Quality control and quality assessment of data from surface-enhanced laser desorption/ionization (SELDI) time-of flight (TOF) mass spectrometry (MS)
© Hong et al; licensee BioMed Central Ltd. 2006
Published: 15 July 2005
Proteomic profiling of complex biological mixtures by the ProteinChip technology of surface-enhanced laser desorption/ionization time-of-flight (SELDI-TOF) mass spectrometry (MS) is one of the most promising approaches in toxicological, biological, and clinic research. The reliable identification of protein expression patterns and associated protein biomarkers that differentiate disease from health or that distinguish different stages of a disease depends on developing methods for assessing the quality of SELDI-TOF mass spectra. The use of SELDI data for biomarker identification requires application of rigorous procedures to detect and discard low quality spectra prior to data analysis.
The systematic variability from plates, chips, and spot positions in SELDI experiments was evaluated using biological and technical replicates. Systematic biases on plates, chips, and spots were not found. The reproducibility of SELDI experiments was demonstrated by examining the resulting low coefficient of variances of five peaks presented in all 144 spectra from quality control samples that were loaded randomly on different spots in the chips of six bioprocessor plates. We developed a method to detect and discard low quality spectra prior to proteomic profiling data analysis, which uses a correlation matrix to measure the similarities among SELDI mass spectra obtained from similar biological samples. Application of the correlation matrix to our SELDI data for liver cancer and liver toxicity study and myeloma-associated lytic bone disease study confirmed this approach as an efficient and reliable method for detecting low quality spectra.
This report provides evidence that systematic variability between plates, chips, and spots on which the samples were assayed using SELDI based proteomic procedures did not exist. The reproducibility of experiments in our studies was demonstrated to be acceptable and the profiling data for subsequent data analysis are reliable. Correlation matrix was developed as a quality control tool to detect and discard low quality spectra prior to data analysis. It proved to be a reliable method to measure the similarities among SELDI mass spectra and can be used for quality control to decrease noise in proteomic profiling data prior to data analysis.
Recent advances in proteomic profiling technologies, such as SELDI-TOF MS (Ciphergen Biosystems, Inc., Fremont, CA, http://www.ciphergen.com), have allowed preliminary profiling and identification of biomarkers in biological fluids for biological, toxicological, and clinical research [1–10]. ProteinChip technology coupled with SELDI-TOF MS is an effective tool for the simultaneous detection of the relative expression levels of proteins over a wide range of molecular weights in biological samples under different conditions. Differences in protein expression level can then be used to identify disease, differentiate different stages of a disease, toxicant treatment versus control, or different time points following toxicant treatment [11–14].
Analysis of SELDI-TOF MS data presents challenges similar to those for gene expression profile analysis from microarray technologies. Global profiling analyses strive to identify reliable and reproducible expression patterns that are signatures specific to each state, such as disease versus healthy control or different experimental conditions (e.g. treated with a toxicant of interest versus untreated). The identification of biomarkers for diagnosis or prognosis is dependent on analysis of the highly dimensional protein expression profiles. Data must be correctly analyzed before valid interpretation and reliable biological conclusions to be drawn from a protein expression profiles. Analysis of poor quality, noise laden protein expression profiles, however, will likely lead to results lacking biological relevance. Therefore, quality assessment of the protein expression profiles and determination of reproducibility of SELDI-TOF MS experiments and profiles prior to data analysis is of critical importance.
Using SELDI-TOF MS coupled with protein chip technologies for biomarker development is a complicated process that involves many steps, including sample collection and preparation, protein chip selection and preparation, matrix selection and application, spectral calibration, loading sample on chip, washing away non-specifically bound proteins, SELDI-TOF MS parameter settings, data recording, and data pre-processing. Any of these many steps could introduce noise, thus, adversely affecting the quality of the experiment and the reliability of the protein expression profile. A high degree of variability of protein expression profiles in SELDI experiments is not infrequent. The coefficient of variation for absolute intensity measures can be as high as 50–60% . Scientists have recently realized that quality control (QC) is an important issue in SELDI experiments and several efforts have been made to apply some QC techniques to improve the reproducibility of SELDI profiling data [3, 16, 17]. For example, QC samples that are pooled from multiple samples have been used to assess the reproducibility of a SELDI experiment [18–20], while technical replicates have been used to assess the reproducibility within the same samples . Because of the complex nature of SELDI-TOF MS – ProteinChip experiments, even with experimental QC, the resultant data must be subjected to stringent quality assessment prior to data analysis. Specifically, low quality spectra should be identified and eliminated from analysis to ensure the reliability of biomarkers and the associated patterns discovered during analysis. For example, systematic variability in experiments may introduce additional error sources into the data and this possibility should be examined prior to data analysis.
We investigated systematic variability for plates, chips, and spot positions in two independent SELDI biomarker studies. All peaks (five in our study) appearing in all QC samples should be used to assess the reproducibility of experiments as recommended by Ciphergen because those peaks are the common proteins for the QC samples and should be in similar levels of expression. The high level of reproducibility of our experiments was demonstrated by low coefficients of variation for the five peaks that appeared in all 144 spectra from QC samples. No systematic bias in the experiments was detected (there is no single source of variation that is consistent when switching samples between spots, plates and chips). To identify spectra of low quality, a Pearson correlation matrix was developed as a QC tool to detect low quality spectra in SELDI profiling data analysis, using all peaks in all spectra. The rationale behind the use of a correlation matrix as a QC tool is the assumption that protein expression profiles from biological replicates and technical replicates should be similar. Thus, the correlation matrix is a measure of the similarities among the spectra and useful for quantifying how consistently the experiments have been conducted. We applied the correlation matrix to the SELDI data from the study of biomarkers for liver cancer and liver toxicity, as well as myeloma-associated lytic bone disease. We found that the correlation matrix was an efficient and reliable means to detect low quality spectra should be removed. Doing so should result in more reliable biomarker identification in the final protein expression profiles.
Liver cancer and liver toxicity study
Quality control samples. Plasma was collected in Li-Heparin tubes from two individuals (one male, one female, one Caucasian, one Oriental). Following isolation of plasma from whole blood, all tubes were pooled, mixed, aliquoted into eppendorf tubes, and stored frozen at -80°C. Each tube contained between 80–100 μl of plasma. Four tubes were used for each 96 well plate, one spot per chip. Once thawed, the remaining sample was discarded.
Myeloma-associated lytic bone disease
Sera from 64 newly diagnosed myeloma patients were collected during routine clinic visit (Institutional Review Board approved and signed informed consents are kept on record). 36 samples were from patients with 1 to 26 lytic bone lesions, determined by X-ray skeletal surveys, and 28 from patients with no evidence of lytic bone disease
SELDI protein profiling
Liver cancer and liver toxicity study: The plasma samples were analyzed by SELDI-TOF MS on a Ciphergen ProteinChip Biology System II (PBS II). Specifically, 10 μl of unfractionated samples were used in duplicate for the nonfractionated analyses and 20 μl was used for the fractionation step. Unfractionated and fractionated samples were examined on a weak cation exchange chip (CM10). Two solubilization steps were used for each of the samples whether or not fractionation was to be performed: 9 M urea, 2% CHAPS, 50 μM Tris, pH 9; and 1 M urea, 0.22% CHAPS, 50 μM Tris, pH9. In addition, each of the plasma samples was fractionated for analysis. The plasma to be fractionated was subjected to anion exchange chromatography with stepwise pH elution into 6 fractions , each of these fractions was analyzed on a CM10 chip. Plasma (20 μl) was centrifuged for 10 minutes and mixed with the chaotropic urea/CHAPS denaturing and solubilization buffer (30 μl). Each sample was run in duplicate. Sinapinic acid (SPA) was used as the matrix for each sample. The chips were read at two different laser energies to permit ionization and visualization of proteins.
Myeloma-associated lytic bone disease: The serum samples were analyzed by SELDI-TOF MS on a Ciphergen PBS II, using IMAC30 ProteinChips (Ciphergen Biosystems, Inc.) activated with CuSO4 according to Ciphergen's protocols. All sera were assayed in quadruplicate and positioned on the 96 spots on each of the 12 chips in a bioprocessor to allow for variability between spots to be determined. For consistency, a BioMek 2000 liquid-handling robot (Beckman-Coulter, Fullerton, CA) was used. SPA was the energy-absorbing matrix. Proteins above 1000 Kda m/z were interrogated using an average of 66 laser shots with laser intensity of 195 and sensitivity of 8. Low molecular weight (MW) standards were assayed with each run (12 chips/run). Calibrations were done for each run.
The raw SELDI mass spectra were pre-processed prior to subsequent analysis of the expression profiles. Ciphergen Express 2.0 software was used to calibrate all the spectra in the myeloma-associated lytic bone disease study. Normalization by Total Ion Current was applied to all spectra in both the liver cancer and liver toxicity study and the myeloma-associated lytic bone disease study to minimize the variability in spectra obtained from different times using Ciphergen ProteinChip 3.2 software. The normalization process took the Total Ion Current used for all the spots, averaged the intensity, and adjusted the intensity scales for all the spots so that all spectra could be displayed on the same scale. Baseline subtraction was conducted prior to normalization, as recommended by Ciphergen. Baseline subtraction offsets in the spectra were the result of both electrical noise and the noise from Energy Absorbing Molecules (EAM). The lowest spectral amplitude was detected and then was used to correct the peak height and area. The Biomarker Wizard function in Ciphergen ProteinChip 3.2 software was used to autodetect the peaks present in all of the spectra. The spectral region from 0 to 2500 Da is unreliable for both normalization and peak detection due to matrix interference and was therefore not included in the analysis. A signal to noise ratio greater than 5 was used for the first pass selection of peaks and signal to noise ratio greater than 3 was used for the second pass. Cluster mass window of 0.3% was used for both studies. A requirement to be in a minimum of 100% of samples was set for clusters generation for quality control samples used in liver cancer and liver toxicity study because the common proteins were expected to be expressed in all of the samples. In myeloma-associated lytic bone disease study the requirement was set to be in a minimum of 5% samples for clusters generation because different patient samples were assayed and different protein expressions for different patients were expected.
After spectral pre-processing using ProteinChip 3.2 software, data was exported as a profile matrix with rows representing spectra and columns representing peak intensities. Principal component analysis (PCA) was performed using Spotfire DecisionSite 7.1 software (Somerville, MA, http://www.spotfire.com). Pearson correlation matrix and other statistical calculations were performed with JMP 5.1 software http://www.jmp.com. Heat maps and correlation matrix surfaces were created with DMVS 2.0 software (Shenzhen Chipscreen Biosciences, Ltd, Shenzhen, P. R. China, http://www.chipscreen.com).
Assessment of systematic variability from plates, chips, and spots
Fig. 1A shows PCA results for data points representing spectra of the 144 QC samples with color denoting bioprocessor plate. The dispersion of the points and lack of clustering indicates that variability across plates is random. Fig. 1B similarly shows the PCA results for all 144 QC samples with color denoting one of the eight different spot positions. Again, dispersion and lack of grouping indicates the absence of systematic variability across position. Fig. 1C similarly shows the PCA results for all 144 QC samples with color denoting each of the 12 chips used in the six bioprocessor plates. Again there is no apparent systematic variance. Thus, no systematic variability across plates, across chips or the spot positions is discernable from PCA analysis.
The samples of interest in our studies (i.e., the study of biomarkers for liver cancer and liver toxicity, as well as myeloma-associated lytic bone disease) were loaded on the spots of the chips in the bioprocessor plates together with the QC samples. As the analysis of the QC samples showed no systematic variability across spot locations, chips and plates, these sources of variability should have minimum affect on the data analysis for the identification of potential biomarkers that differentiate liver cancer patients from the healthy individuals and patients with bone lesions from those without bone lesions. Thus, the biomarkers and the associated patterns in our study were identified from the protein profiles with little systematic variability (results not shown).
Assessment of reproducibility of experiments
Detection of low quality spectra
The reproducibility of experiments measures the overall consistency among a set of spectra obtained from SELDI experiments. However, it does not give any detail on the quality of individual spectra in the dataset. In other words, reproducibility is a good assessment of the overall consistency of a set of experiments, but it can not be used to identify low quality spectra. Low quality spectra need to be removed from the data set prior to analysis in order to reliably discover biomarker proteins and the associated expression patterns that differentiate a disease from control or that distinguish different stages of a disease such as cancer. For assessment of reproducibility, all common peaks have to be (and were) used. But for quality assessment, whole spectra (all possible peaks) must to be used. A Pearson correlation matrix was used for QC of the individual spectra of biological and technical replicates prior to SELDI data analysis. The rationale behind the approach is to assume that the protein expression profiles of biological and technical replicates of a sample are similar and thus that the correlation among spectra of same or similar samples should be very high.
Pearson correlation matrix of four spectra from patient X. Column "Mean" lists the average Pearson correlation coefficients of each spectrum to other spectra.
Pearson correlation matrix of four spectra from patient Y. Column "Mean" lists the average Pearson correlation coefficients of each spectrum to other spectra.
Recently, the scientific community has been using proteomics to improve diagnostic/prognostic capability and to study the underlying biological processes of diseases and toxicology. Because of the complicated nature of the SELDI-TOF MS ProteinChip analysis, stringent and reliable quality assessment (QA) and quality control (QC) methods and procedures undertaken prior to data analysis are critical. The QA/QC analysis should assure that the data has acceptable experimental variability and that incongruent data that are not the result of biology are screened out. Such assurance of data quality is a necessary condition for identification of reliable biomarker expression patterns with clinical utility for diagnosis and prognosis and for toxicity and safety assessment.
Two phases of quality assessment and quality control can be applied in the workflow of SELDI-TOF MS ProteinChip proteomic analysis. The first QA/QC phase is aimed at examining and improving the experimental techniques to produce quality data. This phase incorporates biological and technical replicates to identify and quantify the systematic bias and noise in the experiments; the process spans all experimental steps such as calibration of instruments, proper sample preparation, and so forth. The emphasis of this phase is on improving the quality of the experiments.
The second QA/QC phase emphasises data analysis. Even if stringent procedures are applied in the first phase, it is difficult to guarantee that all spectra from the experiments are of sufficient quality for subsequent data analysis. A thorough quality assessment and quality control process is required in order to ensure that only high quality data are used. Low quality spectra need to be identified and excluded.
A correlation matrix was found to be an effective means to detect low quality spectra among the biological and technical replicates. Use of correlation matrix approach is based on the assumption that spectra of biological and technical replicates should be similar and well correlated, and thus outlying spectra are of relatively lower quality and significance. Our results suggest that inter-spectral correlation coefficient values of R > .95 to .97 are attainable and representative figure-of-merit for quality data. Spectra that are inconsistent with other replicates are likely to have been compromised in one or more of the many complex steps in SELDI, or by such factors as lack of uniformity of the matrix material. While the correlation matrix can identify likely anomalies and noisy spectra for exclusion, it cannot indicate the source of the variability. Even several phase one QC/QA methods such as using pooled QC samples and using standard calibrant to calibrate the spectra were adopted in our studies to make the experiments high quality, the correlation matrix still identified spectra should be excluded from analysis because of low correlation with the other replicate spectra.
The liver cancer and liver toxicity study used 144 QC samples dispersed evenly across six Ciphergen bioprocessor plates and different chips, and randomly located on different spots. The experiments were conducted on different dates. No date effect was observed. The samples were collected into heparinized tubes and stored at -80°C. The period of storage for samples used in this study ranged from several weeks to one year. The experiments and the quality assessment analysis showed that the samples were very stable during sustained storage.
Application of new technologies such as SELDI-TOF MS coupled with the ProteinChip for biomarker identification requires a robust QA/QC assessment process. QA/QC pre-processing is critical for discovery of biomarkers and expression patterns that have potential use in clinical settings, as well as in toxicity studies for safety assessment. In this report, we described a method for monitoring the quality of proteomics profiles from SELDI-TOF MS. The results showed that the reproducibility of our SELDI experiments studies of liver cancer and liver toxicity was high with coefficients of variation less than 20% for five common proteins across all 144 QC samples. In addition, we found no systematic variability across plates or chips, or spot locations. The quality assessment lent confidence that the data obtained met requirements for biomarker identification. To ensure that all data used in the data analysis for the discovery of biomarkers are of high quality with low noise, the low quality spectra should be identified and removed prior to data analysis. We developed a method using a correlation matrix to identify the low quality spectra observed in biological and technical replicates. The correlation matrix approach was applied to the liver cancer and toxicity study and the myeloma-associated lytic bone disease study to efficiently identify low quality spectra prior to post-analysis.
The authors wish to thank the faculty and staff of the Myeloma Institute for Research and Therapy for their dedication to excellence in patient care and research. This study was supported in part by grant CA-55819 from the National Cancer Institute (JE).
- Merchant M, Weinberger SR: Recent advancements in surface-enhanced laser desorption/ionization-time of flight-mass spectrometry. Electrophoresis 2000, 21: 1164–1177. 10.1002/(SICI)1522-2683(20000401)21:6<1164::AID-ELPS1164>3.3.CO;2-SView ArticlePubMedGoogle Scholar
- Kozak KR, Amneus MW, Pusey SM, Su F, Luong MN, Luong SA, Reddy ST, Farias-Eisner R: Identification of biomarkers for ovarian cancer using strong anion-exchange ProteinChips: Potential use in diagnosis and prognosis. Proc Natl Acad Sci USA 2003, 100: 12343–12348. 10.1073/pnas.2033602100PubMed CentralView ArticlePubMedGoogle Scholar
- Petricoin EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg SM, Mills GB, Simone C, Fishman DA, Kohn EC, Liotta LA: Use of proteomic patterns in serum to identify ovarian cancer. The Lancet 2002, 359: 572–577. 10.1016/S0140-6736(02)07746-2View ArticleGoogle Scholar
- Yuan M, Carmichael WW: Detection and analysis of the cyanobacterial peptide hepatotoxins microcystin and nodularin using SELDI-TOF mass spectrometry. Toxicon 2004, 44: 561–570. 10.1016/j.toxicon.2004.07.015View ArticlePubMedGoogle Scholar
- Gretzer MB, Chan DW, van Rootselaar CL, Rosenzweig JM, Dalrymple S, Mangold LA, Partin AW, Veltri RW: Proteomic analysis of dunning prostate cancer cell lines with variable metastatic potential using SELDI-TOF. The Prostate 2004, 60: 325–331. 10.1002/pros.20066View ArticlePubMedGoogle Scholar
- Pusztai L, Gregory BW, Baggerly KA, Peng B, Koomen J, Kuerer HM, Esteva FJ, Symmans WF, Wagner P, Hortobagyi GN, Laronga C, Semmes OJ, Wright GL Jr, Drake RR, Vlahou A: Pharmacoproteomic Analysis of Prechemotherapy and Postchemotherapy Plasma Samples from Patients Receiving Neoadjuvant or Adjuvant Chemotherapy for Breast Carcinoma. Cancer 2004, 100: 1814–1822. 10.1002/cncr.20203View ArticlePubMedGoogle Scholar
- Knigge T, Monsinjon T, Andersen OK: Surface-enhanced laser desorption/ionization-time of flight-mass spectrometry approach to biomarker discovery in blue mussels (Mytilus edulis) exposed to polyaromatic hydrocarbons and heavy metals under field conditions. Proteomics 2004, 4: 2722–2727. 10.1002/pmic.200300828View ArticlePubMedGoogle Scholar
- Yang ECC, Guo J, Diehl G, DeSouza L, Rodrigues MJ, Romaschin AD, Colgan TJ, Siu KWM: Protein Expression Profiling of Endometrial Malignancies Reveals a New Tumor Marker: Chaperonin 10. Journal of Proteome Research 2004, 3: 636–643. 10.1021/pr049975zView ArticlePubMedGoogle Scholar
- Krieg RC, Fogt F, Braunschweig T, Herrmann PC, Wollscheidt V, Wellmann A: ProteinChip Array analysis of microdissected colorectal carcinoma and associated tumor stroma shows specific protein bands in the 3.4 to 3.6 kDa range. Anticancer Res 2004, 24: 1791–1796.PubMedGoogle Scholar
- Papadopoulos MC, Abel PM, Agranoff D, Stich A, Tarelli E, Bell BA, Planche T, Loosemore A, Saadoun S, Wilkins P, Krishna S: A novel and accurate diagnostic test for human African trypanosomiasis. The Lancet 2004, 363: 1358–1363. 10.1016/S0140-6736(04)16046-7View ArticleGoogle Scholar
- Furuta M, Shiraishi T, Okamoto H, Mineta T, Tabuchi K, Shiwa M: Identification of pleiotrophin in conditioned medium secreted from neural stem cells by SELDI-TOF and SELDI-tandem mass spectrometry. Brain Res Dev Brain Res 2004, 152: 189–197. 10.1016/j.devbrainres.2004.06.014View ArticlePubMedGoogle Scholar
- Gravett MG, Novy MJ, Rosenfeld RG, Reddy AP, Jacob T, Turner M, McCormack A, Lapidus JA, Hitti J, Eschenbach DA, Roberts CT Jr, Nagalla SR: Diagnosis of intra-amniotic infection by proteomic profiling and identification of novel biomarkers. Journal of American Medical Association 2004, 292: 462–469. 10.1001/jama.292.4.462View ArticleGoogle Scholar
- Wadsworth JT, Somers KD, Cazares LH, Malik G, Adam BL, Stack BC Jr, Wright GL Jr, Semmes OJ: Serum Protein Profiles to Identify Head and Neck Cancer. Clinical Cancer Research 2004, 10: 1625–1632. 10.1158/1078-0432.CCR-0297-3View ArticlePubMedGoogle Scholar
- Dare TO, Davies HA, Turton JA, Lomas L, Williams TC, York MJ: Application of surface-enhanced laser desorption/ionization technology to the detection and identification of urinary parvalbumin-alpha: A biomarker of compound-induced skeletal muscle toxicity in the rat. Electrophoresis 2002, 23: 3241–3251. 10.1002/1522-2683(200209)23:18<3241::AID-ELPS3241>3.0.CO;2-DView ArticlePubMedGoogle Scholar
- Yusui Y, Pepe M, Thompson ML, Adam BL, Wright GL Jr, Qu Y, Potter JD, Winget M, Thornquist M, Feng Z: A data-analytic strategy for protein biomarker discovery: profiling of high-dimensional proteomic data for cancer detection. Biostatistics 2003, 4: 449–463. 10.1093/biostatistics/4.3.449View ArticleGoogle Scholar
- Grizzle WE, Semmes OJ, Basler J, Izbicka E, Feng Z, Kagan J, Adam BL, Troyer D, Srivastava S, Thornquist M, Zhang Z, Thompson IM: The early detection research network surface-enhanced laser desorption and ionization prostate cancer detection study: a study in biomarker validation in genitourinary oncology. Urologic Oncology 2004, 22: 337–343.View ArticlePubMedGoogle Scholar
- Rogers MA, Clarke P, Noble J, Munro NP, Paul A, Selby PJ, Banks RE: Proteomic profiling of urinary proteins in renal cancer by surface enhanced laser desorption ionization and neural-network analysis: identification of key issues affecting potential clinical utility. Cancer Research 2003, 63: 6971–6983.PubMedGoogle Scholar
- Coombes KR, Fritsche HA Jr, Clarke C, Chen JN, Baggerly KA, Morris JS, Xiao LC, Hung MC, Kuerer HM: Quality control and peak finding for proteomics data collected from nipple aspirate fluid by surface enhanced laser desorption and ionization. Clinical Chemistry 2003, 49: 1615–1623. 10.1373/49.10.1615View ArticlePubMedGoogle Scholar
- Drake RR, Cazares LH, Corica A, Malik G, Schwegler EE, Libby AE, Wright GL Jr, Adam BL, Semmes OJ: Quality control, preparation, and protein stability issues for blood serum and plasma used in biomarker discovery and proteomic profiling assays. Bioprocessing Journal 2004, 3: 45–50.Google Scholar
- Qu Y, Adam BL, Yasui Y, Ward MD, Cazares LH, Schellhammer PF, Feng Z, Semmes OJ, Wright GL Jr: Boosted decision tree analysis of surface-enhanced laser desorption/ionization mass spectral serum profiles discriminates prostate cancer from noncancer patients. Clinical Chemistry 2002, 48: 1835–1843.PubMedGoogle Scholar
- Zhukov TA, Johanson RA, Cantor AB, Clark RA, Tockman MS: Discovery of distinct protein profiles specific for lung tumors and pre-malignant lung lesions by SELDI mass spectrometry. Lung Cancer 2003, 40: 267–279.View ArticlePubMedGoogle Scholar
- Jolliffe IT: Principal Component Analysis. Springer-Verlag; 1986.View ArticleGoogle Scholar
- Gilbert K, Figueredo S, Meng XY, Yip C, Fung ET: Serum protein-expression profiling using the ProteinChip biomarker system. Methods Mol Biol 2004, 264: 259–69.PubMedGoogle 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.