April 1, 2014 (Vol. 34, No. 7)
Richard A. A. Stein M.D., Ph.D.
Despite advances in the understanding of fundamental mechanisms and the development of increasingly refined therapeutic strategies, the field of cancer biology still presents many difficulties.
For example, at the clinical level, there are stratification and therapeutic challenges. With respect to basic research, cellular pathways of malignancy need to be better defined. Also in acute need of advancement and transformation is the understanding of interindividual genomic and molecular differences.
For many years, cancer has been viewed as a condition in which the progressive accumulation of recurrent mutations leads, in time, to the malignant phenotype. “However, sequencing data reveal that this is not the case,” says Henry H. Heng, Ph.D., an associate professor of pathology at Wayne State University’s Center for Molecular Medicine and Genetics.
The progressive accumulation of genetic alterations as a pathogenic model for cancer development—a view that drives the search for common mutations that are causally linked to cancer—fails to explain many lines of experimental and clinical data generated over the years. “We need to have a paradigm shift from gene-centric to the genome-centric thinking,” continues Dr. Heng.
On the basis of multiple layers of experimental data and theoretical synthesis, Dr. Heng and colleagues have developed a genome-mediated evolutionary model of cancer. In this model, cancer is viewed as an evolutionary process where genome reorganization rather than specific mutations is a key. This model underscores the need to shift from the independent characterization of individual molecules and pathways toward capturing the overall dynamics of how cell populations respond to cellular perturbation.
This model tries to account for the variability that is seen across different cancer gene expression profiling datasets. This variability, which has been found in many studies, constitutes one of the major challenges in cancer biology.
“There is no concept of a fixed cancer genome because, as we already know, every tumor that we have studied has a completely different configuration of the chromosomes,” explains Dr. Heng. Experiments conducted by Dr. Heng and colleagues revealed a marked genomic diversity as one of the main characteristics of cancer cell populations.
One of the consequences of this altered chromosomal architecture is reflected in the transcriptome. “Transcriptome profiling at the single-cell level reveals that every cell is different. In fact, the only way for a cancer cell to survive is to keep changing its transcriptome. This is its way of life, and this is the winning ticket for cancer,” asserts Dr. Heng.
However, this heterogeneity opens difficulties in understanding the link between chromosomal dynamics and cancer development. Cancer has traditionally been viewed as a gene mutation disease where genes define genetic inheritance. “When we sequence genomes, we are generating a catalog of the parts and of the pathways, but we have no idea how to put those parts together,” observes Dr. Heng.
In the context of the marked variability that characterizes the genomes of cancer cells, developing a better understanding of the transcriptome emerges as an even more significant challenge. “The same bricks can be used to construct buildings that are functionally very different,” states Dr. Heng, “but it is impossible to build a house from the bricks without knowing the architecture.”
System inheritance is, therefore, a very different type of inheritance, conceptually, than DNA inheritance (parts inheritance). System inheritance is shaped by the topological relationships among genes and genomic regions within the nucleus.
“DNA represents the inheritance of the materials of the house, but architectural information of how to build the house, which is the systems inheritance, lies within the karyotype (whole chromosomal complement) rather than the DNA,” says Dr. Heng. “Clearly, the blueprint is about the relationship among genes, which is defined by the gene distribution pattern along and between chromosomes.”
This concept has multiple ramifications, one of them being the fact that relying on the individual mutation or expression profiles will not provide information about the blueprint of cancer. To overcome this limitation, more attention should be directed toward genome-level changes.
A cardinal feature of the genome-centric view of cancer is its ability to explain and incorporate the contribution of outliers, which historically have been classified as noise, and are often eliminated from measurements, even though they fundamentally shape population characteristics.
By using single-cell and population-based experimental approaches, Dr. Heng and colleagues revealed that outliers play important roles in cancer evolution and the development of cellular heterogeneity. Outliers cannot be adequately characterized by conventional strategies, but they fundamentally impact cell growth and survival and ultimately cancer’s successful evolution.
The genome theory of cancer evolution does not simply view outliers as “noise” in the system. Instead, outliers are thought to ensure system robustness and to shape evolution and adaptation.
“The key is not to study each individual pathway separately but, instead, to dynamically survey the transcriptome in cells,” advises Dr. Heng. This underscores the need to develop a novel framework in which individual transcriptome profiles can be translated into predictions in evolutionary terms. “We need to develop methods to translate information collected from individual cells to the population potential.
To accomplish this, we need to do more than push technology. We also need to push new concepts,” concludes Dr. Heng.
“One of our efforts is to better understand the epigenetic regulation of the transcriptome, because most of the time, the transcriptome is only a readout, which may or may not be of functional relevance,” says Rakesh Kumar, Ph.D., professor and chair of the department of biochemistry and molecular medicine at George Washington University.
As different pathways and signals feed into epigenetic changes, epigenetic signals are thought to assume a fundamental role in the functionality of the transcriptome. Unless a cell is assayed prospectively at multiple time points, the information provided by transcriptomic data is severely limited. It amounts to a kind of snapshot that reveals correlation but it is much less informative about causation.
“Integrating epigenetic and transcriptomic datasets would open the possibility of stating with a somewhat greater confidence that we are visualizing the functional transcriptome, which is likely to be changed in response to a given signal from that cell’s surface,” continues Dr. Kumar. This approach also promises to maximize opportunities to capitalize on therapeutically important aspects of the transcriptome, because epigenetic modifications that modulate the transcriptome could be therapeutically targeted.
“We cannot modify the transcriptome without going through a barrier of epigenetic changes,” adds Dr. Kumar. “In addition, there is a clear need to focus on master regulatory nodes, which may influence a large proportion of transcriptome while following rules that are not understood at the moment.”
“We view the transcriptome as being the functional capacity of the cell,” says James H. Eberwine, Ph.D., Elmer Holmes Bobst professor of pharmacology at the University of Pennsylvania. Years ago, Dr. Eberwine and colleagues developed one of the first approaches for single-cell RNA amplification, called antisense RNA (aRNA) amplification.
This approach revealed the possibility of harvesting RNA using a microelectrode from single cells dissociated from the rat hippocampus and providing, after two rounds of linear amplification, enough amplified RNA for further analyses, opening the possibility of performing single-cell gene expression profiling. “We now understand, from using single-cell studies, that two cells that look identical and sit next to one another may have completely different transcriptomes,” explains Dr. Eberwine.
Single-cell transcriptome analysis becomes important in many circumstances, one of them being drug screening. Such analysis is particularly appropriate when mRNA-encoding receptors are present in low copy numbers.
By using sequencing-based single-cell transcriptomics, Dr. Eberwine and colleagues revealed that single-cell transcriptomics can help identify receptors, such as G-protein-coupled receptors, and identify novel therapeutic targets in an unbiased manner. “If one wants to screen drugs in a rational manner, rather than screen against every drug from a chemical library,” suggests Dr. Eberwine, “it might be reasonable to perform a single-cell analysis, determine what RNAs are expressed, and see if those specific receptors can be modulated by a select subset of compounds from the chemical library.”
Dr. Eberwine and colleagues recently engineered a transcriptome in vivo analysis (TIVA) tag, which allows mRNA from single cells to be captured in live tissues. By using this tool in conjunction with RNA sequencing, they were able to compare single-cell transcriptomes from cells grown in culture with those from in vivo mouse and human tissues. “As compared to cells that are dispersed, transcriptome profiling reveals different patterns of gene expression for a cell that is in its intact microenvironment, in contact with all its neighbors,” remarks Dr. Eberwine.
This was the first time mRNA was noninvasively captured from live single cells in their physiological microenvironment. One of the important implications of this finding is that the capacity of cells to respond to drugs when they are in isolation is different from when they are in their natural environment. “One of the potential problems with drug discovery is that drugs are currently screened primarily in cell cultures or in cell lines, and when they are not in their natural context, they express different mRNAs,” concludes Dr. Eberwine.
As a vibrant area strategically situated at the convergence of several disciplines, transcriptomics provides not only a level of scientific inquiry, but also a platform that fuels novel paradigm shifts. In transcriptomics, two particularly challenging (and potentially rewarding) goals have emerged: one is the interrogation of transcriptome dynamics in cell populations and at the single-cell level; the other is the integration of multiple layers of information.
As progress toward these goals is realized, a deeper knowledge of transcriptomics will stimulate technological development. And more sophisticated technology will enable researchers to enrich their understanding of transcriptomics. This symbiosis promises transformative times in cancer biology.
Assessing Genetic Variants with Bioinformatics Tools
“In collaboration with a number of other groups, we have been developing algorithmic approaches to predict genetic variants that are associated with some forms of genetic diseases,” says Sean D. Mooney, Ph.D., an associate professor at the Buck Institute for Research on Aging.
Tens or hundreds of thousands of inherited genetic variants that cause traits inherited in a Mendelian fashion have been described in every population, and many of these genetic variants were associated with malignant tumors. The molecular and functional effects of these genetic variants has been of abiding interest to Sean D. Mooney, Ph.D., an associate professor at the Buck Institute for Research on Aging.
“In collaboration with a number of other groups, we have been developing algorithmic approaches to predict genetic variants that are associated with some forms of genetic diseases,” says Dr. Mooney. Several years ago, Dr. Mooney and his colleagues developed, and later updated, a bioinformatics tool called MutDB. This tool, which is available online, integrates and annotates single nucleotide polymorphisms from Swiss-Prot and dbSNP with biochemically relevant information.
Recently, many studies have suggested that a large number of previously characterized genetic variants affect proteins not through amino acid substitution, but by affecting splicing enhancers or silencers and creating dysfunction in mRNA splicing regulation. To develop a bioinformatics tool that can differentiate mutations that change amino acids from those that affect splicing, Dr. Mooney and colleagues, in collaboration with Dr. David Cooper’s group from Cardiff University and Dr. Predrag Radivojac’s group from Indiana University, initially collected disease-associated variants that are thought to enhance or silence splicing.
“Matthew Mort, the lead author of the study, then built a supervised method to compare those variants to other variants that are most likely neutral and do not affect splicing and are probably not associated with function or disease,” explains Dr. Mooney. This led to the development of MutPred Splice, a machine-learning model that can classify novel variants that have not been previously associated with splicing diseases.
“The features that we used ranged from prediction of splicing sites, to conservation of sequence and amino acid preferences in a region of the exon,” comments Dr. Mooney. “Perhaps more than 15% of the mutations that cause inherited diseases in the human genome may be associated with splicing.”
“The hope is that someone doing exome or whole-genome sequencing could use this platform in conjunction with other methods to determine whether specific mutations are regulatory variants outside of genes or whether they are missense variants that are likely to cause amino acid substitutions in the protein, and prioritize disease-specific variants for other studies,” Dr. Mooney continues. In comparison with many other tools that are available, MutPred Splice predicts not only whether a genetic variant affects splicing, but also whether it will cause disease.