Histological staining methods preparatory to laser capture microdissection significantly affect the integrity of the cellular RNA
© Wang et al; licensee BioMed Central Ltd. 2006
Received: 23 November 2005
Accepted: 27 April 2006
Published: 27 April 2006
Gene expression profiling by microarray analysis of cells enriched by laser capture microdissection (LCM) faces several technical challenges. Frozen sections yield higher quality RNA than paraffin-imbedded sections, but even with frozen sections, the staining methods used for histological identification of cells of interest could still damage the mRNA in the cells. To study the contribution of staining methods to degradation of results from gene expression profiling of LCM samples, we subjected pellets of the mouse plasma cell tumor cell line TEPC 1165 to direct RNA extraction and to parallel frozen sectioning for LCM and subsequent RNA extraction. We used microarray hybridization analysis to compare gene expression profiles of RNA from cell pellets with gene expression profiles of RNA from frozen sections that had been stained with hematoxylin and eosin (H&E), Nissl Stain (NS), and for immunofluorescence (IF) as well as with the plasma cell-revealing methyl green pyronin (MGP) stain. All RNAs were amplified with two rounds of T7-based in vitro transcription and analyzed by two-color expression analysis on 10-K cDNA microarrays.
The MGP-stained samples showed the least introduction of mRNA loss, followed by H&E and immunofluorescence. Nissl staining was significantly more detrimental to gene expression profiles, presumably owing to an aqueous step in which RNA may have been damaged by endogenous or exogenous RNAases.
RNA damage can occur during the staining steps preparatory to laser capture microdissection, with the consequence of loss of representation of certain genes in microarray hybridization analysis. Inclusion of RNAase inhibitor in aqueous staining solutions appears to be important in protecting RNA from loss of gene transcripts.
Usually, a minimum of 5–50 μg total RNA is required for indirect or direct labeling of cDNA, respectively, with fluorochromes such as Cy3 and Cy5 to perform hybridization on one microarray chip to analyze gene expression . RNA yields of this magnitude are usually impossible from the small number of cells that typically comprise LCM-derived samples, requiring amplification to make cDNA microarray analysis possible. RNA amplification by RNA polymerase-based in vitro transcription is thought to introduce the least bias to gene expression profiling [15–17], although it is possible that amplifications methods have the potential of amplifying the effect of damage to mRNAs introduced during LCM processing.
It has not been determined to what extent each of the steps in LCM sample processing damages the RNA, potentially leading to loss of representation of certain genes when amplified, reverse transcribed, labeled and hybridized to a microarray containing a wide spectrum of cDNA targets on a glass slide. The purpose of the present study was to assess how much the choice of histological staining before LCM would contribute to gene loss, recognized as dropout of hybridized spots, when compared with unstained samples. To render our analysis statistically significant we performed four biological replicates for each staining protocol, starting with 16 individual LCM-derived tissue samples. Each of these 16 was compared with a pool of RNA from an identical pellet of cells. We found that the MGP staining method, which included RNAase inhibitor, was the least detrimental to the RNA, while the Nissl method, which had no RNAase inhibitor, inflicted the most damage to the RNA.
Comparisons between LCM subgroups
Number of genes significantly differently expressed (p < 0.001 and > 2-fold mean expression difference) after LCM and staining.
Number of spots differently hybridizing after staining with stains other than MGP.
Loss of mRNA suitable for hybridizing to some of the cDNAs was the result expected from damage during processing, and this loss was expected to show no gene specificity. The accuracy of this prediction can be visually appreciated in the heat maps presented in Figure 3. Thus, no additional analysis was done to analyze which specific genes were involved in expression changes among the individual chips.
Superimposed upon this experimentally induced spot drop out are the differences inherent in the microarray hybridization method. The number of spots that show increased intensities (always lower than the number of spots with decreased intensities) may be a rough measure of this effect in each subgroup. Thus, we subtracted this value form the number of spots with lower intensities in the stained samples to yield a value that can be used for inter-group comparison.
The cDNA microarray hybridization method is now widely used to analyze simultaneously the expression of thousands of genes in cancer tissues [5, 21], but the contamination of tumors by epithelial, stromal or immune cells presents problems for obtaining accurate expression profiles. Microarray analysis coupled with LCM is a good way to solve this problem. However, cDNA microarray analysis requires a large amount of total RNA (5 μg to 100 μg). It has been shown that one or two rounds of amplification can be used reliably, without compromising RNA quality or gravely skewing patterns of gene expression, when only small initial amounts of total RNA are available [16, 17, 22]. In this study, we used two rounds of amplification to obtain amounts of antisense RNA sufficient for microarray hybridization yielding strong signal intensities, permitting statistical data processing.
To positively identify the specific cells desired for the microdissection, a staining protocol must be used, which permits identification of plasma cells in desiccated cryosections. Such sections, without cover slips, are required for microdissection and may pose a critical problem, since desiccate cryosections are not optically ideal for histological examination (cf. Figure 1). However, staining solutions may react with cellular components, such as RNA, potentially adversely affecting RNA integrity. In addition, the aqueous components of the staining reagents may contain or activate intracellular RNAases and result in RNA degradation. For this reason, it was important to test different conditions for tissue staining and section dehydration to assure high quality of RNA from LCM samples. To assure ourselves of adequate numbers of uniform populations of plasma cells, which are scarce in most tissues, we utilized pellets of cultured plasma cell tumor cells that were sectioned and stained as if they had been collected by LCM. We used this approach, because obtaining large enough sample of plasma cell tumor cells from tissues by LCM would have been impracticable.
Our modification (addition of RNAase inhibitor) to the standard MGP staining method has made it suitable for use with LCM. Sections treated as described here can be used to prepare total RNA of good quality. Though pyronin is a fluorescent molecule that can bind to RNA, we detected only 21 changes when compared with unstained LCM samples. This amounted to less than 0.3% of the analyzed microarray elements, and less than one half of the number changed due to the other staining methods tested.
It appears that inclusion of RNAase inhibitor in staining solutions can prevent RNA from damage by endogenous and exogenous RNAase. The Nissl-stained group was the only staining protocol that did not include RNAase inhibitor. The number of genes in this group with 2-fold changes, 70 spots, was the highest among all four groups, indicating that this staining procedure damaged the most mRNAs.
In summary, we performed a statistical analysis of 16 cDNA microarray chips comparing fluorescent-labeled cDNA generated from amplified RNA from 16 independently isolated and processed, LCM-procured, stained frozen sections with amplified RNA from an identical pellet of plasma cell tumor cells. This showed that a variable amount of spot fall-out was seen after staining of the frozen sections, depending on the staining method, indicating that the choice of staining method can be important in minimizing loss of mRNAs from microarray analysis of LCM-derived RNA. Happily, the degree of mRNA damage was small with the commonly used H&E method, so long as RNAase inhibitor was included in the aqueous hematoxylin solution. Nissl staining solution, which may not be compatible with RNAase inhibitor addition, introduced the greatest degree of damage. Fortunately, for our ongoing study of plasma cell tumor development, the MGP stain, chosen for its unique ability to flag plasma cells in tissue sections, was the least destructive of all. Our analysis suggests that maximizing the use of non-aqueous staining solutions is recommended for other specialty stains that might be advantageous for identification of other cell types in studies that require use of LCM. The inclusion of RNAase inhibitor in aqueous solutions appears to be important in rendering histological staining procedures suitable to extraction of intact RNA.
RNA damage can occur during the staining steps preparatory to laser capture microdissection, resulting in loss of representation of certain genes in microarray hybridization analysis. Inclusion of RNAase inhibitor in aqueous staining solutions appears to be important in protecting RNA from loss of gene transcripts.
The murine plasma cell tumor cell line TEPC 1165  was used both as a source of frozen sections for LCM as well as a source of reference RNA, i.e., RNA extracted without LCM processing. Cells were grown in RPMI 1640 media supplemented with 10% fetal bovine serum, 10 ng/ml IL-6, 4 mM L-glutamine and 100 units/ml penicillin and 100 units/ml streptomycin. Duplicate cell pellets were collected by low-speed centrifugation in 50-ml conical centrifuge tubes and immediately snap frozen for RNA extraction and LCM preparation.
Preparation of LCM samples
The cell pellets were frozen in a dry ice-ethanol bath, embedded with OCT frozen tissue embedding medium, and stored at -80°C. Six-micrometer serial frozen sections were cut on a cryostat at -20°C and mounted on chilled, pre-cleaned, uncoated microscope glass slides that had been heated at 200°C overnight to eliminate RNAase. Slides were kept on dry ice or at -80°C until LCM was completed. Sections were stained individually with four different methods before LCM. RNAase-free (treated with diethylpyrocarbonate) water was used for all solutions. All procedures were carried out at room temperature, unless otherwise stated, and usually on the same day as the frozen sectioning.
Methyl Green Pyronin staining
After fixation in 70% ethanol (10 sec), each slide was rinsed in DEPC-treated water (5 sec). Then, 50 μl of MGP staining solution (Sigma/Aldrich) to which RNAase inhibitor (SUPERase*In, Ambion) had been added to a final concentration of 500 U/ml, was applied to the section and incubated 2 min, followed by a quick dip in diethylpyrocarbonate-treated water, dehydration in 100% ethanol in 2 steps (quick dip, then for 10–15 sec), and xylenes in 2 steps (2 min then 3 min). It is important to note that MGP, like all histological stains, yields significantly less detail on frozen sections than on paraffin-embedded tissue, in part, owing to lack of a cover slip. MGP did enable identification of plasma cells in oil granulomas under these conditions. To be used in this way, however, one must include RNAase inhibitor if RNA extraction is desired after LCM.
Hematoxylin and eosin staining
A modification was incorporated into the manufacturer's (Sigma/Aldrich) protocol; namely, RNAase inhibitor (SUPERase*In, Ambion) was added to the aqueous hematoxylin solution to a final concentration of 500 U/ml. Briefly, slides were stained as follows: 70% EtOH for 10 sec, DEPC-treated water for 5 sec, hematoxylin with RNAase inhibitor for 20 sec, 70% EtOH for 30 sec, eosin Y in 100% EtOH for 20 sec, followed by dehydration with a series of alcohol for 30 sec each, and xylenes for 2 min.
The frozen sections were fixed with 70% ethanol (30 sec), rinsed in DEPC-treated water (30 sec), immersed in NS staining solution (Arcturus Corp.) for 30 sec, dehydrated with graded alcohols (30 sec each) and xylenes (2 min and 3 min).
Sections were fixed in ice-cold acetone for 2 min, rinsed in DEPC-treated PBS for 10 sec, incubated with a 1:250 dilution of goat anti-mouse Igκ conjugated with TXRD (Southern Biotechnology Associates, Inc.) to which RNAase inhibitor (SUPERase*In, Ambion) had been added to a final concentration of 500 U/ml, for 2 min in the dark, dehydrated with 100% ethanol and xylenes (11).
Laser capture microscopy
Cells were lifted from the slide onto CapSure LCM plastic caps (Arcturus Corp.) using a 30-μm spot size of laser at the power of 50 μv using PixCell II LCM system (Arcturus Corp.).
RNA extraction and purification
Total RNA from the unprocessed cell pellet was extracted using TriZol reagent (Sigma), treated with RNAase-free DNAase I (Qiagen, Inc.), followed by RNA purification with Qiagen RNeasy kit according to the manufacturer's protocol. RNA concentration was measured spectrophotometrically.
RNA extraction of 16 LCM-derived samples (4 from each staining group) was performed with the PicoPure RNA extraction kit (Arcturus Corp.), simultaneously treated with RNAase-free DNAase I (Qiagen, Inc) according to the manufacturer's protocol. Briefly, 50 μl of extraction buffer was pipetted onto the tissue fragments that had been collected in the cap of a sterile plastic 1.5-micro-centrifuge tube (CapSure LCM caps, Arcturus Corp.), and they were incubated together in a hot-air oven at 42°C for 30 min. Cells and buffer were collected by centrifugation at 800 × g for 2 min. RNA was DNAase-treated and purified on a preconditioned RNA-purification spin column according to the manufacturer's recommendations. LCM-derived total RNA was quantitated with Ribogreen (Molecular Probes, Inc.) using Victor 2 1420 multilabel counter (Perkin-Elmer/Wallac). To assess the RNA quality and confirm the absence of genomic DNA contamination, the Agilent 2100 Bioanalyzer (Pico RNA chip) was used to perform electrophoresis on picogram amounts of total RNA. Only samples with intact 18S and 28S rRNA peaks were used for gene expression analysis.
We performed antisense RNA amplification of each RNA sample with RiboAmp™ RNA Amplification kit (Arcturus, Inc.) based on T7 RNA polymerase in vitro transcription of cDNA synthesized by reverse transcription primed by oligo-dT attached to the T7 RNA polymerase promoter . To avoid bias caused by different starting amounts of total RNA in the amplification process, 5 ng of total RNA were used in all experimental and reference samples. To obtain sufficient antisense RNA for microarray experiments, two rounds of amplification were performed on all samples. The total yield of amplification products was determined using Ribogreen (Molecular Probes, Inc.), and the sizes of the amplified RNAs were determined with the Agilent Bioanalyzer (nano RNA chip).
Probe labeling and cDNA array hybridization
Probes were synthesized and labeled from 4 μg of amplified RNA (aRNA) by the method of Xiang et al . In brief, 4 μg of amplified RNA were combined with 4 μg amine-modified random primer and 5 units of RNAase inhibitor (SUPERase*In, Ambion). The mixture was incubated at 70°C for 10 min, then chilled on ice for 10 min, and left at room temperature for 10 min. Primer-RNA solution was added to the reverse transcriptase mix (including 0.5 mM dATP, dGTP, dCTP, 0.3 mM dTTP, and 0.2 mM aminoallyl-dUTP) and incubated at 42°C for 2 h. The reaction was terminated by adding 10 μl 0.5 M EDTA, and RNA was hydrolyzed with 10 μl 1 M NaOH at 65°C for 30 min. The solution was neutralized with 10 μl 1 M HCl, and then MinElute PCR purification kits (Qiagen, Inc) were used to purify the cDNAs. The cDNA was eluted three times with 10 μl of H2O. In each staining group 3 of the 4 reference RNA samples were labeled with Cy5, while 1 of the 4 was labeled with Cy3. Three of each group of 4 experimental samples were labeled with Cy3, and 1 of 4 was labeled with Cy5 to test for bias from labeling. Probes were mixed and hybridized overnight at 42°C in hybridization solution (New England Nuclear) to microarray chips with a total of 9998 cDNA elements (InCyte) printed by the NCI microarray facility.
Evaluation of gene-specific dye bias
Four arrays were hybridized for each of the 4 groups (MGP, HE, IF, NS). In each group 3 arrays had cDNA derived from the reference, unsectioned pellet of cells labeled with Cy5 and the LCM-derived test sample labeled with Cy3, and one array had the labeling reversed. We were interested in evaluating the possibility of gene-specific dye bias that was not removed by the normalization process. To evaluate this possibility, we computed the average difference in log2 ratio between normalized forward- and reverse-label experiments for each array element as an estimate of the residual dye bias for the associated gene. Of the 8534 array elements, none had an average difference of log2 ratio in absolute value greater than 1 (i.e., 2-fold difference). Thus, we concluded that after normalization there was no systematic bias favoring either dye.
Array scanning and data processing
Arrays were scanned at 10-μm resolution using a GenePix 4000 (Axon Inc.) array scanner. PMT voltages were varied to gain optimum intensity of the spots. Four duplicate hybridizations/slides were used in each experiment. LCM-derived RNAs were regarded as "experimental" samples, and 3 of the 4 cDNA probes derived from them were labeled with Cy3, whereas the "reference" samples, derived from RNA from unprocessed cell pellets were usually labeled with Cy5. To rule out color bias produced by the labeling process itself, the fluorochromes were reversed once in each group of four repeats. Quantification files were obtained by using the GenePixPro 4.0 software (Axon Inc.). The fluorescence signal intensity was determined as the volume in a fixed-size circle, and background was estimated as the median pixel value in a diamond-shaped region between each adjacent four-spot region.
Data filtering and normalization
Log-ratios of local median background-subtracted intensity levels were analyzed. For each array element, if the intensity was less than 75 in one channel but greater than 75 in the other channel, the intensity <75 was set at 75. Array elements that were flagged as poor quality during image analysis or that had intensities less than 75 in both channels were not reliable and treated as missing . Six percent of the array elements were found to be missing in this way. Those samples with missing values for a particular array element were excluded from all analyses involving that array element. In order to have enough samples in each group to compare expression profiles between the groups, array elements were further filtered out and removed from analysis if their number of non-missing values in each group was less than three. Eleven percent of the array elements were filtered out in this way, resulting in 8534 array elements for analysis. Log-ratios for each microarray were normalized by the locally robust lowess smoother (20) to adjust for any dye bias and variation in PMT voltage settings.
Average linkage hierarchical analysis of differentially expressed genes was performed by using 1-Pearson correlation as the distance metric to group genes based on their patterns of variation across the staining groups. The clusters and associated heat maps were implemented in R.
Enumeration of differentially expressed genes
Identification of array elements that were differentially expressed between each subgroup and the reference sample and among subgroups was done using one-sample t-tests and two-sample t-tests, respectively. Differentially expressed genes were identified as those genes that were significant at the 0.001 level (p < 0.001) and which were at least 2-fold different in the geometric mean of the expression measurements. A treated subgroup was considered to have weaker expression than another treated subgroup if, among the genes that were differentially expressed between these two subgroups, it had more lower-expressed genes than higher-expressed genes. A global permutation test was used as a global test to assess if one subgroup had statistically significant weaker expression than the other subgroup. Specifically, two-sample t-tests were performed by randomly permuting the group labels. In each permutation, the difference between the number of lower-expressed genes and the number of higher-expressed genes was recorded. The p-value of the permutation test was the proportion of permutations with the same or larger difference as those observed in the original data. Two subgroups would be defined as having globally significant different expression profiles if the p-value from the permutation test is less than 0.05.
- LCM (Laser Capture Microdissection):
MGP (Methyl Green Pyronin), H&E (Hematoxylin and Eosin), NS (Nissl Stain), IF (Immunofluorescence).
This research was supported in part by the Intramural Research Program of the NIH and the National Cancer Institute. We thank Dr. Liya Shen for the use of the refrigerated microtome, Dr. Charlie Xiang for assistance with his RNA labeling methodology, Elizabeth Mushinski, Alexander Kovalchuk and Richardo Dreyfus for assistance in photomicroscopy.
- Heller RA, Schena M, Chai A, Shalon D, Bedilion T, Gilmore J, Woolley DE, Davis RW: Discovery and analysis of inflammatory disease-related genes using cDNA microarrays. Proc Natl Acad Sci USA. 1997, 94: 2150-2155. 10.1073/pnas.94.6.2150.PubMedPubMed CentralView ArticleGoogle Scholar
- Duggan DJ, Bittner M, Chen Y, Meltzer P, Trent JM: Expression profiling using cDNA microarrays. Nat Genet. 1999, 21 (Suppl 1): 10-14. 10.1038/4434.PubMedView ArticleGoogle Scholar
- Shaffer AL, Lin KI, Kuo TC, Yu X, Hurt EM, Rosenwald A, Giltnane JM, Yang L, Zhao H, Calame K, Staudt LM: Blimp-1 orchestrates plasma cell differentiation by extinguishing the mature B cell gene expression program. Immunity. 2002, 17: 51-62. 10.1016/S1074-7613(02)00335-7.PubMedView ArticleGoogle Scholar
- Nakazono M, Qiu F, Borsuk LA, Schnable PS: Laser-capture microdissection, a tool for the global analysis of gene expression in specific plant cell types: identification of genes expressed differentially in epidermal cells or vascular tissues of maize. Plant Cell. 2003, 15: 583-596. 10.1105/tpc.008102.PubMedPubMed CentralView ArticleGoogle Scholar
- Sgroi DC, Teng S, Robinson G, LeVangie R, Hudson JR, Elkahloun AG: In vivo gene expression profile analysis of human breast cancer progression. Cancer Res. 1999, 59: 5656-5661.PubMedGoogle Scholar
- Emmert-Buck MR, Bonner RF, Smith PD, Chuaqui RF, Zhuang Z, Goldstein SR, Weiss RA, Liotta LA: Laser capture microdissection. Science. 1996, 274: 998-1001. 10.1126/science.274.5289.998.PubMedView ArticleGoogle Scholar
- Luo L, Salunga RC, Guo H, Bittner A, Joy KC, Galindo JE, Xiao H, Rogers KE, Wan JS, Jackson MR, Erlander MG: Gene expression profiles of laser-captured adjacent neuronal subtypes. Nat Med. 1999, 5: 117-122. 10.1038/4806.PubMedView ArticleGoogle Scholar
- Pedersen TX, Leethanakul C, Patel V, Mitola D, Lund LR, Dano K, Johnsen M, Gutkind JS, Bugge TH: Laser capture microdissection-based in vivo genomic profiling of wound keratinocytes identifies similarities and differences to squamous cell carcinoma. Oncogene. 2003, 22: 3964-3976. 10.1038/sj.onc.1206614.PubMedView ArticleGoogle Scholar
- Krizman DB, Chuaqui RF, Meltzer PS, Trent JM, Duray PH, Linehan WM, Liotta LA, Emmert-Buck MR: Construction of a representative cDNA library from prostatic intraepithelial neoplasia. Cancer Res. 1996, 56: 5380-5383.PubMedGoogle Scholar
- Hiller T, Snell L, Watson PH: Microdissection RT-PCR analysis of gene expression in pathologically defined frozen tissue sections. Biotechniques. 1996, 21: 38-40. 42, 44PubMedGoogle Scholar
- Fend F, Emmert-Buck MR, Chuaqui R, Cole K, Lee J, Liotta LA, Raffeld M: Immuno-LCM: laser capture microdissection of immunostained frozen sections for mRNA analysis. Am J Pathol. 1999, 154: 61-66.PubMedPubMed CentralView ArticleGoogle Scholar
- Mikulowska-Mennis A, Taylor TB, Vishnu P, Michie SA, Raja R, Horner N, Kunitake ST: High-quality RNA from cells isolated by laser capture microdissection. Biotechniques. 2002, 33: 176-179.PubMedGoogle Scholar
- Mount S, Mead P, Cooper K: Chlamydia trachomatis in the endometrium: can surgical pathologists identify plasma cells?. Adv Anat Pathol. 2001, 8: 327-329. 10.1097/00125480-200111000-00002.PubMedView ArticleGoogle Scholar
- Xiang CC, Kozhich OA, Chen M, Inman JM, Phan QN, Chen Y, Brownstein MJ: Amine-modified random primers to label probes for DNA microarrays. Nat Biotechnol. 2002, 20: 738-742. 10.1038/nb0702-738.PubMedView ArticleGoogle Scholar
- Phillips J, Eberwine JH: Antisense RNA Amplification: A Linear Amplification Method for Analyzing the mRNA Population from Single Living Cells. Methods. 1996, 10: 283-288. 10.1006/meth.1996.0104.PubMedView ArticleGoogle Scholar
- Wang E, Miller LD, Ohnmacht GA, Liu ET, Marincola FM: High-fidelity mRNA amplification for gene profiling. Nat Biotechnol. 2000, 18: 457-459. 10.1038/74546.PubMedView ArticleGoogle Scholar
- Feldman AL, Costouros NG, Wang E, Qian M, Marincola FM, Alexander HR, Libutti SK: Advantages of mRNA amplification for microarray analysis. Biotechniques. 2002, 33: 906-914.PubMedGoogle Scholar
- Nordan RP, Potter M: A macrophage-derived factor required by plasma cell tumors for survival and proliferation in vitro. Science. 1986, 233: 566-569.PubMedView ArticleGoogle Scholar
- Simon R, Korn E, McShane LM, Radmacher MD, Wright GW, Zhao Y: Design and analysis of DNA microarray investigations. Springer Verlag.
- Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J: Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res. 2002, 30: e15-10.1093/nar/30.4.e15.PubMedPubMed CentralView ArticleGoogle Scholar
- Kitahara O, Furukawa Y, Tanaka T, Kihara C, Ono K, Yanagawa R, Nita ME, Takagi T, Nakamura Y, Tsunoda T: Alterations of gene expression during colorectal carcinogenesis revealed by cDNA microarrays after laser-capture microdissection of tumor tissues and normal epithelia. Cancer Res. 2001, 61: 3544-3549.PubMedGoogle Scholar
- Xiang CC, Chen M, Kozhich OA, Phan QN, Inman JM, Chen Y, Brownstein MJ: Probe generation directly from small numbers of cells for DNA microarray studies. Biotechniques. 2003, 34: 386-393.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.