Search

We introduce Pathifier, an algorithm that infers pathway deregulation scores for each tumor sample on the basis of expression data. This score is determined, in a context-specific manner, for every particular dataset and type of cancer that is being investigated. The algorithm transforms gene-level information into pathway-level information, generating a compact and biologically relevant representation of each sample. We demonstrate the algorithm's performance on three colorectal cancer datasets and two glioblastoma multiforme datasets and show that our multipathway-based representation is reproducible, preserves much of the original information, and allows inference of complex biologically significant information. We discovered several pathways that were significantly associated with survival of glioblastoma patients and two whose scores are predictive of survival in colorectal cancer: CXCR3-mediated signaling and oxidative phosphorylation. We also identified a subclass of proneural and neural glioblastoma with significantly better survival, and an EGF receptor-deregulated subclass of colon cancers.

During disease progression the cells that comprise solid malignancies undergo significant changes in gene copy number and chromosome structure. Colorectal cancer provides an excellent model to study this process. To indentify and characterize chromosomal abnormalities in colorectal cancer, we performed a statistical analysis of 299 expression and 130 SNP arrays profiled at different stages of the disease, including normal tissue, adenoma, stages 1-4 adenocarcinoma, and metastasis. We identified broad (> 1/2 chromosomal arm) and focal (< 1/2 chromosomal arm) events. Broad amplifications were noted on chromosomes 7, 8q, 13q, 20, and X and broad deletions on chromosomes 4, 8p, 14q, 15q, 17p, 18, 20p, and 22q. Focal events (gains or losses) were identified in regions containing known cancer pathway genes, such as VEGFA, MYC, MET, FGF6, FGF23, LYN, MMP9, MYBL2, AURKA, UBE2C, and PTEN. Other focal events encompassed potential new candidate tumor suppressors (losses) and oncogenes (gains), including CCDC68, CSMD1, POLR1D, and PMEPA1. From the expression data, we identified genes whose expression levels reflected their copy number changes and used this relationship to impute copy number changes to samples without accompanying SNP data. This analysis provided the statistical power to show that deletions of 8p, 4p, and 15q are associated with survival and disease progression, and that samples with simultaneous deletions in 18q, 8p, 4p, and 15q have a particularly poor prognosis. Annotation analysis reveals that the oxidative phosphorylation pathway shows a strong tendency for decreased expression in the samples characterized by poor prognosis.

We have analyzed gene expression in different normal human tissues and different types of solid cancers derived from these tissues. The cancers analyzed include brain (astrocytoma and glioblastoma), breast, colon, endometrium, kidney, liver, lung, ovary, prostate, skin, and thyroid cancers. Comparing gene expression in each normal tissue to 12 other normal tissues, we identified 4,917 tissue-selective genes that were selectively expressed in different normal tissues. We also identified 2,929 genes that are overexpressed at least 4-fold in the cancers compared with the normal tissue from which these cancers were derived. The overlap between these two gene groups identified 1,340 tissue-selective genes that are overexpressed in cancers. Different types of cancers, including different brain cancers arising from the same lineage, showed differences in the tissue-selective genes they overexpressed. Melanomas overexpressed the highest number of brain-selective genes and this may contribute to melanoma metastasis to the brain. Of all of the genes with tissue-selective expression, those selectively expressed in testis showed the highest frequency of genes that are overexpressed in at least two types of cancer. However, colon and prostate cancers did not overexpress any testis-selective gene. Nearly all of the genes with tissue-selective expression that are overexpressed in cancers showed selective expression in tissues different from the cancers' tissue of origin. Cancers aberrantly expressing such genes may acquire phenotypic alterations that contribute to cancer cell viability, growth, and metastasis.

Predicting at the time of discovery the prognosis and metastatic potential of cancer is a major challenge in current clinical research. Numerous recent studies searched for gene expression signatures that outperform traditionally used clinical parameters in outcome prediction. Finding such a signature will free many patients of the suffering and toxicity associated with adjuvant chemotherapy given to them under current protocols, even though they do not need such treatment. A reliable set of predictive genes also will contribute to a better understanding of the biological mechanism of metastasis. Several groups have published lists of predictive genes and reported good predictive performance based on them. However, the gene lists obtained for the same clinical types of patients by different groups differed widely and had only very few genes in common. This lack of agreement raised doubts about the reliability and robustness of the reported predictive gene lists, and the main source of the problem was shown to be the small number of samples that were used to generate the gene lists. Here, we introduce a previously undescribed mathematical method, probably approximately correct (PAC) sorting, for evaluating the robustness of such lists. We calculate for several published data sets the number of samples that are needed to achieve any desired level of reproducibility. For example, to achieve a typical overlap of 50% between two predictive lists of genes, breast cancer studies would need the expression profiles of several thousand early discovery patients.

We have analyzed gene expression data from three different kinds of samples: normal human tissues, human cancer cell lines, and leukemic cells from lymphoid and myeloid leukemia pediatric patients. We have searched for genes that are overexpressed in human cancer and also show specific patterns of tissue-dependent expression in normal tissues. Using the expression data of the normal tissues, we identified 4,346 genes with a high variability of expression and clustered these genes according to their relative expression level. Of 91 stable clusters obtained, 24 clusters included genes preferentially expressed either only in hematopoietic tissues or in hematopoietic and one to two other tissues; 28 clusters included genes preferentially expressed in various nonhematopoietic tissues such as neuronal, testis, liver, kidney, muscle, lung, pancreas, and placenta. Analysis of the expression levels of these two groups of genes in the human cancer cell lines and leukemias identified genes that were highly expressed in cancer cells but not in their normal counterparts and, thus, were overexpressed in the cancers. The different cancer cell lines and leukemias varied in the number and identity of these overexpressed genes. The results indicate that many genes that are overexpressed in human cancer cells are specific to a variety of normal tissues, including normal tissues other than those from which the cancer originated. It is suggested that this general property of cancer cells plays a major role in determining the behavior of the cancers, including their metastatic potential.

Using DNA microarray and cluster analysis of expressed genes in a cloned line (M1-t-p53) of myeloid leukemic cells, we have analyzed the expression of genes that are preferentially expressed in different normal tissues. Clustering of 547 highly expressed genes in these leukemic cells showed 38 genes preferentially expressed in normal hematopoietic tissues and 122 other genes preferentially expressed in different normal nonhematopoietic tissues, including neuronal tissues, muscle, liver, and testis. We have also analyzed the genes whose expression in the leukemic cells changed after activation of WT p53 and treatment with the cytokine IL-6 or the calcium mobilizer thapsigargin. Of 620 such genes in the leukemic cells that were differentially expressed in normal tissues, clustering showed 80 genes that were preferentially expressed in hematopoietic tissues and 132 genes in different normal nonhematopoietic tissues that also included neuronal tissues, muscle, liver, and testis. Activation of p53 and treatment with IL-6 or thapsigargin induced different changes in the genes preferentially expressed in these normal tissues. These myeloid leukemic cells thus express genes that are expressed in normal nonhematopoietic tissues, and various treatments can reprogram these cells to induce other such nonhematopoietic genes. The results indicate that these leukemic cells share with normal hematopoietic stem cells the plasticity of differentiation to different cell types. It is suggested that this reprogramming to induce in malignant cells genes that are expressed in different normal tissues may be of clinical value in therapy.

One's present repertoire of antibodies encodes the history of one's past immunological experience. Can the present autoantibody repertoire be consulted to predict resistance or susceptibility to the future development of an autoimmune disease? Here, we developed an antigen microarray chip and used bioinformatic analysis to study a model of type 1 diabetes developing in nonobese diabetic male mice in which the disease was accelerated and synchronized by exposing the mice to cyclophosphamide at 4 weeks of age. We obtained sera from 19 individual mice, treated the mice to induce cyclophosphamide-accelerated diabetes (CAD), and found, as expected, that 9 mice became severely diabetic, whereas 10 mice permanently resisted diabetes. We again obtained serum from each mouse after CAD induction. We then analyzed, by using rank-order and superparamagnetic clustering, the patterns of antibodies in individual mice to 266 different antigens spotted on the chip. A selected panel of 27 different antigens (10% of the array) revealed a pattern of IgG antibody reactivity in the pre-CAD sera that discriminated between the mice resistant or susceptible to CAD with 100% sensitivity and 82% specificity (P = 0.017). Surprisingly, the set of IgG antibodies that was informative before CAD induction did not separate the resistant and susceptible groups after the onset of CAD; new antigens became critical for post-CAD repertoire discrimination. Thus, at least for a model disease, present antibody repertoires can predict future disease, predictive and diagnostic repertoires can differ, and decisive information about immune system behavior can be mined by bioinformatic technology. Repertoires matter.

The ALL-1 gene is directly involved in 5-10% of acute lymphoblastic leukemias (ALLs) and acute myeloid leukemias (AMLs) by fusion to other genes or through internal rearrangements. DNA microarrays were used to determine expression profiles of ALLs and AMLs with ALL-1 rearrangements. These profiles distinguish those tumors from other ALLs and AMLs. The expression patterns of ALL-1-associated tumors, in particular ALLs, involve oncogenes, tumor suppressors, antiapoptotic genes, drug-resistance genes, etc., and correlate with the aggressive nature of the tumors. The genes whose expression differentiates between ALLs with and without ALL-1 rearrangement were further divided into several groups, enabling separation of ALL-1-associated ALLs into two subclasses. One of the groups included 43 genes that exhibited expression profiles closely linked to ALLs with ALL-1 rearrangements. Further, there were evident differences between the expression profiles of AMLs in which ALL-1 had undergone fusion to other genes and AMLs with partial duplication of ALL-1. The extensive analysis described here pinpointed genes that might have a direct role in pathogenesis.

Using DNA microarray and clustering of expressed genes we have analyzed the mechanism of inhibition of wild-type p53-induced apoptosis by the cytokine interleukin 6 (IL-6) and the calcium mobilizer thapsigargin (TG). Clustering analysis of 1,786 genes, the expression level of which changed after activation of wild-type p53 in the absence or presence of IL-6 or TG, showed that these compounds did not cause a general inhibition of the ability of p53 to up-regulate or down-regulate gene expression. Expression of various p53 targets implicated as mediators of p53-induced apoptosis was also not affected by IL-6 or TG. These compounds thus can bypass the effect of wild-type p53 on gene expression and inhibit apoptosis. IL-6 and TG activated different p53-independent pathways of gene expression that include up-regulation of antiapoptotic genes. IL-6 and TG also activated different differentiation-associated genes. The ability of compounds such as cytokines and calcium mobilizers to inhibit p53-mediated apoptosis without generally inhibiting gene expression regulated by p53 can facilitate tumor development and tumor resistance to radiation and chemotherapy in cells that retain wild-type p53.

We present a coupled two-way clustering approach to gene microarray data analysis. The main idea is to identify subsets of the genes and samples, such that when one of these is used to cluster the other, stable and significant partitions emerge. The search for such subsets is a computationally complex task. We present an algorithm, based on iterative clustering, that performs such a search. This analysis is especially suitable for gene microarray data, where the contributions of a variety of biological mechanisms to the gene expression levels are entangled in a large body of experimental data. The method was applied to two gene microarray data sets, on colon cancer and leukemia. By identifying relevant subsets of the data and focusing on them we were able to discover partitions and correlations that were masked and hidden when the full dataset was used in the analysis. Some of these partitions have clear biological interpretation; others can serve to identify possible directions for future research.