Fibrinogen production is enhanced in an in-vitro model of non-alcoholic fatty liver disease: an isolated risk factor for cardiovascular events?
- Emily N. W. Yeung†1Email author,
- Philipp Treskes†1,
- Sarah F. Martin2,
- Jonathan R. Manning3,
- Donald R. Dunbar3,
- Sophie M. Rogers2,
- Thierry Le Bihan2,
- K. Ann Lockman1,
- Steven D. Morley1,
- Peter C. Hayes1,
- Leonard J. Nelson1 and
- John N. Plevris1
© Yeung et al. 2015
Received: 22 April 2015
Accepted: 29 June 2015
Published: 10 August 2015
The Erratum to this article has been published in Lipids in Health and Disease 2016 15:125
Cardiovascular disease (CVD) remains the major cause of excess mortality in patients with non-alcoholic fatty liver disease (NAFLD). The aim of this study was to investigate the individual contribution of NAFLD to CVD risk factors in the absence of pathogenic influences from other comorbidities often found in NAFLD patients, by using an established in-vitro model of hepatic steatosis.
Histopathological events in non-alcoholic fatty liver disease were recapitulated by focused metabolic nutrient overload of hepatoblastoma C3A cells, using oleate-treated-cells and untreated controls for comparison. Microarray and proteomic data from cell culture experiments were integrated into a custom-built systems biology database and proteogenomics analysis performed. Candidate genes with significant dysregulation and concomitant changes in protein abundance were identified and STRING association and enrichment analysis performed to identify putative pathogenic pathways.
The search strategy yielded 3 candidate genes that were specifically and significantly up-regulated in nutrient-overloaded cells compared to untreated controls: fibrinogen alpha chain (2.2 fold), fibrinogen beta chain (2.3 fold) and fibrinogen gamma chain (2.1 fold) (all rank products pfp <0.05). Fibrinogen alpha and gamma chain also demonstrated significant concomitant increases in protein abundance (3.8-fold and 2.0-fold, respectively, p <0.05).
In-vitro modelling of NAFLD and reactive oxygen species formation in nutrient overloaded C3A cells, in the absence of pathogenic influences from other comorbidities, suggests that NAFLD is an isolated determinant of CVD. Nutrient overload-induced up-regulation of all three fibrinogen component subunits of the coagulation cascade provides a possible mechanism to explain the excess CVD mortality observed in NAFLD patients.
KeywordsSystems biology Proteomics Microarray STRING Non-alcoholic fatty liver disease Cardiovascular disease Metabolic syndrome C3A cells
Non-Alcoholic Fatty Liver Disease (NAFLD) has been traditionally regarded as the consequence of a high-fat western diet and sedentary lifestyle [1, 2]. Its increasing worldwide prevalence, however, suggests that NAFLD is more than a lifestyle disease. Studies have identified susceptibility genes and genetic polymorphisms that associate with development and severity of NAFLD . This may explain the differences in demographic data observed in Asia compared to elsewhere in the world, including a younger age distribution and a higher proportion of patients who are judged lean by Body Mass Index, but show an altered metabolic profile associated with obesity [4, 5].
Cardiovascular disease (CVD) remains one of the major causes of excess mortality in NAFLD patients [6–8]. The term “Metabolic Syndrome” (MetS) describes a clinical cluster of diseases that tend to aggregate in individuals over time, including central obesity, hypertension, impaired fasting glucose and dyslipidemia [9–11]. Together these components create a pro-atherogenic environment that is postulated to accelerate atherosclerosis and increase the risk of cardiovascular diseases. Whilst the definition of MetS does not presently include NAFLD, the association of atherosclerotic markers, such as carotid artery wall thickness, with NAFLD has been reported previously [12, 13]. Further attempts to understand the possible causative role of NAFLD in CVD have used statistical models to exclude atherogenic contributions from traditional CVD risk factors and other components of MetS [14, 15]. However, to date, there are no experimental genomic or proteomic studies that examine specifically whether NAFLD is an isolated risk factor for CVD in absence of other components of MetS, or if these conditions have a common cause.
Previously, we developed an in vitro model of cellular steatosis by exposing hepatoblastoma C3A cells to nutrient overload (treatment with lactate, pyruvate, octanoate and ammonia), which reproduces the characteristic pathophysiological changes found in NAFLD, including intracellular triglyceride accumulation and reactive oxygen species (ROS) formation . This model allows the opportunity to assess the individual contribution of NAFLD to CVD risk factors in the absence of pathogenic influences from other comorbidities often found in NAFLD patients.
In the present study, changes in hepatoblastoma C3A gene transcription and protein expression in response to cellular steatosis and ROS formation induced specifically by nutrient overload were assembled into a custom-built data portal allowing evaluation of integrated transcriptomic and proteomic data to identify gene products potentially involved in pathogenic pathways. Candidates showing consistently greater than two fold alterations in specific nutrient overload-mediated gene transcription and protein expression were subjected to further analysis by the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) v9.1 database (http://string-db.org) and enrichment analysis to identify predicted functional partners and pathogenic pathways contributing potentially to a pro-atherogenic environment.
Cell culture, treatment and sample collection
Hepatoblastoma C3A cells (ATCC® CRL-10741TM, LGC Standards, Teddington, UK) were cultured as previously described . Briefly, cell cultures were treated either with the combination of lactate, pyruvate, octanoate and ammonia (LPON), oleate (OLE), or untreated controls. Both octanoate and oleate readily diffuse into mitochondria to promote efficient ß-oxidation and lipid accumulation, but while OLE treatment causes simple cellular steatosis, LPON treatment induces both ROS formation and mitochondrial dysfunction, in addition to steatosis, typically seen in NAFLD . C3A cells were treated in three separate experiments either with LPON, OLE, or untreated controls and processed for transcriptomic or proteomic analysis as described in the following sections.
Sample preparation and transcriptomics
Cells were washed twice in cold PBS and transferred to cold RNALater® (Life Technologies, Paisley, UK) for overnight incubation at 4 °C. Afterwards, RNA was isolated with an RNAqueous®-4PCR kit (Life Technologies) and subsequently amplified and biotinylated with an Illumina® TotalPrep RNA Amplification kit (Life Technologies), following the manufacturer’s instructions. RNA expression was measured by hybridization to the Illumina® Whole Human Genome BeadChip H12 Microarray (Illumina United Kingdom, Essex, UK). Data were extracted through the GCOS software (Affymetrix UK Ltd., High Wycombe, UK). CELfiles were used for additional data processing and imported into Bioconductor  to examine differences between LPON- and OLE-treated groups and untreated controls. Data were normalized by robust multi-array average (RMA) in the Oligo module (http://www.bioconductor.org/packages/2.0/bioc/html/oligo.html). Gene ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed with the DAVID tool [18, 19] on genes that were significantly differentially expressed. Data was statistically analyzed with the bioconductor Limma package .
Sample preparation and proteomics
Protein extraction was performed as previously described . Briefly, samples were denatured in 8 M urea, reduced by incubating with dithiothreithol prior to cysteine alkylation with iodoacetamide and overnight digestion with 8 μg trypsin. Protein concentrations were estimated by Bradford protein assay (Thermo Scientific, Rockford, IL, USA) on a 10 μl sample, diluted to 2 M Urea and quantified against a BSA standard curve. 4 μg peptide samples were acidified (1 % formic acid), centrifuged and cleaned using Stagetips , dried by SpeedVac, and stored at −20 °C.
2 μg peptide samples were analysed in a randomised sequence by capillary-HPLC- MSMS as described previously , using an on-line system consisting of a micro-pump (1200 binary HPLC system, Agilent, UK) coupled to a hybrid LTQ-Orbitrap XL instrument (Thermo-Fisher, Leicestershire, UK). Acetonitrile and water were HPLC quality (Fisher). Formic acid was Suprapure 98-100 % (Merck, Darmstadt, Germany) and trifluoroacetic acid was 99 % purity sequencing grade. LC-MSMS label-free quantification was performed using Progenesis 4.0 (Nonlinear Dynamics, Newcastle upon Tyne, UK) as described previously . Multicharged (2+,3+,4+) ion intensities were extracted from LC-MS files and MSMSdata were searched using Mascot Version 2.4 (Matrix Science Ltd, London, UK) against the Homo Sapiens subset of the NCBI protein database (12/01/2011; 34,281 sequences) using a maximum missed-cut value of 2, variable oxidation (M), N-terminal protein acetylation and fixed carbamidomethylation (C); precursor mass tolerance was 7 ppm and MSMS tolerance 0.4 Da. The significance threshold (p) was < 0.05 (MudPIT scoring). A minimum peptide cut off score of 20 was set, corresponding to <3 % global false discovery rate (FDR) using a decoy database search. Proteins identified and quantified with two or more peptide sequences were retained. A two-tailed t-test for independent samples or biological triplicates was performed on arcsinh transformed, normally distributed intensity data.
Data mining and candidate gene identification
STRING Network Associations
Predicted functional partners of candidate genes induced specifically in response to nutrient overload were identified using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) v9.1 database (http://string-db.org). Only interactions in Homo sapiens with a probabilistic confidence score ≥0.900, corresponding to a "highest-confidence" network, were considered in this study. STRING PFPs were cross-validated against the original transcriptomic and proteomic data for differential expression and protein abundance, using similar criteria (>2.0 fold increase in LPON-treated cells, excluding PFPs with >2.0 fold increase in OLE-treated cells: see Fig. 1) in order to identify candidate genes that had otherwise been excluded by the more stringent primary search strategy.
To identify dysregulated pathways and biological processes contributing potentially to a pathogenic phenotype, candidate genes differentially expressed in response to nutrient overload and their PFPs were grouped according to their functional annotations with data from the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) database via the Gene Ontology enRIchment anaLysis and visuaLizAtion (GOrilla) tool (http://cbl-gorilla.cs.technion.ac.il) [25, 26]. Only dysregulations with p < 0.05 were considered in this study, with p-values being corrected for multiple testing using the Benjamini and Hochberg method . Enrichment was based on gene ranking, which was indicated by the STRING analysis confidence score.
Microarray data for candidate genes specifically up-regulated by nutrient overload
Entrez gene ID
Max Mean value
Max Mean value
Direction of change wrt control
Max Mean value
Direction of change wrt control
Max Mean value
farnesyl-diphosphate farnesyltransferase 1
Fibrinogen alpha chain
Proteomics data for candidate genes specifically up-regulated by nutrient overload
Farnesyl-diphosphate farnesyltransferase 1
Fibrinogen alpha chain
Fibrinogen alpha chain (FGA)
FGA gene expression was significantly higher in LPON-treated cells compared to untreated controls by 2.3 fold (rank products pfp = 0.0002). The FGA protein product was also more abundant in LPON-treated cells (3.8 fold change, p = 0.009).
Microarray data for predicted functional partners of FGA up-regulated by nutrient overload
Entrez gene ID
Maximum mean value
Maximum mean value
Direction of change wrt control
Maximum mean value
Direction of change wrt control
Fibrinogen gamma chain
Fibrinogen beta chain
Proteomics data for predicted functional partners of FGA up-regulated by nutrient overload
Fibrinogen gamma chain
Enrichment analysis based on the number of predicted functional partners of FGA indicated by STRING analysis
Enrichment (N, B, n, b)
Response to calcium ion
Secretion by cell
Farnesyl-diphosphate farnesyltransferase 1 (FDFT1)
Gene expression of FDFT1, which catalyses the first committed step in sterol synthesis on the pathway to cholesterol, was up-regulated by 2.4 fold (rank products pfp = 0) in LPON-treated cells compared to untreated controls. A concomitant increase in protein abundance by 3.2 fold (p = 0.02) in LPON-treated cells was also recorded. 10 STRING PFPs of FDFT1 in Homo sapiens with a probabilistic confidence score ≥0.900 were identified (Supplementary material; Additional File 2); however none of these genes satisfied the criteria of >2.0 fold change in gene expression (Rank products pfp <0.05) when comparing LPON-treated cells to untreated controls.
CVD remains a major cause of excess mortality in NAFLD patients [6–8]. In the present study, we sought to clarify the possible causative role of NAFLD in CVD by identifying gene and protein dysregulation using an in vivo model of cellular steatosis. Use of an in vitro model is particularly relevant as it enables experimental strategies to be designed to exclude pathogenic influences from conditions, such as diabetes mellitus and hypertension, which commonly co-exist with NAFLD .
Difference between OLE and LPON-treated cells
Up-regulation in the genetic expressions of all 3 fibrinogen chains was consistently more dramatic in LPON-treated compared OLE-treated cells, the latter of which models simple cellular steatosis. In contrast, LPON treatment additionally induced ROS formation and mitochondrial dysfunction  Thus, we hypothesize that dysregulation of gene expression in LPON-treated cells was made more dramatic by oxidative stress due to free radical formation.
Previous cross-sectional studies have demonstrated increased blood fibrinogen concentrations in response to mild pulmonary inflammation in healthy humans caused by diesel exhaust particulates, a major constituent of city air pollution [36, 37]. In a previous double-blinded randomised crossover study, diesel exhaust inhalation has also been shown to mediate excess CVD risk caused by air pollution by increasing thrombus formation following platelet activation secondary to ROS formation and oxidative stress . This supports the notion that multiple mechanisms intersecting at the level of ROS formation may contribute to excess CVD risk. However, while ROS-formation in C3A cells after exposure to nanoparticles has been confirmed in our laboratory, such observations should be taken with caution as the mechanism of free radical generation may be different in particulate-induced and steatosis-induced conditions . Thus, while ROS formation observed after inhalation of diesel exhaust arises from the alteration of redox potential by particulates and inflammatory cells , in our model of NAFLD, ROS are formed endogenously due to nutrient excess .
In conclusion, this study presents evidence supporting the hypothesis of NAFLD being an isolated determinant of CVD, as use of a proven in vitro model of NAFLD eliminates the confounding influences of other comorbidities found in NAFLD patients. This supports the notion that the liver is a source of pro-coagulant molecules which contributes to the risk of thrombosis and enhanced CVD risk in the NAFLD population. However, in using an in vitro model examines only one aspect of the concomitant pathophysiological mechanisms operating in vivo. Other factors operating in concert may play significant roles in modifying the systemic response to the observed cellular effects, for example the role of visceral fat as a systemic pro-inflammatory signal in CVD risk in NAFLD patients. Further work is required to elucidate the pathophysiological mechanisms explaining the association between free radical-induced oxidative stress and CVD in a NAFLD setting. Clinical evidence could be sought by measuring the plasma levels of FGA, FGB and FGG in NAFLD patients, to investigate potential relationships between plasma fibrinogen levels and severity of CVD determined by conventional atherosclerotic markers, such as carotid artery wall thickness. Indeed, verification of FGA, FGB and FGG as early plasma predictors of CVD development would provide a rationale to explore the possibility of prophylactic fibrinolytic therapy in a NAFLD context, similar to that already used in myocardial infarction .
This work was funded internally, from departmental funds of the Hepatology Laboratory, Division of Health Sciences, University of Edinburgh. The decision to submit the article for publication was not influenced by the funding body.
- Browning JD, Szczepaniak LS, Dobbins R, Nuremberg P, Horton JD, Cohen JC, et al. Prevalence of hepatic steatosis in an urban population in the United States: impact of ethnicity. Hepatology. 2004;40:6.View ArticleGoogle Scholar
- Feldstein AE. Novel insights into the pathophysiology of nonalcoholic fatty liver disease. Semin Liver Dis. 2010;30:4.View ArticleGoogle Scholar
- Farrell GC, Wong VWS, Chitturi S. NAFLD in Asia – as common and important as in the West. Nat Rev Gastrolenterol Hepatol. 2013;10:5.Google Scholar
- Denis GV, Hamilton JA. Healthy obese persons: How can they be identified and do metabolic profiles stratify risk? Curr Opin Endocrinol Diabetes Obes. 2013;20:5.View ArticleGoogle Scholar
- Liu CJ. Prevalence and risk factors for NAFLD in Asian people who are not obese. J Gastroenterol Hepatol. 2012;27:10.Google Scholar
- Adams LA, Lymp JF, St Sauver J, Sanderson SO, Lindor KD, Feldstein A, et al. The natural history of nonalcoholic fatty liver disease: a population- based cohort study. Gastroenterology. 2005;129:1.View ArticleGoogle Scholar
- Harrison SA. Day CP Benefits of lifestyle modification in NAFLD. Gut. 2007;56:12.View ArticleGoogle Scholar
- Preiss D, Sattar N. Non-alcoholic fatty liver disease: an overview of prevalence, diagnosis, pathogenesis and treatment considerations. Clin Sci (Lond). 2008;115:5.View ArticleGoogle Scholar
- Eckel RH, Grundy SM, Zimmet PZ. The metabolic syndrome. Lancet. 2005;365:9468.View ArticleGoogle Scholar
- Ekstedt M, Franzen LE, Mathiesen UL, Thorelius L, Holmqvist M, Bodemar G, et al. Long-term follow-up of patients with NAFLD and elevated liver enzymes. Hepatology. 2006;44:4.View ArticleGoogle Scholar
- Soderberg C, Stal P, Askling J, Glaumann H, Lindberg G, Marmur J, et al. Decreased survival of subjects with elevated liver function tests during a 28- year follow-up. Hepatology. 2010;51:2.View ArticleGoogle Scholar
- Anstee QM, Targher G, Day CP. Progression of NAFLD to diabetes mellitus, cardiovascular disease or cirrhosis. Nat Rev Gastroenterol Hepatol. 2013;10:6.Google Scholar
- Targher G, Bertolini L, Padovani R, Rodella S, Zoppini R, Zenari L, et al. Relations between carotid artery wall thickness and liver histology in subjects with nonalcoholic fatty liver disease. Diabetes Care. 2006;29:6.Google Scholar
- Choi SY, Kim D, Kang JH, Park MJ, Kim YS, Lim SH, et al. Nonalcoholic fatty liver disease as a risk factor of cardiovascular disease: relation of non-alcoholic fatty liver disease to carotid atherosclerosis]. Korean J Hepatol. 2008;14:1.View ArticleGoogle Scholar
- Hamaguchi M, Kojima T, Takeda N, Nagata C, Takeda J, Sarui H, et al. Nonalcoholic fatty liver disease is a novel predictor of cardiovascular disease. World J Gastroenterol. 2007;13:10.View ArticleGoogle Scholar
- Lockman KA, Baren JP, Pemberton CJ, Baghdadi H, Burgess KE, Plevris-Papaioannou N, et al. Oxidative stress rather than triglyceride accumulation is a determinant of mitochondrial dysfunction in in vitro models of hepatic cellular steatosis. Liver Int. 2012;32:7.View ArticleGoogle Scholar
- Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol. 2004;5:10.View ArticleGoogle Scholar
- da Huang W, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009;4:1.Google Scholar
- Dennis Jr G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, et al. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 2003;4:5.Google Scholar
- Smyth GK. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol. 2004;3:Article3.PubMedGoogle Scholar
- Le Bihan T, Martin SF, Chirnside ES, van Ooijen G, Barrios-Llerena ME, O'Neill JS, et al. Shotgun proteomic analysis of the unicellular alga Ostreococcus tauri. J Proteomics. 2011;74:10.View ArticleGoogle Scholar
- Rappsilber J, Mann M, Ishihama Y. Protocol for micro-purification, enrichment, pre-fractionation and storage of peptides for proteomics using StageTips. Nat Protoc. 2007;2:8.View ArticleGoogle Scholar
- Martin SF, Munagapati VS, Salvo-Chirnside E, Kerr LE, Le Bihan T. Proteome turnover in the green alga Ostreococcus tauri by time course 15 N metabolic labeling mass spectrometry. J Proteome Res. 2012;11:1.View ArticleGoogle Scholar
- Le Bihan T, Grima R, Martin S, Forster T, Le Bihan Y. Quantitative analysis of low-abundance peptides in HeLa cell cytoplasm by targeted liquid chromatography/mass spectrometry and stable isotope dilution: emphasising the distinction between peptide detection and peptide identification. Rapid Commun Mass Spectrom. 2010;24:7.View ArticleGoogle Scholar
- Eden E, Lipson D, Yogev S, Yakhini Z. Discovering motifs in ranked lists of DNA sequences. PLoS computational biology. 2007;3:3.View ArticleGoogle Scholar
- Eden E, Navon R, Steinfeld I, Lipson D, Yakhini Z. GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC bioinformatics. 2009;10:1.View ArticleGoogle Scholar
- Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. 1995;Series B:Methodological.Google Scholar
- Adams LA, Waters OR, Knuiman MW, Elliott RR, Olynyk JK. NAFLD as a risk factor for the development of diabetes and the metabolic syndrome: an eleven-year follow-up study. The American journal of gastroenterology. 2009;104:4.View ArticleGoogle Scholar
- Kaptoge S, Di Angelantonio E, Pennells L, Wood AM, White IR, Gao P, et al. C-reactive protein, fibrinogen, and cardiovascular disease prediction. N Engl J Med. 2012;367:14.View ArticleGoogle Scholar
- Lisman T, Weeterings C, de Groot PG. Platelet aggregation: involvement of thrombin and fibrin(ogen). Front Biosci. 2005;10:2504.View ArticlePubMedGoogle Scholar
- Danesh J, Lewington S, Thompson SG, Lowe GD, Collins R, Kostis JB, et al. Plasma fibrinogen level and the risk of major cardiovascular diseases and nonvascular mortality: an individual participant meta-analysis. Jama. 2005;294:14.Google Scholar
- Ndrepepa G, Braun S, King L, Fusaro M, Keta D, Cassese S, et al. Relation of fibrinogen level with cardiovascular events in patients with coronary artery disease. Am J Cardiol. 2013;111:6.View ArticleGoogle Scholar
- Hickman IJ, Sullivan CM, Flight S, Campbell C, Crawford DH, Masci PP, et al. Altered clot kinetics in patients with non-alcoholic fatty liver disease. Ann Hepatol. 2009;8:4.Google Scholar
- Mertens I, Van Gaal LF. Obesity, haemostasis and the fibrinolytic system. Obes Rev. 2002;3:2.View ArticleGoogle Scholar
- Russo I. The Prothrombotic Tendency in Metabolic Syndrome: Focus on the Potential Mechanisms Involved in Impaired Haemostasis and Fibrinolytic Balance. Scientifica (Cairo). 2012;2012:525374.Google Scholar
- Ghio AJ, Kim C, Devlin RB. Concentrated ambient air particles induce mild pulmonary inflammation in healthy human volunteers. Am J Respir Crit Care Med. 2000;162:3.View ArticleGoogle Scholar
- Pekkanen J, Brunner EJ, Anderson HR, Tiittanen P, Atkinson RW. Daily concentrations of air pollution and plasma fibrinogen in London. Occup Environ Med. 2000;57:12.View ArticleGoogle Scholar
- Lucking AJ, Lundback M, Mills NL, Faratian D, Barath SL, Pourazar J, et al. Diesel exhaust inhalation increases thrombus formation in man. Eur Heart J. 2008;29:24.View ArticleGoogle Scholar
- Filippi C, Pryde A, Cowan P, Lee T, Hayes P, Donaldson K, et al. Toxicology of ZnO and TiO2 nanoparticles on hepatocytes: Impact on metabolism and bioenergetics. Nanotoxicoloty. 2015;9:1.Google Scholar
- Tornqvist H, Mills NL, Gonzalez M, Miller MR, Robinson SD, Megson IL, et al. Persistent endothelial dysfunction in humans after diesel exhaust inhalation. Am J Respir Crit Care Med. 2007;176:4.View ArticleGoogle Scholar
- Buczko W, Mogielnicki A, Kramkowski K, Chabielska E. Aspirin and the fibrinolytic response. Thrombosis Research. 2003;110:5–6.View ArticleGoogle Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.