webacgh

Appl Bioinformatics 2006; 5 (2): 125-130APPLICATION NOTE 1175-5636/06/0002-0125/$39.95/0

© 2006 Adis Data Information BV. All rights reserved.

WebaCGHAn Interactive Online Tool for the Analysis and Display of ArrayComparative Genomic Hybridisation Data

Casey Frankenberger,1 Xiaolin Wu,1 Jerry Harmon,2 Deanna Church,3 Lisa M. Gangi,1 David J. Munroe1 andUlises Urzua4

1 Laboratory of Molecular Technology, Scientific Application International Corporation (SAIC) – Frederick, Inc., NationalCancer Institute – Frederick, Frederick, Maryland, USA

2 Scientific Application International Corporation (SAIC), Annapolis, Maryland, USA3 National Center for Biotechnology Information, National Institutes of Health, Bethesda, Maryland, USA4 Programa de Biologia Celular y Molecular, Instituto de Ciencias Biomedicas, Facultad de Medicina, Universidad de Chile,

Santiago, Chile

Gene copy number variations occur both in normal cells and in numerous pathologies including cancer andAbstractdevelopmental diseases. Array comparative genomic hybridisation (aCGH) is an emerging technology thatallows detection of chromosomal gains and losses in a high-resolution format. When aCGH is performed oncDNA and oligonucleotide microarrays, the impact of DNA copy number on gene transcription profiles may bedirectly compared. We have created an online software tool, WebaCGH, that functions to (i) upload aCGH andgene transcription results from multiple experiments; (ii) identify significant aberrant regions using a localZ-score threshold in user-selected chromosomal segments subjected to smoothing with moving averages; and(iii) display results in a graphical format with full genome and individual chromosome views. In the individualchromosome display, data can be zoomed in/out in both dimensions (i.e. ratio and physical location) and plottedfeatures can have ‘mouse over’ linking to outside databases to identify loci of interest. Uploaded data can bestored indefinitely for subsequent retrieval and analysis. WebaCGH was created as a Java™-based webapplication using the open-source database MySQL®.Availability: WebaCGH is freely accessible at http://129.43.22.27/WebaCGH/welcome.htmContact: Xiaolin Wu ([email protected]) or Ulises Urzua ([email protected])

Background development of array-based CGH (aCGH),[5-8] resulting in signifi-cantly higher resolution (up to ~150 Kbp) and, when applied tocDNA or oligonucleotide arrays, allowing the direct comparisonAlterations in DNA copy number are attributes of the naturalof gene copy number with expression profiles.[9]genomic variability in the human being[1,2] and are disease-causa-

tive of many types of cancers and developmental pathologies.[3] The major challenge in aCGH data analysis is the ability toComparative genomic hybridisation (CGH) has proved to be an resolve authentic DNA copy number changes from noise. Theinvaluable tool for the low-resolution (5–10 Mbp) detection and higher sequence complexity of genomic DNA compared withanalysis of chromosomal amplifications and deletions. In this RNA affects hybridisation kinetics, specificity, and background.technique, reference and test genomic DNAs are separately la- Moreover, the use of short targets such as cDNA or oligonucleo-belled with different fluorescent nucleotides and co-hybridised to tides generates a poor fluorescent signal-to-background ratio com-normal metaphase chromosome spreads.[4] The general concept pared with bacterial artificial chromosome (BAC) arrays.[10] Earli-behind CGH has recently been refined and improved with the er reports used moving averages to smooth aCGH profiles and

126 Frankenberger et al.

identify chromosomal aberrations.[7,11] More recently, a wide vari- tion takes about 15 sec for a 4× replicate of a 10 000-spot array.ety of more sophisticated bioinformatic tools such as a Smith- All replicates of an experiment are averaged and the resultingWaterman-based algorithm,[12] unsupervised hidden Markov mod- value displayed depending on the settings used. A combinedels,[13] adaptive weights smoothing[14] and expectation-maximisa- maximum of ten gene transcription and aCGH files can be main-tion edge-placement algorithms[15] have been assessed. tained in each project.

In a previous work, we devised stand-alone trial software that The settings for data display request the user to enter the ratioaccurately identified genomic aberrations in mouse ovarian tu- range of gene transcription results and a scaling factor that deter-mour cells using moving averages and a root mean square plus an mines the ratio range for aCGH results. Typically, the aCGH scaleexperimental filter.[16] This approach was validated by comparison should be set shorter than the transcriptional scale. As illustrated inof the aCGH output with conventional CGH. In this application figure 1, the overall data distribution of aCGH data is compressednote, we describe the use and capabilities of an improved online relative to transcriptional data in the test-reference experiments.version of the above-mentioned tool, namely WebaCGH. This is because of the higher sequence complexity of genomic

DNA compared with total RNA, an observation also reported byResource Description others.[7,10,18] Thus, using a scaling factor helps to match aberrant

DNA/RNA profiles and visualise the impact of gain and losses onup- and down-regulation of gene expression, respectively. ThisCapabilitiesfeature is available in both the full genome and the individual

WebaCGH is a multi-user Java™-based web application de- chromosome views (see figure 2).signed for display, analysis and comparison of multiple aCGH and Additional settings include the threshold boundary, the Z-scoregene transcription microarray results. Data can be smoothed using threshold and the moving-average window. The threshold bounda-custom-adjustable moving averages along the entire genome and ry is an experimentally determined ratio cut-off, which can bein each individual chromosome. Additional display settings in- obtained from aCGH hybridisations with samples of known DNAclude a scaling factor to facilitate the simultaneous visualisation of copy number changes, typically chromosome X-based aneup-DNA copy number and transcription levels, an experimentallydetermined DNA copy number threshold, and a Z-score cut-off tohighlight statistically interesting points. Importantly, this tool au-tomatically links all plotted datapoints to external databases con-taining diverse gene annotations.

Implementation

Data format for WebaCGH analysis must be a tab-delimited filecontaining the ratio values and the well identification number foreach feature in the microarray used. The well identification num-ber corresponds to a unique numeric identifier for every singlecDNA clone or oligonucleotide in all microarray sets supported atthe microarray database (mAdb) of the National Cancer Institute,National Institutes of Health (http://nciarray.nci.nih.gov/). Eachdata file may contain from one to ten replicates of a singleexperiment. A previous data normalisation filter and scale adjust-ment is helpful especially when multiple experiments are com-pared.[17] Data uploading times for a single array file in a typicalcomputer set-up vary from 10 to 30 sec (for 10 000 to 80 000datapoints, respectively). For an 8× replicate of a 10 000-spotarray, loading time is 22 sec. Once data are uploaded, a projectmay be created. Each project is defined by the user as the datasetsto be assembled and displayed together graphically. Project crea-

4

2

0

−2

−4

Rat

io (

log 2

)

RNA (r-r) RNA (r-t) DNA (r-r) DNA (r-t)

Fig. 1. Comparison of raw overall data distribution between RNA and DNAhybridisation experiments onto a cDNA microarray platform. ReferenceRNA and reference DNA were co-hybridised against themselves (r-r) andagainst test RNA and test DNA, respectively (r-t). Both test samples wereobtained from the same biological source (mouse ovarian IC5 cells). Val-ues are shown for 13 417 clones from the NIA-15K cDNA mouse collec-tion. Upper and lower box boundaries indicate the 75th and 25th percentile,respectively. Whiskers above and below the box indicate the 90th and the10th percentiles. Lines splitting each box mark the median.

© 2006 Adis Data Information BV. All rights reserved. Appl Bioinformatics 2006; 5 (2)

WebaCGH 127

Fig. 2. Sample views of WebaCGH chromosomal display. Results of a 4× replicate array comparative genomic hybridisation (aCGH) experiment (mousenormal genomic DNA as reference vs genomic DNA from IC5, a mouse ovarian cancer cell) were obtained with NIA-15K cDNA microarrays. Aftervisualisation of the full genome view, data display focused on chromosome 19: (a) aCGH data without smoothing; and (b) aCGH data smoothed, 5 Mbpwindow. aCGH results (fuchsia line) are shown overlaid with gene transcription results (blue line): (c) smoothing with 15 Mbp window; (d) horizontaldisplay, smoothing with 15 Mbp window; and (e) horizontal display, ‘zoomed-in’ between 30 and 40 Mbp, smoothing, 15 Mbp window. Z-score cut-off was1.5 in all cases.

loidies[7,18] or normal female/male comparisons.[16] The Z-score, sion of a large number of adjacent genes, usually matched with

chromosomal aberrations).[22]also called the standard score, expresses the divergence of each

experimental ratio value (x) from the mean (μ), divided by the The moving-average function[7,11] can be used to smooth thestandard deviation (σ). Therefore, the larger the value of Z that is data, providing continuity along each chromosome (figure 2b-eused as a cut-off, the less probable that the experimental result is and figure 3b). Data smoothing is necessitated by the high noisedue to chance. The Z-score transformation is especially useful level inherent in aCGH data collected from cDNA and oligonucle-

otide array platforms (figure 1). When employing a moving aver-when seeking to compare the relative standings of datapoints inage, each datapoint is expressed relative to that of an average ofnoisy distributions (i.e. different means and/or different standardthe datapoints located within a physically defined window pre-deviations across chromosomal location). In WebaCGH,selected by the user. The default window size for WebaCGH hasdatapoints over the Z-score threshold are highlighted as circlesbeen arbitrarily set at 5 Mbp. The window boundaries are defined(see figure 2 and figure 3), and, importantly, μ and σ are calculatedas beginning half of the distance of the window size before the

from the moving-average window set by the user (see followingdatapoint (in the default instance this would be 2.5 Mbp) and

paragraph) rather than from the overall data. Initial work using Z-ending half of the distance of the window size after the datapoint.

score in array analysis computed it directly on fluorescence inten-Given the selected window size, all datapoints are averaged and

sities to normalise and standardise data locally in single two- the moving-average value displayed graphically. Figure 2a showscolour experiments[19] or globally in multiple single-channel DNA input raw (non-smoothed) data whereas figure 2b displays themicroarrays.[20] More recently, Z-scores have been applied to same data smoothed with the default moving-average window.correlate gene transcription profiles with conventional CGH data Note that the number of Z-score tagged points decreases whenin head and neck squamous cell carcinoma[21] and to determine smoothing is applied. Accordingly, if the moving-average window

is further enlarged, a less stringent Z-score threshold should beregional expression biases (i.e. unidirectional changes in expres-



Fig. 3. Empirical demonstration of WebaCGH capability. Array comparative genomic hybridisation (aCGH) data for IF5 (another mouse ovarian cancer cell)were obtained as described in figure 2. (a) Non-smoothed data, tagged with Z-score 2.5; (b) data smoothed with a moving-average window of 22 Mbp,tagged with Z-score 0.75 (an experimental DNA copy number threshold was set at –0.12, +0.12 in the log2-scale ratio); and (c) conventional metaphaseCGH results. For the experimental description, see Urzua et al.[16] n = number of chromosome spreads used to obtain each profile.

applied (see Empirical Demonstration section). Finally, in displayed as a tool-tip box when the datapoint is ‘moused over’.Further, each point may be clicked to take the user to the mAdbtelomeric regions, where the window might extend beyond theFeature Page, which shows details about the selected clone/physical end of the chromosome, its boundary would default to theoligonucleotide and provides connectivity to outside databasesactual end of the chromosome.such as National Center for Biotechnology Information (NCBI)After the display parameters are set, data loaded intoLocusLink (superseded by Entrez Gene; http://www.ncbi.nih.gov/WebaCGH can be visualised as full genome view or individualentrez/query.fcgi?db=gene), Kyoto Encyclopedia of Genes andchromosome view, or they can be ‘zoomed-in’ to a sub-chromo-Genomes (KEGG; http://www.genome.jp/kegg/), Cancer Genomesomal view. In all display modes, datapoints are shown accordingAnatomy Project (CGAP; http://cgap.nci.nih.gov/), GeneCards®

to their relative physical location along the chromosome and(http://www.genecards.org/index.shtml) and SOURCE (http://

within individual cytogenetic bands. In the full genome view, thesource.stanford.edu), among others.

complete dataset is displayed against the entire complement of

chromosomes for the given genome (currently, human andDevelopment

mouse). This visualisation must be displayed first to access the

individual chromosome view. Clicking on the corresponding chro- The overall configuration of WebaCGH is shown in figure 4.mosome ideogram will link the user to another webpage showing The application is composed of JavaServer Pages™ (JSP), whichthe data displayed horizontally along the selected chromosome call data from a MySQL® (http://www.mysql.com) database. This(figure 2d-e). The user can ‘zoom in’ to narrow the genomic database consists of five types of tables: (i) username and pass-position displayed. In figure 2e, the field has been narrowed to word table, containing information for authenticating the userbegin at 30 Mbp and end at 40 Mbp. The gene symbol, chromo- login; (ii) user data tables, which consist of files uploaded by thesomal location, UniGene ID, and file name for each datapoint is user containing experimental results in the form of log2 ratios and


WebaCGH 129

their corresponding well identification numbers; (iii) array data Empirical Demonstrationtables, containing chromosomal location and other well identifica-

A major advantage of array based-CGH over conventionaltion-associated information about the arrayed DNAs; (iv) projectmetaphase CGH is the detection of cryptic genomic imbalances, atables, which are composites of one or more user data tables withfeature having enormous impact in various fields of medicaltheir corresponding array data table, with the purpose of ap-genetics. With the aim of validating the capacity of WebaCGH,pending chromosomal locations to array results; and (v) chromo-real array-CGH data obtained with cDNA microarrays weresome data tables, which are used to dynamically create the imagesanalysed at the WebaCGH server and compared with conventionalfor both the genome and chromosome views.CGH experiments as depicted in figure 3. Despite its elevated

The JSP are dynamic pages that change depending on the users noise, non-smoothed data (figure 3a) was able to show statisticallyand their projects created from the database. Some of the JSP are significant points using a high Z-score threshold. When movinghidden to the user but have code needed for executing the applica- averages (25 Mbp) and a lower Z-score cut-off (0.75) were ap-tion. For example, the genomic and chromosome view jpeg plied, chromosomal imbalances appeared sharply highlighted inimages are created in genomeImage.jsp and chromoImage.jsp, each of the five mouse chromosomes depicted in figure 3b. Usingrespectively. Then, the images are called into the genomeView.jsp these WebaCGH settings, both aCGH and conventional CGHand chromoView.jsp where the tool-tip, which contains the func- profiles of chromosome 15 were shown to be consistent. Impor-tions for both the datapoint ‘mouse over’ information box and the tantly, additional aberrations in chromosomes 11 through 14 wereclickable link, can be overlaid and interact with the image. Addi- detected only with the microarray approach. WebaCGH reliabilitytionally, the moving-average and Z-score functionalities are em- was further supported upon closer examination of chromosomalbedded in genomeImage.jsp and chromoImage.jsp as they also fragment 11B4-11E2, which showed a tendency to amplify withdefine the plotting of datapoints subjected to calculations in the conventional CGH, whereas it appeared clearly amplified withimages generated. aCGH (102 contiguous clones). In contrast, deletion of fragment

Userdata

tables

Projecttables

loginpage.jsp

chromoImage.jsp

genomeImage.jsp

loginconfirm.jsp

createproject.jsp

mAdb page

Username andpassword

table

projectdisplay.jsp

Arraydata

tables

fileupload.jspfiledelete.jsp

Start

Finish

genomeView.jsp chromoView.jsp

Chromosomedata tables

Fig. 4. WebaCGH configuration and interconnectivity. The dynamic JavaServer Pages™ (JSP) run on an Apache Tomcat server that calls data from aMySQL® database using a MySQL® connector. The five types of tables comprising the database are depicted as grey boxes. Thick arrows depictconnectivity between the database and JSP. JSP shown in grey font correspond to pages unseen by the user but containing code necessary to run certainWebaCGH functionalities such as deletion of uploaded data files and creation of chromosomal images (for further description see Developmentsubsection). mAdb = microarray database.



9. Pollack JR, Sorlie T, Perou CM, et al. Microarray analysis reveals a major direct12A3-12C1 (38 contiguous clones) may be considered cryptic torole of DNA copy number alteration in the transcriptional program of human

conventional CGH.breast tumors. Proc Natl Acad Sci U S A 2002 Oct 1; 99 (20): 12963-8

10. Pinkel D, Albertson DG. Comparative genomic hybridization. Annu Rev Ge-

nomics Hum Genet 2005; 6: 331-54Conclusion11. Geschwind DH, Gregg J, Boone K, et al. Klinefelter’s syndrome as a model of

anomalous cerebral laterality: testing gene dosage in the X chromosome

pseudoautosomal region using a DNA microarray. Dev Genet 1998; 23 (3):The WebaCGH tool offers a strong graphical environment215-29designed to store, display and identify aberrant loci in aCGH

12. Price TS, Regan R, Mott R, et al. SW-ARRAY: a dynamic programming solutionexperiments. WebaCGH accepts a simple data format and allowsfor the identification of copy-number changes in genomic DNA using arraythe direct comparison of aCGH datasets with gene transcriptioncomparative genome hybridization data. Nucleic Acids Res 2005 Jun 16; 33results. A help manual is available in the Documentation section at(11): 3455-64

the WebaCGH website (http://129.43.22.27/WebaCGH/13. Chen W, Erdogan F, Ropers HH, et al. CGHPRO: a comprehensive data analysis

welcome.htm).tool for array CGH. BMC Bioinformatics 2005 Apr 5; 6 (1): 85

14. Hupe P, Stransky N, Thiery JP, et al. Analysis of array CGH data: from signal ratio

to gain and loss of DNA regions. Bioinformatics 2004 Dec 12; 20 (18): 3413-22Acknowledgements

15. Myers CL, Dunham MJ, Kung SY, et al. Accurate detection of aneuploidies in

array CGH and gene expression microarray data. Bioinformatics 2004 Dec 12;

This article is dedicated to the memory of Jose Urzua. 20 (18): 3533-43

16. Urzua U, Frankenberger C, Gangi L, et al. Microarray comparative genomicWe thank Seymour Davies and John Powell for help in data management.Dr Urzua was an Exchange Scientist at the Laboratory of Molecular Technolo- hybridization profile of a murine model for epithelial ovarian cancer reveals

gy supported by a fellowship from the Oncology Research Faculty Develop- genomic imbalances resembling human ovarian carcinomas. Tumour Biol 2005

ment Program, Office of International Affairs of the National Cancer Institute. Aug 9; 26 (5): 236-44This work has been funded in whole or in part with federal funds from the 17. Vaquerizas JM, Dopazo J, Diaz-Uriarte R. DNMAD: web-based diagnosis andNational Cancer Institute, National Institutes of Health under contract no. normalization for microarray data. Bioinformatics 2004 Dec 12; 20 (18):N01-C0-12400.

3656-8The authors declare no conflicts of interest regarding the contents of this 18. Lage JM, Leamon JH, Pejovic T, et al. Whole genome analysis of genetic

manuscript.alterations in small DNA samples using hyperbranched strand displacement

amplification and array-CGH. Genome Res 2003 Feb; 13 (2): 294-307

19. Colantuoni C, Henry G, Zeger S, et al. SNOMAD (Standardization and NOrmali-

References zation of MicroArray Data): web-accessible gene expression data analysis.

1. Iafrate AJ, Feuk L, Rivera MN, et al. Detection of large-scale variation in the Bioinformatics 2002 Nov; 18 (11): 1540-1human genome. Nat Genet 2004 Sep; 36 (9): 949-51 20. Cheadle C, Vawter MP, Freed WJ, et al. Analysis of microarray data using Z score

2. Sebat J, Lakshmi B, Troge J, et al. Large-scale copy number polymorphism in the transformation. J Mol Diagn 2003 May; 5 (2): 73-81human genome. Science 2004 Jul 23; 305 (5683): 525-8

21. Masayesva BG, Ha P, Garrett-Mayer E, et al. Gene expression alterations over3. Albertson DG, Pinkel D. Genomic microarrays in human genetic disease and

large chromosomal regions in cancers include multiple genes unrelated tocancer. Hum Mol Genet 2003 Oct 15; 12 Spec. no. 2: R145-52malignant progression. Proc Natl Acad Sci U S A 2004; 101: 8715-204. Kallioniemi A, Kallioniemi OP, Sudar D, et al. Comparative genomic hybridiza-

tion for molecular cytogenetic analysis of solid tumors. Science 1992 Oct 30; 22. Furge KA, Dykema KJ, Ho C, et al. Comparison of array-based comparative258 (5083): 818-21 genomic hybridization with gene expression-based regional expression biases

5. Solinas-Toldo S, Lampel S, Stilgenbauer S, et al. Matrix-based comparative to identify genetic abnormalities in hepatocellular carcinoma. BMC Genomicsgenomic hybridization: biochips to screen for genomic imbalances. Genes

2005; 6: 67Chromosomes Cancer 1997 Dec; 20 (4): 399-407

6. Pinkel D, Segraves R, Sudar D, et al. High resolution analysis of DNA copynumber variation using comparative genomic hybridization to microarrays. Nat

Correspondence and offprints: Dr Ulises Urzua, Programa de BiologiaGenet 1998 Oct; 20 (2): 207-11

Celular y Molecular, Instituto de Ciencias Biomedicas, Facultad de7. Pollack JR, Perou CM, Alizadeh AA, et al. Genome-wide analysis of DNA copy-number changes using cDNA microarrays. Nat Genet 1999 Sep; 23 (1): 41-6 Medicina, Universidad de Chile, Avda. Independencia #1027, Santiago,

8. Bignell GR, Huang J, Greshock J, et al. High-resolution analysis of DNA copyChile.

number using oligonucleotide microarrays. Genome Res 2004 Feb; 14 (2):E-mail: [email protected]


webacgh

Documents