- Open Access
Assessing unmodified 70-mer oligonucleotide probe performance on glass-slide microarrays
© Wang et al.; licensee BioMed Central Ltd. 2003
- Received: 14 August 2002
- Accepted: 8 November 2002
- Published: 6 January 2003
Long oligonucleotide microarrays are potentially more cost- and management-efficient than cDNA microarrays, but there is little information on the relative performance of these two probe types. The feasibility of using unmodified oligonucleotides to accurately measure changes in gene expression is also unclear.
Unmodified sense and antisense 70-mer oligonucleotides representing 75 known rat genes and 10 Arabidopsis control genes were synthesized, printed and UV cross-linked onto glass slides. Printed alongside were PCR-amplified cDNA clones corresponding to the same genes, enabling us to compare the two probe types simultaneously. Our study was designed to evaluate the mRNA profiles of heart and brain, along with Arabidopsis cRNA spiked into the labeling reaction at different relative copy number. Hybridization signal intensity did not correlate with probe type but depended on the extent of UV irradiation. To determine the effect of oligonucleotide concentration on hybridization signal, 70-mers were serially diluted. No significant change in gene-expression ratio or loss in hybridization signal was detected, even at the lowest concentration tested (6.25 μm). In many instances, signal intensity actually increased with decreasing concentration. The correlation coefficient between oligonucleotide and cDNA probes for identifying differentially expressed genes was 0.80, with an average coefficient of variation of 13.4%. Approximately 8% of the genes showed discordant results with the two probe types, and in each case the cDNA results were more accurate, as determined by real-time PCR.
Microarrays of UV cross-linked unmodified oligonucleotides provided sensitive and specific measurements for most of the genes studied.
- Additional Data File
- Oligonucleotide Probe
- Probe Type
- cDNA Array
- Hybridization Signal Intensity
The advent of microarray technology has enabled scientists to investigate biological questions in a more global fashion. Instead of studying genes individually, the expression of thousands of genes can be analyzed simultaneously using probes attached to the surface of a microscope slide [1,2,3,4,5,6]. The cDNA microarray represents a popular array type in which double-stranded PCR products amplified from expressed sequence tag (EST) clones are spotted onto glass slides [7,8], allowing gene-expression profiles to be determined with high reproducibility and efficiency. However, construction of cDNA microarrays presents a number of challenges, largely related to costs associated with clone validation, tracking and maintenance. The laborious and problematic tracking of cDNA clones and PCR amplicons may lead to 10-30% misidentification of clones . For all practical purposes, sequence verification of array elements is an ongoing necessity. Other limitations of cDNA microarrays are their difficulty, because of cross-hybridization, in discriminating expression patterns of homologous genes, alternative splice variants and antisense RNAs.
Alternatively, microarrays can be composed of short oligonucleotides (25 bases) synthesized directly onto a solid matrix using photolithographic technology (Affymetrix) [2,9] or constructed from long oligonucleotides (55-70 bases) spotted onto glass slides [10,11,12]. To mimic the Affymetrix design of freely moving probes tethered at one end onto a solid support, in-house manufactured or commercially available long oligonucleotides are modified by the addition of a 5' amino group for covalent attachment onto pre-activated glass slides [5,10]. This oligonucleotide design strategy has been widely viewed as a prerequisite for accurate gene-expression measurements. However, there is no clear evidence that other covalent attachments do not form. With oligonucleotide arrays, problems related to clone tracking, handling of glycerol stocks and failed PCR amplifications are avoided. The completion of numerous microbial, plant and eukaryotic genomes, as well as extensive EST data, provides sufficient sequence information to design unique oligonucleotides capable of distinguishing homologous genes and alternative splice variants. As such, oligonucleotide probes have an added flexibility over PCR amplicons.
Comprehensive studies comparing the Affymetrix approach with cDNA arrays have only recently appeared in the literature [13,14]. Studies comparing long oligonucleotides to cDNA arrays have not been as forthcoming. In the only example to date, 5'-amino-modified 50-mers representing prokaryotic genes were compared to corresponding PCR amplicons . Analysis of the hybridization signals derived from these two probe types, while providing important insights pertaining to sensitivity and specificity, were limited in scope (total of eight genes) and design (interrogation was carried out with complementary targets derived from synthetic RNA as opposed to cellular RNA). A drawback to using modified oligonucleotides is the significant cost associated with the addition of the 5'-amino linker. An alternative strategy is to utilize unmodified oligonucleotides spotted onto glass slides, where attachment is believed to be primarily ionic in nature . However, a comparison of this approach to standard cDNA arrays has yet to be provided. It is imperative that comparisons be carried out on all probe types in the light of conflicting reports regarding the correlation between Affymetrix and cDNA array-based expression measurements [13,14]. Whereas one study shows both approaches correctly identifying 16 out of 17 differentially regulated genes , a second study found a correlation of r = 0.328 between matched results from the same two platforms . Discordant results were not resolved in the latter study. Here we test the performance of unmodified 70-mers printed alongside PCR amplicons. Using this unique study design, both probe types can be simultaneously interrogated with a complex target composed of both cellular and synthetic RNA.
Optimal attachment parameters for 70-mers and PCR amplicons on the same slide
The success of microarray assays requires stable binding and retention of probes throughout the entire printing/blocking/hybridization/washing process. Oligonucleotides were spotted alongside PCR amplicons onto TeleChem SuperAmine aminated slides, and immobilized by ultraviolet (UV) cross-linking. To determine the optimal UV cross-linking energy required for efficient oligonucleotide immobilization, a series of spotted arrays from the same printing session were subjected to increasing UV energy (70, 150, 250, and 450 mJ/cm2). Deposition and retention of the probe onto aminated slides were assessed by: staining with Vistra Green solution and subsequent fluorescence scanning at 532 nm; and hybridization with Cy-labeled targets derived from rat brain and heart RNA. Using these two methods, optimal retention of both oligonucleotide and PCR amplicon probes was determined to occur between 250 and 450 mJ/cm2 as described below.
These results show that immobilization of unmodified 70-mer oligonucleotides to SuperAmine aminated slides by high-UV cross-linking energy is sufficient and comparable to PCR amplicons. Clearly, our oligonucleotide immobilization protocol should be sufficient to sustain routine microarray hybridization and wash procedures, which are much less stringent than the overnight wash at 42°C with 0.2% SDS and no salt.
Sensitivity of unmodified 70-mers on aminated slides
Arabidopsis probe elements serve as excellent negative controls when exogenous cRNA is not added to the labeling reaction. In the absence of cRNA spiking, cross-hybridization of Cy-labeled rat targets to Arabidopsis probe elements was negligible (data not shown).
Concordance between probe types when measuring differential gene expression in biological samples
Validation of microarray results with real-time PCR
There were five notable discrepancies between the two probe types, as compared to the 60 that were in agreement. A discrepancy was defined as a change equal to or greater than twofold measured with one probe type and no change (or a change in the opposite direction) measured with the other probe type. The resolution of the discordant results for inositol-1,4,5-trisphosphate receptor, H+-ATPase, branched aminotransferase and epoxide hydrolase is presented in Figure 5b. In each case, real-time PCR results were in agreement with the PCR amplicon-derived expression ratios. Interestingly, each of the oligonucleotide-derived expression ratios erroneously suggested that these genes were not differentially expressed in heart and brain tissues.
Effect of oligonucleotide probe concentration on signal intensity
In the study reported here, we systematically compared the performance of unmodified 70-mer oligonucleotides to traditional PCR amplicons, both probe types printed and UV cross-linked onto glass slides coated with primary amine groups. Direct comparisons are best accomplished when both probes are printed alongside each other, allowing for simultaneous interrogation with a complex target. Hence, analysis is not confounded by uneven aminosilane coating in different batches of slides, inconsistencies in the array resulting from different print sessions, differences in day-to-day label incorporation, or variations in day-to-day hybridization and wash procedures. A correlation coefficient (r) of 0.80 was obtained from our analysis, indicating that the two probe types gave comparable expression ratios. One variable that was not controlled for in our study was the number of cross-links per DNA molecule. Given a constant UV exposure, many more cross-links per molecule of cDNA probe are presumably formed compared to the shorter oligonucleotide probe. It is possible that the correlation coefficient was not higher as a result of the differential reaction of the two probe types to UV irradiation.
We designed our arrays to contain 75 different probes corresponding to mammalian signal transduction genes with a wide range of expression levels. In heart versus brain comparisons, oligonucleotide probes, like their cDNA probe counterparts, could reproducibly discern differences in mRNA populations as low as twofold (namely, 14-3-3 protein gamma) and as high as around 90-fold (namely, creatine kinase). Hence, the dynamic range of unmodified oligonucleotides is at least two orders of magnitude in fold-change measurements.
In the course of our work, we generated a resource of 10 Arabidopsis spiking control cRNAs along with their corresponding 70-mer oligonucleotide and PCR amplicon probes. As part of our quality-control procedures, all microarray assays routinely incorporate the spiking controls. These reagents will allow the microarray user to add specific concentrations of known transcripts into a complex mix of mammalian target RNA in order to assess, for example, hybridization kinetics, intra-slide variability, inter-slide variability, sensitivity and effectiveness of normalization algorithms. On the basis of experiments with the spiking controls, unmodified oligonucleotides can be used to detect twofold changes in transcript number at a level of 2-20 mRNA copies per cell. It is important to note that our protocol for generating first-strand cDNA target involves the use of random primers. At the outset, the Arabidopsis cRNAs were engineered to contain a 3' poly(A) tail. Hence, alternative protocols using oligo(dT) to prime mRNA for the synthesis of labeled target [15,16] can still take advantage of our spiking control set.
In our initial assessment of cDNA and 70-mer oligonucleotide probe types, the latter was printed at a concentration of 50 μM. Even at a printing concentration as low as 6 μM, oligonucleotide probes were capable of discerning twofold expression differences in complex cellular RNA mixtures and in synthetic spiked cRNAs. In fact, decreasing the oligonucleotide printing concentration from 50 to 6 μM had the effect of increasing the hybridization signal around two- to sixfold for a number of the probes (Figure 7a). The reason is unclear, but it is possible that high-density packing of an oligonucleotide probe within the confines of a small spot interferes with fluorescence emission of the target or hybridization efficiency. Alternatively, the higher spotting concentrations may favor cross-linking of the oligonucleotide probes to each other following UV irradiation. In either case, this phenomenon appears to be sequence dependent as not all probes exhibited this behavior. The present study also demonstrates that longer probes are not necessarily associated with higher hybridization signals, as the hybridization signals from half of the 70-mer oligonucleotide probes were actually higher than or equivalent to their corresponding PCR amplicons, which have an average length of 1 kilobase (kb). Taken together, the combination of unmodified oligonucleotides and low printing concentrations has resulted in an approximately 16-fold reduction in reagent costs. An issue not evaluated in the present study, but one that has significant cost-saving potential, is the effect of reducing the length of unmodified oligonucleotides on microarray sensitivity. Clearly, this is an area for future investigation.
The mechanism of the adherence of unmodified oligonucleotides to glass slides has been addressed . Attachment involves noncovalent interactions such as electrostatic interactions, where the negatively charged phosphate backbone of the oligonucleotide is attracted to the positively charged surface of the glass slide (for example, a surface containing protonated alkyl amines). Whereas noncovalent interactions appear to be the predominant mechanism for oligonucleotide attachment, covalent linkage is likely to have an important supplementary role in UV-irradiated microarrays. This seems plausible as our stringent overnight washes in strong detergent did not appreciably detach unmodified oligonucleotides from the slide surface. The importance of UV cross-linking cannot be overemphasized. Under-irradiation of cDNA arrays is known to cause insufficient binding of DNA and over-irradiation results in over-nicking of DNA samples . A further complicating factor is our finding that oligonucleotides printed onto different slide chemistries (or slides with similar chemistries from different vendors) will have very different optimal UV titration curves. In our hands, optimal UV cross-linking occurred at 450 and 70 mJ/cm2 for oligonucleotides printed onto TeleChem SuperAmine™ and Corning GAP II™ slides, respectively. For TeleChem slides, under-irradiation (70-150 mJ/cm2) causes insufficient oligonucleotide attachment. For Corning slides, over-irradiation (150-450 mJ/cm2) results in a decrease in the hybridization signal that may reflect excessive covalent attachment of oligonucleotides. As UV cross-linking may adversely affect oligonucleotide accessibility to labeled target during hybridization, we cannot discount the possibility that alternative attachment strategies (for example, 5'-amino-modified oligonucleotides) may provide greater sensitivity and specificity. This issue needs to be explored in the future.
In summary, the present study provides evidence that the performance of unmodified 70-mer oligonucleotides is comparable to cDNAs printed on glass slides. Optimal conditions were identified for oligonucleotide attachment and hybridization/wash conditions, resulting in high assay sensitivity and reproducibility. Our results show that unmodified oligonucleotides can provide an accurate, reproducible and cost-effective means to measure gene-expression profiles. Of interest is the fact that our hybridizations were successfully carried out on slides that simultaneously contained both PCR amplicons and oligonucleotides. Hence, future microarrays can be constructed in a modular fashion, with oligonucleotide-based elements being added to existing PCR amplicons as more genomic sequence information is gathered, in the absence of readily available cDNA clones. Lastly, our findings have broader implications, suggesting that the combination of expression measurements across different platforms (for example, Affymetrix and cDNA arrays, unmodified long oligonucleotides and cDNA arrays) within a single analysis maybe feasible .
Constructing exogenous spiking cRNA controls and a PCR amplicon printing set to assess oligonucleotide sensitivity
Arabidopsis 70-mer oligonucleotide probes
Complete gene name
GenBank accession number
Root cap 1 protein
Photosystem I chlorophyll
Lipid transfer protein 4
Lipid transfer protein 6
Ribulose-1,5-bisphosphate carboxylase/oxygenase, large subunit
Papain-type cysteine endopeptidase
Rat 70-mer oligonucleotide design
To minimize cross-hybridization, oligonucleotides of 70 bases (unmodified) were designed using the computer program, Pick70 . Oligonucleotide design considerations included uniqueness, avoidance of internal self-annealing structures, narrow Tm range (75-80°C) over the entire oligonucleotide set and masking of low-complexity regions. The TIGR Rat Gene Index containing a non-redundant set of expressed mRNA sequences  was used as the 'complete genome source' for selecting 70-mer oligonucleotide sequences with Pick70. Oligonucleotides were chosen to represent 'housekeeping' and signal transduction genes, while other 70-mers were designed to detect tissue-specific transcripts from either brain or heart (for example, those for SCG10, creatine kinase, and desmin). The sequences of the 70-mers and corresponding GenBank accession numbers of the genes are available as and additional data file. Oligonucleotides were synthesized at a 50 nmol scale by Invitrogen/Life Technologies (Carlsbad, CA) or Operon Technologies (Alameda, CA), and resuspended in sterile milliQ water to a final concentration of 100 μM. We selected individual cDNA clones from the TIGR Rat Gene Index whose EST sequences corresponded to the same gene from which the 70-mers were designed. Rat cDNA clone inserts were amplified by PCR with M13F (5'-GTTTTCCCAGTCACGACGTTG-3') and M13R (5'-TGAGCGGATAACAATTTCACACAG-3') primers [15,16]. Insert size ranged from 0.5 to 1.5 kb.
Oligonucleotides (50 μM, except where indicated otherwise) and PCR amplicons (100-200 nM) in 50% DMSO were printed onto SuperAmine slides (TeleChem International, Sunnyvale, CA) using an Intelligent Automation Systems (IAS) arrayer (Cambridge, MA) with a 12-pen print head . The rat 70-mers and PCR amplicons were printed in quadruplicate while the 10 Arabidopsis 70-mers and PCR amplicons were spotted into six different sectors on the slide. After printing, DNA was cross-linked to the slides by UV irradiation with a Stratalinker UV Crosslinker (Stratagene, La Jolla, CA) and stored in a vacuum chamber until use. To assess oligonucleotide retention, slides were UV cross-linked and stained for 10 min in Vistra Green Nucleic Acid staining solution (Amersham Pharmacia, Piscataway, NJ) at a 1:10,000 dilution. Afterwards, slides were washed at least five times, 1 min each, in milliQ water at room temperature, centrifuged to dryness (500 rpm × 5 min), and scanned at 535 nm using a dual laser GenePix 4000B scanner (Axon Instruments, Foster City, CA). Subsequently, slides were gently agitated in 0.2% SDS (no salt) overnight at 42°C, washed extensively with water, scanned to ensure that the dye was completely removed, restained with Vistra Green and rescanned.
Target labeling and array hybridization
To generate labeled single-stranded cDNA target, 10 μg total RNA from rat heart or brain (Clontech, Palo Alto, CA) was reverse transcribed for 2-3 h at 42°C in the presence of 6 μg random primers (Invitrogen/Life Technologies), 1x first-strand synthesis buffer (Invitrogen/Life Technologies), 10 mM DTT, dNTP mix (25 mM dATP, 25 mM dCTP, 25 mM dGTP, 15 mM dTTP, 10 mM amino allyl-dUTP), and 200 units Superscript II reverse transcriptase (Invitrogen/Life Technologies). RNA was hydrolyzed with 200 mM NaOH and 100 mM EDTA for 15 min at 65°C, then neutralized with 200 mM HCl. First-strand cDNA was purified from unincorporated amino allyl-dUTPs on QIAquick PCR purification columns (Qiagen, Valencia, CA) according to manufacturer's instructions, except that QIAquick wash buffer was replaced with 5 mM K+ phosphate buffer (pH 8.5) containing 80% ethanol, and cDNA was eluted with 4 mM K+ phosphate buffer (pH 8.5). Eluted cDNA was lyophilized, resuspended in 4.5 μl 0.1 M Na2CO3 buffer (pH 9), mixed with either Cy3 or Cy5 NHS-ester (Amersham Pharmacia), and incubated for 1 h in the dark at room temperature. Cy3- and Cy5-labeled cDNA targets were then purified on QIAquick PCR purification columns, combined and concentrated by lyophilization, and hybridized to the microarray at 42°C for 16 h in hybridization solution containing 50% formamide, 5x SSC, 0.1% SDS, 20 μg mouse Cot-1 DNA and 10 μg poly(dA). Reverse dye labeling of samples was employed in separate experiments to account for any bias in dye coupling or emission efficiency of Cy dyes. After hybridization, microarray slides were washed by immersion into 2x SSC, 0.2% SDS for 5 min at 42°C, 0.2x SSC, 0.1% SDS for 1 min at room temperature, 0.2x SSC for 1 min at room temperature, and 0.05x SSC twice for 1 min at room temperature, dried by centrifugation, and immediately scanned. Different hybridization and wash conditions were tested (data not shown). The procedures described above have been optimized for both PCR amplicon and oligonucleotide probes regardless of whether the two probe types are printed together or separately. We chose a random primer labeling scheme so that oligonucleotide probe design would not be restricted to any particular region of the mRNA molecule. In contrast, oligo(dT)12-18 priming protocols  limit design considerations to the 3' end of the mRNA molecule.
Array image processing and data analysis
Cy3 and Cy5 fluorescence on microarray slides were scanned at 10 μm resolution using a GenePix 4000B scanner and saved as two single TIFF images. The intensities of spots on the two images were subsequently analyzed with GenePix Pro 3.0 software and a dataset was output. We used the following criteria to flag bad or extremely weak spots from the array dataset: spot area < 70 pixels, % saturated pixels > 50%, and sum of the median signal intensity < 1,000. Normalization of the array dataset was based on total median background subtracted intensities from the Cy3 and Cy5 channels and linear regression of the median signal intensities generated from the Arabidopsis control cRNA set spiked into the query RNA samples at a 1:1 ratio . After normalization, expression ratios were calculated for each non-flagged spot and log2 transformed.
The sequences of the 70-mers and corresponding GenBank accession numbers of the genes are available as an additional data file.
We thank Nnenna Nwokekeh for her excellent technical assistance. We also thank members of the TIGR microarray team for technical assistance and helpful comments. This work was supported by a Programs for Genomic Applications grant from the National Heart Lung and Blood Institute.
- Harkin DP: Uncovering functionally relevant signaling pathways using microarray-based expression profiling. Oncologist. 2000, 5: 501-507.View ArticleGoogle Scholar
- Lipshutz RJ, Fodor SP, Gingeras TR, Lockhart DJ: High density synthetic oligonucleotide arrays. Nat Genet. 1999, 21: 20-24. 10.1038/4447.View ArticleGoogle Scholar
- Schulze A, Downward J: Navigating gene expression using microarrays - a technology review. Nat Cell Biol. 2001, 3: E190-E195. 10.1038/35087138.View ArticleGoogle Scholar
- Schena M, Heller RA, Theriault TP, Konrad K, Lachenmeier E, Davis RW: Microarrays: biotechnology's discovery platform for functional genomics. Trends Biotechnol. 1998, 16: 301-306. 10.1016/S0167-7799(98)01219-0.View ArticleGoogle Scholar
- Southern E, Mir K, Shchepinov M: Molecular interactions on microarrays. Nat Genet. 1999, 21: 5-9. 10.1038/4429.View ArticleGoogle Scholar
- Watson A, Mazumder A, Stewart M, Balasubramanian S: Technology for microarray analysis of gene expression. Curr Opin Biotechnol. 1998, 9: 609-614. 10.1016/S0958-1669(98)80138-9.View ArticleGoogle Scholar
- Schena M, Shalon D, Davis RW, Brown PO: Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science. 1995, 270: 467-470.View ArticleGoogle Scholar
- Harrington CA, Rosenow C, Retief J: Monitoring gene expression using DNA microarrays. Curr Opin Microbiol. 2000, 3: 285-291. 10.1016/S1369-5274(00)00091-6.View ArticleGoogle Scholar
- Lockhart DJ, Dong H, Byrne MC, Follettie MT, Gallo MV, Chee MS, Mittmann M, Wang C, Kobayashi M, Horton H, et al: Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol. 1996, 14: 1675-1680.View ArticleGoogle Scholar
- Zhao X, Nampalli S, Serino AJ, Kumar S: Immobilization of oligodeoxyribonucleotides with multiple anchors to microchips. Nucleic Acids Res. 2001, 29: 955-959. 10.1093/nar/29.4.955.View ArticleGoogle Scholar
- Call DR, Chandler DP, Brockman F: Fabrication of DNA microarrays using unmodified oligonucleotide probes. Biotechniques. 2001, 30: 368-372.Google Scholar
- Kane MD, Jatkoe TA, Stumpf CR, Lu J, Thomas JD, Madore SJ: Assessment of the sensitivity and specificity of oligonucleotide (50mer) microarrays. Nucleic Acids Res. 2000, 28: 4552-4557. 10.1093/nar/28.22.4552.View ArticleGoogle Scholar
- Yuen T, Wurmbach E, Pfeffer RL, Ebersole BJ, Sealfon SC: Accuracy and calibration of commercial oligonucleotide and custom cDNA microarrays. Nucleic Acids Res. 2002, 30: e48-10.1093/nar/30.10.e48.View ArticleGoogle Scholar
- Kuo WP, Jenssen TK, Butte AJ, Ohno-Machado L, Kohane IS: Analysis of matched mRNA measurements from two different microarray technologies. Bioinformatics. 2002, 18: 405-412. 10.1093/bioinformatics/18.3.405.View ArticleGoogle Scholar
- Guo QM, Malek RL, Kim S, Chiao C, He M, Ruffy M, Sanka K, Lee NH, Dang CV, Liu ET: Identification of c-myc responsive genes using rat cDNA microarray. Cancer Res. 2000, 60: 5922-5928.Google Scholar
- Hegde P, Qi R, Abernathy K, Gay C, Dharap S, Gaspard R, Hughes JE, Snesrud E, Lee N, Quackenbush J: A concise guide to cDNA microarray analysis. Biotechniques. 2000, 29: 548-550.Google Scholar
- Cheung VG, Morley M, Aguilar F, Massimi A, Kucherlapati R, Childs G: Making and reading microarrays. Nat Genet. 1999, 21 (Suppl): 15-19.View ArticleGoogle Scholar
- Malek RL, Irby RB, Guo QM, Lee K, Wong S, Ruffy M, Tsai J, Kwong KY, Frank B, Liu ET, et al: Identification of Src transformation fingerprint in human colon cancer. Oncogene. 2002, 21: 7256-7265. 10.1038/sj.onc.1205900.View ArticleGoogle Scholar
- Pick70. [http://se.osxgnu.org/pub/mirrors/sourceforge/arrayoligosel/]
- TIGR rat gene index. [http://www.tigr.org/tdb/tgi/rgi]
- Quackenbush J: Computational analysis of microarray data. Nat Rev Genet. 2001, 2: 418-427. 10.1038/35076576.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.