IDENTIFICATION OF TARGETED CANCER THERAPIES 
USING GENE MODULE MAPS



Jessi Blackburn

based on Wong et al. (1)

"Revealing Targeted Therapy for Human Cancer by Gene Module Maps"




I. INTRODUCTION


i. Overview

 

Cancer is a very complex disease, resulting from a diverse collection of genetic gains, chromosomal losses, and epigenetic modifications, the full effects of which on tumor biology are still largely unknown.  However, within the last decade, array-based molecular profiling has revolutionized the field of cancer research.  Researchers can now utilize microarray technology to analyze the expression of thousands of genes simultaneously in tumor cells or tissue samples, allowing them to study the complex biology of cancer much more comprehensively than ever before. Besides basic biology, the gene expression profiles of human cancers have been found to contain crucial information on clinical behavior, where they can accurately predict patient prognosis, response to treatment and overall survival. However, while microarrays have improved the predictive aspects of cancer diagnosis, they have been under-utilized in the design of targeted therapies. Recent work by Wong et al. applies gene module maps to tumor microarray data in order to identify a novel therapy for breast cancer patients whose tumors expressed a very specific poor-prognosis signature. This type of work brings the cancer field one step closer to the ultimate goal of using gene expression profiles to match drug with disease, thereby individualizing cancer treatment.      



ii. Microarrays


The principle behind microarray analysis is that mRNA from a single sample is labeled for use as a target, and then hybridized to a large number of single-stranded DNA sequences, termed probes.  The expression level of a given gene can then be determined by the intensity of the signal, or the amount of mRNA binding to a particular DNA sequence. There are 2 types of microarrays: spotted cDNA arrays and oligonucleotide arrays, detailed in Figure 1. Current microarray technology allows for analysis of the entire genome; for example, Affymetrix provides a microarray spanning ~28,000 genes, utilizing over 750,000 different probes. 


Microarray data are typically assembled into heat-maps (Figure 2), showing genes that are up- or down-regulated in experimental samples compared to control. Heat-maps are then sorted based on cluster analysis, where genes with similar expression levels are grouped together; the genes may or may not have any functional relation.  See reviews by Schulze and Downward, and Lee and Saeed for more information.

 

iii. Clinical Use of Microarrays


There is much enthusiasm for the application of microarray technology to the clinical setting. However, if microarrays are to contribute to both patient diagnosis and tailored medicine, there can be no questions regarding their accuracy in determining the gene expression profile of a tumor sample. For this reason, microarrays are not yet considered appropriate for clinical use.  Several reports have been published detailing contradictory results when the same RNA sample was hybridized to different microarray platforms (2). Other studies have shown that the same clinical problem, such as metastatic breast cancer, generated completely different gene expression profiles, calling into question the reliability of micoarrays in disease diagnosis (3). Because many of these discrepancies can be explained by technological difficulties, such as use of different probe sets, inadequate normalization and small sample size, and because of the great potential benefit of microarray technology in the clinic, the FDA has established the Microarray Quality Control (MAQC) project, in which 137 different participants from across the country work together to reach a consensus both on the technical procedures used in microarray analysis, and on the procedures involved in creating and validating prognostic signatures for disease diagnosis.    



iv. Prognostic Signatures in Breast Cancer


Breast cancer has been one of the most extensively studied tumor types in the microarray era. Gene expression data has been used to define new subtypes of the disease, to designate treatment sensitivity, and to predict patient prognosis.  Because breast cancer is the most common cancer, and accounts for ~16% of all cancer deaths prognostication of early breast cancer is key in providing the correct treatment to produce best outcome for the patient (4). Currently, however, the only prognostic indicators clinically available for breast cancer that are over-expression of the estrogen receptor (ER) and human epidermal growth factor receptor 2 (HER2), both of which are associated with more aggressive disease.     


Within the last 5 years, numerous groups have use gene expression profiling to improve upon the quality and quantity of prognostic markers, and have found several gene expression signatures that are consistently and significantly associated with a poor prognosis in breast cancer. All of these signatures were derived from gene expression data from hundreds of individual tumors, and some are currently undergoing clinical validation.  The 70-geneAmsterdam signature, which can predict the likelihood of metastasis 5 years after diagnosis in seemingly good-prognosis (lymph-node negative) patients, has recently been FDA approved. It is currently marketed as the Mammaprint by Agilent. The goal of the 70-gene Mammaprint is to save patients defined as good prognosis (less than a 5 percent chance of developing metastases 5 years after diagnosis) from unnecessary chemotherapy.  To date, this is the only FDA approved test for gene expression profiling in the clinic.  




However, despite the development of such excellent prognostic indictors, there are very few suggestions for how to use these signatures to define better treatments for poor-prognosis patients. Prognostic signatures are simply a list of genes consistently over- or under-expressed in poor-prognosis cancer; unfortunately, these signatures reveal very little about the biology driving thee aggressive nature of the tumor.  



v. Gene Module Maps


In 2008, Wong et al. applied gene module maps to breast cancer microarray data, allowing them to discover a functional pathway active in tumors expressing a specific poor-prognosis signature. They then targeted this pathway to kill tumor cells expressing the prognostic signature, efficiently demonstrating that the application of module maps to gene expression data can provide a rapid way to translate a prognostic signature into a targeted therapy.    


A module map is a simple bioinformatics strategy based on the principle that genes typically function together, such as in a signal transduction cascade or an enzymatic pathway.  Additionally, prior biological knowledge of genes, such as a similar structure, tissue-specificity, or induction by a certain stimulus, can be used to group genes together.  Gene modules are therefore defined a priori, and gene module analysis of microarray data searches for coordinate regulation of these modules within the microarray (5).  Expression profiles, which can contain 25,000 genes, are thereby simplified into several hundred up- and down-regulated functional groups. There are several resources available on the internet that can be used to created gene modules, such as GeneXpress, Gene Ontology, GeneHopping, and Onto-Tools.


As with all other bioinformatics strategies, there are advantages and disadvantages to the application of gene modules to microarray data.  As described in Figure 3, gene module maps can detect subtle but significant changes in gene expression that would otherwise go unnoticed in typical cluster analysis. Also importantly, because gene module maps are based on biologic function, applying module maps to array data allows the researcher to better understand the biology underlying any changes in gene expression.  However, it is this basis on biological function that is the major limitation of gene module analysis. The functions of many human genes are either unknown or poorly defined, and thus are not included in module maps, which can cause researchers to miss genes that are potentially vital to disease development and progression (5).        




II. Revealing Targeted Therapy for Human Cancer by Gene Module Maps

Wong et al. used the software tool Genomica to apply gene sets derived from Gene Ontology to gene expression data from 295 breast tumors.  The goal of the study was to use module maps to identify a new, targeted therapy to treat breast cancers with a poor-prognosis wound signature.  Chang et al. has previously assoicated the wound signature with increased risk of metastasis and death in breast, lung and gastric cancers. Wong et al. found several related gene sets coordinately regulated with the wound signature (Figure 4A), which they combined into 2 larger mitochondria and proteasome modules (Figure 4B and 4C). It is important to note that there is very little overlap in the genes expressed in these modules and those expressed in the wound signature itself, indicating these genes are coordinately regulated, and not part of the prognostic signature.


When expression of these modules was applied to survival data, it was found that the over-expression of the proteasome module was very significantly correlated with less metastasis-free and overall survival (Kaplan-Meier curves shown in Figure 5).  Further, meta-analysis of multiple patient populations, containing over 1000 different tumors at all stages of progression, demonstrated that the wound signature and the proteasome module are very strongly correlated in human breast cancer (Table II). 



Genetic linkage analysis by Adler, et al. has previously shown tumors expressing the wound signature have gains in chromosome 8q, resulting in over-expression of MYC, a transcription factor, and CSN5, an activator of ubiquitin ligases (6).  Transfection of these genes into normal breast epithelial cells induces a gene expression profile very similar to that of the wound signature (Figure 6).


Wong et al. found stable over-expression of these genes in MCF10As, a normal breast epithelial cell line, also coordinately induced expression of the proteasome modules (Figure 7).  Cells transfected with MYC and CSN5 were therefore treated with Bortezomib, a proteasome inhibitor. In Figure 8, it was shown that Bortezomib selectively kills the cell lines over-expressing MYC and CSN5, suggesting that Bortozomib may target those cells expressing the wound signature. This was demonstrated explicitly in Figure 9, where several breast cancer cell lines were used to correlate expression of the wound signature to sensitivity to Bortezomib.



III. Conclusion


There is currently a diverse array of drugs targeting many different molecular targets, but the challenge in cancer treatment has always been matching the right drug with the right disease. For example, Bortezomib is an FDA approved drug already in clinical use for multiple myloma, and is currently undergoing trials for use in solid tumors. These data demonstrate that it may be very effective against breast cancers expressing the poor-prognosis wound signature.  This work is among the first to demonstrate a strategy for identifying targeted therapies for subtypes of cancer identified by a specific prognostic gene expression signature.


In this study, the use of module maps also revealed new consequences of oncogenic transformation.  MYC is a well known oncogenic transcription factor, and has been found to induce genes associated with the wound signature in breast cancer. MYC was also found to induce genes associated with the proteasome module. This was not entirely unexpected, as the proteasome is thought to act as a co-activator of certain MYC target genes.  However, this work demonstrated that breast cancer cells expressing MYC, and thus the wound signature, have come to rely heavily on proteasome function for survival; these cells were quickly killed when treated with a proteasome inhibitor. In this instance, module maps yielded new insight into oncogenic transformation and breast cancer biology.


Gene expression data are being generated daily on many different cancer types. Gene module maps streamline data analysis to provide biologic and mechanistic insights on particular gene expression profiles or prognostic signatures. This may facilitate the rapid translation of gene expression data into targeted therapies for patients. To this end, the software Genomica, used by Wong et al. in the work detailed above, is available for public use, and provides detailed tutorials on generating module maps from gene expression data and gene sets.        



 

IV. References

1) Wong D. et al. Revealing Targeted Therapy for Human Cancer by Gene Module Maps.  Cancer Res. 2008; 68:2: 369-378

2) Shi L. et al. The MicroArray Quality Control project shows inter- and intraplatform  reproducibility of gene expression measurements. Nature Biotech.
    2006; 24: 1151-1161

3) Dupuy A.et al. Critical review of published microarray studies for cancer outcome and guidelines on statistical analysis and reporting. J Natl Cancer Ins
    2007; 99: 147-157

4)  Sortiriou et al. Taking gene-expression profiling to the clinic: when will molecular signatures become relevant to patient care? Nature Rev Cancer.
     2007; 7: 545-553   

5) Wong et al. Learning more from microarrays: Insights from modules and networks. J Invest Dermatology. 2005; 125: 175-182

6) Adler et al. Genetic regulators of large-scale transcriptional signatures in cancer. Nature Genetics. 2006; 38;4: 421-430