Microarrays and Cancer Segal et al. CS 466 Saurabh Sinha

Genomics and pathology Genomics provides high-throughput measurements of molecular mechanisms –Microarrays, ChIP-on-chip, etc. Genomics may provide the molecular underpinnings of pathology, in a highly comprehensive manner –Revolutionize the diagnosis and management of diseases, including cancer

Prior applications to cancer Gene expression measurements have been applied to cancer diagnosis Measure each gene’s expression in several normal tissue samples, and several pathological (diseased) samples Find subset of genes differentially expressed in the two sample groups If such “gene signatures” of particular cancer types are found, they can become the basis of tests for malignancy

We want better … Genes may be differentially expressed, but not enough to cross certain thresholds used in the analysis Analyzing the data on a gene-by-gene basis is error prone -- microarray data has inherent noise Finding the genes involved in one type of cancer is only the first step; it does not reveal the underlying processes

Part 1: Cancer modules

A “module” level view Many methods use “gene modules” (sets of genes) as basic blocks for analysis Instead of trying to find changes in individual gene expression profiles, look out for entire sets of genes with changing expression profiles

The study of Mootha et al. Showed that expression of “oxidative phosphorylation” genes (a particular set of genes) is reduced in diabetic muscle Signal not very strong when looking at individual genes, but highly significant when looking at the “gene module”

Source: Nature Genetics 37, S38 - S45 (2005) Disease tissue (Diabetes mellitus type 2) Normal tissue (Normal tolerance to glucose) Grey: all genes Red: oxidative phosphorylation genes

Segal et al.: Methodology Compile a large collection of cancer-related microarrays –microarrays measuring gene expression in cancer tissues or normal tissue Compile a large collection of gene sets (modules) from earlier studies Identify gene set (modules) induced or repressed in a microarray Identify modules induced in several arrays, or repressed in several arrays Check if these arrays are enriched in some clinical annotation

Identify gene set (modules) induced or repressed in a microarray Given expression value E g,m of each gene g in the microarray experiment m Compute average expression E g of the gene g over all microarrays If E g,m is 2-fold greater than E g, call the gene g as induced in array m Categorize each gene as being induced or not-induced in the array. Source: Nature Genetics 36, (2004)

Identify gene set (modules) induced or repressed in a microarray |All genes| = N |Module| = n |Induced| = m |Intersection| = k Hypergeometric test(N,n,m,k): If a set of m genes was chosen at random (sampling w/o replacement), what is the probability that the intersection would be larger than or equal to k? All genes Module Induced Intersection

Identify gene set (modules) induced or repressed in a microarray |All genes| = N |Module| = n |Induced| = m |Intersection| = k Hypergeometric test(N,n,m,k): Sum over i>=k: If a set of m genes was chosen at random (sampling w/o replacement), what is the probability that the intersection would be equal to i? All genes Module Induced Intersection

Identify gene set (modules) induced or repressed in a microarray |All genes| = N |Module| = n |Induced| = m |Intersection| = k Hypergeometric test(N,n,m,k): All genes Module Induced Intersection “p-value” of the Hypergeometric test

Identify gene set (modules) induced or repressed in a microarray |All genes| = N |Module| = n |Induced| = m |Intersection| = k Hypergeometric test(N,n,m,k) If the “p-value” is very small, then we infer that the intersection is “statistically significant”, i.e., the module is induced in the microarray Similarly define module repressed in microarray All genes Module Induced Intersection

Segal et al.: Methodology Compile a large collection of cancer-related microarrays –microarrays measuring gene expression in cancer tissues or normal tissue Compile a large collection of gene sets (modules) from earlier studies Identify gene set (modules) induced or repressed in a microarray Identify modules induced in several arrays, or repressed in several arrays Check if these arrays are enriched in some clinical annotation

Source: Nature Genetics 36, (2004)

Segal et al.: Methodology Compile a large collection of cancer-related microarrays –microarrays measuring gene expression in cancer tissues or normal tissue Compile a large collection of gene sets (modules) from earlier studies Identify gene set (modules) induced or repressed in a microarray Identify modules induced in several arrays, or repressed in several arrays Check if these arrays are enriched in some clinical annotation

Identify modules induced in several arrays, or repressed in several arrays Source: Nature Genetics 36, (2004)

Segal et al.: Methodology Compile a large collection of cancer-related microarrays –microarrays measuring gene expression in cancer tissues or normal tissue Compile a large collection of gene sets (modules) from earlier studies Identify gene set (modules) induced or repressed in a microarray Identify modules induced in several arrays, or repressed in several arrays Check if these arrays are enriched in some clinical annotation

Source: Nature Genetics 36, (2004)

Segal et al: Cancer “module maps” Source: Nature Genetics 37, S38 - S45 (2005) Red(m,c): Microarrays in which module m was overexpressed (induced) are enriched in condition c Green: Microarrays in which module m was underexpressed (repressed) are enriched in condition c Rows and columns are not in an arbitrary order. They have been “clustered” to display similar rows (or columns) together

Insights from cancer module map Some modules activated or repressed across many tumor types. Such modules could be related to general tumorogenic processes Some modules specifically activated or repressed in certain tumor types or stages of tumor progression

From modules to regulation A module map shows the transcriptional changes underlying cancer Transcriptional changes are a result of transcription factors and their binding sites A deeper understanding of cancer would come from finding out which transcription factors and binding sites led to the transcriptional changes

Part 2: Cis-regulatory elements

Genomics and gene regulation Such knowledge comes from genomics data ChIP-chip studies identify which transcription factors bind which DNA sequences Analysis of DNA sequence, using known binding site motifs, gives us putative binding sites Cross-species conservation also tells us something about possible locations of binding sites

Cis-regulatory analysis Identify a set of genes whose promoters contain the same binding sites –Such a set of genes is likely to be regulated by the same TF –Often called a “regulatory module” Earlier studies mined microarrays for “co- expressed” genes, then used motif finding algorithms to discover their shared binding sites

Cis-regulatory analysis Another approach (Segal et al. 2003) tried to solve the problem in an integrated manner Find a set of genes such that –their expression profiles are similar (microarrays) –they share the same binding sites (sequence) Joint learning of “regulatory module” from two very different types of data: microarray and sequence –An important theme in current bioinformatics

Cis-regulatory analysis Connection between gene expression and cis-regulatory elements (binding sites) also explored in Beer & Tavazoie. Found rules on combinations and locations of binding sites that would cause the gene to be over- or under- expressed

The binding sites “RRPE” and “PAC” must occur within 240 bp and 140 bp of gene start Genes containing both motifs, following certain rules on location, are tightly co-regulated Genes containing any one motif, or both in incorrect positional configuration, have close to random expression Source: Nature Genetics 37, S38 - S45 (2005)

Eukaryotes These studies have mostly focused on yeast (which is a eukaryote, but has a small, compact genome) Not much work of this type in the longer, more complex genomes of metazoans (e.g., humans, rodents, fruitflies) The genome is not compact; may not suffice to look at sequence right next to a gene. Intergenic regions are long, and cis- regulatory signals may not be close to gene

One study in humans HeLa cells are an “immortal” cell-line derived from cervical cancer cells in a person who died in –Used extensively in studying cancer Method of Segal et al. (joint learning of regulatory modules from gene expression and sequence data) applied to these cells

One study in humans Gene expression data used: microarrays measuring genes during cell cycle in HeLa cells Sequence: 1000 bp promoters (upstream) of human genes

Result of analysis: Two motifs found to be shared by this set of genes. The genes have similar expression profiles. One of the identified motifs (NFAT) known to be involved in cell-cycle Source: Nature Genetics 37, S38 - S45 (2005)

Summary The common theme is to analyze sets of genes, and relate their common expression patterns to cancer types or to presence of cis-regulatory motifs Search algorithms may be required to identify some of these features