyte whether the filaments are cross-linked to MPs or not. Just by counting the number of actin filaments or measure the increase in fluorescence intensity due to actin filament binding to HMM on the surface we could detect the presence of analyte in the 10–100 nM concentration range (Additional file 1: Figure S1). This is in the relevant range for several clinically important disease markers, e.g. the inflammation marker C-reactive protein (CRP) . However, while detection of 50–60 nM antigens gave consistent results between experiments in terms of the number of observed filaments in step IV of Figure 3, attempts to consistently detect antigen at lower concentration (0.5 - 5 nM) was unsuccessful (Additional file 1: Figure S1) as further discussed below.
To summarize, we have demonstrated a conceptually useful approach for capturing analytes from serum for subsequent motor driven separation/detection. The usefulness of the method was clear from the consistent signal in the final step in Figure 3 (corresponding to a certain number of actin filaments) for a sample antigen concentration of 50–60 nM. However, clearly, our results also suggest that there is appreciable room for improvement. In order to understand complicating factors to facilitate future optimizations, we used a combined microscopic and fluorescence spectroscopic approach to analyze the different steps in the procedure in Figure 3.
For further optimization, it is of interest to know to what extent that antigens are actually captured by the antibody-conjugated MPs in step I of the procedure in Figure 3. First, in order to quantify the number of antibodies per MP in the reaction in Figure 2a we measured (Figure 5b) the intrinsic protein fluorescence prior to, and after coupling. The decrease in the antibody concentration in the MP-free solution suggested by these measurements indicated antibody coupling corresponding to about 40 antibodies per MP i.e. an antibody concentration of 1.2 μM assuming a MP concentration of about 30 nM in step I of the procedure in Figure 3.
We also searched for similar aggregates in step IV (Figure 3), i.e. after completion of the entire procedure but very few were observed (data not shown) in spite of a large number of actin filaments compared to the control situation. Provided that all MPs could be detected by the combined dark-field and fluorescence microscopy, this suggests that very few MPs are required for an appreciably increased number of actin filaments in step IV.
Therefore, the consistent increase in the number of actin filaments at 50 nM antigen, is strong support of the usefulness of the described approach. Furthermore, it would be trivial to substitute polyclonal antibodies with monoclonal antibodies in a real device thereby eliminating complexities due to MP-MP aggregate formation. In the case with monoclonal antibodies, those on the magnetic particles would be directed against another epitope of the antigen than those on the actin filaments. In addition to increased sensitivity, this approach would also increase the specificity due to the involvement of two different specific antibodies in creating the diagnostically relevant link between magnetic particle and actin filament.
Sensitivity and detection rate should be possible to increase further by substituting filament counting and whole frame intensity measurements by an automated process whereby a concentrator device [7, 29] concentrates the motile actin filaments in a nanoscale detector area. This would give a measure of the fluorescence intensity attributed to HMM propelled actin filaments insensitive to e.g. changes in background fluorescence or altered illumination conditions that may severely affect the full-frame fluorescence intensity . Moreover, the concentration approach is automated and fast compared to the manual filament counting procedure.
A third modification of the procedure that is likely to cause significant improvements, both with regard to limit of detection and specificity, would be the use of magnetic forces to pull actin-MP aggregates to the surface in the final step in Figure 3. In this case, only those actin filaments that are linked to MPs would be pulled by the magnetic field whereas only those MPs that are bound to an actin filament would be specifically linked to the surface via HMM. Under these conditions the flow cell height should be increased, from the current value of 0.1 mm, in order to increase the volume from which actin-MP aggregates are recruited and increase the ratio of magnetically driven to diffusional transport.
As indicated above, the procedure in Figure 3 would be readily interfaced with a molecular motor driven concentrator device [29, 36] for additional 100–1000 fold signal amplification. Finally, we have presented evidence that the number of rinsing steps may be appreciably reduced. Also the incubation times may be reduced. For instance only 10 min incubation period was used in a recent immunoassay  for binding of analyte at attomolar concentration to magnetic microparticles in solution. However, optimizations of the detailed assay procedure would require a large number of time consuming tests. These would not add much of conceptual interest and are therefore outside the scope of the present study.
We have described a novel approach for capturing analytes from serum by antibody conjugated magnetic particles adapted for use in molecular motor driven concentrator devices [7, 29]. Importantly, this method eliminates the recently discovered  deleterious effects of serum on the actin filaments and on actomyosin motor function. Subsequent to the analyte capture, magnetic separation thus allows exchange of body fluids (e.g. blood serum) for an optimized biological buffer and subsequent mixing with antibody-conjugated actin filaments. Now, large aggregates tend to form between actin filaments and micrometer sized antibody coated particles, opening for conventional magnetic concentration of aggregates between actin filaments and magnetic particles. On the other hand, magnetic nanoparticles form smaller aggregates with actin filaments that may be transported by heavy meromyosin motor fragments. Pending optimizations, particularly the use of monoclonal antibodies as suggested by our analysis of the described procedure, opens for use of a molecular motor driven concentrator device to achieve further signal amplification.
Magnetic nanoparticles (MPs) with mean diameter of 30 nm (according to manufacturer) and surface-coated with primary carboxylic groups was purchased from TurboBeads (Zurich, Switzerland). C6-succinimidyl 6-hydrazinonicotinate acetone hydrazone (C6-SANH) and C6-succinimidyl 4-formylbenzoate (C6-SFB) were purchased from Solulink, San Diego, CA, USA. Anti-rabbit IgG (H&L, Goat, a-rIgG) was purchased from Rockland Immunochemicals, Gilbertsville, USA. Zeba desalt spin columns and bicinchoninic acid (BCA) protein assay kits were purchased from Pierce Rockford, IL, USA. Rabbit IgG (rIgG) and all other chemicals were of analytical grade and, unless otherwise stated, purchased from Sigma-Aldrich Sweden AB, Stockholm, Sweden.
Actin was prepared from rabbit skeletal muscle  and actin filaments were fluorescently labeled with Alexa-488 phalloidin (APh) (Molecular Probes Invitrogen, Eugene, OR) [32, 38, 39]. Myosin II was purified from rabbit fast skeletal muscle and heavy meromyosin (HMM) was prepared by digestion of myosin with α-chymotrypsin [38, 40]. All experiments using animal material were performed in accordance with national and EU-legislation and were approved by the Regional Ethical Committee for Animal experiments (reference # 96–11), Linköping, Sweden.
Magnetic nanoparticles (30 mg mL-1 corresponding to 83 nM; according to manufacturer´s specifications) were dispersed by sonication in an ultrasonic bath for 15 minutes, then collected using a magnet (Neodymium Magnet, 14 × 8 mm, pulling strength 5 ± 0,5 kg. Svenska Magnet Fabriken AB, Hallstahammar, Sweden) and washed with 1 mL of activation buffer (54 mM (2-(N-morpholino)ethanesulfonic acid (MES), pH 5.5). These steps were repeated three times. One mL of MP solution (30 mg mL-1 in activation buffer) was mixed with 1 mL of activation solution (10 mg EDC and 10 mg NHS dissolved in 1 mL activation buffer) and incubated for 20 min at room temperature (RT) on a shaker. The activated MPs were collected using the magnet and the supernatant was discarded. One mg of a-rIgG was mixed with 1 mL of the activated MPs in activation buffer and incubated for 30 minutes at RT on a shaker. This was followed by washing four times with 1 mL of PBST (10 mM sodium phosphate, 0.15 M NaCl with 0.5% Tween-20, pH 7.4) and finally two times with 1 mL of buffer A (1 mM MgCl2, 10 mM 3-morpholinopropane-1-sulfonic acid (MOPS), 0.1 mM K2-ethylene glycol tetraacetic acid (EGTA), pH 7.4). The a-rIgG-conjugated MPs were then mixed with 1 mL of storage buffer (buffer A with 0.05% sodium azide) and stored at 4°C over night.
F-actin, conjugated with a-rIgG was prepared essentially as described previously . First, F-actin (3.5 mg mL-1 in modification buffer 1; 100 mM (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid) (HEPES), 150 mM KCl, 5 mM MgCl2, 1 mM Na2ATP, pH 8.0) was incubated with C6-SANH linker (dissolved in Dimethyl sulfoxide, DMSO) at RT for 2 h with a molar ratio of 1:2 (actin:linker). Excess linker was removed using ultracentrifugation (Beckman-Optimamax-XP ultracentrifuge, 100,000 g, 4°C, 25 min) through a 10% glycerol cushion. Pellets were re-suspended in modification buffer 1 with 5% glycerol cushion. A second ultracentrifugation step was applied with subsequent re-suspension in conjugation buffer (100 mM MES, 150 mM KCl, 5 mM MgCl2, 1 mM Na2ATP, pH 6.0). The actin solutions were sonicated gently on ice to disperse the filaments and these were then stored on ice until the conjugation with a-rIgG. Our previous results  suggest that the covalent modification of actin with the C6-SANH linker does not modify actin function.
The a-rIgG (2 mg mL-1 in PBS buffer; 0.1 M sodium phosphate, 0.15 M NaCl, pH 7.4) solution was incubated with 10-fold molar excess of C6-SFB linker (dissolved in DMSO) at RT for 2 h. Modified a-rIgG was desalted using zeba desalt spin column against conjugation buffer (see above) to remove excess linker and to exchange buffer. The modified antibodies were stored on ice until further use.
The C6-SANH modified F-actin sample was incubated with C6-SFB modified a-rIgG solutions. Addition of 10 mM catalyst buffer from stock solution (100 mM aniline in conjugation buffer) started the reaction that was allowed to proceed for 6 h at RT. The molar ratio of modified F-actin and modified a-rIgG was 2:1 respectively.
In order to remove catalyst and exchange the buffer, samples were dialyzed at 4°C against G-actin buffer (2 mM Tris base, pH 8.5, 0.2 mM Na2ATP, 0.5 mM dithiothreitol (DTT), 0.2 mM CaCl2, 3 mM NaN3). For long-term storage, samples were flash frozen in liquid nitrogen and stored at -80°C.
The conjugated G-actin monomers (actin-a-rIgG) were co-polymerized with non-conjugated monomers to form filaments with 1:3 molar ratio of conjugated to non-conjugated actin. The co-polymerization reaction was performed at 4°C (3 h) by addition of KCl, MgCl2 and ATP to final concentrations of 100 mM, 2 mM and 3.3 mM, respectively. The co-polymer was fluorescence-labeled with APh or RhPh (molar ratio: 1:1.5; actin:APh or RhPh) at 4°C over night (see above).
Figure 3 illustrates the principle for magnetic capture of antigen from blood and the subsequent separation from other serum components and integration with the actomyosin motor system. Serum was first obtained from human blood drawn into Vacutainer™ glass tubes. The blood was from volunteers who consented to the procedures under a protocol approved by the regional ethical committee in Linköping, Sweden. After 1 h at RT the sample was centrifuged (3000 g for 20 minutes at RT; 20°C) to remove blood cells and solidified fibrin thus giving blood serum. Three different concentrations of serum were prepared by dilution in buffer B (buffer A with 1 mM DTT and 50 mM KCl): 100%, 10% and 1%. For the subsequent magnetic separation process the different serum samples were aliquoted into six (two for each concentration), tubes (200 μl per tube). Magnetic particles conjugated with a-rIgG were added to all six tubes (2.5 mg magnetic particles per tube).
One additional tube contained 0% serum (100% buffer B). The rIgG (60 nM) was added to four of the vials (100%, 10%, 1% and 0%) and all seven tubes were put on a shaker for 1 h at RT. Samples were then washed 10 times each with 500 μl of buffer B while applying magnetic forces to keep magnetic particles and their bound analytes (rIgG) in the tubes. This allowed removal of serum and exchange for optimized biological buffer (buffer B). Each tube was then incubated with 250 nM a-rIgG conjugated actin filaments (final volume 250 μl) for 1 h at RT on a shaker in the dark followed by gentle wash 10 times in buffer B under magnetic separation forces. Finally 250 μl of buffer B was added to the tube before the content was infused into a motility assay flow cell.
In another set of experiments, 2.5 mg MP-a-rIgG was mixed with 100% Serum and 0% Serum with 50 nM of rIgG for 30 min at RT on a shaker in the test samples. Alternatively, 0.5-50 nM Rh-rIgG was used as antigen. Control samples were treated similarly but were without rabbit IgG. After stipulated times the samples were washed three times with 1 mL of PBST (not used with Rh-rIgG) and three times with 1 mL of buffer A, while applying magnetic forces to keep magnetic particles and bound analytes (rIgG) in the tubes. This was expected to allow effective removal of serum components. Subsequently, 300 μL of 1% bovine serum albumin (BSA) was added to each tube followed by 30 min incubation at RT under gentle shaking. This step when used (not with Rh-rIgG) was followed by two times wash with 1 mL of buffer A. Each tube was then incubated with 250 nM a-rIgG conjugated filaments for 30 min at RT on a shaker in the dark. Samples were then washed gently seven times (five times with Rh-rIgG) in buffer B under magnetic pelleting of complexes with magnetic particles, e.g. MP-a-rIgG-rIgG-a-rIgG-actin and final addition of 250 μl buffer B to re-disperse the pellet (see above).
Flow cells were constructed from two cover-slips with double-sided sticky tape as spacers. The motility supporting surface (bottom of flow cell) was derivatized with trimethylchlorosilane (TMCS) as described previously . All solutions that were added to the flow cell were based on buffer A (see above) and all proteins were diluted in buffer B (see above). The flow cell was pre-incubated according to standard procedures [15, 39, 42]: (i) HMM (120 μg mL-1) for 2 min, (ii) 1 mg mL-1 BSA for 30 s (iii) wash with buffer B and (iv) addition of re-dispersed pellet from the magnetic separation (described above). After an incubation period of 3 min, flow cells were washed with buffer B, and (vi) incubated with r60 assay solution (buffer A with 10 mM DTT, 35 mM KCl, ionic strength 60 mM). The r60 solution was also supplied with an anti-bleach system (final activity concentrations of 3 mg mL-1 glucose, 20 U mL-1 glucose oxidase and 870 U mL-1 catalase). Finally, (vii) a60 assay solution was added to the flow cell to induce filament sliding. The a60 solution was similar to the r60 solution but with addition of 1 mM MgATP and an ATP regenerating system (2.5 mM creatine phosphate and 3.5 U mL-1 creatine phosphokinase).
Fluorescence spectroscopy was performed using a FluoroLog FL3-22 Spectrophotometer (Instruments S.A Inc. New Jersey, USA). For estimation of changes in antibody and serum protein concentrations, we used an excitation wavelength of 295 nm (slit 2 nm) and read emission in the range from 310 nm to 400 nm (slit 6 nm). Antigens labeled with rhodamin (~2 rhodamin per antibody) were studied using an excitation wavelength of 555 nm (slit 3) and emission was read at 570–700 nm (slit 5 nm).
Fluorescently labeled actin filaments were observed using a Nikon Eclipse TE300 inverted fluorescence microscope (Nikon Corp., Tokyo, Japan) equipped with a temperature-regulated Nikon (100 × 1.4 NA) oil immersion objective, TRITC (Ex. 540/25, DM 565, and BA 605/25) and FITC (Ex. 465–495, DM 505, and BA 515–555) filter sets. Image sequences were recorded using a cooled Hammamtsu EMCCD camera as described previously . Actin velocities were measured using a manual tracking program developed in a Matlab enviroment (The MathWorks Inc, Natick, MA) [43, 44]. The number of actin filaments and the total intensities of the different frames were measured using Image J (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://imagej.nih.gov/ij/, 1997–2012.).
Dark field microscopy was performed using a Nikon oil immersion dark field condenser together with a Nikon oil immersion objective (100 ×, 0.5-1.3 NA). The latter was also used for fluorescence microscopy in Figure 6.
Statistical analyses were implemented in GraphPad Prism (v. 6.01; GraphPad Software, San Diego, CA). Unless otherwise stated, data are given as mean ± standard error of the mean (SEM) and statistical hypothesis testing was performed using two-tailed t-test (paired when appropriate) or repeated measures ANOVA.
Images from the individual experiments were analyzed manually by counting the number of filaments in one frame from each flow cell. These filaments were then categorized into stationary or motile by observation in 25 frames (5 seconds of real time). To verify the manual counting method the average background-subtracted intensity from 50 consecutive image frames of stationary filaments was also obtained and used as an alternative measure of the relative number of filaments. Average background pixel intensity was calculated as the mean pixel intensity value in four filament-free areas, one in each quadrant, of the averaged intensity image.
This work was funded by the European commission (FP7) under the contract MONAD (NMP4-SL-2009-228971), The Carl Trygger Foundation, The Swedish Research Council (Project # 621-2010-5146), The Crafoord Foundation and the Faculty of Natural Sciences and Engineering at Linnaeus University.
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.