Network structure and thermal properties of bioactive (SiO2–CaO–Na2O–P2O5) glasses
- 929 Downloads
Ca- and P-based bioactive glasses are excellent candidates for design and manufacture of biomaterials. Understanding the structure and physico-chemical–thermal behaviour of bioactive glasses is a fundamental step towards the design of a new generation of biocompatible materials. In this study, the structure of SiO2–CaO–Na2O glasses and its derivatives, obtained by substituting Na2O with P2O5 and prepared by melt–quench technique, was studied with neutron and electron diffraction techniques combined with thermal analysis, high-resolution electron microscopy and X-ray photoelectron spectroscopy. Neutron and electron diffraction data were analysed with reverse Monte Carlo simulation and pair distribution function analysis, respectively. Bioactivity of P2O5 substituted glasses was also investigated and proven in vitro using simulated body fluid. Based on the structural analysis, it was found that Si and P atoms are in well-defined tetrahedral units with a bond distance of 1.60 Å for both Si–O and P–O bonds, although P exhibits a higher average coordination number than Si. With increasing phosphate content, tendentious changes in the glass behaviour were observed. Linear increase in Tg, supported by the changes in the average coordination numbers of Si and P, indicates strengthening of network structure with increasing P content and formation of P–O–Ca atomic linkages, which lead to Ca–P-rich atomic environments in the silicate network. These Ca–P-rich environments trap volatile elements and thus decrease the total weight loss during heating at higher P concentrations. In the case of the highest investigated P2O5 content (5 mol%), nanoscale structural inhomogeneity and the formation of Ca–P-rich clusters were also revealed by electron diffraction and atomic resolution imaging. This type of Ca–(Na)–P clustering has a key role in the behaviour of phosphate-substituted silicate glasses under physiological conditions.
Bioactive glasses are mostly silicate-based glasses which, besides exhibiting biocompatibility, are able to bond actively to host living tissue . Their main practical advantages with respect to other bioactive or biocompatible materials like ceramics, bulk metallic glasses or other metallic materials are based on their metastable amorphous structure, which induces high reactivity under physiological conditions. At the same time, their composition, in contrast to metal-based implant materials, allows biomedical applications without additional coating.
Besides silicates, several glassy compositions based on B2O3 or P2O5 have been found suitable for different kinds of biomedical applications [2, 3]. Each compositional series has its advantage: the SiO2-based group comprises a wide range of glass formulations, the B2O3-based glasses are characterized by higher reactivity than silicate materials, which leads to faster bioactive kinetics, and P2O5-based systems are better resorbable materials under physiological conditions, and their dissolution rate can be tuned by changing their compositions . Various concentrations of other oxides can be incorporated in the basic glass composition to enhance particular properties of the glassy material, like the CaO and Na2O are useful to adjust the surface reactivity in biological environments.
However, common disadvantages of all compositions, namely poor load bearing performance and manufacturing difficulties, impede their widespread application in clinics. One way to overcome these disadvantages is the synthesis of glass–ceramics nanocomposite materials from bioactive glass as a starting material either by heat treatment or melt–quench technique . As both bioactivity and the high temperature behaviour of glasses depend on composition and structure, the structure of bioactive glasses has been intensively investigated.
It has been established that bioactive silicate glasses are characterized by a network structure of SiO4 polyhedra with some orthophosphate (PO4) substitution . Subsequently, molecular dynamics simulations  suggested that the distribution of the network modifier cation Ca2+ might not be homogeneous in the random network structure but has a strong affinity to the orthophosphates. Combined neutron and X-ray diffraction study of commercial Bioglass® provided experimental evidence for the non-uniform distribution of Ca2+ in the network structure . At the same time, nuclear magnetic resonance spectroscopy proved the existence of phosphate clusters up to six PO4 tetrahedra units [9, 10]. These results indicate that local fluctuations in the atomic structure and composition comprise an inherent property of bioactive silicate glasses that has to be considered when thermal and mechanical properties or bioactive behaviour is explained.
This integrated bulk-to-nanoscale study is motivated by a better understanding of the connection between the nanostructure and thermal behaviour of bioactive silicate glasses. To achieve this goal, we measure and model the average network structure of a bioactive glass series in the compositional range of SiO2(45)CaO(25)Na2O(30 − x)P2O5(x), where x = 0,1,3,5, using neutron diffraction and reverse Monte Carlo simulations, and then study the nanoscale inhomogeneities by transmission electron microscopy methods and X-ray photoelectron spectroscopy. In parallel, thermal characterization is carried out as a function of composition. Finally, merging bulk and nanoscale structural data, interpretation of thermal behaviour is presented.
The requirements for synthetic bone substitute materials summarize the influence of the respective material towards the biocompatibility and bioactivity . In the present work, the biological properties of studied materials were evaluated using in vitro settings.
Nominal composition in mole% of the studied glass samples
Neutron diffraction experiments
Neutron diffraction (ND) measurements were performed in a broad momentum transfer range, combining the data measured by the 2-axis PSD monochromatic neutron diffractometer (λ0 = 1.069 Å; Q = 0.45–9.8 Å−1)  at the 10 MW Budapest research reactor and by the 7C2 diffractometer at the LLB, Saclay (λ0 = 0.726 Å; Q = 0.52–19 Å−1) . The powder specimens of about 3–5 g/each were filled in cylindrical V-sample holders of 8 and 6 mm diameter for the two experiments, respectively. The structure factors, S(Q)s, were evaluated from the raw experimental data, using the programme packages available at the two facilities. As the statistics of the data is better for the PSD diffractometer at relatively low Q values (below ~ 4 Å−1) and the statistics of 7C2 diffractometer data is much better above 8 Å−1, the S(Q) data were combined by normalizing the data of the PSD diffractometer to that of the 7C2 diffractometer in the 4–8 Å−1 range by least squares method. The agreement of the corresponding S(Q) values was within 1% in the overlapping Q-range. The combined values of the two spectra were used for further data treatment. (i.e. for Q < 4 Å−1 only the PSD data and for Q > 8 Å−1 only the 7C2 data were used.)
Electron microscopy and electron diffraction
For high-resolution transmission electron microscopy (HRTEM) and electron diffraction (ED), the sample was gently crushed under ethanol in an agate mortar, and a drop of the resulting suspension was deposited onto a lacey carbon-covered Cu grid (Ted Pella). HRTEM was carried out using a JEOL 3010 at 300 keV (LaB6 cathode, 0.17 nm point-to-point resolution, UHR pole piece) equipped with GIF. Electron diffraction measurements were carried out in nanobeam ED mode using a Philips CM20 (operating at 200 keV) and a JEOL 3010 (operating at 300 keV) TEMs, both with LaB6 cathode. The diameter of the analysed area was 1.5 µm and 25 nm for CM20 and JEOL 3010 microscopes, respectively. The applied camera length was 500 mm. Special care was taken to avoid background in nanobeam ED, by selecting appropriate areas for diffraction. The scattered intensity was recorded using Ditabis Imaging Plates (17.5 µm pixel size) and Gatan Orius CCD (7.4 µm pixel size) for CM20 and JEOL3010 microscopes, respectively. The use of Imaging Plates allows to record linear response to the electron dose over six orders of magnitude, which is ideal for measuring scattered intensity by amorphous structures. In the case of CCD, the dynamic range was extended by merging together diffraction patterns taken with different exposure times using the Digital Micrograph script written by Bernhard Schaffer (Gatan). The diffraction measurement procedure was based on the protocol summarized for selected area ED mode , which ensured reproducibility of camera length within 0.5%. Camera length was calibrated for the applied lens currents using a self-supporting random nanocrystalline Al thin film.
DTA measurements were performed using a SETARAM 92-16.18 DTA equipment with a TGA-92 set-up under Ar atmosphere. The DTA was calibrated by melting of high-purity In, Zn and Al metals, and each DTA scan consisted of a controlled heating with 10 °C/min constant heating rate and a subsequent instantaneous uncontrolled cooling of the sample (with an approximate initial cooling rate of > 50 °C/min).
X-ray diffraction (XRD) measurements were performed using a Bruker AXS D8 Discover diffractometer equipped with Göbel mirror and a scintillation detector with Cu Kα radiation. The X-ray beam dimensions were 1 mm * 5 mm, the 2θ step size was 0.02°, and the scan speed 0.1°/min. We used the Diffrac.EVA program and the ICDD PDF database for phase identification.
Approximately 1-cm-sized specimens were selected for XPS analysis under ultrahigh vacuum conditions (2 * 10−9 mbar). The specimens were exposed to 70 °C heat treatment for 48 h, which is the standard baking procedure of the applied vacuum system. The photoelectron spectra were obtained using X-ray radiation (Al anode with water cooling, 15 keV excitation). Constant energy resolution of 1.5 eV was provided by a special cylindrical mirror analyser with a retarding field (type DESA 150, Staib Instruments Ltd.). All spectra were recorded with 0.1 eV energy steps. The XPS measurement yielded information on the average surface composition of an area of ca. 5 mm diameter. Detected lines: O 1s (532 eV); Ca 2p (347 eV); C 1s (285 eV); Na Auger (497 eV); P 2s (189 eV); Si 2p (100 eV). Low concentration elements such as P were recorded with longer detection time to improve the statistics and hence the reliability of the quantification. Spectra were evaluated by applying the usual Shirley background subtraction, and compositions were calculated from peak areas assuming homogeneous model. Sensitivity factors were calculated from the reference area in Ref. . Concentration versus depth curves (called depth profiles) were recorded by alternating ion sputtering (1 keV Ar ion beam) and XPS measurement.
To test the bioactivity of the bulk glasses, samples were incubated in simulated body fluid (SBF) for 30 min, 3 h, 3 days, 7 days and 21 days at 37 °C and p(CO2) = 0.05 atm (5%) as human serum is in equilibrium with such a partial pressure . The SBF and the validation of the apatite-forming ability were done according to the procedure of Kokubo and Takadama . After soaking, the samples were removed from the SBF, gently washed with deionized water and dried at room temperature. The surface morphology of the dried samples was explored using a LEO 1540XB scanning electron microscope (SEM).
Structure factor calculation based on neutron diffraction
ND weighting factors of the partial interatomic correlations in the glasses
Weighting factor, wijND (%)
Reverse Monte Carlo modelling
For the starting RMC model, a disordered atomic configuration was built up with a simulation box containing 10.000 atoms with density data 0.076, 0.0752, 0.0741 and 0.0714 atoms/Å−3 and rmax = 26.39, 26.52, 26.65 and 26.93 Å for the S45P0, S45P1, S45P3 and S45P5 glasses, respectively. In the RMC simulation procedure, two types of constraints were used, for the minimum interatomic distances between atom pairs (cut-off distances) to avoid unreasonable atom contacts and coordination constraints. For the starting configuration, we have used the results based on the literature and our previous results for binary SiO2–Na2O  and CaO–P2O5  glasses. We apply two types, a positive and a negative coordination constraint for the network former Si atoms. Based on the literature and our previous results [22, 23, 24], it is reasonable to suppose that silicon has a fourfold oxygen coordination as a positive constraint, and the lack of onefold and/or twofold oxygen coordination as negative constraints.
Structure factor calculation based on electron diffraction
Comparison of the isotropic ND weighting factors wijND and the angle-dependent ED weighting factors for zero scattering angle wijED for the S45P5 composition
From Table 3, it is also seen that, while during ND experiment almost all of the overall scattering (ca. 90%) is from O–O and cation-O contribution, in the case of electron scattering this contribution comprises only 50% of the overall scattering, implying a significant cation–cation (mostly Si–Si) contribution for this composition, which carries structural information on the medium-range order (MRO).
Results and discussion
Characteristic temperatures of the thermally activated processes for the different bioactive glass compositions
TG peak (°C)
As shown in Fig. 4a, structural strengthening due to the PO4 content increases also the stability of the supercooled liquid state, so that Tx is also increased gradually with the PO4 content up to 3 mol% of phosphate concentration. However, SiO2 and P2O5 have a large miscibility gap in crystalline state and, according to the phase diagram , at low temperature only eutectic forms and practically no miscibility exists. At higher temperature, above cc. 1000 °C, in the liquid state phosphate is miscible in liquid silica. Presumably, this better miscibility prevails in the glassy state as well. The continuously increasing Tg indicates the increase in the cohesion in the glass structure with phosphate content. It is an unexpected behaviour, because the melting point of phosphate glasses is typically lower than that of silica glass. The same trend can be observed in the case of the Tx crystallization temperature up to 3% PO4 content. Above this concentration, the phosphate component becomes saturated and phase separation is preferred. The phase separation in the supercooled liquid state is indicated by the drop of the Tx for S45P5 sample.
At the same time, the investigated glasses show another minor transition, namely a 0.02–0.25% weight loss during the heat treatment. This process also indicates the appearance of some mobility in the glass structure with the increasing temperature. For S45P0, the peak of the weight loss process coincides with the glass transition temperature, indicating a close relation between the two processes. The increase in phosphate content reduces the total weight loss and also the temperatures of the weight loss process. The large melting point difference between the silica and phosphate phases  indicates that the PO4-containing structure is less stable at elevated temperatures; thus, we anticipate that partial melting of the PO4-containing environments induces mobility in the glass structure. This process results in the lowering of weight loss temperature at higher phosphate concentration.
The decrease in total weight loss with phosphate content implies that the amount of volatile component is decreasing. Previous studies indicate the clustering of phosphate  and the non-uniform distribution of Ca2+ in the network structure . We suppose that the volatile component becomes bonded by the formation of more stable Ca(Na?)–phosphate clusters in the glass. These clusters are able to trap the volatile component and thus remove it from the channels of the silica-dominated network structure, leading to the decrease in the total weight loss during heating.
Network structure from neutron diffraction and reverse Monte Carlo simulation
Most important oxygen-linked atomic distance, rij (Å) obtained from RMC simulation
Interatomic distances, rij (Å)
1.60 ± 0.01
2.30/2.55 ± 0.05
2.25/2.65 ± 0.05
2.52 ± 0.05
1.60 ± 0.01
2.30/2.60 ± 0.05
2.25/2.60 ± 0.05
1.60 ± 0.05
2.55 ± 0.05
1.60 ± 0.01
2.30/2.55 ± 0.05
2.25/2.55 ± 0.05
1.60 ± 0.05
2.30/2.55 ± 0.05
1.60 ± 0.01
2.33/2.55 ± 0.05
2.25/2.55 ± 0.05
1.60 ± 0.05
2.30/2.56 ± 0.05
For the Ca–O network, we have established two distinct first-neighbour distances at 2.30 ± 0.05 Å and 2.55 ± 0.05 Å, in agreement with the high-resolution neutron diffraction study . Taking into consideration the medium resolution of the present ND experiment, the agreement is reasonably reported for CaSiO3 glasses [8, 24]. For the studied glassy compositions of the Ca–O average coordination number, distributions close to threefold coordinated oxygen atoms are obtained, that is 2.99, 2.68, 2.86 and 2.56 (± 0.15) atoms for the S45P0, S45P1, S45P3 and S45P5 compositions, respectively (see Fig. 7b).
The P–O first-neighbour distributions show a characteristic peak at 1.60 ± 0.05 Å and a small second peak at 2.00 ± 0.1 Å, where the second peak intensity changes in the function of phosphorous concentration. The agreement is reasonably reported for CaP glasses [8, 21]. The P–O average coordination number varies as 3.97, 3.93 and 3.94 (± 0.05) atoms for the S45P1, S45P3 and S45P5, respectively (see Fig. 7c.).
Unfortunately, gNa–O(r) and gO–O(r) overlap with each other; thus, the results have to be handled carefully. It can be established that for the S45P3 and S45P5 glasses the distributions are similar. In the case of S45P0 and S45P1, the O–O distribution shows a wide distribution with one peak. This indicates that the modifier Na ions prefer the neighbourhood of Si-oxide units, indicating that Na2O acts as a modifier as it breaks up the silicate network, similarly to any alkali oxides, producing non-bridging oxygen atoms within the network. For Na–O, a double peak has been revealed at 2.27–2.35 ± 0.03 Å and 2.62 ± 0.03 Å for all samples [34, 35]. Two well-defined peaks have been obtained for the O–O at 2.30 ± 0.03 Å and 2.65 ± 0.04 Å [8, 22, 36]. Note that the overlapping character of Na–O and O–O distributions should be taken into account.
Si–Si correlation functions show distributions at 2.95/3.05 ± 0.02 Å, which is in good agreement with the literature data [20, 22] and supports the formation of well-defined SiO4 units. In the case of S45P1 sample, the Si–P correlation functions consist of a double peak with distances at 2.85/3.10 ± 0.05 Å, these are melting to the one wider peak at 3.10 ± 0.05 Å for the S45P3 and S45P5 samples, and the P concentration dependence can be observed. The shortest second-neighbour distances obtained for Si–P as ~ 3 Å show a connection between the SiO4 and PO4 tetrahedral units, which indicates that in the building of the basic glass network take parts the Si–O–P linkages.
The short-range order of the glassy network structure is determined by two parameters, the bond length and the oxygen–cation–oxygen bond angles. Si–O and P–O bond lengths, both 1.60 Å, coincide for all compositions within the error of the measurement (Table 1) and bond angle distributions, specific of the silicate structure, fall fairly close to each other as well (Fig. 8). This implies similar average network structures for all compositions, in good agreement with the literature data [38, 39]. However, careful observation of the average coordination numbers allows us to make some considerations regarding the changeability in the network structure. In the case of S45P0 glass, the average coordination number of Si is 3.67, in agreement with the literature data for soda–lime glasses , indicating the presence of non-bridging oxygen, i.e. Si–O–Ca- and Si–O–Na-type connections, which break up the network structure . By adding phosphate to the system, the coordination number of Si slightly increases, but still remains below 4 (3.78, 3.77 and 3.78, for S45P1, S45P3 and S45P5, respectively), while that of P fairly coincides with the ideal value of the tetrahedral PO4 units (3.97, 3.93 and 3.94 for S45P1, S45P3 and S45P5, respectively).
Similarly, different coordination numbers were found for the network formers Al and Si in Cax/2 Alx Si1−x O2 glasses  and explained by the presence of Si–O–Ca-type bonds, which reduces the coordination number of Si without creating vacancy in the network structure . This way, two types of atomic environments are forming in Ca–Al–Si-oxide glasses, namely Si–O and Al–O tetrahedral, and the former one will promote the disintegration of the network structure by the non-bridging oxygen-type Si–O–Ca linkages.
In our case, silica-rich and phosphate-rich environments were deduced from thermal measurements. The linear increase in Tg with phosphate content indicates the strengthening of network structure, which is also reflected in the increase in the average coordination number of Si by ca. 3%. We propose that this strengthening of the network structure relies on the removal of Ca (and probably Na) from the environment of Si by the phosphate clusters. In this way, Ca- and P-rich environments form, which, as indicated by the coordination numbers of P, are well ordered, independently of the overall phosphate content of the glass. These Ca–P clusters are similar in bonding to the building blocks of crystalline Ca–phosphates which can be stable up to above 1000 °C. Thus, the increase in thermal stability reflected in the slopes of the Tg can be explained by the formation of P–O–Ca instead of non-bridging oxygen-type Si–O–Ca linkages. As phosphate traps Ca (and probably Na) by incorporating them into a more ordered, and thermally more stable structure, the weight loss should seemingly decrease with phosphate content, which is in agreement with thermal data (Fig. 4b).
Nanostructure: ePDF analysis, HRTEM and XPS
As both thermal properties and RMC indicated nanoscale inhomogeneities in the network structure, an attempt was made to prove directly the presence of these inhomogeneities using local analytical methods. Electron diffraction-based pair distribution function (ePDF) analysis was carried out on S45P5 sample. As this sample has the highest P2O5 content, structural deviations from the SiCaNa glasses, if there are any, are expected to be the most enhanced.
As the crystalline (apatite-like) character has been enhanced under the electron beam, we cannot state steadily that the crystalline domains presented in Fig. 11a are inherent in the S45P5 glass structure. However, the fact that the structural rearrangement under the electron beam resulted in nanocrystalline apatite serves as an indirect proof of the chemical inhomogeneity of the glass structure, manifested in the formation of Ca- and P-rich environments. The sample volume analysed in Fig. 11 originally contained undoubtedly more Ca and P than silica, as proven by its ability to transform into nanocrystalline apatite. Sample volumes, which did not show this type of structural rearrangement, are supposed to build up by dominantly silica.
On the surface of all three PO4-containing glasses, a phosphorus-containing reaction layer was formed already after 30-min soaking . This reaction layer was not continuous, and its composition also changed from place to place. Morphologically, similar layer was formed on the surface of the S45P0 sample as well; however, this layer was phosphorus free. As the SBF is supersaturated with respect to apatite, this observation is of key importance and supports that the reaction layer is indeed the product of the incipient ion exchange between SBF and glass and not heterogeneously precipitated phosphate phase on the surface of the glass.
Based on morphological and compositional criteria , the apatitic reaction layer on the surface of all three samples can be recognized. A detailed nanostructural analysis of the bioactive layer of the glass samples is a subject of another paper.
The structure of SiO2(45)CaO(25)Na2O(30 − x)P2O5(x) (x = 0, 1, 3, 5) glasses has been studied using diffraction methods, RMC simulation and HRTEM and compared to the composition-dependent thermal properties.
Neutron diffraction-based RMC simulation revealed that the average Si–O and P–O network structures of the different glasses with Ca and Na modifiers are similar on the short-range scale, since both partial pair correlation functions and first neighbours nearly coincide for the different glasses within the error of the measurement. However, the differences observed in the coordination of the Si and P atoms indicate varying medium-range order for the different glass compositions.
Based on the tendentious change of thermal stability with increasing P content, we presume that weakly bounded Ca (and Na) was removed from the environment of Si in the silicate network by forming phosphate clusters which are more stable as they melt above 1000 °C in the bulk form. Thus, the enhanced thermal stability, reflected in the linear increase in the glass transition temperature with phosphate content, can be explained by the formation of these P–O–Ca linkages instead of non-bridging oxygen-type Si–O–Ca linkages. Therefore, Ca–P-rich clusters are similar in bonding to crystalline Ca–phosphates and may trap additional volatile cations, like Na, in agreement with the total weight loss of the glasses on thermal activation. The tendency for Na–P association in weakly bounded atomic environments was demonstrated by the Na- and P-rich surface chemistry revealed by XPS, while, at the larger P content (x = 5), the formation of Ca–P-rich clusters as nanometre-range heterogeneities was observed in HRTEM. We propose that these atomic-scale heterogeneities have a key role in the behaviour of phosphate-doped bioactive glasses under physiological conditions.
Open access funding provided by MTA Wigner Research Centre for Physics (MTA Wigner FK, MTA EK). This work was supported by the Centre for Energy Research, Hungarian Academy of Sciences, under the Project “Investigation of biocompatible glasses for biomedical applications” (125/2017 and 109/2018), and in minor part by the National Research, Development and Innovation Fund Office, Hungary, under the project “Investigation of the nanostructural background of functionality in case of biogenic and biocompatible mineral apatite”, grant number K-125100. V.K.K. is indebted to the János Bolyai Fellowship of the Hungarian Academy of Sciences and the ÚNKP-19-4 New National Excellence Program of the Ministry for Innovation and Technology. Work of Zs.K. was completed in the ELTE Institutional Excellence Program (1783-3/2018/FEKUTSRAT) supported by the Hungarian Ministry of Human Capacities.
- 11.Karadjin M, Essers Ch, Tsitlakidis S, Reible AM, Boccaccini AR, Westhauser F (2019) Biological properties of calcium phosphate bioactive glass composite bone substitutes: current experimental evidence. Int J Mol Sci 305:1–22Google Scholar
- 15.Crist BV (ed) (2000) Handbook of monochromatic XPS spectra. Wiley, New YorkGoogle Scholar
- 18.Hannon AC (2006) ISIS Disordered materials database. http://www.isis2.isis.rl.ac.uk/Disordered/Database/DBMain.htm
- 19.Gereben O, Jovari P, Temleitner L, Pusztai L (2007) A new version of the RMC++ Reverse Monte Carlo programme, aimed at investigating the structure of covalent glasses. J Optoelectron Adv Mater 9:3021–3027Google Scholar
- 23.Vedishcheva NM, Shakhmatkin BA, Wright AC (2004) The structure of sodium borosilicate glasses: thermodynamic modelling vs. experiment. J Non-Cryst Solids 39:345–346Google Scholar
- 24.Mead RN, Mountjoy G (2005) The structure of CaSiO3 glass and the modified random network model. Phys Chem Glasses 46:311–314Google Scholar
- 30.Fanderlik I (1991) Silica glass and its application. Glass science and technology, vol 11. Elsevier, Amsterdam. ISBN 0-444-98755-XGoogle Scholar
- 40.Ha MT, Garofalini SH (2016) Local structure of network modifier to network former ions in soda-lime alumina-borosilicate glasses. J Am Cer Soc 14565:1–9Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.