The promoters of human cell cycle genes integrate signals from two tumor suppressive pathways during cellular transformation

Yuval Tabach^1,2,a, Michael Milyavsky^1,a, Igor Shats^1,a, Ran Brosh^1,a, Or Zuk², Assif Yitzhaky², Roberto Mantovani³, Eytan Domany², Varda Rotter¹ & Yitzhak Pilpel⁴

^aThese authors contributed equally to this work

Article highlights

Synopsis

A major challenge of systems biology is to decipher the structure of regulatory networks. This amounts to identifying the constituent genes, factors that control their expression, and cellular and extra-cellular signals that affect them. While previous studies have been successful in such tasks in lower organisms, extension to higher organisms, especially mammals, remain a significant challenge.

A particularly challenging case is the multi-layered networks that control mammalian cell cycle. These networks evolved to ensure that cells are committed to proliferation only if the appropriate signals were received from the environment and from within the cell. In healthy conditions proliferation is possible only if suppressive alerts were not received. When such signaling pathways are not functioning well, the control of proliferation is lost and the development of cancer is a likely outcome.

It has thus long been understood that cell cycle control and cancer are intimately related. Nonetheless, understanding the precise molecular events that lead to cancerous transformation, and the architecture of the networks that transmit such changes to the cell cycle machinery still represents a great challenge. Genome-wide monitoring of molecular signatures of various cancers, that are now becoming increasingly available through expression chip technologies, have the potential to reveal key molecular events that lead to cancer. Yet, while highly informative, the approach of profiling genome-wide mRNA expression of patients has several critical disadvantages: samples are usually taken from patients at a stage that may be far from the initial onset of the tumorigenesis and knowledge about the history of the process, and ability to manipulate its progression, are lacking. Additional difficulties stem from different genetic backgrounds of patients, variable mutations in tumors, and the uncontrolled contaminations by additional cell types.

Thus, in order to obtain novel and reliable insights into genetic networks associated with oncogenesis, we have recently developed an in-vitro model for cellular transformation ( Milyavsky et al, 2003). This transformation process started with normal cells that would normally enter senescence. In order to overcome senescence and transform the cells they were manipulated by inactivation of tumor suppressors and over-expression of oncogenes. These manipulations gave rise to cells that formed tumors in mice (Milyavsky et al, 2005). Using DNA chips we took snapshots of the transformation process by monitoring mRNA expression at 12 points during transformation (Figure 1). This allowed us to record the history of the process from its onset to the final cancerous state.

Figure 1

(A) Outline of the malignant transformation process. Schematic representation of the spontaneous (young, senescent, immortal, tumorigenic, INK4A methylation) and induced (hTERT, H-Ras, p53 inactivation) modifications of the WI-38 cells along the process of malignant transformation. The stages chosen for microarray profiling are indicated by boxes with numerals corresponding to columns in the expression matrix shown in (B). The time scale of the process is depicted by a horizontal axis, and the corresponding population doublings are represented by PDLs. (B) The normalized expression levels of the 168 genes in the proliferation cluster at 12 stages spanning the transformation process. Normalized expression level is color-coded according to the color bar on the right. The table below the matrix contains the following information on each sample: days in culture, geometric mean and standard deviation of expression level of the cluster's genes, doubling rate (cell cycle doublings/day) of cells at selected stages, activity of hTERT (designated as '+' for all samples following hTERT overexpression), activity of p53, as inferred from the application of its dominant-negative peptide, GSE56 ('-' indicates expression of GSE56). Here and throughout the paper, the following cell line designations are introduced: cells are either young or senescent; grow slow or fast; a sample name followed by 'G' denotes the application of GSE56; T before sample names indicates the presence of the immortalizing telomerase; R following the sample name indicates the insertion of Ras.

Full figure and legend (261K)Figures & Tables index

In the present study we aimed at deciphering the architecture of the control networks that derives this malignant transformation. We focused on one cluster of genes that is mainly composed of constituents of the cell cycle machinery (with genes related to processes such as DNA replication, mitotic spindle organization, chromosome segregation etc.). We started by analysis of the promoters of these genes and mainly detected binding sites of known cell cycle transcription factors. Subsequent analysis revealed that these factors are not working in isolation, but rather form combinatorial interactions among themselves that likely allow the integration of multiple sources of information into the control of expression of the regulated genes.

But the true power of our experimental setup, that allowed analysis of the entire tumorigenesis process, manifested itself when we analyzed the activity, throughout the process, of two main tumor suppressive channels centered around p53 and p16 genes that are known to be inactivated in many naturally occurring cancers. These two tumor suppressors mainly transmit tumor suppressive signals from within the cell or from its environment, respectively. We found that many of the genes in the cluster are subject to a combined regulation of the two tumor suppressors; the activity of either of the suppressive pathways alone did not contain enough information for modeling the expression of the majority of the genes in the cluster. In many classical genetic switches with two regulators, a simple 'logical gate' may describe the dependence of the regulated gene on its regulators. In some cases both regulators have to be active in order for the gene to be affected (an 'AND gate), in other cases either regulator is sufficient (an 'OR gate'). In our case we found a more complex behavior - the expression level of many of the genes in the cluster was not proportional to the AND-gated, or OR-gated activity of p53 and p16, yet they showed marked (negative) correlation with the summated expression profiles of the two regulators. Nonetheless, not all genes appeared to share this response. Further promoter motif analyses revealed that only genes that have a particular combination of regulatory motifs show an apparent ability to sum up the activity from the two channels and produce an analogous output. This has allowed us to establish a 3-way connection between gene expression, promoter architecture, and state of up-stream signaling pathway (Figure 5). But a fourth layer was still missing. What is the biological significance of the expression pattern of the genes in this cluster? As may be expected from their role in cell cycle progression we found that the doubling time of the cells in various stages of the transformation process was correlated with the average expression level of the genes in the cluster. What this means is that the promoter motif architecture identified here serves to integrate activity from two prime tumor suppressors and map them onto corresponding rates of cellular proliferation. At a stage where the two tumor suppressors are inactivated the expression level of the cluster's genes is maximal and so is their proliferation rate. We have then conducted experimental tests to check the computational hypotheses and confirmed many of them on a sample of the genes. As a result we have substantially increased the list of indirect repression targets of p53.

Figure 5

Three-way linkage between expression profiles, promoter architecture, and tumor suppressor pathways. (A) The entire expression matrix of the proliferation cluster genes, sorted with SPIN (Tsafrir et al, 2005), revealing an 'elongated' shape for this cluster. (B) Correlation coefficient between p21 and p16 expression profiles to each of the genes in the cluster; color-coded according to color bar shown. (C) Cumulative distribution of CHR, ELK1, and NFY along the sorted list of genes in (A). (D) Bars depicting main areas of density of regulatory motifs along the sorted expression matrix in (A) are based on cumulative appearance in (C). (E) Arrows in the networks represent positive (green) or negative (red) interactions. Reviewed: 1—Ullrich et al (1990); 2—Gille et al (1995); 3—Serrano et al (1997), Lin et al (1998), Zhu et al (1998); 4, 5, 7, 9–11—Weinberg (1995), Sherr (1996), Sherr et al (1999), Hahn et al (2002); 6—el-Deiry et al (1993); 8—Yun et al (2003); 12—newly proposed interaction (this paper).

Full figure and legend (173K)Figures & Tables index

Lastly, a crucial question is the relevance of these genes, which were manipulated here artificially, to naturally occurring cancers. Expression profiles of mRNAs derived from cancer patients revealed that the transcription levels of the cluster's genes are good predictors of diseases prognosis with high levels of the cluster's genes observed in patients with poor outcome. This is an indication that the genes whose regulation was deciphered here are key players in naturally occurring cancers.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.