Pathways of urothelial cancer progression suggested by Bayesian network analysis of allelotyping data

Abstract

Urothelial cancers of the bladder (UC) comprise biologically heterogeneous group of tumors and display complex genetic alterations. Several genetic changes have been analyzed in detail and some of them are associated with the development and progression of UCs. Only a few studies, however, are focused on identifying the order in which the aberrations may appear during UC tumorigenesis. We have analyzed 123 papillary UCs of the bladder by microsatellites for each of the chromosomal regions that have been suggested to be specifically involved in this type of tumor. We used Bayesian network modeling that enables to uncover multivariate probabilistic dependencies between variables. This methodology applied to LOH data allowed us to discover patterns of losses in UCs. Exploiting the mechanism of probabilistic reasoning in Bayesian networks we suggest primary and secondary events in tumor pathogenesis and reconstruct the possible flow of progression of allelic changes. Key words: urothelial cancer; loss of heterozygosity; Bayesian networks; network inference; genetic pathwaysUrothelial carcinoma of the bladder (UC) comprise biologically and morphologically heterogenous groups of neoplasms. During the last decade a broad spectrum of genetic alterations has been described in UCs. Cytogenetic, CGH and microsatellite analyses showed loss, gain and amplification of DNA sequences at several chromosomal regions. 1 Hemi-and homozygous deletion at and methylation/mutation of the CDKN2A gene at chromosome 9p21 is considered to be an early genetic event. 2 The vast majority of UCs acquire several additional genetic alterations during progression, including deletion of chromosome 2q, 5q and 8p, deletion/ mutation of the p53 and Rb genes or amplification and overexpression of the ERBB-2 gene. [3][4][5][6][7][8] Although some of the genetic alterations occur at random, the recurrent changes may refer to a network of genes that are specifically involved in tumor development and progression.Several models indicating a step-by-step order of genetic changes as a single pathway from normal urothelial cell to malignant tumor have been proposed. [9][10][11] Desper et al. 12 and Schäffer et al. 13 applied mathematical models to infer the order of genetic changes during progression of UCs. They described the progression of alterations as a tree with a root representing the normal cell and used 2 different tree models. In a distance-based tree leaf nodes represent genetic aberrations. The distance function between the nodes in the tree was defined based on probabilities of the co-occurrence of genetic events. A distance-based phylogenetic tree-building algorithm was used to infer the tree structure that fits the pairwise distances at best. Alternatively, a maximum weight branching algorithm was used to reconstruct a tree in which both internal nodes and leaf nodes correspond to aberrations. A major limitation of these models is that they describe the progression of genetic events as trees, whereas the biological intuition suggests that this...

Molecular pathogenesis of bladder cancer

“…The latter group appeared to contain more aggressive tumors (T1-T3). Bulashevska et al 106 analyzed LOH data for 17 chromosomes. As this does not detect increased copy number, it is not possible to compare their fi ndings with those of Hoglund et al 105 However, the network obtained again showed 9p and 9q as the most probable primary event, with 8p-and 17-as major subsequent events that lead to progression.…”

“…gov/Chromosomes/Mitelman), 105 CGH data, 79 or LOH data. 106 All have predicted loss of 9p/9q as "early" events. Hoglund et al 105 used cytogenetic data to predict two potential pathways, one initiated by −9, followed by −11p and 1q+ and a second initiated by +7 followed by 8p-and +8q.…”

Downregulation of Dicer, a component of the microRNA machinery, in bladder cancer

“…Several reports have indicated a deletion in chromosome 14q in TCCs, especially in more invasive diseases. Micro-deletion in this subtelomeric region 14q32 may be a very early event in the pathogenesis of TCCs (25)(26)(27)(28).…”

Mathematical modeling of carcinogenesis based on chromosome aberration data

“…B5 is associate with both B1 and B2, suggesting that the presence of both B1 and B2 increases the possibility of B5. Bulashevska et al [18] applied Bayesian network models to LOH data from 123 papillary urothelial cancers of the bladder. By allowing only pairwise dependencies and partition aberrations into modules so as to reduce the number of parameters, Hjelm et al [29] constructed Bayesian networks based on a dataset of colorectal cancer tumor karyotypes.…”