Increasing gene discovery and coverage using RNA-seq of globin RNA reduced porcine blood samples
- Igseo Choi†1,
- Hua Bao†2,
- Arun Kommadath†2,
- Afshin Hosseini2, 4,
- Xu Sun2,
- Yan Meng2,
- Paul Stothard2,
- Graham S Plastow2,
- Christopher K Tuggle3,
- James M Reecy3,
- Eric Fritz-Waters3,
- Samuel M Abrams1,
- Joan K Lunney1Email author and
- Le Luo Guan2Email author
© Choi et al.; licensee BioMed Central Ltd. 2014
Received: 1 October 2014
Accepted: 16 October 2014
Published: 4 November 2014
Transcriptome analysis of porcine whole blood has several applications, which include deciphering genetic mechanisms for host responses to viral infection and vaccination. The abundance of alpha- and beta-globin transcripts in blood, however, impedes the ability to cost-effectively detect transcripts of low abundance. Although protocols exist for reduction of globin transcripts from human and mouse/rat blood, preliminary work demonstrated these are not useful for porcine blood Globin Reduction (GR). Our objectives were to develop a porcine specific GR protocol and to evaluate the GR effects on gene discovery and sequence read coverage in RNA-sequencing (RNA-seq) experiments.
A GR protocol for porcine blood samples was developed using RNase H with antisense oligonucleotides specifically targeting porcine hemoglobin alpha (HBA) and beta (HBB) mRNAs. Whole blood samples (n = 12) collected in Tempus tubes were used for evaluating the efficacy and effects of GR on RNA-seq. The HBA and HBB mRNA transcripts comprised an average of 46.1% of the mapped reads in pre-GR samples, but those reads reduced to an average of 8.9% in post-GR samples. Differential gene expression analysis showed that the expression level of 11,046 genes were increased, whereas 34 genes, excluding HBA and HBB, showed decreased expression after GR (FDR <0.05). An additional 815 genes were detected only in post-GR samples.
Our porcine specific GR primers and protocol minimize the number of reads of globin transcripts in whole blood samples and provides increased coverage as well as accuracy and reproducibility of transcriptome analysis. Increased detection of low abundance mRNAs will ensure that studies relying on transcriptome analyses do not miss information that may be vital to the success of the study.
Blood is a valuable resource to probe an animal’s physiological and pathological status as well as to obtain repeated samples before harvest, for example, monitoring the dynamic change of gene expression in response to disease, treatment, or aging, for which the onset of gene expression response is not known. However, transcriptomic analysis of blood samples is a challenge since blood is composed of heterogeneous cell types including red blood cells (99%), platelets (1%) and white blood cells (<1%; e.g., neutrophils, monocytes, basophils, lymphocytes and eosinophils) [1, 2]. In human blood, HBA and HBB are the most abundant transcripts (~52-76%) [3, 4]. The high level of globin transcripts in blood was reported to be the most limiting factor for accurate and sensitive detection of gene expression, especially for the less abundant transcripts [3–5]. This issue is a great concern for sequence-based approaches, in which the globin transcripts will be highly abundant and limit the potential coverage and detection of other transcripts from blood .
To date, several globin RNA reduction protocols have been successfully applied to gene expression studies in human [6–9]. GLOBINclearTM (Ambion, Austin, TX, USA), a commercial product widely used in human clinical research, removes up to 95% of the HBA and HBB transcripts in human whole blood samples and improves the efficacy of gene expression assays [4, 10, 11]. Further approaches developed by Affymetrix (Affymetrix Inc., Santa Clara, CA, USA) [5, 11] or PNA Bio Inc. (Thousand Oaks, CA, USA) [9, 10] also have differential reduction rates of globin transcripts in human blood. Globin RNA reduction improved the sensitivity and reproducibility of high throughput mRNA expression analysis of whole human blood samples [3–5, 7, 9, 10]. There is, however, neither a commercial GR product available nor any literature demonstrating the efficiency and effects of GR at global level for porcine whole blood .
Our objectives were to develop a porcine specific GR protocol and to evaluate the effects of GR treatment on gene discovery and coverage in RNA-seq experiments for swine.
Results and discussion
Comparisons of globin reduction methods
Porcine specific globin oligonucleotides used in RNase H-mediated globin reduction assay
Final Conc. (10X)
5′- GAT CTC CGA GGC TCC AGC TTA ACG GT -3′
5′- TCA ACG ATC AGG AGG TCA GGG TGC AA -3′
5′- AGG GGA ACT TAG TGG TAC TTG TGG GC -3′
5′- GGT TCA GAG GAA AAA GGG CTC CTC CT -3′
Performance of GR protocol in an RNA-Seq experiment
After determining the valid and highly efficient GR method, we evaluated the effects of the RNase H GR treatment on gene discovery and coverage in an RNA-seq experiment. Since the above study on the GR method included samples collected using different blood collection tubes and RNA isolation methods, we evaluated the effects of the RNase H GR treatment on gene discovery and coverage in an RNA-seq experiment with a different set of 12 porcine blood samples collected in TempusTM tubes and for which the RNA was isolated by a magnetic bead based MagMaxTM kit.
RNA-seq mapping statistics for pre- and post-globin reduction samples
Classification of samples based on RNA integrity number (RIN)
We examined the RIN changes after GR treatment of pig blood RNA and its effect on sequencing results. The quality of RNA was not changed overall (p >0.1); though 8 samples showed a reduction in RIN after GR treatment, only 3 samples showed a marked decrease of RIN (0.4-0.6) after GR. However, there was a reduction in RNA yield following GR treatment with only 33.3-78.2% of total RNA being recovered. Studies on GR treatment in humans also reported the reduction of RNA yields ranging from 52-95% of total RNA [3, 4, 7, 15]. The reasons for the significant reduction and the wide variation in RNA yield are not clear. To offset the RNA loss accompanying GR treatment, it would be desirable to prepare sufficient amounts of initial RNA. Because we identified possible bias introduced by RIN from the preliminary sequencing results (data not shown), we empirically classified the samples into three categories based on RIN after GR treatment: high (RIN ≥7), moderate (5 ≤ RIN <7), and low (RIN <5) representing ideal, critical and inferior RNA integrity for RNA-seq experiments, respectively.
Increased coverage of non-globin genes in post-GR samples
The list of 34 genes that showed decreased expression level after globin reduction
polymerase (RNA) II (DNA directed) polypeptide A, 220 kDa
arginine-glutamic acid dipeptide (RE) repeats
superoxide dismutase 3, extracellular
MHC class II, DR beta-like 3 pseudogene
Retinoic acid receptor RXR-gamma
ankyrin repeat domain 52
endo/exonuclease (5′-3′), endonuclease G-like
EFR3 homolog B (S. cerevisiae) [Source: HGNC Symbol; Acc:29155]
PHD finger protein 21A
lysine (K)-specific methyltransferase 2A
poliovirus receptor-related 1 (herpesvirus entry mediator C)
crumbs homolog 3 (Drosophila)
proline and serine rich 1
tubulin, beta 4A class IVa
synaptic Ras GTPase activating protein 1
The lower detection levels of a small number of genes in post-GR samples could also be due to the effect of RIN. We investigated all genes with decreased level of detection after GR (fold change <0) from each sample independently, regardless of statistical significance (Additional file 1: Figure S3). We observed that samples with the most notable RIN change after GR (RIN reduction ≥0.4) had the highest number of genes with decreased expression level (samples 4, 7 and 8; Additional file 1: Figure S3). In addition to the effect of RIN, technical variations or sampling effects could also contribute to differences in detection levels of genes.
Increased number of non-globin genes identified in post-GR samples
The porcine specific GR protocol described here successfully removed a significant proportion of the HBA and HBB transcripts prior to sequence analysis. The range of gene discovery from RNA sequencing was extended with significant increases in number of identified genes via improved coverage. Our DE analyses using the GR samples showed increased sensitivity, with no observed strong negative effects as a result of the GR protocol. We also demonstrated the effects of RIN on blood RNA-seq analyses. Thus, the GR protocol incorporated into porcine blood transcriptomics will help advance pig physiological, pathological and blood biomarker studies, by providing more targets for drug development and disease resistance research.
Blood samples and RNA isolation
Animal protocols were approved by the Kansas State University and University of Alberta Animal Care and Use Committees. A total of 24 blood samples were used to conduct two independent studies: comparisons of three GR methods to select the best method and evaluating the effects of the selected GR method on an RNA-seq experiment. For the first study, 3 mL of blood samples from 9 pigs of 1-2 months age produced from Landrace x Large White selected from a commercial populations used in the Porcine reproductive and respiratory syndrome Host Genetics Consortium (PHGC) studies  were collected in TempusTM Blood RNA tubes (Life Technologies, Carlsbad, CA, USA) and 2.5 mL of blood samples from 3 crossbred pigs of Large White x Landrace were collected in PAXgeneTM Blood RNA tubes (PreAnalytiX, Qiagen, Germany) at the University of Alberta. Total RNA was isolated with PAXgeneTM Blood RNA Kit (PreAnalytiX, Qiagen) for PAXgeneTM tubes and either TempusTM Spin RNA Isolation Kit (Life Technologies, Carlsbad, CA, USA) or magnetic bead based MagMaxTM for Stabilized Blood Tubes RNA Isolation Kit for Tempus tubes (Life Technologies), according to the respective manufacturer’s instructions.
For the second study evaluating the effects of RNase H mediated GR protocol on RNA-seq, another set of 12 blood samples were drawn from crossbred pigs of Duroc x (Landrace x Yorkshire) in a PHGC population. Three mL of blood from each pig at 1-2 months of age was collected into TempusTM Blood RNA Tubes at Kansas State University. Total RNA was isolated using the MagMaxTM for Stabilized Blood Tubes RNA isolation kit according to the manufacturer’s protocol.
RNA concentration was quantified using a NanoDrop ND-1000 spectrophotometer (Nano-Drop Technologies, Wilmington, DE, USA) and RNA quality was assessed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Inc., Santa Clara, CA, USA). To determine an accurate 28S/18S rRNA ratio in the pig, we aligned the human 28S sequence against pig genome build 10.2 using BLAST and identified 97-100% of similarity on pig chromosome 6: 871128-866484 (Ensembl release 73). The sizes of the 28S and 18S genes in pig were estimated to be 4645 bp and 2302 bp, respectively, yielding an rRNA ratio of 2.02, whereas the rRNA ratios in human and mouse are known to be 2.69 and 2.53, respectively (the ratio obtained from Genbank database; M11167 and X03205 in human and NR003279 and NR003278 in mouse).
Design of porcine specific oligonucleotides
We first tested GLOBINclearTM Human Kit (Ambion, Austin, TX, USA) which hybridized biotinylated oligonucleotides with globin transcripts by binding to Streptavidin Magnetic Beads. Second, we designed porcine specific biotinylated Peptide Nucleic Acids (PNA) oligos to inhibit reverse transcription of globin transcripts (HBA: 5′-CGAGGCTCCAGCTTA-3′ and HBB: 5′-CACCAGCCACCACCT-3′). Third, we designed four porcine specific antisense oligonucleotides for HBA and HBB using Primer 3 (v. 0.4.0) (Table 1) to hybridize with globin transcripts prior to digestion with RNase H. To design porcine specific oligonucleotides, we first used Clustal Omega (http://www.ebi.ac.uk/Tools/msa/clustalo/) to align the porcine HBA (ENSSSCT00000008741) and HBB (ENSSSCT00000016076) transcript sequences in the current assembly of the pig genome (build 10.2) with their orthologues from human, mouse, cow and pig obtained from the Ensembl database (http://www.ensembl.org) and then checked the similarity of the 3′ end hybridization sites (Additional file 1: Figure S2).
Globin reduction treatment
GR treatment with porcine specific oligonucleotides was performed using a modified Affymetrix GR protocol . In brief, 10X GR oligonucleotides mix was prepared adding 100 uL each of two HBA Oligos at 30 uM, two HBB Oligos at 120 uM per reaction, yielding a final concentration of 7.5 uM HBA Oligos and 30 uM HBB Oligos. Three ug of denatured total RNA was hybridized in a thermal cycler at 70°C for 2 min with the 400 uL 10X GR oligonucleotides mix in hybridization buffer (100 mM Tris-HCL, pH 7.6; 200 mM KCl) at 70°C for 5 min, then cooled to 4°C. The RNA-DNA hybrids were digested with 2 U RNase H (Ambion) in the reaction buffer (100 mM Tris–HCl, pH 7.6, 20 mM MgCl2, 0.1 mM DTT, SUPERase-In) at 37°C for 10 min and cooled to 4°C. The reaction was stopped by addition of 1.0 ul 0.5 M EDTA. RNase H treated RNA was immediately purified with the RNeasy MinElute Cleanup Kit (Qiagen, Toronto, Canada, Cat. No.: 74204) according to manufacturer’s instructions. RNA quality of GR treated samples was assessed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Inc.).
Quantitative real-time PCR (qPCR) analysis
We quantified the mRNA level of the porcine HBA and HBB transcripts by SYBR Green I based qPCR using a StepOneTM Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). First strand cDNA was synthesized using SuperScript® II reverse transcriptase (Invitrogen) and random hexamer primers in a final volume of 20 μL following the manufacturer’s instruction. SYBR Green I based qPCR was performed in a total volume of 10 μL per reaction comprising 2 μL of template, 1 μL of the assay-specific primer mix, 5 μL of the Fast SYBR® Green Master Mix Bulk Pack (Applied Biosystems) and 2 μL of water. The reaction conditions used were one cycle of 95°C for 3 min for initial denaturation, 23 cycles of 95°C for 30 s and 60°C for 30 s. The primer sequences are shown in Additional file 1: Table S1.
Library preparation for sequencing
Poly-A + fractions from the GR treated samples and respective non GR treated samples (1.5 μg RNA each) were purified by using oligo-dT magnetic beads (Illumina, Inc., San Diego, USA), and used to construct cDNA libraries. The Poly (A+) RNA was primed with random hexamers and fragmented at 94°C for 8 min. Second strand cDNA was synthesized after the first strand cDNA using SuperScript II (Invitrogen). The cDNA fragments were end-repaired and a single ‘A’ nucleotide was added to 3′-ends to prevent them from cross ligation during the adapter ligation step. Then individual RNA adapter index oligos were ligated to the end-repaired cDNA and subsequently amplified using Veriti Thermo cycler (Applied Biosystems). The initial denaturation was performed at 98°C for 30 seconds, followed by 15 cycles at 98°C for 10 seconds, annealing at 60°C for 30 seconds and extension at 72°C for 30 seconds. The final extension was followed at 72°C for 5 minutes, and held at 10°C.
The quality and size (~260 bp) of the resulting cDNA libraries were assessed using the High sensitivity DNA Kit (Agilent Technologies, Inc.) in an Agilent Bioanalyzer 2100 (Agilent Technologies, Inc.). The quantification was performed using StepOneTM Real-Time PCR System (Applied Biosystems), as suggested in the Sequencing Library qRT-PCR Quantification Guide (Illumina, Inc.). The KAPA SYBR® FAST ABI Prism qPCR Kit (Kapa Biosystems, Inc., Woburn, USA) was used for the qPCR reactions. The individual libraries were pooled into 2 nM after quantification.
Sequencing was performed on the HiSeq System (Illumina, Inc.). The pooled 10 μL of the 2 nM libraries were diluted and denatured. The pooled cDNA libraries (12 pM) were loaded on the cBot (Illumina, Inc.) for clustering on a flow cell, and single-read cluster generation proceeded using the TruSeqTM SR Cluster Generation Kit v3 (Illumina, Inc., Cat.: FC-930-3001). A portion of each library was diluted to 10 nM and stored at -20°C. Fifty cycles of sequencing-by-synthesis using the paired-end protocol was performed on a HiSeq (Illumina, Inc.) according to manufacturer’s instructions. Real-time analysis and base calling was performed using the HiSeq Control Software Version 1.4.8 (Illumina, Inc.).
Sequence reads with base quality scores were produced by the Illumina sequencer. Raw reads were processed using the Illumina CASAVA (v. 1.8) to filter out the low-quality reads. Sequence reads were then aligned to the pig genome reference assembly (build 10.2; ) using TopHat 2.0.8  with default parameters. The number of reads uniquely mapped to each gene (Ensembl 71 annotation) was determined using Htseq-count (v0.5.3.p3; ). To determine number of genes identified in each sample, we required a read count >5.
To identify genes detected at decreased or increased levels between the globin reduced and non-reduced samples, the read count data were analysed using edgeR (version 3.0.8)  in R (version 2.15.2), as described . Count data was normalized by the library size to account for different numbers of reads obtained from each sample. To determine differences in detection levels between the two groups, an exact test for the negative binomial distribution was used. The genes were considered to be differentially detected at FDR <0.05. RSeQC (v2.3.3)  was used for read distribution over gene body to check 5′/3′ bias. We used BlastN (v2.2.25)  to perform the alignment between globin oligos and the genes with decreased levels after GR treatment.
This project was supported by Applied Livestock Genomics Program (ALGP) grants #13 and #29, Genome Canada grant #2209_F, the USDA ARS project #1245-32000-098, and the NRSP-8 Bioinformatics Coordination project. The authors gratefully acknowledge the suggestions and advice of Dr. James Koltes at Iowa State University.
- Klem TB, Bleken E, Morberg H, Thoresen SI, Framstad T: Hematologic and biochemical reference intervals for Norwegian crossbreed grower pigs. Veterinary clinical pathology/American Society for Veterinary Clinical Pathology. 2010, 39 (2): 221-226.PubMedView ArticleGoogle Scholar
- Takahashi J, Misawa M, Iwahashi H: Oligonucleotide microarray analysis of age-related gene expression profiles in miniature pigs. PLoS One. 2011, 6 (5): e19761-10.1371/journal.pone.0019761.PubMed CentralPubMedView ArticleGoogle Scholar
- Mastrokolias A, den Dunnen JT, van Ommen GB, 't Hoen PA, van Roon-Mom WM: Increased sensitivity of next generation sequencing-based expression profiling after globin reduction in human blood RNA. BMC Genomics. 2012, 13: 28-10.1186/1471-2164-13-28.PubMed CentralPubMedView ArticleGoogle Scholar
- Field LA, Jordan RM, Hadix JA, Dunn MA, Shriver CD, Ellsworth RE, Ellsworth DL: Functional identity of genes detectable in expression profiling assays following globin mRNA reduction of peripheral blood samples. Clin Biochem. 2007, 40 (7): 499-502. 10.1016/j.clinbiochem.2007.01.004.PubMedView ArticleGoogle Scholar
- Wu K, Miyada G, Martin J, Finkelstein D: Globin reduction protocol: A method for processing whole blood RNA samples for improved array results. Affymetrix Technical Note. 2007, Available at: http://media.affymetrix.com:80/support/technical/technotes/blood2_technote.pdfGoogle Scholar
- Tian Z, Palmer N, Schmid P, Yao H, Galdzicki M, Berger B, Wu E, Kohane IS: A practical platform for blood biomarker study by using global gene expression profiling of peripheral whole blood. PLoS One. 2009, 4 (4): e5157-10.1371/journal.pone.0005157.PubMed CentralPubMedView ArticleGoogle Scholar
- Vartanian K, Slottke R, Johnstone T, Casale A, Planck S, Choi D, Smith J, Rosenbaum J, Harrington C: Gene expression profiling of whole blood: Comparison of target preparation methods for accurate and reproducible microarray analysis. BMC Genomics. 2009, 10 (1): 2-10.1186/1471-2164-10-2.PubMed CentralPubMedView ArticleGoogle Scholar
- Wright C, Bergstrom D, Dai H, Marton M, Morris M, Tokiwa G, Wang Y, Fare T: Characterization of Globin RNA Interference in Gene Expression Profiling of Whole-Blood Samples. Clin Chem. 2008, 54 (2): 396-405. 10.1373/clinchem.2007.093419.PubMedView ArticleGoogle Scholar
- Debey S, Zander T, Brors B, Popov A, Eils R, Schultze JL: A highly standardized, robust, and cost-effective method for genome-wide transcriptome analysis of peripheral blood applicable to large-scale clinical trials. Genomics. 2006, 87 (5): 653-664. 10.1016/j.ygeno.2005.11.010.PubMedView ArticleGoogle Scholar
- Liu J, Walter E, Stenger D, Thach D: Effects of globin mRNA reduction methods on gene expression profiles from whole blood. JMD. 2006, 8 (5): 551-558. 10.2353/jmoldx.2006.060021.PubMed CentralPubMedView ArticleGoogle Scholar
- Whitley P, Moturi S, Santiago J, Johnson C, Setterquist R: Improved microarray sensitivity using whole blood RNA samples. Ambion Tech Notes. 2005, 12: 20-23.Google Scholar
- Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, Lopez R, McWilliam H, Remmert M, Söding J, Thompson JD, Higgins DG: Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol Syst Biol. 2011, 7 (1): 539-544.PubMed CentralPubMedView ArticleGoogle Scholar
- Miller ER, Ullrey DE, Ackermann I, Schmidt DA, Luecke RW, Hoefer JA: Swine hematology from birth to maturity. II. Erythrocyte population, size and hemoglobin concentration. J Anim Sci. 1961, 20: 890-897.PubMedGoogle Scholar
- Ramirez CG, Miller ER, Ullrey DE, Hoefer JA: Swine hematology from birth to maturity. II. Erythrocyte population, size and hemoglobin concentration. J Anim Sci. 1963, 22 (4): 1068-1074.Google Scholar
- Shin H, Shannon CP, Fishbane N, Ruan J, Zhou M, Balshaw R, Wilson-McManus JE, Ng RT, McManus BM, Tebbutt SJ: Variation in RNA-Seq Transcriptome Profiles of Peripheral Whole Blood from Healthy Individuals with and without Globin Depletion. PLoS One. 2014, 9 (3): e91041-10.1371/journal.pone.0091041.PubMed CentralPubMedView ArticleGoogle Scholar
- Lunney J, Steibel J, Reecy J, Fritz E, Rothschild M, Kerrigan M, Trible B, Rowland R: Probing genetic control of swine responses to PRRSV infection: current progress of the PRRS host genetics consortium. BMC proceedings. 2011, 5 (Suppl 4): S30-10.1186/1753-6561-5-S4-S30.PubMed CentralPubMedView ArticleGoogle Scholar
- Groenen MAM, Archibald AL, Uenishi H, Tuggle CK, Takeuchi Y, Rothschild MF, Rogel-Gaillard C, Park C, Milan D, Megens H-J, Li S, Larkin DM, Kim H, Frantz LAF, Caccamo M, Ahn H, Aken BL, Anselmo A, Anthon C, Auvil L, Badaoui B, Beattie CW, Bendixen C, Berman D, Blecha F, Blomberg J, Bolund L, Bosse M, Botti S, Bujie Z, et al: Analyses of pig genomes provide insight into porcine demography and evolution. Nature. 2012, 491 (7424): 393-398. 10.1038/nature11622.PubMed CentralPubMedView ArticleGoogle Scholar
- Kim D, Pertea G, Trapnell C, Pimentel H, Kelley R, Salzberg SL: TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol. 2013, 14 (4): R36-10.1186/gb-2013-14-4-r36.PubMed CentralPubMedView ArticleGoogle Scholar
- Anders S, Pyl PT, Huber W: HTSeq – A Python framework to work with high-throughput sequencing data. Bioinformatics. 2014, ᅟ: ᅟ-Sep 25, doi:10.1093/bioinformatics/btu638. [Epub ahead of print]Google Scholar
- Robinson MD, McCarthy DJ: Smyth GK: edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010, 26 (1): 139-140. 10.1093/bioinformatics/btp616.PubMed CentralPubMedView ArticleGoogle Scholar
- Wang L, Wang S, Li W: RSeQC: quality control of RNA-seq experiments. Bioinformatics. 2012, 28 (16): 2184-2185. 10.1093/bioinformatics/bts356.PubMedView ArticleGoogle Scholar
- Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ: Basic local alignment search tool. J Mol Biol. 1990, 215 (3): 403-410. 10.1016/S0022-2836(05)80360-2.PubMedView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/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.