Self organizing map based molecular signature representing the development of hepatocellular carcinoma Iizuka N et al FEBS Letters 2005 579 5 p 1089 Erin Bridgeford Nancy Guill n BE 450 April 13 2005 Microarrays to investigate problems in cell biology Data from transcription state of the cell under certain conditions Each experiment produces lots of data Finding single change in gene expression Look at overall patterns of gene expression Hypothesis driven vs fishing expedition Image removed due to copyright reasons GeneChip Microarray Courtesy of Affymetrix Used with permission Probe is 25 mer oligonucleotide for high specificity Multiple probes for each expression or genotype measurement Optimized probe set Analysis of gene expression profiling data Due to high volume of data systematic methods for organization are required to convert data into a manageable set Strategies grouped in two categories Discrimination or supervised learning Clustering or unsupervised learning k means self organizing maps Underlying biological phenomena might get lost in abstraction Self organizing maps for clustering of expression data SOM is a similarity graph and a clustering diagram Converts complex nonlinear statistical relationships between high dimensional data items into simple geometric relationships on a low dimensional display SOM has a series of partitions with a predefined geometrical configuration and initially their reference vectors are random Genes or samples are mapped to the relevant partitions depending on which reference vector they are most similar to Demo Gene expression profiles in hepatocellular carcinoma Microarray studies aimed at translating molecular information into clinical practice Studies for breast cancer and large B cell lymphoma Studies generally include cohort of patients followed for years after treatment Link gene clusters with good or poor prognosis survival recurrence HCC outcome complicated by the fact that cirrhosis pre neoplastic compromises liver functionality Heterogeneous nature of human HCC Results have to provide rationale for a molecular classification of the tumor to be able to predict outcomes and guide treatments Hepatocellular carcinoma and hepatitis B HBV and C HCV viruses Mutagenic effect of virus Chronic inflammation and disease leads to malignant neoplastic event Molecular basis not well understood What are they trying to do with this Goal Understand the relation between development and dedifferentiation of HCC Hypothesis Disease progression chronic HCV infection well differentiated HCC moderately differentiated HCC poorly differentiated HCC Approach Perform a comprehensive analysis of gene expression levels and identify discriminatory genes for each stage to elucidate the molecular basis of HCC using a global picture of expression patterns Materials and Methods Sample Selection Samples taken from 76 HCC patients 50 seropositive for HCVAb 26 seronegative for HCV All seronegative for HBV surface antigen Histopathology on HCV samples 7 well differentiated HCC group G1 35 moderately differentiated HCC G2 8 poorly differentiated G3 Control Groups Two control groups Group L0 comprised of 6 nontumorous histologically normal liver samples from patients with benign or metastatic liver tumors 1 focal nodular hyperplasia 2 hemangiomas 3 metastatic tumors 2 from colon cancer 1 gastric All seronegative for HCV Ab and HBVsAg Group L1 Five HCV infected nontumorous samples from 5 HCC patients Two chronic hepatitis 3 liver cirrhosis Concerns in sample selection No normal samples of liver as baseline No samples from HCV patients without HCC Materials and Methods DNA Microanalysis Resected specimens divided in two groups One frozen immediately after surgery for later RNA extraction One preserved in 10 formaldehyde and embedded in paraffin Used to demonstrate that non necrotic tissues were source of RNA RNA extraction performed Quality control of RNA Look for genomic DNA contamination Check for RNA decay by agarose gel electrophoresis If ratio of 28S 18S rRNA is around 2 0 suggests RNA had not decayed before or during extraction Reduced 28S 18S ratios indicate poor quality RNA Materials and Methods Microarray Analysis Synthesis of cDNA and cRNA see Iizuka et al Cancer Research 62 2002 Oligonucleotide microarray screening huU95A DNA Chips 12 600 probes that correspond to 8900 named genes for initial screen Image removed due to copyright reasons Materials and Methods Gene Selection At first pass 3559 genes selected Expression levels were greater than 40 arbitrary units arbitrary units intensity brightness of probed spot over brightness of local background Fisher ratio applied to evaluate which genes could help discriminate among the groups Measures the difference between two means normalized by the average variance ie estimates signal to noise ratio Larger Fisher ratio suggests a stronger likelihood for a gene s ability to discriminate between groups Gene selection cont d Random permutation test perfomed to validate Fisher ratio Looks to find undesired structure in random data If original result is due to chance then randomly relabelling data should achieve similar ratios Genes with P 0 005 were selected Different numbers of genes for each group deemed discriminatory L0 to L1 152 genes L1 to G1 191 genes G1 to G2 54 genes G2 to G3 40 genes Materials and Methods Identifying Discriminatory Genes Percentage of genes identified by chance false discovery rate calculated Ratio of false positives total positives A high FDR value can still be meaningful Materials and Methods Comparing Classes For class comparison Self organizing map minimum distance Algorithm used for clustering data classifier designed Provides visualization with top 40 genes of multi dimensional from each class Finds centers of classes and measures between those centers and a test image s center data Materials and Methods Some Concerns No indication that laser capture microdissection LCM or any more precise method of tissue selection was used analyzing stroma and vasculature as well Technology does not always generate reproducible or consistent results even with optimized samples Per Stearns The current state of the art provides 5 10 variation in signal intensities among replicate array elements on the same microarray and 10 30 variation among corresponding array elements on different microarrays 1 1Stears et al Trends in Microarray Analysis Nature Medicine 9 140 145 2003 Materials and Methods Concerns