- freely available
Pathogens 2013, 2(3), 544-555; doi:10.3390/pathogens2030544
Published: 10 September 2013
Abstract: The prevalence of tick-borne diseases is increasing worldwide. One such emerging disease is human anaplasmosis. The causative organism, Anaplasma phagocytophilum, is known to infect multiple animal species and cause human fatalities in the U.S., Europe and Asia. Although long known to infect ruminants, it is unclear why there are increasing numbers of human infections. We analyzed the genome sequences of strains infecting humans, animals and ticks from diverse geographic locations. Despite extensive variability amongst these strains, those infecting humans had conserved genome structure including the pfam01617 superfamily that encodes the major, neutralization-sensitive, surface antigen. These data provide potential targets to identify human-infective strains and have significance for understanding the selective pressures that lead to emergence of disease in new species.
Human exposure to vectors carrying disease agents has been increased by climate and land-use changes causing more contact between humans and domestic animals with wildlife reservoirs . One such recently emerging disease is tick-borne anaplasmosis that causes infections in multiple animal species . These include cattle, sheep, goats, horses, dogs, foxes, cats, rodents and most recently, humans. Different strains of the causative organism, Anaplasma phagocytophilum, have different host predilections and not all strains can infect all hosts. Small mammals are thought to be a reservoir, infecting immature stages of ticks which act as bridge vectors transferring infection to humans and domestic animals. Cases reported to the U.S. Centers for Disease Control and Prevention comprise a flu-like febrile illness with some severe sequelae such as multiple organ failure, or severe acute respiratory distress syndrome. In the U.S., case reports increased from 348 cases in 2000 to 1,761 cases in 2010 . The reported hospitalization rate is 36% . Granulocytic anaplasmosis (GA) can be treated with antibiotics, but the symptoms, such as headache, fever, and muscle aches are non-specific and can be confused with other common diseases such as Lyme, transmitted by the same species of tick. If left untreated GA can be severe , resulting in a case fatality rate in the U.S. of up to 3% (CDC data for 2003). Increasingly, there are reports of infections transmitted by blood transfusions in the U.S. and Europe [6,7]. Here we show that, in contrast to the extensive worldwide genomic diversity of A. phagocytophilum strains, human-infective strains are a conserved subset. This has implications for understanding the selective pressures that lead to emergence of disease in new species and for control of this infection.
2. Results and Discussion
2.1. Comparative Genomics of Nine Strains of A. phagocytophilum
Underlying the ecological complexity and variable host tropism of A. phagocytophilum is a large degree of genetic variability. European strains of A. phagocytophilum are highly pathogenic in sheep and cattle, whereas North American strains do not cause disease in domestic ruminants. Some U.S. strains, defined as Ap-variant 1 because of a 2 base pair difference in 16S rRNA, are infectious to deer and other ruminants, but do not infect mice and are thought not to infect humans [8,9]. Highly unusual for an obligate intracellular pathogen with a small, 1.5 Mb genome, A. phagocytophilum contains approximately 100 “functional pseudogenes” of the major surface protein 2 gene (msp2/p44) that codes for the major antigen on its surface . These hypervariable pseudogene cassettes are recombined into a single expression site which provides N- and C-terminal conserved sequences, allowing the bacterium to serially express variable antigens and evade host immunity [11,12]. We applied high-throughput genome sequencing to nine strains, derived from humans and different host animal species from the U.S. and Europe. These strains are: two human origin strains from New York and Minnesota, a rodent- and dog-origin strain from Minnesota, two Ap-variant 1 strains from Minnesota, a horse-origin strain from California and two sheep-origin strains from Norway, for which no in vitro cultures are available. In Figure 1 the complete genome sequence of a strain derived from rodents in an area of high prevalence for human disease (Camp Ripley, MN, USA) is compared with that of the human ApHZ strain (from the state of New York). The figure shows Blast comparisons of the rodent and human strains, particularly the positions of members of the pfam01617 superfamily, to which msp2/p44 belongs. Overall, despite widely separated geographic and species origins, the two genomes are conserved and almost completely syntenic. The majority of the highly repetitive msp2/p44 genes (visualized using partial opacity) are located in the third of the genome closest to the origin of replication (top of figure, changing sign of GC skew), as reported previously . However, using high-stringency Blast comparisons (fifth circle from the outside) differences between the two genomes were primarily localized to this region also and to msp2/p44 genes in particular, showing the propensity of this gene family to undergo rapid diversification.
This comparative genomics analysis was extended to all nine strains (Figure 2). Except for Ap-variant 1 strains, the U.S. strains had high identities across their entire genomes, with the ApDog strain most similar to human ApHZ by Blast analysis. The Norwegian sheep and U.S. Ap-variant 1 strains had the lowest identities with ApHZ in Blast comparisons, especially in members of the pfam01617 superfamily.
The differences in this superfamily were examined more closely in order to determine their msp2/p44 genomic repertoires and if differences were primarily located in the known pseudogene hypervariable regions (Figure 3).
Similar to the rodent strain ApJM, the msp2/p44 repertoires of a U.S. human strain, ApHGE1 and the ApDog strain from the same area of Minnesota were closely related and slightly different from the New York human strain ApHZ (8/95 or 9/95 different msp2/p44 pseudogenes determined as <90% identical, Table 1). In contrast, strains infecting horses or ruminants (Ap-variant 1 strains) in the U.S. or Norway (representing “classical” strains long known to be pathogenic to sheep  shared fewer msp2/p44 pseudogenes with human strains, with the two Norwegian sheep strains having almost totally different repertoires from human strains. Differences between individual members of the msp2/p44 repertoire were located in the known central hypervariable region between flanking consensus amino acids LAKT.. LAKT (Figure 3).
|Different msp2/p44 pseudogenes from ApHZ||0/95||8/95||8/95||9/95||49/95||71/95||75/95||92/95||89/95|
The top panel compares strains with ApHZ, the lower panel compares strains and species with AmFL. Abbreviations: ANIm, Average % genome nucleotide identity using Mummer; Tetra, correlation coefficient of tetranucleotide signature frequencies; AmFL, Anaplasma marginale Florida strain; AmStM, A. marginale St. Maries Idaho strain; AcIs, Anaplasma centrale Israel strain; Ap, A. phagocytophilum strains.
Two additional quantitative measures of genome structural relationships compare the average nucleotide identities or the tetranucleotide frequencies between two genomes. The latter method is independent of alignment algorithms. The proposed threshold for prokaryotic species separation is 94% for average nucleotide identity and 0.990 for the correlation coefficient of tetranucleotide signature frequencies . It is evident from Table 1 that the existing taxonomy of Anaplasma species meets these definitions. However, it is apparent that by these measures also, the U.S. human, dog and rodent strains are closely related (average nucleotide identity 98.79%–98.84% with ApHZ). The Norwegian ruminant strains are again most distinct from ApHZ (average nucleotide identity 94.92%–95.87%), agreeing with the similar divergence observed in their pfam01617 superfamily genes.
2.2. Comparison of the msp2/p44 Family among Strains of A. phagocytophilum
Although these data showed that two U.S. A. phagocytophilum strains infecting humans, as well as a dog and rodent strain were similar to one another and different from strains infecting ruminants, we wanted to analyze this more globally using a larger dataset. To do this, we took advantage of the fact that conserved regions of msp2/p44 have been used as sensitive PCR diagnostic targets because of the large copy number of the pfam01617 superfamily. More than 500 partial pseudogene sequences are present in GenBank, from humans, ticks and animal strains derived from multiple regions throughout the U.S., Europe and Asia. From these sequences it is possible to extract the hypervariable region of msp2/p44 flanked by sequence encoding a consensus LAKT on either end, facilitating alignment . Importantly, for this analysis it is necessary to recognize that multiple different hypervariable regions exist in a single genome because of the ~100 distinct pseudogene cassettes present. Therefore, it is not sufficient to simply analyze phylogenetic trees and apparent evolutionary relationships between all msp2/p44 sequences. If one wishes to compare global repertoires with the genome-sequenced human ApHZ strain, it is necessary to align each msp2/p44 sequence with all ~100 ApHZ strain genomic pseudogene cassettes to find the best fit (maximum sequence identity). If strains are related to ApHZ, one expects to find one or more ApHZ pseudogenes with high identity to other strain msp2/p44s. Our analyses, therefore, employed a matrix where every available msp2/p44 polypeptide (661 total including those encoded by ApHZ pseudogene cassettes) was aligned with every other, generating 218,791 alignments having a mean amino acid sequence identity of 48.2%. These percentage sequence identities can be rapidly analyzed using a 661 (row) × 661 (column) spreadsheet. Surprisingly, despite the differences in msp2/p44 repertoires observed previously, all 91 available msp2/p44 human-origin sequences from widely dispersed U.S. locations (states of New York, Massachusetts, Wisconsin and Minnesota) encoded polypeptides with a mean maximum percentage identity with ApHZ of 97.2% (Table 2). Further, in analyzing the available msp2/p44 sequences it was notable that, world-wide, other human-origin strains also achieved similar levels of sequence identity with ApHZ (e.g., 98.1% mean maximum percentage amino acid identity of a dataset from Japan comprising 27 human-origin msp2/p44 variants, no significant difference with ApHZ). In contrast, A. phagocytophilum strains from Europe and Asia from non-human sources had mean maximum identities of <76% (significantly different from ApHZ, Table 2).
|Table 2. Comparison of msp2/p44 variants from different geographic locations with the human ApHZ strain.|
|Country||Source||Number of msp2/p44 Variants Analyzed||Mean Maximum % a.a. Identity with ApHZ (+/−Std.Dev.)||Significantly different from U.S. Human *|
|U.S.A.||Ruminant, tick (Ap-variant 1)||34||86.5 (14.7)||Yes|
|Czech Republic||Human, Roe deer, Perdix, Ixodes ricinus||6||86.6 (14.5)||N/A|
|Japan||Sika deer||17||39.3 (3.0)||Yes|
|Japan||Ixodes persulcatus||87||69.0 (6.9)||Yes|
|Japan||Ixodes ovatus||22||71.8 (15.5)||Yes|
|Japan||Haemaphysalis formosensis||9||75.6 (14)||N/A|
* Kruskal-Wallis One Way Analysis of Variance on Ranks (p ≤ 0.001) followed by Dunn’s method for Multiple Comparisons versus a Control group (p ≤ 0.05). N/A: Not applied to sample sizes <10.
3.1. Origin of A. phagocytophilum Strains
The origins of the A. phagocytophilum strains used in this study are as follows: ApHZ, human, Westchester County, New York; HGE1, human, Minnesota; ApJM, meadow jumping mouse (Zapus hudsonius), Camp Ripley, Minnesota; ApDog, Minnesota; ApCRT35, Camp Ripley tick (Ixodes scapularis), Minnesota (defined previously as Ap-variant 1); ApCRT38-1, Camp Ripley tick (I. scapularis), Minnesota (Ap-variant 1); ApMRK, horse, California; ApNorV1, sheep, Norway; ApNorV2, sheep, Norway. A. phagocytophilum genomic DNA was prepared from in vitro cultured organisms or directly from infected sheep (Norwegian strains), as described previously .
3.2. Ethics Statement
The experimental study in sheep was approved by the Norwegian Animal Research Authority.
3.3. Genome Sequencing and Bioinformatics
Genomic DNA was sequenced on the Roche/454 Genome Sequencer using non-paired and 3 kb paired-end libraries, also as described . Mean genome coverage with respect to ApHZ varied between 31.3X and 72.1X. The ApJM sequence was finished using manual inspection for conflicts and mismatched paired ends and PCR to fill gaps. All sequences were compared with ApHZ for regions of identity using Blastn analysis in CGView [18,19]. Differences in the msp2/p44 repertoires between strains were defined using a method previously validated using the pfam01617 superfamily and two completely Sanger-sequenced genomes of Anaplasma marginale . Briefly, this method uses Mosaik to align individual reads and generate BAM format files to detect gaps in alignment (no coverage) with respect to the annotated reference sequence. In A. marginale, this method detected all different pseudogenes having <90% nucleotide identity with the reference genome. In homologous comparisons between ApHZ Roche/454 reads and the ApHZ genome all msp2/p44 pseudogenes were detected as present (Table 1). Comparisons of average nucleotide identities and the correlation coefficients of tetranucleotide signature frequencies between genomes were conducted using Jspecies software, as described .
3.4. Analysis of msp2/p44 Repertoires
To analyze global msp2/p44 repertoires, deposited msp2/p44 sequences were downloaded from GenBank following Blast searches; several large datasets are also available from published studies [16,21,22,23,24,25]. The sequences were each trimmed to that encoding the hypervariable region, where possible using the consensus flanking LAKT residues, and the polypeptides were aligned (all against all) with MATGAT . This generates the alignments that can be inspected for accuracy, and allows export of all percent identity values to a spreadsheet. This spreadsheet was analyzed in Excel for the relatedness of the different datasets. The percent identity of each variant msp2/p44 sequence with every ApHZ msp2/p44 polypeptide encoded by a pseudogene was reported and, from that, the mean best match with ApHZ (maximum percent identity) determined for each dataset. As the percentage identities were not normally distributed and the population variances were unequal, significant differences between groups were determined by nonparametric methods (Table 2).
These data show that there is genome diversity worldwide within the A. phagocytophilum species that extends close to some proposed guidelines for species discrimination. However, strains infecting humans are a subset with more conserved genome structure than the species overall, and this includes their repertoires of msp2/p44 genes. This subset of A. phagocytophilum strains does not differ significantly from the U.S. ApHZ strain and is closely related to strains infecting U.S. domestic dogs. The mean seroprevalence (using a conserved msp2/p44 peptide as antigen) in the U.S. among 479,640 dogs was 4.8% with prevalence >50% in some counties in the Northeast and Midwest, corresponding to the location of the majority of human cases . These data have significance in several areas. First, msp2/p44 encodes a protein that induces neutralizing antibodies against homologous strains of A. phagocytophilum [28,29]. The repertoire differences between strains reflect an evolving antigenic system with adaptive pressures imposed by an array of different persistently infected hosts. In contrast, humans are incidental dead-end hosts that do not impose significant pressure for change. The similarities between human-origin strains in the U.S., Europe and Asia suggest that humans may not be susceptible to many of the circulating wildlife strains. They may become susceptible when selection pressures in small mammal reservoir hosts cause evolution of novel strains that allow invasion and survival in humans. Rodents are important reservoirs for a multitude of human disease agents, vector-borne or not, and are known to be infected with strains of A. phagocytophilum having different species tropisms . Rodent control has historically been emphasized as a means to control human disease outbreaks. The underlying evolutionary drivers remain to be identified but could reflect the long-standing association between “men and mice”. Second, these data have practical significance for control of this emerging infection. Despite the ubiquity of A. phagocytophilum strains and positive serology world-wide, it is necessary to focus on the prevalence and transmission of a smaller, human-infective subset. Differentiation of such strains from the high-prevalence background may be possible using genomic PCR targets or selected msp2/p44 polypeptides for serology. In order to accomplish this goal it is necessary to acquire genome sequences from multiple human-origin strains on different continents. The technology now exists to do this.
We thank Anna M. Lundgren for technical support, Savita Shanker for high-throughput DNA sequencing, David Allred and Dan Brown for reading the manuscript and to Dan Brown for suggesting the use of Jspecies software. This investigation received support from grants RO1 GM081714 and GM081714-03S1.
Draft and complete sequences obtained in this project are being made available through the NCBI Bioproject portal  as bioprojects PRJNA158483, 163167, 163169, 171710, 183838, 216999, 217002, 217033, and 217037.
Conflicts of Interest
The authors declare no conflict of interest.
- Patz, J.A.; Olson, S.H.; Uejio, C.K.; Gibbs, H.K. Disease emergence from global climate and land use change. Med. Clin. North Am. 2008, 92, 1473–1491. [Google Scholar] [CrossRef]
- Dumler, J.S.; Choi, K.S.; Garcia-Garcia, J.C.; Barat, N.S.; Scorpio, D.G.; Garyu, J.W.; Grab, D.J.; Bakken, J.S. Human granulocytic anaplasmosis and Anaplasma phagocytophilum. Emerg. Infect. Dis. 2005, 11, 1828–1834. [Google Scholar] [CrossRef]
- CDC. Statistics and Epidemiology. Annual Cases of Anaplasmosis in the United States, Available online: www.cdc.gov/anaplasmosis/stats/ (accessed on 5 September 2013).
- Dahlgren, F.S.; Mandel, E.J.; Krebs, J.W.; Massung, R.F.; McQuiston, J.H. Increasing incidence of Ehrlichia chaffeensis and Anaplasma phagocytophilum in the United States, 2000–2007. Am. J. Trop. Med. Hyg. 2011, 85, 124–131. [Google Scholar] [CrossRef]
- Zhang, L.; Wang, G.; Liu, Q.; Chen, C.; Li, J.; Long, B.; Yu, H.; Zhang, Z.; He, J.; Qu, Z.; et al. Molecular analysis of Anaplasma phagocytophilum isolated from patients with febrile diseases of unknown etiology in China. PLoS One 2013, 8, e57155. [Google Scholar]
- Annen, K.; Friedman, K.; Eshoa, C.; Horowitz, M.; Gottschall, J.; Straus, T. Two cases of transfusion-transmitted Anaplasma phagocytophilum. Am. J. Clin. Pathol. 2012, 137, 562–565. [Google Scholar] [CrossRef]
- Jereb, M.; Pecaver, B.; Tomazic, J.; Muzlovic, I.; Avsic-Zupanc, T.; Premru-Srsen, T.; Levicnik-Stezinar, S.; Karner, P.; Strle, F. Severe human granulocytic anaplasmosis transmitted by blood transfusion. Emerg. Infect. Dis. 2012, 18, 1354–1357. [Google Scholar]
- Massung, R.F.; Priestley, R.A.; Miller, N.J.; Mather, T.N.; Levin, M.L. Inability of a variant strain of Anaplasma phagocytophilum to infect mice. J. Infect. Dis. 2003, 188, 1757–1763. [Google Scholar] [CrossRef]
- Massung, R.F.; Mather, T.N.; Levin, M.L. Reservoir competency of goats for the Ap-variant 1 strain of Anaplasma phagocytophilum. Infect. Immun. 2006, 74, 1373–1375. [Google Scholar] [CrossRef]
- Dunning Hotopp, J.C.; Lin, M.; Madupu, R.; Crabtree, J.; Angiuoli, S.V.; Eisen, J.A.; Seshadri, R.; Ren, Q.; Wu, M.; Utterback, T.R.; et al. Comparative genomics of emerging human ehrlichiosis agents. PLoS Genet. 2006, 2, e21. [Google Scholar] [CrossRef]
- Barbet, A.F.; Meeus, P.F.; Bélanger, M.; Bowie, M.V.; Yi, J.; Lundgren, A.M.; Alleman, A.R.; Wong, S.J.; Chu, F.K.; Munderloh, U.G.; et al. Expression of multiple outer membrane protein sequence variants from a single genomic locus of Anaplasma phagocytophilum. Infect. Immun. 2003, 71, 1706–1718. [Google Scholar] [CrossRef]
- Lin, Q.; Rikihisa, Y. Establishment of cloned Anaplasma phagocytophilum and analysis of p44 gene conversion within an infected horse and infected SCID mice. Infect. Immun. 2005, 73, 5106–5114. [Google Scholar] [CrossRef]
- Foley, J.E.; Nieto, N.C.; Barbet, A.F.; Foley, P. Antigen diversity in the parasitic bacterium Anaplasma phagocytophilum arises from selectively-represented, spatially clustered functional pseudogenes. PLoS One 2009, 4, e8265. [Google Scholar]
- Stuen, S. Anaplasma phagocytophilum—The most widespread tick-borne infection in animals in Europe. Vet. Res. Commun. 2007, 31, 79–84. [Google Scholar] [CrossRef]
- Richter, M.; Rosselló-Móra, R. Shifting the genomic gold standard for the prokaryotic species definition. Proc. Natl. Acad. Sci. USA 2009, 106, 19126–19131. [Google Scholar] [CrossRef]
- Barbet, A.F.; Lundgren, A.M.; Alleman, A.R.; Stuen, S.; Bjöersdorff, A.; Brown, R.N.; Drazenovich, N.L.; Foley, J.E. Structure of the expression site reveals global diversity in MSP2 (P44) variants in Anaplasma phagocytophilum. Infect. Immun. 2006, 74, 6429–6437. [Google Scholar] [CrossRef]
- Al-Khedery, B.; Lundgren, A.M.; Stuen, S.; Granquist, E.G.; Munderloh, U.G.; Nelson, C.M.; Alleman, A.R.; Mahan, S.M.; Barbet, A.F. Structure of the type IV secretion system in different strains of Anaplasma phagocytophilum. BMC Genomics 2012, 13, e678. [Google Scholar] [CrossRef]
- Grant, J.R.; Stothard, P. The CGView Server: A comparative genomics tool for circular genomes. Nucleic Acids Res. 2008, 36, W181–W184. [Google Scholar] [CrossRef]
- Grant, J.R.; Stothard, P. Comparing thousands of circular genomes using the CGView Comparison Tool. BMC Genomics 2012, 13, e202. [Google Scholar] [CrossRef]
- Dark, M.J.; Al-Khedery, B.; Barbet, A.F. Multistrain genome analysis identifies candidate vaccine antigens of Anaplasma marginale. Vaccine 2011, 29, 4923–4932. [Google Scholar] [CrossRef]
- Casey, A.N.J.; Birtles, R.J.; Radford, A.D.; Bown, K.J.; French, N.P.; Woldehiwet, Z.; Ogden, N.H. Groupings of highly similar major surface protein (p44)-encoding paralogues: A potential index of genetic diversity amongst isolates of Anaplasma phagocytophilum. Microbiology 2004, 150, 727–734. [Google Scholar] [CrossRef]
- Gaowa; Wuritu; Wu, D.; Yoshikawa, Y.; Ohashi, N.; Kawamori, F.; Sugiyama, K.; Ohtake, M.; Ohashi, M.; Yamamoto, S.; et al. Dection and characterization of p44/msp2 transcript variants of Anaplasma phagocytophilum from naturally infected ticks and wild deer in Japan. Jpn. J. Infect. Dis. 2012, 65, 79–83. [Google Scholar]
- Wuritu; Gaowa; Kawamori, F.; Aochi, M.; Masuda, T.; Ohashi, N. Characterization of p44/msp2 multigene family of Anaplasma phagocytophilum from two different tick species, Ixodes persulcatus and Ixodes ovatus, in Japan. Jpn. J. Infect. Dis. 2009, 62, 142–145. [Google Scholar]
- Ohashi, N.; Gaowa; Wuritu; Kawamori, F.; Wu, D.; Yoshikawa, Y.; Chiya, S.; Fukunaga, K.; Funato, T.; Shiojiri, M.; et al. Human granulocytic anaplasmosis, Japan. Emerg. Infect. Dis. 2013, 19, 289–292. [Google Scholar] [CrossRef]
- Caspersen, K.; Park, J.H.; Patil, S.; Dumler, J.S. Genetic variability and stability of Anaplasma phagocytophila msp2 (p44). Infect. Immun. 2002, 70, 1230–1234. [Google Scholar] [CrossRef]
- Campanella, J.J.; Bitincka, L.; Smalley, J. MatGAT: An application that generates similarity/identity matrices using protein or DNA sequences. BMC Bioinformatics 2003, 4, e29. [Google Scholar] [CrossRef]
- Bowman, D.; Little, S.E.; Lorentzen, L.; Shields, J.; Sullivan, M.P.; Carlin, E.P. Prevalence and geographic distribution of Dirofilaria immitis, Borrelia burgdorferi, Ehrlichia canis, and Anaplasma phagocytophilum in dogs in the United States: Results of a national clinic-based serologic survey. Vet. Parasitol. 2009, 160, 138–148. [Google Scholar] [CrossRef]
- Kim, H.Y.; Rikihisa, Y. Characterization of monoclonal antibodies to the 44-kilodalton major outer membrane protein of the human granulocytic ehrlichiosis agent. J. Clin. Microbiol. 1998, 36, 3278–3284. [Google Scholar]
- Wang, X.; Kikuchi, T.; Rikihisa, Y. Two monoclonal antibodies with defined epitopes of P44 major surface proteins neutralize Anaplasma phagocytophilum by distinct mechanisms. Infect. Immun. 2006, 74, 1873–1882. [Google Scholar] [CrossRef]
- Foley, J.E.; Nieto, N.C.; Massung, R.; Barbet, A.; Madigan, J.; Brown, R.N. Distinct ecologically relevant strains of Anaplasma phagocytophilum. Emerg. Infect. Dis. 2009, 15, 842–843. [Google Scholar] [CrossRef]
- The NCBI Bioproject Portal. Available online: www.ncbi.nlm.nih.gov/bioproject (accessed on 5 September 2013).
© 2013 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).