Application of hybridization control probe to increase accuracy on ligation detection or minisequencing diagnostic microarrays
© Ritari et al; licensee BioMed Central Ltd. 2009
Received: 22 June 2009
Accepted: 14 December 2009
Published: 14 December 2009
Nucleic acid detection based on ligation reaction or single nucleotide extension of ssDNA probes followed by tag microarray hybridization provides an accurate and sensitive detection tool for various diagnostic purposes. Since microarray quality is crucial for reliable detection, these methods can benefit from correcting for microarray artefacts using specifically adapted techniques.
Here we demonstrate the application of a per-spot hybridization control oligonucleotide probe and a novel way of computing normalization for tag array data. The method takes into account the absolute value of the detection probe signal and the variability in the control probe signal to significantly alleviate problems caused by artefacts and noise on low quality microarrays.
Diagnostic microarray platforms require experimental and computational tools to enable efficient correction of array artefacts. The techniques presented here improve the signal to noise ratio and help in determining true positives with better statistical significance and in allowing the use of arrays with poor quality that would otherwise be discarded.
Even though enzyme aided recognition provides good accuracy as such, ultimately sensitivity and reliability are dependent on microarray quality and successful hybridization. The fidelity of recorded intensities of hybridized detection probes might be adequate for diagnostics provided the array spot quality is constantly good throughout the array as is often the case with in situ synthesized and other high quality microarrays. However, aberrant spots that vary in morphology or DNA content can occur for instance in contact printed arrays, causing problems with accuracy of spot finding and quantification. In addition, the printing process can introduce additional background noise impeding the read-out of results. Therefore, diagnostic microarrays may benefit from information processing steps to remove biases and noise, but this requires additional experimental measures to determine the source of variance. Methods used in gene expression normalization are not directly applicable because they typically rely on the bulk signal of all array spots assuming only that a small minority of genes are differentially expressed (reviewed in [15, 16]). In diagnostic microarrays, this assumption does not generally hold and also in some applications the number of spots per array is much lower.
To avoid generating false positives in the analysis, we have first computationally adjusted the control channel intensities to the level of detection probe channel median of all spots on a subarray in order to prevent the typically stronger control probe signal from dominating over the detection probe signal. The assumption behind this is that the median value represents empty detection probe signal. Next, the probe to control ratios are computed and used in logarithm as weighting coefficients in computing adjusted values for detection probe spot signals (figure 2). The advantage of this approach is that the probe signal value is multiplied by the weighting coefficient value over zero only if the detection probe signal is stronger than the control in that particular spot. If the spot is empty in the detection probe channel, the result of multiplication by the log ratio gives a small number even if the log ratio would be positive, unlikely to cause any false calls. In addition, as the log ratio and consequently the output is 0 with equal probe and control values, it provides a common reference point for all spots within and between microarrays.
The probe set used in the normalization experiments consisted of ligation detection probes similar to a previous study . Forty-two different probes (the sequences are to be published elsewhere) were multiplexed in ligation reactions. The ligation reactions and hybridization conditions were as described previously . Briefly, the ligations were carried out in a final volume of 20 μl containing 1× ligation buffer (TAQ ligase buffer, New England Biolabs, MA, USA), 30 mM tetramethylammonium chloride (TMAC), 250 fmol of each discriminating probe, 250 fmol of each common probe, 5 pmol of the complementary hybridisation control probe, a variable amount of purified PCR products and 4 U of Taq DNA ligase (New England Biolabs). The reaction was cycled for 40 rounds at 94°C for 30 s and at 60-64°C for 4 min in a thermocycler (MJ Research). The LDR mix (20 μl) was diluted to obtain 40 μl of hybridization mixture containing 5× SSC and 0.1 mg/ml herring sperm DNA. After heating the mix to 94°C for 2 min and chilling on ice, ligation control probe was added and the mix was applied onto the slide. The microarray slides were produced by contact printing by Telechem (CA, USA) or by university core facility (Biomedicum Biochip Center, University of Helsinki, Finland). The microarrays had 16 subarrays each, consisting of 119 tag oligos in triplicates [see Additional file 1]. Scanning were done as described previously . All computations were done in R-software environment (2.8.0) , using the Bioconductor package Marray (1.20.0)  [see Additional file 2] for reading GenePix result (gpr) files [see Additional file 3]. In order to evaluate the effect of normalization, no filtering or outlier removal procedures were applied to the data.
The normalization procedure was tested with two different sets of detection probes on microarrays containing either three or five spot replicates. Some subarrays had printing artefacts like background noise or low quality spots. Figure 3 shows an example of a typical poor quality subarray. In these kind of arrays, the spot morphology varied considerably and background noise was high in places resulting in signal-to-noise ratio much lower than in regular microarrays. Comparison of results before and after applying the control probe and normalization demonstrates the impact of these procedures on the signal-to-noise ratio. Clearly, just subtracting background signal from each spot raw value in the detection probe channel is not enough to provide reliable results (figure 3C). Computing signal ratio of the detection probe to the adjusted control (figure 3D) greatly reduces variation in the background distribution but at the same time weakens true positive signal and causes some spots to falsely reach too high values. The normalization procedure presented here, however, avoids these pitfalls and is able to correct for the variation and keep the true positive signal clear (figure 3E and 3G). We tested 8 microarray slides with altogether 128 subarrays. Furthermore, the procedure was applied on data from previously published microarray hybridizations [see Additional file 4]. In this set, the ligation probes were designed against environmental fungi  and detected on a tag microarray with five spot replicates. Also in this case the normalization was capable of correcting microarray noise. As expected, the computations had little or no effect on results from good quality microarrays [see Additional file 5].
DNA microarrays, while being a potential tool for diagnostics, typically have irregularities in spot quality causing problems with accuracy of detection. Efficient background correction is required in ligation, minisequencing and similar diagnostic systems where standard gene expression analysis methods may not apply. Although high quality microarrays are preferable for diagnostics, they may not always be economically feasible for routine use due to high costs. The prices of customizable commercial high-density in situ synthesized microarrays are likely decrease but these kind of platforms can still benefit from noise correction as human error in performing the hybridizations can not be ruled out. In addition, some emerging diagnostic platforms such as integrated microfluidic systems can take advantage of microarrays for detection of probes on a small scale [19, 20]. The mass fabrication process of the devices and hybridization conditions might bring about variation in spot quality and available space for replicates in microsystems is likely to be limited as well. In diagnostic applications in general, accuracy of the results is highly important and proper correction procedures can help deal with noise to increase statistical reliability, making the system practicable even if the detection platform is not fully optimal.
We found the normalization procedure based on an internal control for each assay spot to be useful when working with tag microarrays having compromised spot quality and background noise. A similar approach has been used by others before to monitor array spot quality and to correct for printing variations. In a study by Ye and coworkers, amplicons were first generated by forward and reverse primers carrying different labels . One strand of a amplicon served as target for SNP recognition probes while the other strand was used as a internal control hybridizing to a complementary control probe in the same spot. In another work, a 25 mer control oligo was spotted alongside with 70 mer recognition probes to monitor spot quality with a complementary labeled 25 mer . In both of these studies, a given spot control signal was first compared with the mean control signal and the obtained value was then used to divide the detection signal of the same spot. Similarly, Akhras and coworkers computed signal ratios of detection probe vs. an all-positive control on a padlock probe tag array . However, problematic microarrays or abnormal spots were not the focus of these studies and correcting microarray defects was not discussed in further detail. The method presented here uses a similar principle of internal control but different computational procedure to effectively reduce noise on low quality spots, emphasizing signal extraction rather than mere elimination of problematic spots. It is important not only to monitor the array spot quality but also to take into account the possible bias in probe to control ratio to effectively process aberrant spots on low quality arrays.
We have also demonstrated earlier with cDNA microarrays that effective measurement of spot quality with an additional dye improves the reliability of detection . Using an all-positive control probe serves a similar idea in assisting spot quantification which is highly dependent on the accurate determination of the true spot area to estimate spot and background signal intensities. This is especially relevant if the signal intensity of a spot is low in the detection channel. The control channel, having a relatively high intensity in all spots, helps in locating the spots and capturing their areas more accurately. However, it should be noted that our method does not use all the information available in the control channel. For instance, the control signal could be used to model the detection probe channel intensity profile. This approach could potentially increase the signal-to-noise-ratio of the detection probe channel on weak spots, opening the possibility for further development of the method.
The computation presented here can be easily implemented in any freely programmable system used for microarray data analysis. It should be noted however, that application of the control probe requires that each spot on the tag microarray harbors the complementary control sequence, along with the actual tag sequence. One way to overcome the need to synthesize novel long probes is to mix control oligonucleotide serving as an internal control with each of the speficic oligonucleotides and deposit these mixture on arrays . In addition, if it is expected that over 50% of detection probes should be positive in an experiment, negative control spots are needed to compute the normalization. This is because the procedure assumes that the detection probe channel median represents empty signal in computing adjusted signal values for the control channel.
Much of the noise introduced by variable spot quality on detection probe read-outs can be corrected by applying simple computations. This involves adjusting the control probe signals to detection probe signal levels and taking into account absolute value of the detection probe signal and variability in the control probe signal. The method is potentially advantageous in various diagnostic microarray platforms.
The Finnish Cultural Foundation is acknowledged for supporting JR financially. PA and JH were supported financially by the European regional development fund (ERDF) and Maj and Tor Nessling Foundation. We are also grateful to docent Eeva Auvinen from the Haartman Institute of the University of Helsinki for contributing ligation probes, and to Rita Fingerroos for performing microarray hybridizations.
- Landegren U, Kaiser R, Sanders J, Hood L: A ligase-mediated gene detection technique. Science. 1988, 241 (4869): 1077-1080. 10.1126/science.3413476.View ArticlePubMedGoogle Scholar
- Gerry NP, Witowski NE, Day J, Hammer RP, Barany G, Barany F: Universal DNA microarray method for multiplex detection of low abundance point mutations. J Mol Biol. 1999, 292 (2): 251-262. 10.1006/jmbi.1999.3063.View ArticlePubMedGoogle Scholar
- Nilsson M, Malmgren H, Samiotaki M, Kwiatkowski M, Chowdhary BP, Landegren U: Padlock probes: circularizing oligonucleotides for localized DNA detection. Science. 1994, 265 (5181): 2085-2088. 10.1126/science.7522346.View ArticlePubMedGoogle Scholar
- Syvänen AC, Aalto-Setalä K, Harju L, Kontula K, Soderlund H: A primer-guided nucleotide incorporation assay in the genotyping of apolipoprotein E. Genomics. 1990, 8 (4): 684-692. 10.1016/0888-7543(90)90255-S.View ArticlePubMedGoogle Scholar
- Pastinen T, Kurg A, Metspalu A, Peltonen L, Syvänen AC: Minisequencing: a specific tool for DNA analysis and diagnostics on oligonucleotide arrays. Genome Res. 1997, 7 (6): 606-614.PubMedGoogle Scholar
- Sigurdsson S, Hedman M, Sistonen P, Sajantila A, Syvänen AC: A microarray system for genotyping 150 single nucleotide polymorphisms in the coding region of human mitochondrial DNA. Genomics. 2006, 87 (4): 534-542. 10.1016/j.ygeno.2005.11.022.View ArticlePubMedGoogle Scholar
- Hardenbol P, Yu F, Belmont J, Mackenzie J, Bruckner C, Brundage T, Boudreau A, Chow S, Eberle J, Erbilgin A, Falkowski M, Fitzgerald R, Ghose S, Iartchouk O, Jain M, Karlin-Neumann G, Lu X, Miao X, Moore B, Moorhead M, Namsaraev E, Pasternak S, Prakash E, Tran K, Wang Z, Jones HB, Davis RW, Willis TD, Gibbs RA: Highly multiplexed molecular inversion probe genotyping: over 10,000 targeted SNPs genotyped in a single tube assay. Genome Res. 2005, 15 (2): 269-275. 10.1101/gr.3185605.PubMed CentralView ArticlePubMedGoogle Scholar
- Zeng F, Ren ZR, Huang SZ, Kalf M, Mommersteeg M, Smit M, White S, Jin CL, Xu M, Zhou DW, Yan JB, Chen MJ, van Beuningen R, Huang SZ, den Dunnen J, Zeng YT, Wu Y: Array-MLPA: comprehensive detection of deletions and duplications and its application to DMD patients. Hum Mutat. 2008, 29 (1): 190-197. 10.1002/humu.20613.View ArticlePubMedGoogle Scholar
- Akhras MS, Thiyagarajan S, Villablanca AC, Davis RW, Nyren P, Pourmand N: PathogenMip assay: a multiplex pathogen detection assay. PLoS ONE. 2007, 2 (2): e223-10.1371/journal.pone.0000223.PubMed CentralView ArticlePubMedGoogle Scholar
- Kostic T, Weilharter A, Rubino S, Delogu G, Uzzau S, Rudi K, Sessitsch A, Bodrossy L: A microbial diagnostic microarray technique for the sensitive detection and identification of pathogenic bacteria in a background of nonpathogens. Anal Biochem. 2007, 360 (2): 244-254. 10.1016/j.ab.2006.09.026.View ArticlePubMedGoogle Scholar
- Szemes M, Bonants P, de Weerdt M, Baner J, Landegren U, Schoen CD: Diagnostic application of padlock probes--multiplex detection of plant pathogens using universal microarrays. Nucleic Acids Res. 2005, 33 (8): e70-10.1093/nar/gni069.PubMed CentralView ArticlePubMedGoogle Scholar
- Busti E, Bordoni R, Castiglioni B, Monciardini P, Sosio M, Donadio S, Consolandi C, Rossi Bernardi L, Battaglia C, De Bellis G: Bacterial discrimination by means of a universal array approach mediated by LDR (ligase detection reaction). BMC Microbiol. 2002, 2: 27-10.1186/1471-2180-2-27.PubMed CentralView ArticlePubMedGoogle Scholar
- Hultman J, Ritari J, Romantschuk M, Paulin L, Auvinen P: Universal ligation-detection-reaction microarray applied for compost microbes. BMC Microbiol. 2008, 8: 237-10.1186/1471-2180-8-237.PubMed CentralView ArticlePubMedGoogle Scholar
- Rantala A, Rizzi E, Castiglioni B, de Bellis G, Sivonen K: Identification of hepatotoxin-producing cyanobacteria by DNA-chip. Environ Microbiol. 2008, 10 (3): 653-664. 10.1111/j.1462-2920.2007.01488.x.View ArticlePubMedGoogle Scholar
- Quackenbush J: Computational analysis of microarray data. Nat Rev Genet. 2001, 2 (6): 418-427. 10.1038/35076576.View ArticlePubMedGoogle Scholar
- Quackenbush J: Microarray data normalization and transformation. Nat Genet. 2002, 32 (Suppl): 496-501. 10.1038/ng1032.View ArticlePubMedGoogle Scholar
- R Development Core Team: R: A language and environment for statistical computing. 2008, R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0, [http://www.R-project.org]Google Scholar
- Yee Hwa (Jean) Yang with contributions from Agnes Paquet and Sandrine Dudoit: marray: Exploratory analysis for two-color spotted microarray data. 2007, R package version 1.20.0Google Scholar
- Hashimoto M, Hupert ML, Murphy MC, Soper SA, Cheng YW, Barany F: Ligase detection reaction/hybridization assays using three-dimensional microfluidic networks for the detection of low-abundant DNA point mutations. Anal Chem. 2005, 77 (10): 3243-3255. 10.1021/ac048184d.View ArticlePubMedGoogle Scholar
- Hashimoto M, Barany F, Soper SA: Polymerase chain reaction/ligase detection reaction/hybridization assays using flow-through microfluidic devices for the detection of low-abundant DNA point mutations. Biosens Bioelectron. 2006, 21 (10): 1915-1923. 10.1016/j.bios.2006.01.014.View ArticlePubMedGoogle Scholar
- Yin BC, Li H, Ye BC: A dual-probe hybridization method for reducing variability in single nucleotide polymorphism analysis with oligonucleotide microarrays. Anal Biochem. 2008, 383 (2): 270-278. 10.1016/j.ab.2008.09.003.View ArticlePubMedGoogle Scholar
- Peterson G, Bai J, Narayanan S: A co-printed oligomer to enhance reliability of spotted microarrays. J Microbiol Methods. 2009, 77 (3): 261-266. 10.1016/j.mimet.2009.02.014.View ArticlePubMedGoogle Scholar
- Gupta R, Ruosaari S, Kulathinal S, Hollmen J, Auvinen P: Microarray image segmentation using additional dye--an experimental study. Mol Cell Probes. 2007, 21 (5-6): 321-328. 10.1016/j.mcp.2007.03.006.View ArticlePubMedGoogle Scholar