Genetic Regulators of Large-scale Transcriptional Signatures in Cancer Presented by Mei Liu September 26, 2007

Introduction Over the years Global gene expression profiles of thousands of disease specimens, especially cancer, have been analyzed Hundreds of gene expression signatures associated with disease progression, prognosis, and response to therapy have been described The signatures encompass genes that are associated with many important parameters of cancer, but their control mechanisms are still largely unknown Limitations Each signature contains large number of genes, so it is technically infeasible to study the function of an expression signature as a whole Forced to study candidate genes individually or a handful of genes in multiplex fashion

Introduction Limited assessment of the functional consequences of a signature Hampered development of specific therapies that may target cancer on the basis of their gene expression signatures Gene expression signatures may arise in cancer samples for many reasons: Variations in the composition of cell types Responses to different host environment Accumulated effects of aneuploidy and epigenetic changes (acting in cis) Response to altered activities of key transcriptional regulators in cancer (acting in trans)

Introduction To experimentally reproduce and functionally assess the consequences of gene expression signatures Regulators offer an efficient approach Encodes transcriptional factors or signal proteins that controls hundreds of downstream genes Disadvantages: A signature may be controlled by one or more regulators that act in a conditional or combinatorial manner Regulator itself may not be part of the expression signature  Need an unbiased genome-wide method to identify functional regulators of gene expression signatures

Introduction Proposed a general method based on genetic linkage Identify functional regulators that drive large-scale transcriptional signatures in cancer Intersect genome-wide DNA copy number and gene expression data Used the method to identify genetic regulators of the ‘wound respond signature’ in human breast cancers Based on the concept that molecular programs of normal wound healing might be reactivated in cancer metastasis The wound signature might be genetically determined because it is expressed in tumor cells and is a consistent feature in repeat sampling of tumors

Results – Linkage Analysis Genotype – genetic makeup (particular set of genes an organism possesses) Phenotype – actual physical properties (i.e. height, weight, hair color) Linkage analysis aims to associate the pattern of genotype dist. with the pattern of phenotype dist. in a group of individuals in order identify the likely genes that control the phenotype In this case, phenotype is the presence or absence of the gene expression signatures in cancer samples Difficulty Genes involved in linkage analyses << # of samples ~ 10,000 genes vs. ~50 samples in typical microarray studies of cancer Insufficient statistical power to map the linkage to each gene

Results – Linkage Analysis SLAMS (Stepwise Linkage Analysis of Microarray Signatures) Initially map linkage of prospective regulator genes to large chromosomal regions Then refine and validate the list of candidate regulators within the linked region using additional sources of data Overcome inherent noise in gene expression and DNA copy number data Define phenotype in the linkage analysis by the coordinate behavior of many genes within a gene expression signature Establish linkage to chromosomal regions by coordinate amplification or deletion of several neighboring loci

Results – Linkage Analysis SLAMS (four-step strategy) #1: Sort tumors into two groups by presence or absence of the signature #2: Rank the change in DNA copy number of each gene by association with the signature #3: Filter candidate genes encoded within the linked chromosomal locus by their transcriptional regulation #4: Validate based on ability of their expression levels to predict the signature in additional tumor samples

Results – Linkage Analysis SLAMS Application Analyzed 37 breast tumors for gene expression patterns and mapped for DNA copy number change at 6,692 loci Observed amplification of 57 DNA probes in association with the wound signature 32 probes represent chromosome 8q 132 probes representing 8q out of total 6,692 probes Probability of encountering 32 of 132 probes from one chromosomal arm in 57 random trials is 3.4 x Strong linkage between amplification of a large region of 8q with the wound signature

Results – Linkage Analysis Filtered the 32 amplified genes in 8q based on their mRNA expression patterns in 85 breast tumors Tested association between level of mRNA expression of candidate genes and the wound signature CSN5 showed the strongest positive correlation with the wound signature among tumor samples Pairwise and iterative analysis of CSN5 with candidate regulators suggested the combination of CSN5 with MYC mRNA was significantly associated with the wound signature (P = 6.6 x ) Predict CSN5 and MYC function together to activate the wound signature

Results – Linkage Analysis Identify the optimal regulatory model of the wound signature in tumor samples mRNA expression levels of MYC and CSN5 Two-tiered DT assigned tumor samples to 2 groups Group 1: low CSN5 or MYC mRNA level Group 2: moderate or high levels of CSN5 & MYC

Results – Linkage Analysis Substantial different wound score 80% samples with an activated wound signature (wound score  0.2) are captured in group 2 High expression level of CSN5 & MYC is a significant predictor of poor patient survival in breast tumors CSN5 & MYC function together to induce poor-prognosis program in human breast cancers

Results – Linkage Analysis Verify the association between wound signature and amplification of CSN5 and MYC Quantified DNA copy number of CSN5 & MYC loci using an independent set of 41 early breast tumor samples Tumors with the wound signature had significantly higher copy number of CSN5 & MYC MYC & CSN5 as candidate regulators of the wound signature is further supported by additional sources of information

Results – Linkage Analysis Would signature is based on the sustained transcriptional response of fibroblasts to serum stimulation MYC was strongly induced during serum response, as were CSN5 and other CSN components CSN6 is a bona fide member of wound signature MYC & CSN5 can activate a subset of normally serum responsive genes MYC is required for transcriptional response of fibroblast in response to serum Although wound signature genes are not enriched for chromosome 8q localization, they overlapped significantly with MYC target genes (P < ) Suggest direct regulation of wound signature by MYC

Results – Regulator Validation Validation of wound signature regulation by MYC and CSN5 Experimentally validated roles of MYC & CSN5 in wound signature and cancer progression Induced 201 of 255 genes representing ‘activated’ wound signature Repressed 114 of 257 genes representing ‘quiescent’ wound signature Magnitude of wound signature activation induced by MYC & CSN5 co-expression corresponds to 7.3-fold increased risk of death 5.2-fold increased risk of metastatis Confirmed that MYC & CSN5 are causative genetic lesions in breast cancers with the wound signature

Results – Functional Consequences Co-expression of MYC & CSN5 Increased cell proliferation compared with either gene alone Altered cell shape: appeared round less polarized loss of actin stress fibers and focal adhesion contacts Increased the ability of the cells to invade  MYC & CSN5 cooperate functionally to confer several properties associated with invasive tumor cells

Results – Regulation Mechanisms Mechanisms of gene regulation via interplay of MYC and CSN5 Over-expression of CSN5 increased the rate of MYC ubiquitination by 3-fold CSN5 strongly increased turnover of MYC protein No MYC target genes were repressed by CSN5 coexpression, suggesting that CSN5 specifically promotes transcription of select MYC target genes All results together show that CSN5 is an essential activator of MYC transcriptional activity CSN5 increases the transcriptional potency of MYC toward select target genes to promote proliferation, survival, and invasion

Conclusion Developed an integrated genomic approach to identify genetic regulators of large-scale transcriptional signatures in human cancers Method is general and may be used to identify linkage between gene expression signatures and other types of genetic data SNPs or DNA methylation maps Limitation: requires human interpretation, which may introduce subjective bias

Conclusion The wound signature application illustrates several advantages of finding genetic regulators Simplify the application of diagnostic signatures in the clinical setting Knowledge of the regulators allowed us to activate the wound signature in untransformed breast epithelial cells to an extent seen in cancer samples SLAMS method and functional validation can clarify the regulatory architecture of expression signatures and resolves signatures that are causally related vs. those merely occur at the same time

Conclusion The method may be generally useful as a starting point in understanding the regulation and functions of gene expression signatures in cancer Inhibition of CSN5-mediated regulation of MYC may be a useful therapeutic strategy for high-risk breast cancers