Construction of evolutionary tree models for renal cell carcinoma from comparative genomic hybridization data.

Renal cell carcinoma is characterized by an accumulation of complex chromosomal alterations during tumor progression. Chromosome 3p deletions are known to occur early in the carcinogenesis, but the nature of subsequent events, their interrelationships, and their sequence is poorly understood, as one usually only obtains a single "view" of the dynamic process of tumor development in a particular cancer patient. To address this limitation, we used comparative genomic hybridization analysis in combination with a distance-based and a branching-tree method to search for tree models of the oncogenesis process of 116 conventional (clear cell) renal carcinomas. This provides a means to analyze and model cancer development processes based on a more dynamic model, including the presence of multiple pathways, as compared with the fixed linear model first proposed by Vogelstein et al. (N. Engl. J. Med., 319: 525-532, 1988) for colorectal cancer. The most common DNA losses involved 3p (61%), 4q (50%), 6q (40%), 9p (35%), 13q (37%), and Xq (21%). The most common gains were seen at chromosome 17p and 17q (20%). The tree model derived from the distance-based method is consistent with the established theory that -3p is an important early event in conventional (clear cell) renal cancer and supports the prediction made from the branching tree that -4q is another important early event. Both tree models suggest that there may be two groups of clear cell renal cancers: one characterized by -6q, +17q, and + 17p, and another by -9p, -13q, and -18q. Putative prognostic parameters were -9p and -13q. The distance-based tree clarifies that -8p (present in 12% of tumors) is a late event, largely independent of other events. In summary, tree modeling of comparative genomic hybridization data provided new information on the interrelationships of genetic changes in renal cancer and their possible order, as well as a clustering of these events. Using tree analysis, one can derive a more in-depth understanding of the renal cancer development process than is possible by simply focusing on the frequencies of genetic events in a given cancer type.

Distance-based reconstruction of tree models for oncogenesis.

Comparative genomic hybridization (CGH) is a laboratory method to measure gains and losses in the copy number of chromosomal regions in tumor cells. It is hypothesized that certain DNA gains and losses are related to cancer progression and that the patterns of these changes are relevant to the clinical consequences of the cancer. It is therefore of interest to develop models which predict the occurrence of these events, as well as techniques for learning such models from CGH data. We continue our study of the mathematical foundations for inferring a model of tumor progression from a CGH data set that we started in Desper et al. (1999). In that paper, we proposed a class of probabilistic tree models and showed that an algorithm based on maximum-weight branching in a graph correctly infers the topology of the tree, under plausible assumptions. In this paper, we extend that work in the direction of the so-called distance-based trees, in which events are leaves of the tree, in the style of models common in phylogenetics. Then we show how to reconstruct the distance-based trees using tree-fitting algorithms developed by researchers in phylogenetics. The main advantages of the distance-based models are that 1) they represent information about co-occurrences of all pairs of events, instead of just some pairs, 2) they allow quantitative predictions about which events occur early in tumor progression, and 3) they bring into play the extensive methodology and software developed in the context of phylogenetics. We illustrate the distance-based tree method and how it complements the branching tree method, with a CGH data set for renal cancer.

Inferring tree models for oncogenesis from comparative genome hybridization data.

Comparative genome hybridization (CGH) is a laboratory method to measure gains and losses of chromosomal regions in tumor cells. It is believed that DNA gains and losses in tumor cells do not occur entirely at random, but partly through some flow of causality. Models that relate tumor progression to the occurrence of DNA gains and losses could be very useful in hunting cancer genes and in cancer diagnosis. We lay some mathematical foundations for inferring a model of tumor progression from a CGH data set. We consider a class of tree models that are more general than a path model that has been developed for colorectal cancer. We derive a tree model inference algorithm based on the idea of a maximum-weight branching in a graph, and we show that under plausible assumptions our algorithm infers the correct tree. We have implemented our methods in software, and we illustrate with a CGH data set for renal cancer.

Tree models for dependent copy number changes in bladder cancer.

We analyzed comparative genomic hybridization data on a collection of 237 bladder cancer tumors with the aim of identifying sets of copy number aberrations (CNAs) that tend to occur together. A test based on Fisher's exact test for pairs, but taking into account multiple testing, showed strong dependencies amongst several pairs of aberrations including (+1q, -11p), (+17q, +20q), (+10p, -17p), (-8p, -17p), (+5p, +10p). To determine whether co-occurrence of CNAs may characterize tumor subtypes, we used two recently proposed methods to construct tree models of tumor progression. We constructed tree models for all the tumors, the tumors of stage pT1, and the tumors of stages pT2-4. The tree models confirmed that most of the non-random events and the associations are the same for different stages. We conclude that the combination of large data sets and tree models provide a useful approach to systematically identifying tumor subgroups characterized by more than a single chromosomal aberration.

Graph models of oncogenesis with an application to melanoma.

We describe several analytical techniques for use in developing genetic models of oncogenesis including: methods for the selection of important genetic events, construction of graph models (including distance-based trees, branching trees, contingency trees and directed acyclic graph models) from these events and methods for interpretation of the resulting models. The models can be used to make predictions about: which genetic events tend to occur early, which events tend to occur together and the likely order of events. Unlike simple path models of oncogenesis, our models allow dependencies to exist between specific genetic changes and allow for multiple, divergent paths in tumor progression. A variety of genetic events can be used with the graph models including chromosome breaks, losses or gains of large DNA regions, small mutations and changes in methylation. As an application of the techniques, we use a recently published cytogenetic analysis of 206 melanoma cases [Nelson et al. (2000), Cancer Genet. Cytogenet.122, 101-109] to derive graph models for chromosome breaks in melanoma. Among our predictions are: (1) breaks in 6q1 and 1q1 are early events, with 6q1 preferentially occurring first and increasing the probability of a break in 1q1 and (2) breaks in the two sets [1p1, 1p2, 9q1] and [1q1, 7p2, 9p2] tend to occur together. This study illustrates that the application of graph models to genetic data from tumor sets provide new information on the interrelationships among genetic changes during tumor progression.

Chromosome abnormalities in ovarian adenocarcinoma: III. Using breakpoint data to infer and test mathematical models for oncogenesis.

Cancer geneticists seek to identify genetic changes in tumor cells and to relate the genetic changes to tumor development. Because single changes can disrupt the cell cycle and promote other genetic changes, it is extremely hard to distinguish cause from effect. In this article we illustrate how 7 techniques from statistics, theoretical computer science, and phylogenetics can be used to infer and test possible models of tumor progression from single genome-wide descriptions of aberrations in a large sample of tumors. Specifically, we propose 4 tree models for tumor progression inferred from the large ovarian cancer data set described in the first 2 articles in this series. The models are derived from 2 different methods to select the non-random genetic aberrations and 2 different methods to infer the trees, given a set of events. Various aspects of the tree models are tested and extended by 5 methods: overall tests of independence, likelihood ratio tests, principal components analysis, directed acyclic graph modeling, and Bayesian survival analysis. All our methods lead to strikingly consistent conclusions about chromosomal breakpoints in ovarian adenocarcinoma, including (1) the non-random breakpoints in ovarian adenocarcinoma do not occur independently; (2) breakpoints in regions 1p3 and 11p1 are important early events and distinguish a class of tumors associated with poor prognosis; and (3) breakpoints in 1p1, 3p1, and 1q2 distinguish a class of ovarian tumors, and the breaks at 1p1 and 3p1 are associated with poor prognosis.

Evaluation of the clonal relationship between primary and metastatic renal cell carcinoma by comparative genomic hybridization.

The outcome of patients with renal cell carcinoma is limited by the development of metastasis after nephrectomy. To evaluate the genetic basis underlying metastatic progression of human renal cell carcinoma in vivo, we performed a comparative genomic hybridization analysis in 32 clear-cell renal-cell carcinoma metastases. The most common losses involved chromosomes 3p (25%), 4q (28%), 6q (28%), 8p (31%), and 9p (47%). The most common gains were detected at 17q (31%) and Xq (28%). There was one high-level gene amplification at chromosome 11q22-23. The mean number of aberrations in lymph node (4.8 +/- 2.8) and lung metastases (6.2 +/- 4.0) was lower than in other hematogenous metastases (11.5 +/- 8.7, P < 0.05), suggesting that hematogenous dissemination is linked to an acquisition of complex genomic alterations. As genetic differences between primary tumors and metastases give information on genetic changes that have contributed to the metastatic process, relative DNA sequence copy number changes in 19 matched tumor pairs were compared. Genomic changes, which frequently occurred in metastases but not in the corresponding primary tumor were losses of 8p and 9p and gains of 17q and Xq. An abnormal function of genes in these regions may contribute to the metastatic process. According to a statistical analysis of shared genetic changes in matched tumor pairs, a high probability of a common clonal progenitor was found in 11 of 19 patients (58%). Six metastases (32%) were genetically almost completely different from the primary, suggesting that detection of genomic alterations in primary tumors gives only a restricted view of the biological properties of metastatic renal cell carcinoma.

Somatic deletions in hereditary breast cancers implicate 13q21 as a putative novel breast cancer susceptibility locus.

A significant proportion of familial breast cancers cannot be explained by mutations in the BRCA1 or BRCA2 genes. We applied a strategy to identify predisposition loci for breast cancer by using mathematical models to identify early somatic genetic deletions in tumor tissues followed by targeted linkage analysis. Comparative genomic hybridization was used to study 61 breast tumors from 37 breast cancer families with no identified BRCA1 or BRCA2 mutations. Branching and phylogenetic tree models predicted that loss of 13q was one of the earliest genetic events in hereditary cancers. In a Swedish family with five breast cancer cases, all analyzed tumors showed distinct 13q deletions, with the minimal region of loss at 13q21-q22. Genotyping revealed segregation of a shared 13q21 germ-line haplotype in the family. Targeted linkage analysis was carried out in a set of 77 Finnish, Icelandic, and Swedish breast cancer families with no detected BRCA1 and BRCA2 mutations. A maximum parametric two-point logarithm of odds score of 2.76 was obtained for a marker at 13q21 (D13S1308, theta = 0.10). The multipoint logarithm of odds score under heterogeneity was 3.46. The results were further evaluated by simulation to assess the probability of obtaining significant evidence in favor of linkage by chance as well as to take into account the possible influence of the BRCA2 locus, located at a recombination fraction of 0.25 from the new locus. The simulation substantiated the evidence of linkage at D13S1308 (P < 0.0017). The results warrant studies of this putative breast cancer predisposition locus in other populations.