This chapter is from the book
This chapter is from the book
This chapter is from the book
THE TRANSCRIPTOME
If we were to take an inventory of all the messages involved in the creation of a single protein, we would need to include those that code for enzymes that catalyze post-translational modifications—methylases, acetylases, carboxylases, glycosylases, and other enzymes that remove amino acids from the amino or carboxy terminus. We would also need to account for the various regulatory messages, mentioned earlier, in addition to a large number of mRNA species that code for basic cellular processes. The messages present in the cell at any given time span a broad range of maturation; some represent early prespliced messages, and some have been spliced into final transcripts. These different versions would also need to be included in our analysis. If we were to continue this process for every transcribed coding region, we would eventually account for every message in the cell—both coding and regulatory. The complete inventory of transcribed elements within a cell has come to be known as the transcriptome.
Despite the fact that at any given moment the complete set of existing RNA transcripts represents only a small percentage of those coded for by the genome, the transcriptome is far more complex in diversity and structure. Figure 5-1 displays the various species that comprise the transcriptome.
)
Figure 5-1 Diagrammatic representation of the transcriptome, including various species of RNA: messenger RNA, miRNA, siRNA, prespliced
mRNA, and double-stranded RNA. Many processes contribute to the variety of RNA species present in a cell, and the production
of a single protein involves expression of several different genes, including some that code for enzymes involved in post-translational
modifications.
Unlike the genome, which remains mostly static throughout the life of a cell, the transcriptome varies tremendously over time and between cells that have the same genome. Moreover, various disease states have a dramatic impact on the gene-expression profiles of different cells. Tumor cells, for example, would naturally be expected to display characteristic gene-expression profiles that could ultimately act as signatures for various disease subtypes.
One of the most important goals of the emerging discipline known as information-based medicine is to use these profiles to create more precise patient stratifications. Many diseases that are currently classified according to their clinical symptoms (phenotype) will soon be defined by a combination of gene-expression profile and clinical symptoms, or gene-expression profile alone. Furthermore, expression profiles for many cell types will be helpful in establishing safe doses even after a disease is well characterized and the correct treatment is established. The ability to monitor the state of the transcriptome, both in health and disease, is paving the way for a new era in medicine. Remarkably, this new era will benefit from a mix of technologies that cross both basic science and clinical medicine. The tools of information-based medicine include disease-specific databases of expression profiles linked to clinical and demographic information in addition to sophisticated search and data-mining tools.
Mapping a Complete Transcriptome
Because the transcriptome displays tremendous variability, both between individuals of the same species and between different cells within a single individual, mapping a complete transcriptome is a tremendously complex undertaking—more ambitious, perhaps, than mapping a complete genome. Such a project has been launched. As in the genome mapping project, the effort is based on a technical collaboration; in this case the alliance is between Affymetrix, Inc. and the U.S. National Cancer Institute (NCI). The stated goal of the project is to build a complete transcriptional map of the genome at a resolution approaching individual nucleotides. The core infrastructure for the project is composed of DNA microarray technology from Affymetrix in conjunction with sophisticated gene-clustering algorithms and data-mining software. By marrying the results to information gleaned from whole genome sequencing and annotation projects, researchers will be able to gain a thorough understanding of all issues related to transcription, including structural modifications of the chromosomal material, identification of transcription factor binding sites, DNA modifications such as methylation, and origins of replication. Accomplishing these impressive goals will require interrogating the genome with millions of small, very closely spaced probes, which can reveal subtle differences within single coding regions. At the time of this writing, an important milestone in the project had recently been achieved: the release of a complete transcriptional map of human chromosomes 21 and 22 [[1]].
The results are illustrative of this discussion. Collectively, the two chromosomes contain 770 well-characterized and predicted genes. Coding and noncoding regions were interrogated using DNA microarrays containing small probes spaced approximately 35 base pairs apart across the entire chromosome. Surprisingly, it was discovered that the number of RNA species present in the cell that corresponded to sequences on these chromosomes exceeded the number of characterized and predicted coding regions by an order of magnitude. Approximately 50% of these positive probes were found to be located less than 300 base pairs distant from the nearest identified coding region. Table 5-1 summarizes the results.
Table 5-1. Relative Proportion of Probes Found in Coding and Noncoding Regions on Chromosomes 21 and 22
|
Cell Lines |
Total Positive Probes |
Positive Probes Found in Exons |
|---|---|---|
|
1 of 11 |
268,466 |
17,924 |
|
5 of 11 |
98,231 |
10,903 |
|
Source: Kapranov P., et. al. (2002) "Large-Scale Transcriptional Activity in Chromosomes 21 and 22." Science 296, 916–919 |
||
The fact that the total number of transcripts dramatically exceeds the number of protein-coding regions is consistent with the emerging view of gene expression that we have been describing. It is important to remember that the additional transcripts represent regulatory messages in various stages of splicing (including double-stranded RNA) in addition to prespliced protein-coding messages. In some sense, the problem can be simplified by defaulting to a completely mathematical representation that compares expression profiles in the form of images rather than sequences. The expression profile for a given metabolic state could be envisioned as a collection of mathematical entities that are up or down regulated in a reproducible pattern. After the pattern is discovered, a representation of the pattern linked to a phenotypic description of the state, disease, or clinical symptoms is stored in a database. Eventually such a database would be populated with millions of patterns and phenotypic descriptions. Medical researchers would then be able to search for matches to their patients' own profiles and clinical records with the goal of predicting the course of a disease or possible treatment outcomes. It is important to point out that a complete description of even a single pathway is complex, and a complete system-level model remains beyond the capabilities of today's information technologies.
Expression Profiling
Expression profiling has already yielded important results and academic research centers are beginning to populate large databases with specific patient information. At the time of this writing, a project of this nature had recently been announced between NuTec Sciences in Atlanta, the Winship Cancer Center associated with Emory University Medical School, and IBM in Armonk New York. Four specific areas of cancer were selected for study: breast, colorectal, lung, and prostate. Information gathered from expression profiling of various tissues and body fluids of affected patients is being analyzed in the context of clinical and demographic history. The goal of the project is to make treatment decisions based on the newly discovered correlations. Such projects can serve a multitude of purposes, including drug rescue for compounds that might fail clinical trial if tested on a patient population that is genetically too broad. Herceptin is a perfect example of a compound that is only effective for certain patient populations. Overexpression of the HER2/neu oncogene (also known as c-erbB2) is a frequent molecular event in many human cancers. The humanized anti-HER2/neu antibody, Herceptin has proven to be effective in patients with metastatic breast cancer who overexpress the HER2/neu oncogene. Herceptin could not pass a clinical trials test in a patient population that was not genetically defined [[2]].
The linkage between databases of expression data, target identification, and clinical trials is important because it has the potential to revolutionize the drug discovery and delivery process. The same databases that support target identification, a pure research activity, will certainly support clinical trials as well as medical diagnostics for clinical use. It is likely that diagnostics companies of the future will deploy the same databases as researchers and use them to make treatment recommendations. In the not-too-distant future, treatment decisions will be made through a process that combines initial genetic data (DNA sequence information) with mRNA expression data and information from an individual's medical record—the exact same information that will be used by researchers to identify targets and test the efficacy of new therapeutics. The trend toward molecular-based personalized medicine will, therefore, drive closer ties between each step of the drug discovery pipeline and clinical medicine.
Some of the RNA species contained within the stored patterns would represent immature unspliced versions of other transcripts also contained within the image. Others would represent regulatory messages that appear in the transcriptome because they modulate the translation of newly expressed messages. The expression levels of these associated messages, especially the regulatory ones, can be important data points. In each case, the balance between a regulatory or immature unspliced message and its corresponding protein-coding sequence has potential as an important diagnostic. For example, a highly up regulated message might not result in an increased level of expression for a particular protein if increased appearance of the message is matched with a high level of regulatory sequences that prevent translation. Likewise, an increase in the appearance of immature unspliced messages has true biological significance because it implies an expression ramp up that exceeds the cell's RNA-processing capabilities. After a recurring pattern is identified, chemical analysis can be used to unravel expression-level details for each coding region represented in the pattern [[3]]. New algorithms are also being used to identify statistical correlations between genes and clusters of genes to help identify related traits and further refine the profiling technique.
Gene-expression levels are time-dependent in a very sensitive way, and in many contexts it will become important to store transcriptional information as a function of time. An example might include drug response at a metabolic level for a specific population of cells. Medical science has historically relied on time-dependent assays in many areas (e.g., endocrinology), and time-dependent transcriptional analysis could certainly become part of the diagnostic arsenal.
As data-analysis tools become more sophisticated, researchers are beginning to take a systems approach to understanding the complex web of interactions that define cellular pathways. Recent approaches have included modeling a pathway where components were analyzed using DNA microarrays, quantitative proteomics, and databases of known physical interactions [[4]]. Transcription maps are central to such analysis. In one specific project conducted at the Institute for Systems Biology, a global model was constructed based on 20 systematic perturbations of a system containing 997 messenger RNAs from the yeast galactose-utilization pathway. The experiments provided evidence that 15 of 289 detected proteins are regulated post-transcriptionally, and the model that emerged identified explicit physical interactions governing the cellular response to each perturbation [[5]]. Such experiments are an important milestone in the history of drug discovery because they demonstrate that it is possible to develop and test complete systems models, which have the potential to rapidly advance the cause of predictive medicine. It is also important to note that the effort brings together contemporary tools of molecular biology and the latest information technologies—structure and pathway databases, expression array profiles, and data-mining/cluster-analysis software. The construction of a complete system-level description of a cell is the ultimate goal of systems biology. Such a description is considered a "grand challenge" problem in molecular biology. One prerequisite for completion of this project is a fully descriptive map of the transcriptome. Describing a single pathway is complex, and a complete system-level model remains beyond the capabilities of today's information technologies.
Because the transcriptome is composed of all messages within the cell—coding, noncoding, and immature—its content is highly dynamic. A complete map of the transcriptome is a collection of millions of expression profiles that represent snapshots of the cell as it responds to various stimuli over time. The map for an entire organism would be comprised of a similar collection of snapshots across thousands of different classes of cells. It is reasonable to assume that time-dependent multicell transcription profiling will ultimately form the basis of any complete analysis of gene expression—the ultimate goal being a complete system-level description of a cell. Efforts to build such maps have become a principal driver of the development of tools for transcriptional profiling.
In the next section, we will begin to examine the technologies behind transcriptional profiling. In addition to their role as a research tool, these technologies are rapidly becoming important diagnostic tools. As with any diagnostic tool, successful deployment depends on the availability of standards. Research efforts that create RNA profile maps for various cellular states are beginning to form the basis of this new diagnostic science.