Large Scale Gene Expression Analysis Using DNA Microarrays

May 10-15, 2004, in Turku, Finland

Organised by Turku Centre for Biotechnology, University of Turku and Åbo Akademi University.

Report

1. Programme and organisers
2. List of participants (May 10, minisymposium)
3. Minisymposium lecture presentations, synopsis and conclusions
4. Workshop groups and participants
5. Concluding remarks of the workshop

1. Programme and organizers

Minisymposium Session I: Microarray techniques and applications
9.00-10.00 Jorg Hoheisel (DKFZ, Heidelberg, Germany): Use of complex DNA- and
antibody-microarrays as tools in functional analyses
10.00-10.30 Coffee
10.30-11.0 Laszlo Puskas, (Biological Resarch Center, Hungarian Academy of
Sciences, University of Szeged, Szeged, Hungary): Gene expression profiling
the effects of dietary omega-3 polyunsaturated fatty acids in brain
11.30-12.00 Tapio Visakorpi (University of Tampere): Microarrays and prostate cancer
12.00-13.00 Lunch
13.00-13.30 Olli Kallioniemi (VTT Biotechnology, Turku, Finland) : Development and
applications of novel biochip technologies
13.30-14.00 Riitta Lahesmaa (Turku Centre for Biotechnology, Turku, Finland):
Microarrays in the analysis of lymphocyte response
14.00-14.30 Coffee
Session II: Bioinformatics and microarray data analysis
14.30-15.00 Mark Reimers (Genomics and Bioinformatics Group, Laboratory of Molecular
Pharmacology, Bethesda, MD): Bio-statistics and network analysis
15.00-15.30 Stephen Rudd (Bioinformatics, Turku Centre for Biotechnology, Turku,
Finland): Bioinformatics and microarray experiments: is this more than just
statistics?
15:30-16:00 John N Weinstein Title to be announced (Bioinformatics Group, Laboratory
of Molecular Pharmacology, Bethesda, MD)
16.00 Closing remarks

Workshop programme
Lecture day 1, May 11
- Welcome
- DNA Microarrays
- Oligonucleotide microarrays (Affymetrix GeneChips)
- Oligonucleotide microarrays versus cDNA Microarrays
- Experimental design, specials issues to be taken into account when working with
microarrays
- Microarrays in research, Case study of research done with microarrays
- Introduction of hands on training day, what to expect

Experiment Day, May 12
Carrying out microarray experiment. Hands on day, making one slide microarray
experiment. Participants can have their own RNA's (control and sample) for the array
or they can obtain it from us. Both human and mice arrays are available.

Data Extraction, May 13-14
Data extraction
Scanning of the slides, and transforming the images into numbers. Introduction to
hybridization quality control software.

Lecture Day 2, May 15
- Data filtering
- Microarray Normalization
- Statistics in microarrays
- Finding differentially expressed genes
- Clustering and classification of genes

Organisers
Riitta Lahesmaa (Director of the Turku Centre for Biotechnology, University of Turku and Åbo Akademi
University)
Tapio Salakoski (Head of the Department of Information technology, University of Turku)
Stephen Rudd (Head of Bioinformatics Laboratory, Turku Centre for Biotechnology, University of Turku and Åbo Akademi University )
Annika Brandt (Microarray Team Group-Leader, DNA Microarray Centre, Turku Centre for Biotechnology, University of Turku and Åbo Akademi University)

2. List of participants (minisymposium May 10, 2004)
Altogether the number of participants 96 plus organizers was 101 that had preregistered and signed the list below:

3. Lectures May 10, minisymposium

Dr Joerg Hoheisel, Head of Functional Genome Analysis Group, DKFZ German Cancer Research Centre, Germany
The Division of Functional Genome Analysis at the DKFZ is involved in the development of technologies for the analysis of DNA-encoded function and its regulation. Current work emphasizes DNA-, protein-, and peptide-microarrays. Apart from addressing chemical and biophysical issues, such as highly parallel in situ peptide synthesis and optimisation of surfaces of protein arrays, for example, the resulting methods are immediately put to the test in relevant, biologically driven projects. Besides analyses on all genes of various model organisms, the system is being developed toward acting as a tool for early diagnosis, prognosis and evaluation of the success of disease treatment.

Use of complex DNA- and antibody microarrays as tools in functional analyses

Microarray technology has become a long way, and is applied in biological and biomedical research as a routine method. Transcriptional profiling and detection of single nucleotide polymorphisms (SNPs) are by far the most applied forms of analyses. Raw information on SNPs is required before associations can be identified between polymorphisms and phenotypic variations in epidemiological studies. Epigenomics: C to T conversions, where 4% of all cytosines in the human genome are methylated at C-5 postion. DNA oligonucleotide microarrays have, however, not been successful in dealing with epigenetic analyses concerning e.g. all methylation sites in a single experiment. Though, methods like on-chip primer extension and minisequencing were mentioned to have improved this type of analysis considerably.
The application of peptide nucleic acids in an array formate could offer an alternative to the DNA oligonucleotide array technology. PNA oligomers are synthetic DNA mimics with an amide backbone and have several advantageous features: they are stable in acidic conditions and they are resistant to nucleases and proteases, plus that their neutral backbone increases the binding strength to complementary DNA compared to the corresponding DNA:DNA duplex. Therefore PNA probe can be shorter than DNA oligonucleotide probes. In addition, mismatches have more destabilising effect in PNA probes than in DNA oligonucleotide probes. PNA hybridizations can be performed in low salt or no salt conditions due to their neutral structure, which leads to less secondary structure of the target DNA and better accessibility to the probe molecules. Most important feature of the PNA probes is contributed by the way of detection, where PNA:DNA or PNA:RNA duplexes can be visualized by time-of-flight secondary ion mass spectrometry (TOF-SIMS), by the detection of the phosphates that constitute DNA but not PNA molecules. By combining the PNA microarrays with TOF-SIMS detection has a potential for a highly sensitive method for the detection of unlabelled DNA or RNA.
Production methods of PNA-arrays: 1)in situ Spot method,or 2) application of prefabricated PNA molecules. Expensive to produce, since PNA chemistry is not widely used. 3) There exists a Fmoc chemistry developed by J. Hoheisel's group, that permit fully automated process. Only the full length molecules after synthesis are attached to the microarray surfaces by selective binding of the terminal thiol or biotin groups, while shorter incomplete reaction products are washed away.

Antibody microarrays could have an enormous impact on the functional analysis by expression profiling. This type of analysis could become invaluable also in disease diagnosis. One of the major problems is created by the fact that the array surface has a profound influence on the results. Also the issue of antibody attachment creates problems, since this influences their functional properties. Antigens in a mixture should all bind to their cognate antibody receptors regardless of their distinct structural features. The arrays can be varied by the use of modification of the glass surface, the kind and length of crosslinkers, and the composition and pH of the spotting buffer, the type of blocking reagents, antibody concentration, and antibody storage buffer.

In conclusion, Dr Hoheisel summarized that enormous impact of the array formate expression profiling is currently finding its way, and it is impossible to predict, at present, where the development is going to be directed. The main stream is concentrating on gene expression profiling and SNP-genotyping experiments as well as on the epigenetic analyses. The protein arrays are more difficult to optimize and await further development.


Laszlo Puskas (Biological Research Center, Hungarian Academy of Sciences, University of Szeged, Szeged, Hungary):
Gene expression profiling the effects of dietary omega-3 polyunsaturated fatty acids in brain
Dietary effects regarding the n-3 polyunsaturated fatty acid structure was analysed in rat brains by using self made rat brain, liver and ganglion cDNA microarrays. 1. Study: Rats were fed either a high linolenic acid (perilla oil) or high eicosapentaenoic + docosahexaenoic acid (fish oil) diet (8%), and the fatty acid and molecular species composition of ethanolamine phosphoglycerides was determined. Gene expression pattern resulting from the feeding of n-3 fatty acids also was studied. Perilla oil feeding, in contrast to fish oil feeding, was not reflected in total fatty acid composition of ethanolamine phosphoglycerides. Levels of the alkenylacyl subclass of ethanolamine phosphoglycerides increased in response to feeding. In the sam fashion, the levels of diacyl phosphatidylethanolamine molecular species containing docosahexaenoic acid (18:0/22:6) were higher in perilla-fed or fish oil-fed rat brains, in constrast to those in ethanolamine plasmalogens, which remained unchanged. Using cDNA microarrays, 55 genes were found to be overexpressed and 47 were suppressed relative to controls by both dietary regimens. The altered genes included those controlling synaptic plasticity, cytosceleton and membrane association, signal transduction, ion channel formation, energy metabolism, and regulatory proteins. The effect seems to be independent of the chain length of fatty acids, but the n-3 structure appears to be important. 2. Study: Rats were fed from conception till adulthood either with normal rat chow with a linoleic (LA) to linolenic acid (LNA) ratio of 8.2:1 or a rat chow supplemented with a mixture of perilla and soy bean oil giving a ratio of LA to LNA of 4.7:1. Fat content of the feed was 5%. Fatty acid and molecular species composition of ethanolamine phosphoglyceride was determined. Effect of this diet on gene expression was also studied. There was an accumulation of docosahexaenoic (DHA) and arachidonic acids (AA) in brains of the experimental animals. Changes in the ratio sn-1 saturated, sn-2 docosahexaenoic to sn-1 monounsaturated, sn-2 docosahexaenoic were observed. Twenty genes were found overexpressed in response to the 4.7:1 mixture diet and four were found down-regulated compared to normal rat chow. Among them were the genes related to energy household, lipid metabolism and respiration.
It was concluded that brain sensitively reflects of the fatty acid composition in the diet. It was suggested that "alteration in membrane architecture and function coupled with alterations in gene expression profiles may contribute to the observed beneficial impact of n-3 type polyunsaturated fatty acids on cognitive functions". 3. Study: Advanced age is associated with reduced brain levels of long-chain polyunsaturated fatty acids, arachidonic acid (AA) and docosahexaenoic acid (DHA). Memory impairment is also a common phenomenon in this age. Two-year-old, essential fatty acid-sufficient rats were fed with fish oil (11% DHA) for 1 month, and fatty acid as well as molecular composition of the major phospholipids, phosphatidylcholine and phosphatidylethanolamine (PE), was compared with that of 2-month-old rats on the same diet.
DHA but not AA was significantly reduced in brains of old rats but was restored to the level of young rats when with fish oil included in the regular chow. This effect was pronounced with diacyl 18:0/22:6 PE species, whereas levels of 18:1/22:6 and 16:0/22:6 remained unchanged in all of the three PE subclasses. Fish oil reduced the AA in the old rat brains, diacyl and alkenylacyl 18:0/20:4 PE were most affected. Phosphatidylcholines gave less pronounced response. Six genes were up-regulated, whereas no significant changes were observed in brains of old rats receiving fish oil for 1 month. None of them except synuclein in young rat brains could be related to mental functions. Old rats on the fish-oil diet did not perform better in Morris water maze test than the control ones. A 10% increase in levels of diacyl 18:0/22:6 PE in young rat brains resulted in a significant improvement of learning ability. The results are interpreted in terms of the roles of different phospholipid molecular species in cognitive functions coupled with differential responsiveness of the genetic machinery of neurons to n-3 polyunsaturated fatty acids.

In conclusion, the rat brain responds to n-3 fatty acids in an age- and feeding-time dependent manner. It is still unclear, whether DHA interacts directly on gene expression of neurons or if gene expression changes and membrane architecture are unrelated events.

Tapio Visakorpi (Institute of Medical Technology, University of Tampere):
Microarrays and prostate cancer
The rational to study molecular mechanisms of malignancies is that they may provide means to provide better tools for diagnostics, prognostics and treatment of cancer. Good examples are trastumab, an antibody against ERBB2 oncoprotein and imanitib, which is a tyrosine kinase inhibitor, which suppresses the activity of ABL oncogene. Neither drugs are effective in protate cancer. Recently Visakorpi's group has shown amplification of uPA gene, and PC-3 cells carrying this amplification are sensitive to uPA inhibitors. The amplification of the uPA gene seems to indicate that invasion property of these cells is dependent on the uPA activity. The amplifilication of uPA is not frequent in prostate cancer, and not a relevant target in the common form of the disease.
2. Genetic predisposition by mono- or dizygous twin studies. The genetic predisposition may be attributable to high- and low-penetrance genes, which increase risk of cancer several fold (hereditary cancers). Linkage analyses have revealed several chromosomal loci and three putative susceptibililty genes: ELAC2, RNASEL, and MSR1. Some susceptibility genes (RB1, PTEN, TP53, APC) are mutated also in sporadic malignancies. Therefore ELAC2, RNASEL and MSR1 were screened in 50 unselected prostate carcinomas for somatic mutations. No evident functionally mutations were found and are thus rare in prostate cancer. Polymorphisms of many gene have been suggested to be in association with risk of prostate cancer such as the androgen receptor (AR) gene At present none of the sequence variations can be regarded as definitely associated with prostate cancer
3. Somatic alterations.Cytogenetical studies and analysis of LOH and comparative genomic hybridization have been used to detect chromosomal aberrations in prostate cancer. But due to the fact that prostate cancer cells do not grow well in vitro, the traditional cytogentetics has been uninformative. Only a few whole genome-wide LOH analyses have been performed. CGH has been mainly used detecting gains and losses of DNA sequency copy numbers, and has revealed that losses are more common than gain or amplifications. The chromosomal losses are detected early stages of prostate cancer, whereas gains and amplifications are seen in hormony refractory tumors. Chromosomal losesse in prostate cancer are 6q, 8p, 10q, 13q, 16q and 18q indicating the locations of tumor suppressor genes in prostate cancer. The hormony-refractory and metastatic tumors show gains in 7p/q, 8q and Xq in CGH.
Two most common deletion are 8p and 13q. Minimally deleted regions 8p21 and 8p22 with the most promising target NKX3.1(homeobox gene), and N33, FEZ1 and PRTLS. Second most common deletion occurs in 13q, associated with aggressiveness of prostate cancer. The strongest target has been identified as RB1 and ENDRB(endothelin receptor gene reported to be hypermethylated and downregulated in prostate cancer).
Gain of 8q is most frequent in the hormone-refractory and metastatic tumors. Most intensily studied gene in this region is MYC, also amplified in prostate carcinomas. By using suppression substraction and cDNA microarrays Visakorpi's group has identified 4 putative target genes for 8q gain: Elongin C, EIF3S3, KIAA0196, and RAD21. They seem to be amplified in 20-30% of the hormone-refractory prostate carcinomas. EIF3S3 is associated with hogh Gleason score and advanced clinical stage of the disease. Other suggested target genes include PSCA and TRPS1.
GSTP1 hypermetylation has been suggested as a diagnostic marker for prostate cancer (Glutathione S-transferases are detoxifying enzymes that protect cells from carcinogenic factors). The human homeobox gene NKX3.1 is frequently detected in prostate cancer. Homozygous and heterozygous mutant mice develop prostate cancer. PTEN (PTEN functions as a lipid phosphatase and targets PIP-3. By dephosphorylating PIP-3, PTEN downregulates the Akt/PKB pathway that promotes cell survival and inhibits apoptosis). Deletions and mutations of PTEN gene have infrequently reported in prostate carcinomas. The frequency of LOH at PTEN locus has been reported to be higher (40%) than the rate of mutations in prostate cancer. Alternative mechanisms could involve haploinsufficiency. TP53 is the most commonly mutated gene in human cancers. Under DNA damage, TP53 can either induce apoptosis or arrest cell cycle for DNA repair. Mutated TP53 has prolonged half-life leading to nuclear accumulation of the abnormal protein. Immunohistochemical detection. Rare in early but in advanced prostate carcinomas found in 20-40% of cases. Nuclear localization is associated with poor prognosis. AR-signalling pathways are re-activated during progression of hormone-refractory prostate cancer. AR mutations are detected in 20-25% of patients treated with anti-androgens. Visakorpi's group has demonstrated that AR gene is amplified in 30% of hormone-refractory prostate carcinomas from patients treated with androgen withdrawal, which selects the gene amplification. The mechanisms of the AR overexpression without amplification is unknown.

Summary: Genetic predisposition of prostate cancer is an extremely complicated issue. It is likely that no high-penetrance prostate cancer genes exist (like BRCA1 and BRCA2 in breast cancer). ELAC", RNASEL and MSR1 contribute to a minute fraction of prostate carcinomas. The traditional epidemiological studies have not been able to identify the major environmental risk factors. Is the hypermethylation of GSTP1 a primary event? NKX3.1 and PTEN have been shown to posses haploinsufficiency characteristics. How common is this mechanism? AR signaling is the best known aberrant pathway in prostate cancer. How to utilize this mechanism? Model system are also limited. Xenografts have been recently established. Hopefully these models will boost the efforts to develop novel targeted treatments for prostate cancer.

Riitta Lahesmaa (Turku centre for Biotechnology, University of Turku and Åbo Akademi University, Turku, Finland)
Defects in the polarization of T helper subtypes Th1 and Th2 can result in various immune-mediated diseases such as asthma. To understand the development of these diseases it is essential to know the process at the at the molecular level. Both Th1 and Th2 originate from common precursor cell, Thp. The differentiation is initiated in response to activation through T cell receptor, costimulatory molecules and cytokine receptors. The main cytokine that directs the Th1 commitment is IL12, when Il4 drives the Th2 polarization. The effects of IL12 and Il4 are mediated through Stat4 and Stat6, respectively. Other key regulatory factors involved in Th1 and Th2 differentioation are T-bet and Gata-3, respectively.
We and others have conducted large-scale gene expression analysis to identify genes involved in differentiation process after 2 days or later using oligonucleotide arrays. In order to solve the molecular mechanisms leading to Th1 and Th2 lineage commitment, it is crucial to define the upstream factors at the very earliest phase of initiation of Th cell commitment. The aims of our study include the indetification of the immediate early genes that are differentially regulated in response to activation- and Th1- and Th2 -inducing cytokines IL12 and IL4.
Affymetrix U95Av2 arrays containing probes for ~9300 genes were used to study the changes in gene expression profiles after 2 and 6 hours of CD3/CD28-activation and induction of Th1 and Th2 polarization. After 2 h of activation alone upregulated expression of 437 and downregulated 361 probe sets as compared to the Thp cells. The peak for activation-mediated changes was seen after 6 h, when activation had induced upregulation of 832 and downregulation of 856 probe sets.
To identify the genes differentially expressed by the cells induced to polarize to Th1 or Th2 direction, the gene expression profiles of the cells cultured for 2 h or 6 h in Th1 polarizing conditions (anti-CD3/anti-CD28/IL12) were compared with Th2 polarizing conditions (anti-CD3/anti-CD28/IL4). In result, total of 63 genes were identified as differentially expressed by the cells induced to polarize to Th1 or Th2 direction.
To further characterize the genes regulated by IL12 or IL4, the expression profiles of cells induced to polarize to Th1 or Th2 direction were compared to expression profiles of cells of the CD3/CD28 activated cells without polarizing cytokines. This comparison revealed that the early changes in the gene expression were mainly driven by IL4. The only genes that were regulated by IL12 after 6h were IFN-gamma (1.87 fold) and GBP1 (1.62-fold). Of the 63 differentially regulated genes in Th1 and Th2 conditions, for 26 genes the regulation by IL4 were seen at both 2h and 6h. the early polarization is mainly driven by IL4, since the activated Th cells seem unresponsive to IL12.
To illustrate the putative functional roles of the newly identified immediate targets of IL4, the genes were grouped to functional gategories base on Gene Ontology annotations. The dominating functional groups consisted of transcription factors, cell adhesion molecules and receptors, enzymes, and other intracellular signaling molecules. The most of genes in these groups were induced by IL4 by 2 hours of polarization.
As expected based on previous studies, genes that displayed constant changes thoughout the early polarization were the well-known mediators of Th1 and Th2 differentiation GATA3, MAF and IFNG. Functional classification of the immediate target genes of IL4 revealed that one of the dominating functional droup consisted of transcription factors. Although role of many of these factors in Th1 and Th2 polarization is currently unknown, genes such as SATB1 and TCF7 have similar functions as GATA3 and have been associated with pathogenesis mediated by Th1 or Th2 responses.
In a distinct setting where early polarization of CD4+ lymphocytes was studied, the effects of the presence and absence of TGF- were studied About 20 novel genes were identified during Th1/2 polarization, and further, target genes associated with the function of IL-12, IL-4 and TGF- A subset of identified target genes were observed to be coregulated by IL-12, IL-4 and TGF- TNFSF9, E4BP4/NFIL3, CTLA1/GZMB, ID2, Cox-2, GNAI1, PLA2G4A, and BCL2A1). The antagonizing effect of TGF- on the expression of these genes regulated by IL-12 or IL-4 could in part explain the inhibitory effect of TGF- on Th differentiation. In the mouse models TGF- 1 has suppressed the airway inflammation associated with asthma. Therefore TGF- target genes in Th2 differentiation could also serve as potential drug targets for therapeutic approaches to treat asthma and allergy. The mice studies have indicated thet the inhibitory actions of TGF- involves suppression of T-bet and Gata3, respectively. In Lahesmaa's studies these genes were not characterized among the numerous primary genes regulated in human cells by TGF- The basis of these differences in gene regulation during early Th differentiation in the mouse and in the human remain to be further elucidated.
In summary, to resolve the precise role of genes for T helper cell differentiation identified by Riitta Lahesmaa's group, the following questions remain to be answered: 1) hierarchial order of the target genes, 2) upstream factors, 3) interacting factors in the signaling pathways involved in the differentiation process, 4) functions of the target genes, 5) their significance for Th differentiation, and 6) can they be used to modulate Th1 and Th2 responses. About further studies of the potential target genes in Th differentiation, the use of RNAi method was mentioned as a method of choice to study the function of these genes.

Olli Kallioniemi (VTT Medical Biotechnology, Turku, Finland)
Development and applications of novel biochip technologies
Olli Kallioniemi presented extensive overviews of different DNA microarray technologies, their use, applications and comparison between distinct platforms. The applicapability of protein arrays, their major problems of usage and applications. Olli Kallioniemi's group has developed and applied three general strategies to facilitate "genome-scale" translational cancer research, as well as validation and extension of traditional microarray experiments:
1) Application of CGH microarrays or NMD microarrays (non-sense mediated RNA decay) to guide towards genes that could be primary genetic alterations or causative events in the multi-step progression of cancer. These could also represent attractive drug targets. Parallel CGH and cDNA microarray studies have revealed 270 such candidate targets in breast cancer. NMD microarrays in turn, may highlight mutated transcripts with a possible tumor suppressive function.
2) Cell-based microarrays using reverse transfection approach (Ziauddin & Sabatini, 2001), which are based on printing cDNAs, siRNAs, drugs or other reagents on microscope slides and plating cells to grow on top of the array to establish a highdensity molecular matrix for exploring cell function. Functional cell-based microarray studies provide fundamentally different data as compared to traditional microarrays. Most importantly, they establish cause-and-effect relationships. This enables cell biological studies in a highly parallel, in a miniaturized genome-wide scale.
3) Sample-based microarray strategies (tissue microarrays), facilitate the analysis of individual DNA, RNA, or protein targets in thousands of samples. For example, a large-scale clinical study of 1000s of patients can be carried out on a single microscope slide in order to establish definitive clinical correlations for molecular targets, or to assess drug target distributions at the population level.
The DNA microarray platforms are based on distinct protocols for manufacturing, hybridization and imaging analysis with proprietary data analysis steps making comparison of the data between platforms difficult. This inevitably restricts the effective use of publicly available large data sets. At present, there are only a limited number of publications, where distinct microarray technologies have been compared.
Olli Kallioniemis group has compared the results from Affymetrix, Agilent, and custom-made microarrays, to determine the comparability between these platforms. They compared four breast cancer cell lines: BT-474, MCF-7, MDA-MB-436, and MDA-MB-361 and the reference cell line HBL-100 (ATCC). Total RNA was isolated with trizol with subsequent Qiagen RNAesy column purification.
Affymetrix U95-Av2 arrays were applied for expression analysis without technical replicates. The data was analysed in addition to MAS 5.0 with Robust Multichip average (RMA) method with quantile normalization and fitted model by median polish.
Agilent cDNA arrays contained 13,156 clones from Incytes human cDNA library, which was analysed with a dye-swap method. The slides were analysed by Feature Extraction software (version A.4.0.45). Variation within platform was calculated using GenBank accession number as an identifier.
The custom printed microarray contained 11,520 clones from Incyte Genomics IRAL cDNA library and 1136 clones from research genetics. The cDNA clones were spotted on poly-L-lysine slides with an OmniGrid arrayer. Samples were hybridized on three replicate slides. The slides were scanned using scanner by Agilent Technologies, and the image analysis was performed with DeArray software. The data analysis was performed with 1) ratio quality value below 0.5 were discarded (1=good, 0=poor quality). Withinslide normalization was calculated by ratio statistics method using all spots in the array. 2) Quality filterrrring was used with Bayesian networks for determination of good or bad spots. Lowess normalization calculation was performed for print-tip groups.
The comparisons were made with Unigene cluster ID as identifier: 2340 were found in each platforms, with 1147 common and possible to evaluate.
The correlation coefficients for ratio values between custom-made and either of the two commercial microarrays were 0.62-0.76 and those of the two commercial ones 0.78-0.86.
The estimation of variability between platform the percentage of genes showing more than twofold change was determined. 5.0% of genes in the commercial arrays showed this type of variation and 9.0% of the custom arrays as compared to Affymetrix and 11.5% as compared to Agilent. However, The biological difference between the cell lines was more prominent than the variation of the platform, and was clustered correctly on all platforms.
The variantions observed in this study lists the following challenges of the MA analysis:
1) Results of one MA analysis can not necessarily be reproduced with another platform
2) Data in public databases difficult to integrate
3) Clone-errors
4) Other methods for validating studies
5) Varying results with distinct MA platforms?
Notably there exists about 16% of the clones in custom-made arrays are associated with wrong annotation information. It would therefore be important for the self-made arrays to determine the percentage of wrong clones and report this in publications.
It is of importance also to pay attention that short oligonucleotides will hybridize more specificly than long probes or cDNA fragments to their respective targets. The estimated presence of alternatively spliced transcripts in the human genome is between 30-50%, and the hybridization properties of short and long cDNAs may lie behing the variying results. Also, the cDNA arrays may give misleading information due to the fact that both sense and antisense strands may react, whereas oligonucleotides give results of the correct strand only.
When different analyzing methods were compared, the results emphasised the importance of deposition of primary data, which is necessary to re-evaluation of the data.
n summary, Optimisation of data preprocessing, QC and normalization for each platform is necessary, and should be considered when choosing the platform for a long term study, as well as when comparing the data available in public data bases. The expression profiling analysis was however, found robustly redundant, for diagnostic classifications. Sources of differences can reside upon clone errors, annotation mistakes, and technical differences(oligos vs cDNA) may be of importance to take into consideration when choosing the proper platform for a particular study.
In summary, microarray strategies can be applied in a large number of molecular, cell biological and clinical studies, thereby expanding the traditional concept of "microarray analysis". It is likely, that the integration of the various types of microarrays will be needed in systems biology studies, translational cancer research and drug target discovery, since no single analysis platform will provide a complete answer to complex biological and clinical questions. This poses new challenges to bioinformatics analyses in the future.
Before closing his speech Olli showed a figure of the publications with DNA-microarrays: the amount of DNA-microarray articles have been on exponential track for the past years, and strongly increasing in numbers.
Second heartwarming picture was about the comparative quality of cDNA slides manufactured commercially (e.g. MWG) and the slides manufactured at the Turku Centre for Biotechnology: OUR SLIDES WERE RANKED TO BE OF BEST QUALITY OF ALL SLIDES TESTED BY KALLIOMAKI'S GROUP !!!

Stephen Rudd (Turku Centre for Biotechnology, University of Turku and Åbo Akademi University, Turku, Finland)
Bioinformatics and microarray experiments: is this more than just statistics?
Stephen has been involved in development of a web based tool for comparative genomics at his previous post in the Institute for Bioinformatics at the GSF Research Centre, Munich. The definitions of the openSputnik are cited below:
openSputnik forms a core environment for the address of specific genomics questions. The core infrastructure for comparative genomics has already been implemented within the area of "reconstructomics". With the ability to perform large scale analysis, archival and interpretation within a single framework and utilising some of the most contemporary bioinformatics methods (everything is XML defined and external transformation methods can be defined) - openSputnik surely has some potential within a genomics pipeline. openSputnik has a rather interesting pedigree. Before Sputnik was born there was a GABI online resource called miniPEDANT (it may still exist somewhere). The idea was a multi-user, flexible, online resource for up-to-date and contemporary bioinformatics methods without the need to know what is really hot. openSputnik needs to present and display some rather complicated data to a less computationally aware audience. While I feel that XML is the best way to display everything - I believe that most biologists will disagree. I have chosen to use Zope as a web-application server system and have implemented a Zope product, openZputnik (does anyone have a better name?), that allows for the administration of the core openSputnik server as well as the display of all contained data. Since this is Zope it is relatively trivial to setup and maintain.
Why openSputnik? Sputnik is a pipeline and infrastructure aimed at both plant genomics and comparative genomics that was written and is maintained at the Institute for Bioinformatics at the GSF Research Centre near Munich. The platform was originally implemented to satisfy the needs of a consortium of German sugarbeet researchers (GABIBEET), and was later adapted to allow a more generic but high throughput analysis of plant-based biological data. Sputnik was implemented as a collection of loosely interacting Python scripts, a PostgreSQL database and a simple Apache webserver. The last release of Sputnik (version 4.0) can still be viewed at MIPS.
"I have the feeling that Sputnik is of more value to the scientific community as a core computing infrastructure than as just a collection of pre-digested results. With the transition from Munich to Turku I decided to recreate the essence of Sputnik in a different langauge while solving some of the problems and rethinking the rationale underlying the computational platform. As a result I have started the openSputnik project and hope that this may make some form of impact in other research groups."
Sputnik was first written to automate the analysis of EST annotation for comparative genomics. openSputnik both maintains the concept behind Sputnik and continues to develop as an optimal solution for the processing of large EST collections.
The core concept behind successful EST annotation is to create an object relational infrastructure where for example a unigene cluster inherits the attributes of the underlying EST sequence data. Such annotative attributes include information such as e.g. the mouse strain from which the ESTs were sequenced, the developmental stage at which the clones were sequenced and so on. With the derivation and annotation of a peptide sequence we can consider the other extreme. With a single EST sequence that we have previously shown to stem from a candidate gene we can associate annotation that stems from peptide domains that are not associated with this EST, but rather from ESTs that assemble either directly or indirectly with this sequence.
The focus of my research group is firmly embedded in comparative genomics - a wide variety of methods have been implemented in openSputnik that allow for the selection of lineage specific transcripts, transcript families, lineage specific domains or domain architectures. We have all plant EST collections with more than 5,000 ESTs clustered, assembled and placed within the openSputnik comparative framework. The next step (the hard one) is to make some sense of this data and to present it in a meaningful manner.
In addition to plant EST analysis we have on-going collaborations with various research groups working on molecular markers in pig, mouse, chickpea and barley. We are also looking at EST collections from some of the more exotic genomes including Hydra magnipapillata, Bombyx mori, Cycas rumphii and Ginkgo biloba.
In the context of ESTs and molecular markers Sputnik and openSputnik have been mentioned in publications
Brenner, E. D., Stevenson, D. W., McCombie, R. W., Katari, M. S., Rudd, S. A., Mayer, K. F., Palenchar, P. M., Runko, S. J., Twigg, R. W., Dai, G., et al. (2003). Expressed sequence tag analysis in Cycas, the most primitive living seed plant. Genome Biol 4, R78.
Kota, R., Rudd, S., Facius, A., Kolesov, G., Thiel, T., Zhang, H., Stein, N., Mayer, K., and Graner, A. (2003). Snipping polymorphisms from large EST collections in barley ( Hordeum vulgareL.). Mol Genet Genomics.
Rudd, S. (2003). Expressed sequence tags: alternative or complement to whole genome sequences? Trends Plant Sci 8, 321-329.
Rudd, S., Mewes, H. W., and Mayer, K. F. (2003). Sputnik: a database platform for comparative plant genomics. Nucleic Acids Res 31, 128-132.

Mark Reimers (Genomics and Bioinformatics Group, Laboratory of Molecular Pharmacology, Bethesda, MD)
Microarray data analysis
"The design of scientific experiments is an art of balancing considerations: skill, cost, equipment, and accuracy." For a comparative analysis care should be taken in planning to keep hybridisation conditions constant. Conditions such as RNA preservation medium, the protocols of hybridisation, and even regional ozone levels, can introduce systematic biases comparable in size to the biological differences you wish to detect. Taking a great deal of care to standardize conditions will pay off in much higher discovery rates. To do a series of two-color hybridisations, you want to prepare enough common reference to serve for all experiments. Chip failures are common, and it is wise to prepare more labelled cDNA than you expect to use.
How many microarrays is enough? If an exploratory study aims to find large (more than two-fold) differences between two conditions, then a design with three samples per condition is usually adequate. If the aim is to find smaller differences, or almost all of the large differences, then five samples per group are necessary to obtain sufficiently reliable enough estimates of variation among samples within conditions, in order to distinguish true differences between conditions. Six samples per condition allows meaningful permutation tests, which can give more accurate, and less conservative, estimates of p-values and false discovery rates. If there are more than two conditions, and the treatments do not drastically alter the cell physiology, then the number of samples within any one condition can be somewhat less; with four or more conditions, one can obtain reasonable estimates of within-condition variation with only four samples per condition. All of these suggestions assume that there are no outlying samples, which should be discarded; it is wise to do one or two more per condition in clinical situations, where outliers occur commonly, and it is safer to do one more for animal experiments, where sometimes one animal in a condition appears very different than all the others.
"To do meaningful clustering requires at least 20 samples, and generally many more. The key issue for clustering genes is how many different types of samples there are, because the different conditions expose the correlations in gene regulation. It is not useful to try to cluster genes from only two groups, as is sometimes done, and rarely useful to cluster genes from a study of fewer than five groups."
Pooling?
There is considerable disagreement about whether to pool individual samples, among practitioners and also among statisticians. Sometimes the amount of sample from any one individual sample is insufficient for hybridization and in that case, pooling is a practical necessity. In theory, if the variation of a gene among different individuals is approximately normally distributed, then pooling n independent samples would result in reduction of variance given by the formula:
where 2 is the variance of the expression estimates of any one gene across samples. In principle we could then reduce further the variation by making replicates of the pool, and hybridising to replicate arrays. Since technical variation is usually less than (roughly half of) individual variation, this strategy would in theory give us more accurate estimates of the group means for each gene. See also Prichard et al "Project Normal", PNAS (2002)).
Dye swap experiments
Reference sample:
i. it extends easily to other experiments, if the common reference is preserved;
ii. is robust to multiple chip failures; and
iii. reduces incidence of laboratory mistakes, because each sample is handled the same way.
A reliable alternative is a common reference obtained by pooling all samples. This enables samples to be compared with each other indirectly. A pooled reference sample reduces the number of extreme gene ratios (which have large errors) on each chip. Some labs take this further and create a 'universal reference': a pool of mRNA derived from several standard cell lines, which they use most often in their experiments. Using a universal reference enables them to compare results for all their experiments.
One complication in two-color arrays is that the two dyes don't get taken up equally well, so that the amount of label per amount of RNA differs (dye bias).
However the dye-swap is the basis for most other efficient designs: the general principles of a good two-color design are that
i. it should be balanced: every sample appears equally often in red and green;
ii. the samples whose ratios are most interesting should appear on the same chips most often.
A good design for KO studies (e.g. KO-receptor with ligand) is to hybridise several dye-swap pairs between the treatment and control within each group, and perhaps to hybridise one or two slides between WT treated and KO treated, and between WT control and KO control. This design gives fairly accurate estimates of both effects of treatment vs. control (in WT and Mutant), which enables accurate comparisons between the effects; there is less accurate information about the direct comparison between WT and mutant.
Microarrays often give many parallel measures for the same target and there is usually a good deal of cross-talk between measures of different targets, also discrepancy between measures of the same target. A classical statistical strategy is a general linear model.
Biological measures give a large numbers of outliers, for which purpose a robust linear model can be applied.
The application of robust linear model for the Affymetrix system, seems to give five- to ten-fold increases in accuracy for RNAs of signaling proteins (as measured by variance between replicate arrays). The hybridization raw results give surprisingly uneven hybridization results within the Affymetrix arrays. The hybridization is also influence how the sequences are located in the array. Notably, in the novel 2.0 plus arrays the design is different to the U133 design, which results in differential hybridization results due to distinct local sequence environment of each PM probe.

John N. Weinstein (Bioinformatics group, Laboratory of Molecular Pharmacology, Bethesda, MD)
Integrating data from microarrays at the DNA, RNA, and protein levels
The new genomic and proteomic technologies are immense powerful, but they pose novel challenges to the researchers. Partly these challenges relate to what is happening after finishing the experiment. The first challenge is the statistical analysis of the microarray data. The second challenge would be the biological interpretation of the results. The third challenge would be to be able to integrate the microarray data with other types of molecular and pharmacological information, which could be called as "Integromics". Our group has developed a number of practical software tools for meeting those challenges:
MedMiner=speeds up 5-10 fold the organization of biomedical literature on genes and and drugs, searches and organizes the biomedical literature on genes, gene-gene relationships, and gene-drug relationships. It uses GeneCards, PubMed, syntactic analysis, truncated-keyword filtering of relationals, and user-controlled sculpting of Boolean queries to generate key sentences from pertinent abstracts. Abstracts selected can be automatically entered to EndNote (Biotechniques 1999;27:1210)
CIM Maker; which produces flexible Clustered Image Maps ("heat maps"), generates color-coded Clustered Image Maps (CIMs or heat maps) to represent "highdimensional" data sets such as gene expression profiles. We introduced CIMs for data on drug activity, target expression, gene expression, and proteomic profiles. Clustering of the axes brings like together with like to create patterns of color (Weinstein et al., Science 1997;275:343-349).
MatchMiner=translates fluently among the many types of gene and protein identifiers, translates among gene identifier types for lists of hundreds or thousands of genes. Includes: GenBank accession numbers, Image clone IDs, common gene names, HUGO names, gene symbols, Unigene clusters, FISH-mapped BAC clones, Affymetric identifiers, and chromosome locations (Bussey et al., Genome Biology 2003;4:R27)
GoMiner_which leverages the Gene Ontology for discovery of functional order in lists of genes, addresses the question, "Now that I've done the gene expression experiment and identified a set of 'interesting' genes, what do these mean biologically?" GoMiner batch-processes and organizes lists of thousands or tens of thousands of genes and provides two fluent, robust visualizations of the genes in the framework of the Gene Ontology hierarchy. (Zeeberg et al., Genome Biology 2003;4:R28)
MethMiner=organisezes patterns of sequence information from DNA methylation studies
LeadScope/LeadMiner=links genomic and proteomic information to the molecular substructures of potential drugs
GEEVS=Genomic Exploration and Visualization System)=integrates multiple types of molecular information at the DNA, RNA, Protein, functional and pharmacological levels.
NCI-60 cancer cell databases=gene expression data from 9,700-gene cDNA array & 6,800-gene Affymetrix oligonucleotide array
http://discover.nci.nih for information and access
Development of these computer resources has been motivated by our studies of the 60 cancer cell lines (the NCI-60) used by the NCI DTP to screen >100,000 chemical compounds since 1990. These cells provide detailed information about mechanisms of drug action and reistance. Based on these studies a multifaceted molecular target profiles of the NCI-60 using 2-D gel electrophoresis. 'reverse-phase' protein microarrays, cdNA microarrays, Affymetrix chips, array-CGH, SKY, SNP-chips, and DNA-methylation sequencing. The clinical molecular markers identified have been validated by tissue microarrays. The integrated databases will have a great impact on diagnosis, prognosis, and therapy of cancer.

4. Workshop May 11-15: Programme, Groups and Participants
Lecture day 1, May 11
(University of Turku, IT-department auditorium, Lemminkäisenkatu 14 A 2th floor)
9:00-9:15 Welcome (Annika Brandt)
9:15-10:00 DNA Microarrays (Annika Brandt)
10:00-10:45 Oligonucleotide microarrays (Affymetrix GeneChips) (Riikka Lund)
10:45-11:00 Coffee
11:00-11:45 Experimental design, specials issues to be taken into account when working
with microarrays (Katja Kimppa)
11:45-13:00 Lunch
13:00-13:45 Microarray manufacturing (Rolf Sara)
13:45-14:30 Microarrays in research, Case study of research done with microarrays
(Tuomas Nikula)
14:30-15:00 Introduction of hands on training day, what to expect (Annika Brandt)

Experiment Day, May 12
(Turku Polytechnic School, Department of Biotechnology, Lemminkäisenkatu 30)
Microarray experiments will be done in pairs, see the last page for your pair.
9:00-10:00 Walk through of the training day
10:00- 11:00 RNA labeling
11:00-12:00 Microarray slide pretreatment
12:00- 13:00 Lunch
13:00- 15:30 cDNA purification and hybridization
19:00- 21:00 Dinner cruise in the Turku archipelago

Data Extraction, May 13
(Scanning of microarrays in Turku Centre for Biotechnology, Tykistökatu 6, 5th floor.
Data extraction in University of Turku, IT department, Lemminkäisenkatu 14 A 2th floor
and 5th floor)
9:00-10:00 Post hybridization washes, Turku Polytechnic School, Department of
Biotechnology, Lemminkäisenkatu 30
Scanning and Data extraction will be done in 4 groups. See the last page for your group.
Group1: Slide Washes + Scan Thu morning: 9.00 - 12.30
Quantitation Thu afternoon: 13.00 - 15.00
Group2: Scan Thu afternoon: 13.00 - 15.00
Quantitation Thu afternoon: 15.00 - 17.00

Data extraction, May 14
(Scanning of microarrays in Turku Centre for Biotechnology, Tykistökatu 6, 5th floor.
Data extraction in University of Turku, IT department, Lemminkäisenkatu 14 A 2th floor
and 5th floor)
Scanning and Data extraction will be done in 4 groups. See the last page for your group.
Group3: Scan Fri morning: 9.00 - 11.00
Quantitation Fri afternoon: 11.30 - 13.30
Group4: Scan Fri afternoon: 12.00 - 14.00
Quantitation Fri afternoon: 14.00 - 16.00
Lecture Day 2, May 15
(University of Turku, IT-department auditorium, Lemminkäisenkatu 14 A 2th floor)
9:00-11:00 Basics of microarray data analysis (Katja Kimppa)
11:00 -12:00 Microarray databases (Tero Raitanen)
12:00-13:00 Lunch
13:00-14:30 Advanced data analysis (Katja Kimppa)
14:30- 15:00 Closing remarks (Annika Brandt)

Group1:
Slide Washes + Scan Thu morning: 9.00 - 12.30
Quantitation Thu afternoon: 13.00 - 15.00
Pair 1 - G.O. Elkhabbuli & J.Kelemen own RNA
Pair 2 - S. Korkut & Ozgur Gul
Pair 3 - A. F. Duartes & P. Maciel
Pair 4 - A. Rodrigues & R. Martinez-Artiaga
Pair 5 - Piotr Bielecki & A.Chini
Pair 6 - Claudina Perez Novo & D. Leinberger

Group2:
Scan Thu afternoon: 13.00 - 15.00
Quantitation Thu afternoon: 15.00 - 17.00
Pair 7 - Evelyn Hatzenbichler & S. Gigliotti
Pair 8 - Deirdre Corcoran & Grace O'Gorman
Pair 9 - Anette Knudsen & Sissel Monstad
Pair 10 - Laila Stordrange & Monica Sebastiana
Pair 11 - Benny Abraham & Rosanna Asselta
Pair 12 - Tiina. Tomperi & Miia. Antikainen & Heikki Koskinen


Group3:
Scan Fri morning: 9.00 - 11.00
Quantitation Fri afternoon: 11.30 - 13.30
Pair 13 - Irina Corin & Berit Eitrem
Pair 14 - Signe Indahl & A. Balde
Pair 15 - Andreia Figuieredo & A. Fortes
Pair 16 - Matti Sankinen & Florence Naillat
Pair 17 - Tove Jansen & Dominic Kurian
Pair 18 - Sippy Kaur & Carola Saloranta


Group 4:
Scan Fri afternoon: 12.00 - 14.00
Quantitation Fri afternoon: 14.00 - 16.00
Pair 19 - Sanna Käkönen & Sirkku Pollari
Pair 20 - Juha Lemmetyinen & O. Ugur Sezerman
Pair 21 - Jing-Jiang Zhou & Silvia Barth
Pair 22 - Celine Claubaut & Mari Nyyssönen
Pair 23 - Halit Canatan & Bartholomeau Acioli
Pair 24 - Andrey Yakovenko & Heikki Koskinen *

5. Concluding remarks of the workshop

ESF Workshop on "Large scale gene expression analysis using DNA microarrays"
Conclusing remarks

The Microarray workshop included symposia, lab training and 2 days of lectures on microarrays.

First lecture day contained information about the different DNA microarray technologies, including cDNA microarrays and Affymetrix oligonucleotide microarrays, the manufacturing and usage of the techniques, Experiment design issues and a case study of research done with both microarray techniques. The last part of this day was the introduction to the laboratory training day.

In the microarray systems introduction, the differences and the pros and cons of different techniques were emphasized.

Experiment design focused on the bases of how to do good and reproducible microarray experiment with reliable results. Discussion about different replicates and their usage was broad amongst the participants and organizers. Discussion about selecting correct samples and controls was also vivid.

As a case study, we had the type I diabetes study that's ongoing in Finland. In this study they try to enlighten the origin of diabetes and the early steps of diabetes. The group is using both cDNA and Affymetrix microarrays as a part of their study.

Microarray experiments were performed on laboratory day. Those participants who did have their own RNA with them were using those and others got RNA from the organizers. The experiments were performed well and large amounts of specific questions were asked during the lab day.

Data extraction days were constructed so that there were ~12 participants extracting their data simultaneously. This was due to the fact that we couldn't find large enough computer class, where the computers would have had enough memory, to do the analysis in bigger group. This also leaded to some restrictions on the amount of time used to the extraction and preliminary analysis of the data. Otherwise the days were functioning and the results from the arrays were looking good.

Last day of the course was about microarray data-analysis. This was constructed as a serie of lectures, starting from the filtering of the data, going through different normalization techniques, following to statistical analysis of microarray results and searching for differentiation in gene-expression with different methods and concluding to basics of clustering techniques and other more sophisticated data mining methods. During the day we also had a lecture about microarray databases and MIAME standard.
This day was constructed as a lecture day because of the amount of people and the fact that the softwares for microarray data-analysis that we are using in the Microarray Centre are not freeware softwares. In all of the softwares that can be used to analyse microarray data the same basic elements (like filtering, normalization, statistics etc.) are implemented. This day was full of all kinds of questions and data-analysis seemed to be an area where there could have been even more lectures and more time used.