Authorʼs reply

Abstract

To Dr. Xie: First, thank you for your questions and suggestions. I wish my following clarifications would provide sufficient answers to your questions. In the first question, you asked why the false discovery correction was not used for screening differentially expressed genes between parental cell line and chemo-resistant sublines. The following are our explanations. Firstly, statistical significance is just one of the three criteria for filtering genes. The expression level of genes may be more important, and fold change was used in most of microarray analysis up-to-date. For published articles concerning microarray analysis, few have included statistical significance as a criterion. In order to make result more convincing, we tried to use ANOVA to help to find difference. Secondly, for adequate statistical analysis, large sample size is suggested. But for microarray analysis, it is not practical because of high expense for each microarray. If more replication can be acquired, the more standard statistical analysis will be available. So let's hypothesize that false discovery correction was used and set cutoff of P <0.05, it would exclude 2370 genes which had large fold change between parental and resistant cell lines, and only several hundred genes which were judged to be of statistical significance would be screened. When these genes were filtered by fold change, a few differential genes would be acquired. In this way we would lose much important information. The last, our research targets were constructed cell lines, duplications were almost the same with each other, so we could try to draw conclusion from the limited samples and replications according to the criteria listed in the article. For further analysis, the samples from tumor tissue may be used and acquire large sample size, in that case, the method and parameters of statistical analysis will be modified. As to the second question, there were reports about the usage of leave-one-out cross validation.1,2 In our research, the leave-one-out cross validation was a method to validate the microarray analysis, the real-time PCR and western blot. All these experiments helped to validate our microarray. Because of limited samples used, large independent samples for external validation were not practical until now. But in our research we also analyzed an independent pancreatic cancer cell line, Aspc-1. The result showed that the filtered gene may be related with MDR. For the third question. The genes for real-time PCR were selected according to the following criteria: (1) genes must be selected from filtered result; (2) the expression value of genes must be greater than 100 (see normalization information), in case it can not be detected by PCR analysis; (3) genes with both known and unknown ESTs should be considered for further analysis. The genes for western blot were selected according to the following criteria: (1) available monoclonal or multiclonal antibody for detecting protein expression; (2) by bioinformatic analysis, related information in the internet is checked and the genes which will be used for further analysis are selected. As to the last question, we are continuing our studies on constructing and finding more pancreatic cancer resistant cell lines and culturing new pancreatic cell lines from specimens. When sufficient material becomes available, more advanced research that may also include animal model will be carried out. Thanks a lot.