Selection and reliability of internal reference genes for quantitative PCR verification of transcriptomics during the differentiation process of porcine adult mesenchymal stem cells
© Monaco et al.; licensee BioMed Central Ltd. 2010
Received: 21 July 2009
Accepted: 30 March 2010
Published: 30 March 2010
The objective of this study was to find highly reliable internal-control genes (ICGs) for normalization of qPCR data from porcine adult mesenchymal stem cells induced to differentiate toward adipogenic and osteogenic lineages.
Stem cells were acquired from subcutaneous back fat and bone marrow of three castrated Yorkshire crossbred male pigs. Adipose and bone marrow-derived stem cells (ADSCs and BMSCs) were cultured in vitro with specific osteogenic or adipogenic differentiation medium for 4 weeks. Total RNA was extract for microarray (13,000 oligonucleotides) and qPCR analyses. Microarray data were used to uncover the most stably expressed genes (that is, potential ICGs). Co-regulation among potential ICGs was evaluated with Ingenuity Pathway Analysis. qPCR was performed on the non-coregulated ICGs candidates and on specific osteogenic (COL1A1) and adipogenic (DBI) genes. geNorm was used to uncover the most reliable ICGs by using qPCR data and the optimal number of ICGs to be used to calculate the normalization factor.
Microarray data analysis revealed 27 potential ICGs. Among those, 10 genes without known co-regulation were selected to perform qPCR. geNorm performed on qPCR data uncovered high stability in expression ratio among the selected ICGs. However, especially reliable normalization was obtained by geometric mean of NSUN5, TIMM17B, and VPS4A. The effect of normalization, assessed on specific osteogenic (COL1A1) and adipogenic (DBI) genes, was apparent for the adipogenic and less apparent for the osteogenic differentiation.
The combination of microarray data and pairwise gene analysis allowed identification of novel and highly reliable ICGs for qPCR data normalization of adult porcine stem cells induced to differentiate to adipogenic and osteogenic lineages.
The quantity and accessibility of subcutaneous adipose tissue in humans makes it an attractive alternative to bone marrow as a source of adult stem cells for therapeutic purposes . However, little is known about similarities and differences between adipose- (ADSCs) and bone marrow-(BMSCs) derived stem cells at the genetic level and during their differentiation into major mesenchymal cell types. Uncovering transcriptomic adaptations of porcine adult mesenchymal stem cells during in vitro differentiation toward a target tissue can provide crucial information for human adult stem cell therapeutic applications.
The two most common techniques currently used for gene-expression analysis are real-time RT PCR (qPCR) and microarray. qPCR is the method of choice for accurate mRNA expression analysis of selected genes, and it is used for verification of microarray results. Although qPCR can provide accurate estimates of mRNA abundance in experimental samples, the technique is subject to errors during the steps from RNA extraction to mRNA quantitative analysis . The qPCR data must be normalized to take those errors into account. To date, the most frequently used method to normalize qPCR data is through the use of internal control genes (ICGs), sometimes referred to as housekeeping genes (HKGs) . Housekeeping-gene expression should not vary in the cells under investigation, or in response to a treatment. However, the literature shows that HKGs expression, although occasionally constant in a given cell type or experimental condition, can vary considerably [4–6]. As a consequence, the presumed stability of expression of these genes must be validated under the particular experimental conditions being investigated. In many studies, traditional HKGs (such as ACTB, GAPDH, HPRT, β-actin) are used to normalize qPCR data without a proper evaluation of their presumed stability of expression [7–9]. The use of appropriate ICGs to normalize qPCR data is an absolute prerequisite for reliable results.
The choice of IGCs using traditional approaches has been based strictly on the absence of statistical effects of a treatment or physiologic state on qPCR data. This criterion is not sufficiently stringent for studies in which a treatment or physiologic adaptation results in a large transcriptomic modification, as in case of cells undergoing differentiation . The protocols used for qPCR require use of the same amount of total RNA for RT of each sample; and in the case of a large change in expression of some transcripts, the relative mRNA abundance of other transcripts may appear concentrated or diluted . Genes with stable expression (that is, the ideal ICGs), also may appear to have a significant change in expression. Those genes would be considered unreliable as ICGs, even though the observed modification in expression is an artifact of the consistency of amount of starting RNA in each reaction [11, 12]. The use of pair-wise expression ratio-stability comparisons among multiple candidate genes  has been found to be an appropriate method to overcome these limitations. This method, developed by Vandesompele et al.  is based on the use of multiple control genes for accurate normalization of qPCR data. The procedure can be applied in many biologic contests because it takes into account large transcriptomic modification that can follow to a treatment or physiologic adaptation.
Differentiation of stem cells may be expected to result in significant changes in gene expression. Evaluation of expression of genes used as markers for differentiated cells requires identification and validation of appropriate ICGs before conclusions can be drawn relative to transcriptomic adaptations of the differentiation process. The objectives of this study were (a) to identify highly reliable ICGs for normalization of qPCR data for porcine ADSCs and BMSCs, each induced to differentiate toward the adipogenic and osteogenic lineages; and (b) to propose a protocol implemented from the method developed by Vandesompele et al. , which can be used in many, if not all, cell/tissue-type experiments using qPCR. In addition, we provided a list of genes that can be considered suitable candidate ICGs to test in any experiment using porcine mesenchymal stem cells.
Materials and methods
ADSCs and BMSCs isolation and culture in vitro
Subcutaneous back fat and bone marrow were acquired from three castrated Yorkshire crossbred male pigs, at ~6 months of age, under protocols approved by the University of Illinois Institutional Animal Care and Use Committee (IACUC). Pigs were killed at the University of Illinois Meat Science Laboratory abattoir. Details on tissue harvesting and stem cells primary culture in vitro are as previously reported . In brief, subcutaneous back fat and bone marrow from femurs were harvested and processed in sterility to avoid contamination of the samples. They were minced with sterile scalpel blades and digested with 0.075% collagenase type I-A (Sigma Aldrich C2674, St. Louis, MO) at 37°C for 90 min. After digestion, cells were separated by centrifugation at 200 g for 10 min at room temperature. The buoyant cell fraction and supernatant were discarded, and 2 ml of red blood cell lysis buffer (Sigma R7757) was added to the pellet and gently mixed for 2 min. Subsequently, 20 ml of Dulbecco's phosphate-buffered saline (DPBS, Sigma D5773) containing 1% penicillin G-streptomycin (Sigma P3539) and 5.0 mg/L of amphotericin B (Sigma A9528) were added to the tubes that were centrifuged at 200 g for 5 min, at room temperature, to obtain a cell pellet that was then resuspended in culture medium. The culture medium used was high-glucose Dulbecco's modified Eagle's medium (DMEM; Sigma D5648) supplemented with 10% fetal bovine serum (FBS, BenchMark; Gemini Bio Products, West Sacramento, CA) plus 1% penicillin G-streptomycin and 5.0 mg/L of amphotericin B. Cells were counted and plated in 75-cm2 Corning cell culture flasks at 7.5 × 105 cells for ADSCs and 1.6 × 106 cells for BMSCs in 15 ml of culture medium and incubated at 39°C and 5% CO2 (passage 0 cells). Medium was changed every other day until cells reached 80% confluence (day 10 of culture). To keep the cells at a sufficiently low density to stimulate further growth, they were trypsinized with 0.25% Trypsin (Sigma T4799), 0.04% EDTA (Sigma E6753), and replated at the density indicated earlier (passage 1 cells). At 80% confluence, cells were trypsinized again and frozen and stored in liquid N2 until further use.
ADSCs and BMSCs osteogenic and adipogenic differentiation in vitro
Frozen ADSCs and BMSCs were thawed and plated (passage 2) at a density of 7.5 × 105 cells/75 cm2. Cells were cultured for 4 days, and then they were trypsinized, counted, and plated (passage 3) again for 4 days before induction of differentiation. To induce the differentiation of ADSCs and BMSCs into osteogenic or adipogenic cells in vitro, ADSCs and BMSCs were cultured with specific osteogenic or adipogenic differentiation medium. Osteogenic medium consisted of basic high-glucose DMEM supplemented with 100 nmol/L dexamethasone (Sigma D4902), 10 mmol/L β-glycerophosphate (Sigma G6376), 0.05 mmol/L ascorbic acid-2-phosphate (Sigma A4403), 10% FBS, 1% penicillin G-streptomycin, and 5.6 mg/L of amphotericin B. Adipogenic medium consisted of basic high-glucose DMEM supplemented with 1.0 μmol/L dexamethasone, 0.5 mmol/L isobutylmethylxanthine (IBMX, Sigma I5879), 10 μmol/L insulin (Sigma I9278), 200 μmol/L indomethacin (Sigma I7378), 10% FBS, 1% penicillin G-streptomycin, and 5.6 mg/L of amphotericin B. Cells were cultured in these media for 3 weeks, with media changed every 3 days.
RNA Extraction, qPCR, and primer design and testing
At 0, 2, 7, and 21 days of differentiation (dd) cells were trypsinized, as described earlier and counted by hemocytometer. Total RNA was immediately isolated by using the RNeasy Mini Kit (Qiagen Sciences, Germantown, MD) according to the manufacturer's protocol. Any residual genomic DNA was eliminated by treatment with RNase-Free DNase Set (Qiagen). RNA concentration was measured with a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE). A portion of the highly pure RNA (A260/A280 = 2.13 ± 0.05) was diluted to 100 mg/L by using DNase-RNase-free water before reverse transcription. Sufficient cDNA was prepared to run all selected genes. Each cDNA was synthesized by reverse transcription and used for qPCR, as described by Bionaz and Loor . Each sample was run in triplicate along with a 6-point relative standard curve plus the nontemplate control. The fivefold dilution standard curve was made by using cDNA from an RNA pool of four swine tissues: liver, mammary gland, jejunum, and kidney. The amplification reactions were conducted in an ABI Prism 7900 HT SDS instrument (Applied Biosystems) as follows: 2 min at 50°C, 10 min at 95°C, 40 cycles of 15 s at 95°C, and 1 min at 60°C. The presence of a single PCR product was verified by the dissociation protocol by using incremental temperatures to 95°C for 15 seconds plus 65°C for 15 seconds. Data were analyzed with the 7900 HT Sequence Detection Systems software (version 2.2.3, Applied Biosystems). Primer design and evaluation were performed as previously described by Tramontana et al. .
Selection and evaluation of ICGs
For the purpose of identifying potential ICGs, porcine microarray (13,000 oligonucleotides) data generated by a dye-swap reference design  were mined to uncover stable expressed genes during osteogenic and adipogenic differentiation in vitro of ADSCs and BMSCs. The protocol to select and evaluate ICGs by using microarray was as previously described .
The most stably expressed genes among all samples in the microarray data set were chosen by following these criteria: (a) only genes with median intensity ≥3 standard deviations above the median of the background were used; (b) filtered data were uploaded into GeneSpring GX7 software (Agilent Technologies) and normalized by using Lowess; (c) only genes with expression ≥100 relative fluorescent units (that is, medium-large mRNA abundance) were chosen; and (d) only genes with expression ratio ~1.0 between sample and reference (RNA mixture from porcine mammary gland, liver, kidney, and jejunum) and with complete annotation were selected.
To uncover reliable ICGs, we used geNorm software, which is based on a pairwise expression ratio stability comparison method . geNorm measures the stability (M) of the ICGs- as the average pairwise expression ratio variation of qPCR data of a particular gene with all other control genes tested. Genes with the lowest M values have the most stable expression. geNorm also determines the optimal number of ICGs and calculates a normalization factor (NF) to be used for normalization . It uses a pairwise variation analysis between the normalization factors NFn and NFn+1 (V) to determine the number of control genes required for accurate normalization. The pairwise comparison requires evaluation of multiple genes to select appropriate internal controls. Critical for the use of the pairwise comparison method in selecting appropriate ICGs is the absence of co-regulation (common upstream regulator factor(s) or direct regulation of transcript expression between gene products) among the genes tested . Co-regulation was assessed by using Ingenuity Pathways Analysis (IPA; Ingenuity System). IPA is a web-based software that generates relations among human and rodent genes or proteins by using information from the published literature.
The evaluation of the appropriate number of ICGs to calculate the NF and the effect of normalization versus nonnormalization was evaluated by running qPCR for collagen type I alpha 1 (COL1A1) and diazepam-binding inhibitor (acyl-CoA-binding protein; DBI) which represented osteogenic and adipogenic gene markers, respectively.
Statistical effect of each variable and combination of variables on 10 potential internal-control genes
Tissue* Different* Day
Results and Discussion
ICGs selection and evaluation
The use of microarray data  allowed the identification of 27 stably expressed genes among >10,000 annotated transcripts (Table S1 in Additional file 1). The 27 genes were highly stable across all samples (ratio sample/reference = 0.99 ± 0.14) in at least 70 of 96 microarrays and represented a novel potential set of ICGs for qPCR analysis of porcine adult mesenchymal stem cells differentiated in vitro.
This preliminary screening allowed the identification of genes with similar mRNA abundance between samples (ADSCs and BMSCs) and the mixture of porcine tissues (reference in the microarray). They were genes with potentially stable expression, similar to those that are considered HKGs . However, as previously suggested [11, 12, 15], the sole reliance on microarray data for evaluation of ICGs for qPCR normalization is inappropriate and should be considered only as a preliminary screening to identify genes with a potential for stable expression. The qPCR and array techniques differ in analytic procedure and sensitivity . To overcome these differences and optimize the evaluation of ICGs for qPCR normalization, several tools have been developed, based on specific algorithms [3, 17, 18]. In particular, the geNorm software is a valuable tool to evaluate ICGs under diverse experimental conditions, and it is one of the most widely used softwares for evaluation of ICGs to date.
Description of 10 potential internal-control genes
Barrier to autointegration factor 1
Dihydroxyacetone kinase 2 homologue (S. cerevisiae)
KTI11 homologue (S. cerevisiae)
Excision repair cross-complementing repair deficiency, complementation group 3
NOL1/NOP2/Sun domain family, member 5
Nucleotide-binding protein 1 (MinD homologue, E. coli)
RNA polymerase II CTD phosphatase homologue (S. cerevisiae)
Translocase of inner mitochondrial membrane 17 homologue B (yeast)
Vacuolar protein sorting 4 homologue A (S. cerevisiae)
Collagen, type I, alpha 1
Diazepam-binding inhibitor (acyl-coenzyme A-binding protein)
ICGs expression pattern
Correlation between number of cell, RNA, and raw qPCR data of internal control genes tested.
Considerations for selecting optimal number of ICGs
The effect of normalization on qPCR data is highlighted by COL1A1 and DBI expression patterns before and after normalization (Figure 4, Table S3 for overall statistical results). For COL1A1, the Tissue × Day interaction was not significant when the gene was not normalized, but normalization with three or five ICGs gave a significant overall Tissue × Day interaction (Table S3); differences were observed at 0 and 2 days. Similarly, overall Tissue × Day interaction of DBI was significant before normalization, whereas after normalization, using three or five ICGs was not significant (Table S3), particularly for 0 days.
Accurate normalization of qPCR data is an unavoidable step for obtaining reliable results. Many published studies dealing with the use of qPCR, however, do not report or follow a proper method to test for suitable ICGs. In this study, we indentified highly reliable ICGs for qPCR normalization data of porcine ADSCs and BMSCs. We also propose a protocol to select ICGs that can be used in many other cell/tissue-type experiments. The traditional criterion to select appropriate ICGs by using absence of time and/or treatment effect on qPCR data, would have rejected our candidate ICGs as inappropriate because of observed statistical effects on their expression. However, the use of IPA to select genes without known co-regulation, and the pairwise comparison method (geNorm), uncovered a very high stability in expression ratio among selected ICGs. This indicates that the relative mRNA abundance of those genes is similarly subjected to the errors before (for example, dilution/concentration effect) and during the qPCR procedure (from RNA extraction to qPCR analysis). The robustness of such an approach relies on the fact that the purpose for normalization of qPCR data is to correct for those errors. In addition, although the use of nine ICGs resulted in the optimal (that is, most stable) NF, the use of NF calculated by the geometric mean of NSUN5, TIMM17B, and VPS4A resulted in IGCs that are highly reliable for normalization of qPCR data from porcine adult mesenchymal stem cells induced to differentiate toward adipogenic and osteogenic lineages. Moreover, because it has been demonstrated that the expression patterns of commonly used ICGs can vary widely in different experimental conditions [23, 24], the suitability of the ICGs uncovered in this study must be validated (for instance, by the simple use of geNorm), as the experimental conditions differ from those reported here. In this regard, the present work provides a large set of genes (Table S1 in Additional file 1) that can be used as potential ICGs, to be tested in similar experimental conditions in laboratories where the microarray data are not readily available.
adipose-derived stem cells
bone marrow-derived stem cells
Dulbecco's Modified Eagle's Medium
Dulbecco's phosphate-buffered saline
fetal bovine serum
internal control genes
ingenuity pathways analysis
The authors thank Synthia Lane and Tara VanEtten for technical assistance with this work and Glenn Bressner for management of the animals used in these studies. This work was supported by a grant from the Illinois Regenerative Medicine Institute IDPH 63080017 through the State of Illinois Department of Public Health.
- Fraser JK, Wulur I, Alfonso Z, Hedrick MH: Fat tissue: an underappreciated source of stem cells for biotechnology. Trends Biotechnol. 2006, 24 (4): 150-154. 10.1016/j.tibtech.2006.01.010.View ArticlePubMedGoogle Scholar
- Valasek MA, Repa JJ: The power of real-time PCR. Adv Physiol Educ. 2005, 29 (3): 151-159. 10.1152/advan.00019.2005.View ArticlePubMedGoogle Scholar
- Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, Speleman F: Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol. 2002, 3: RESEARCH0034-10.1186/gb-2002-3-7-research0034.PubMed CentralView ArticlePubMedGoogle Scholar
- Warrington JA, Nair A, Mahadevappa M, Tsyganskaya M: Comparison of human adult and fetal expression and identification of 535 housekeeping/maintenance genes. Physiol Genomics. 2000, 2: 143-147.PubMedGoogle Scholar
- Thellin O, Zorzi W, Lakaye B, De Borman B, Coumans B, Hennen G, Grisar T, Igout A, Heinen E: Housekeeping genes as internal standards: use and limits. J Biotechnol. 1999, 75: 291-295. 10.1016/S0168-1656(99)00163-7.View ArticlePubMedGoogle Scholar
- Suzuki T, Higgins PJ, Crawford DR: Control selection for RNA quantitation. Biotechniques. 2000, 29: 332-337.PubMedGoogle Scholar
- Ringe J, Kaps C, Schmitt B, Buscher K, Bartel J, Smolian H, Schultz O, Burmester GR, Haupl T, Sittinger M: Porcine mesenchymal stem cells. Induction of distinct mesenchymal cell lineages. Cell Tissue Res. 2002, 307: 321-327. 10.1007/s00441-002-0525-z.View ArticlePubMedGoogle Scholar
- Wislet-Gendebien S, Hans G, Leprince P, Rigo JM, Moonen G, Rogister B: Plasticity of cultured mesenchymal stem cells: switch from nestin-positive to excitable neuron-like phenotype. Stem Cells. 2005, 23: 392-402. 10.1634/stemcells.2004-0149.View ArticlePubMedGoogle Scholar
- Jude CD, Climer L, Xu D, Artinger E, Fisher JK, Ernst P: Unique and independent roles for MLL in adult hematopoietic stem cells and progenitors. Cell Stem Cell. 2007, 1: 324-337. 10.1016/j.stem.2007.05.019.PubMed CentralView ArticlePubMedGoogle Scholar
- Monaco E, Lima A, Wilson S, Kim D, Bionaz M, Hurley WL, Wheeler MB: Temporal transcript profiling of porcine adult mesenchymal stem cells during osteogenic and adipogenic differentiation using a 13K porcine microarray. ISSCR 6th Annual Meeting: Philadelphia. 2008, 322, 323Google Scholar
- Bionaz M, Loor JJ: Identification of reference genes for quantitative real-time PCR in the bovine mammary gland during the lactation cycle. Physiol Genomics. 2007, 29: 312-319. 10.1152/physiolgenomics.00223.2006.View ArticlePubMedGoogle Scholar
- Tramontana S, Bionaz M, Sharma A, Graugnard DE, Cutler EA, Ajmone-Marsan P, Hurley WL, Loor JJ: Internal controls for quantitative polymerase chain reaction of swine mammary glands during pregnancy and lactation. J Dairy Sci. 2008, 91: 3057-3066. 10.3168/jds.2008-1164.View ArticlePubMedGoogle Scholar
- Monaco E, Lima A, Bionaz M, Maki A, Wilson S, Hurley WL, Wheeler MB: Morphological and transcriptomics comparison of adipose and bone marrow derived porcine stem cells. Open Tissue Eng Regen Med J. 2009, 2: 20-33. 10.2174/1875043500902010020.View ArticleGoogle Scholar
- Synnergren J, Giesler TL, Adak S, Tandon R, Noaksson K, Lindahl A, Nilsson P, Nelson D, Olsson B, Englund MC, Abbot S, Sartipy P: Differentiating human embryonic stem cells express a unique housekeeping gene signature. Stem Cells. 2007, 25: 473-480. 10.1634/stemcells.2006-0247.View ArticlePubMedGoogle Scholar
- Piantoni P, Bionaz M, Graugnard DE, Daniels KM, Akers RM, Loor JJ: Gene expression ratio stability evaluation in prepubertal bovine mammary tissue from calves fed different milk replacers reveals novel internal controls for quantitative polymerase chain reaction. J Nutr. 2008, 138: 1158-1164.PubMedGoogle Scholar
- Wang Y, Barbacioru C, Hyland F, Xiao W, Hunkapiller KL, Blake J, Chan F, Gonzalez C, Zhang L, Samaha RR: Large scale real-time PCR validation on gene expression measurements from two commercial long-oligonucleotide microarrays. BMC Genomics. 2006, 7: 59-10.1186/1471-2164-7-59.PubMed CentralView ArticlePubMedGoogle Scholar
- Andersen CL, Jensen JL, Orntoft TF: Normalization of real-time quantitative reverse transcription-PCR data: a model-based variance estimation approach to identify genes suited for normalization, applied to bladder and colon cancer data sets. Cancer Res. 2004, 64: 5245-5250. 10.1158/0008-5472.CAN-04-0496.View ArticlePubMedGoogle Scholar
- Pfaffl MW, Tichopad A, Prgomet C, Neuvians TP: Determination of stable housekeeping genes, differentially regulated target genes and sample integrity: BestKeeper-Excel-based tool using pair-wise correlations. Biotechnol Lett. 2004, 26: 509-515. 10.1023/B:BILE.0000019559.84305.47.View ArticlePubMedGoogle Scholar
- Doll A, Grzeschik KH: Characterization of two novel genes, WBSCR20 and WBSCR22, deleted in Williams-Beuren syndrome. Cytogenet Cell Genet. 2001, 95: 20-27. 10.1159/000057012.View ArticlePubMedGoogle Scholar
- Bauer MF, Gempel K, Reichert AS, Rappold GA, Lichtner P, Gerbitz KD, Neupert W, Brunner M, Hofmann S: Genetic and structural characterization of the human mitochondrial inner membrane translocase. J Mol Biol. 1999, 289: 69-82. 10.1006/jmbi.1999.2751.View ArticlePubMedGoogle Scholar
- Goto Y, Kimura H: Inactive X chromosome-specific histone H3 modifications and CpG hypomethylation flank a chromatin boundary between an X-inactivated and an escape gene. Nucleic Acids Res. 2009, 37 (22): 7416-7428. 10.1093/nar/gkp860.PubMed CentralView ArticlePubMedGoogle Scholar
- Scheuring S, Bodor O, Rohricht RA, Muller S, Beyer A, Kohrer K: Cloning, characterisation, and functional expression of the Mus musculus SKD1 gene in yeast demonstrates that the mouse SKD1 and the yeast VPS4 genes are orthologues and involved in intracellular protein trafficking. Gene. 1999, 234 (1): 149-159. 10.1016/S0378-1119(99)00163-8.View ArticlePubMedGoogle Scholar
- Lee PD, Sladek R, Greenwood CM, Hudson TJ: Control genes and variability: absence of ubiquitous reference transcripts in diverse mammalian expression studies. Genome Res. 2002, 12: 292-297. 10.1101/gr.217802.PubMed CentralView ArticlePubMedGoogle Scholar
- Murphy CL, Polak JM: Differentiating embryonic stem cells: GAPDH, but neither HPRT nor beta-tubulin is suitable as an internal standard for measuring RNA levels. Tissue Eng. 2002, 8: 551-559. 10.1089/107632702760240472.View ArticlePubMedGoogle 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/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.