TECHNOLOGIES FOR TRANSCRIPTIONAL PROFILING
As previously discussed, the development of a complete molecular-level picture of health and disease involves understanding metabolic processes at four distinct levels: genome (DNA sequence), transcriptome (mRNA profile), proteome (protein structure and interactions), and metabolic pathway. The research toolkit required to build this picture has evolved significantly over the past several years and currently includes a variety of technologies that complement each other:
DNA sequence information is directly relevant to the development of genetic tests for illnesses that exhibit a pattern of inheritance. DNA sequence information also provides a context for understanding polygenic diseases and complex phenotypes. Complexities associated with the splicing process make it important to interpret DNA sequence information in the context of phenotypea single mutation can affect many proteins to create a complex phenotypic change.
Messenger RNA profiles are important diagnostics that can delineate a pattern of up and down regulation of a large number of related genes. This information can be used experimentally to help drive the target-discovery process or as a clinical diagnostic to predict the onset or progression of disease. It is important to note that many messages serve a regulatory function and are not translated into protein.
Proteomics is a discipline focused on understanding the structures and interactions of the millions of proteins encoded in the genome. Final protein structures are normally achieved only after post-translational modifications (acetylation, glycosylation, methylation, removal of n-terminal amino acids, etc.); it is not currently possible to predict these modifications from DNA or mRNA sequence information. However, computational modeling of protein structure can be a useful tool for predicting potential sites for post-translational modification, as well as the extent of change to protein folding that is likely to result from those modifications.
Systems biology is an attempt to study the complex and subtle relationships between the millions of proteins that make up an organism. Systems experiments often involve perturbing a system at the mRNA level and measuring the complex downstream results. Modeling of biological systems is a complex bioinformatic challenge that has recently become feasible because of rapid advances in information technology [].
Messenger RNA profiling technologies are emerging as standard tools in all the areas mentioned above, both for research and clinical use. On the research side, transcription profiling has five major applications:
Clarification of details surrounding the splicing process for sequenced coding regions
Identification of key messages for previously unidentified proteins (Many of these protein are minor messages present in small copy numbers and previously unidentified.)
Delineation of metabolic pathways through "perturbation" experiments
Identification of regulatory sequences that are not translated into protein
Analysis of molecular-level responses to various stimulii.e., stress, drugs, hormone signals, genetic mutations, etc.
On the clinical side, transcription profiling is beginning to play a role in identifying patients for clinical trial, predicting the onset of disease, and customizing treatment regimens for individual patients. Central to these advances are pattern-matching technologies that facilitate rapid large-scale database-similarity searches. The task is similar to that of scanning a database containing millions of fingerprint records for patterns that are similar to a reference fingerprint. In the vast majority of situations, in which thousands of genes are involved, it is necessary, for a given set of expressed genes, to correlate up and down regulation patterns with treatment outcomes as measured by phenotypic changes. Over the next several years, such techniques are likely to become core components of clinical medicine.
Over the past several years, it has become apparent that the "one gene at a time" approach to understanding complex metabolic events is simply not adequate. Some estimates indicate that as many as 10% of the 10,000 to 20,000 mRNA species in a typical mammalian cell are differentially expressed between cancer and normal tissues. As a result, several technologies have been developed for quantifying the expression of many genes in parallel. One of the many technologies, the DNA hybridization array, has become dominant because of its low cost and flexibility. (DNA hybridization arrays are also referred to as microarrays, expression arrays, and gene chips.)
Expressed Sequence Tags (ESTs)
Identification of the expressed messages within a particular population of cells has always been an important research goal for cell and molecular biologists. Prior to the advent of high-throughput techniques such as microarrays, researchers typically purified and sequenced complementary DNA (cDNA) for this purpose. Because it is reversely transcribed directly from mRNA, cDNA represents a direct source of spliced and coding sequences of genes. Therefore, sequencing of cDNA has become a well-established and accepted technique that is complementary to genome sequencing efforts in many important ways.
In 1991, the first application of high-throughput sequencing of cDNA clones was described, where clones from human brain were randomly selected and partially sequenced. These partial sequences, which represented genes expressed in the tissue at key time points, were referred to as expressed sequence tags or ESTs. Similar approaches for different tissues and organisms were soon to follow. ESTs provide the largest amount of information possible per sequenced base. The vision of rapid identification of all human genes has led to the development of several commercially financed data banks containing EST sequences [].
EST sequences taken from a random cDNA pool can also yield a statistical picture of the level and complexity of gene expression for the sample tissue. The influence of environmental factors and tissue-specific gene expression can therefore be studied. Furthermore, the gene sequences obtained can be efficiently used for physical mapping by determining their chromosomal position. They also contribute to an understanding of intron and exon boundaries and are often used to predict the transcribed regions of genomic sequences.
Because microarray analysis is highly parallel in nature, typical experiments produce thousands, sometimes millions, of data points. Microarrays are often referred to as gene chips because they are built on technologies adapted from the semiconductor industryphotolithography and solid-phase chemistry. Each array contains densely packed oligonucleotide probes whose sequences are chosen to maximize sensitivity and specificity, allowing consistent discrimination between closely related target sequences. A typical pharmaceutical microarray experiment involves the following steps [, ]:
A microarray is constructed (or purchased) containing thousands of single-stranded gene fragments, including known or predicted splice variants and potential polymorphisms. The sequences are selected to support a large number of cross comparisons to confirm complex results.
mRNA is harvested from selected cells in treated and untreated individuals. (Untreated samples will be used as an internal control in the array.)
mRNA is reverse transcribed into more stable cDNA with the addition of fluo rescent labels; green for cDNA derived from treated cells, and red for cDNA derived from untreated cells. (The labels are composed of 5-amino-propargyl-2'-deoxyuridine 5'-triphosphate coupled to either Cy3 or Cy5 fluorescent dyes: Cy3-dUTP or Cy5-dUTP).
Samples of fluorescently labeled cDNA are applied to the array and exposed to every spot. A sequence match results in binding between the cDNA test sequence and a complementary DNA sequence on the array. Each match contains a double-stranded fluorescently labeled spot that results from the combination of the two fluorescent dyes and the amount of dye containing cDNA in each of the samples.
A laser fluorescent scanner is used to detect the hybridization signals from both fluorophores, and the resulting pattern of colored spots is stored in a database: red for strongly expressed genes in the treated sample, green for weakly expressed genes in the treated sample, yellow for genes that are equally expressed in both samples, black for sequences that are not expressed in either sample. Because the sequence of every spot in the chip is known, the identity of each expressed cDNA can be determined, and the amount and source (treated or untreated sample) inferred from the color and intensity of the spot.
Differences in intensity correspond both to expression levels of the genes in the sample and to exactness of the match. Similar sequences containing various combinations of single and multiple base changes are used as internal controls to provide more precise sequence information about genes expressed in the test sample. For example, the identity of a single base can be deduced by measuring the binding affinity of a test sequence to four slightly different probes that vary only at the position of the base in question. (Each contains one of the four possible bases.) Figure 5-2 outlines the steps involved in a typical two-color microarray experiment.
Figure 5-2 Diagrammatic representation of a typical two-color microarray experiment. DNA derived from treated cells is labeled with a green fluorescent marker. DNA from control cells is labeled with a red fluorescent marker. Both samples are applied to every cell in the array. Complementary sequences bind to the array forming fluorescent spots, which are detected using a laser scanner. The color and intensity of each spot reveals information about the relative strength of the match between each test sequence and the sequence stored in the array.
Microarray production has evolved from a low-density technology based on robotically placed spots to a high-density technology based on the same masking and photolithography technologies used by the computer industry in the manufacture of microprocessors. Consequently, there is little need to conserve on the number of genes to be scanned in a single experiment. For example, the Affymetrix Human Genome U95 (HG-U95) set consists of five microarrays, the first of which contains nearly 63,000 probe sets interrogating approximately 54,000 clusters derived from the UniGene database. These probes are estimated to represent approximately 10,000 full-length genes. As costs continue to decline, high-density oligonucleotide arrays will become the standard tool for comprehensive transcriptional profiling.
Efforts to directly compare samples are almost always complicated by the fact that each profiling experiment requires a separate array. Results must therefore be normalized in order to make meaningful interarray comparisons. Additionally, fabrication techniques limit the length of each probe to approximately 25 bases. To adequately represent each gene, Affymetrix microarrays typically include a set of 16 to 25 25-mers that uniquely represent each coding region and can be expected to hybridize under the same general conditions. Every set of perfect-match (PM) probes for an mRNA sequence has a corresponding set of mismatch (MM) probes. An MM probe is constructed from the same nucleotide sequence as its PM counterpart, with the exception that one base (usually number 13 in the sequence) has been replaced to cause a mismatch. The combination of a PM and its MM counterpart is referred to as a probe pair. There are two reasons for including probe pairs for each sequence. First, at very low concentrations (low copy counts) of the target, the PM probes are operating near the lower limit of their sensitivity, and MM probes display a higher sensitivity to changes in concentration. Second, MM probes are thought to bind to nonspecific sequences with the same affinity as PM probes. This property allows MM probes to serve as internal controls for background nonspecific hybridization.
One drawback of microarray analysis is related to its inability to distinguish very low-abundance transcripts, those present in single-digit copy counts where the copy range is very large. A more common problem involves the quantification of mRNA species from a large population of cells where the species of interest is present only in a small subpopulation. In such situations, the species of interest is likely to be diluted beyond the detection limit by more abundant transcripts appearing across the broader population. Increasing the absolute amount of the hybridized target is not usually helpful because it is the relative abundance of each transcript within the RNA pool, coupled with probe characteristics, that determines the sensitivity of the array for each sequence []. Unfortunately, it is also difficult to identify transcripts that are up regulated by less than 50%, a significant problem for researchers in areas such as oncology and neuroscience where subtle changes in gene expression are critical to understanding the differences between disease and health.
However, the downstream effects of subtle changes in gene regulation are often more dramatic and straightforward to measure. Many of the minor messages that are difficult to detect are likely to be regulatory genes, and microarray analysis can still reveal the more pronounced levels of up and down regulation associated with more downstream members of these gene pathwaysthe messages most likely to code for druggable targets. Additionally, several new techniques are being developed to solve the minor message problem with the goal of detecting and precisely counting the number of molecules of every transcript in the cell, including those with copy counts in the single-digit range. The next section discusses other RNA quantification techniques.
The major goal of microarray data analysis is to identify statistically significant differences between genes expressed in the control and test samples. One of the most straightforward and commonly used analysis techniques involves construction of a simple scatterplot where each point represents the expression level of a specific gene in two samples, one assigned to the x-axis and one to the y-axis. For each point, its position relative to the main diagonal (the identity line) directly relates the ratio of expression levels for the test and control sequences. Messages with identical expression levels in both samples appear on the identity line, whereas differentially expressed sequences appear at some point above or below the diagonal as determined by the level of expression on each axis. Points that appear above the diagonal are overexpressed in the sample represented by the y-axis; conversely, points that appear below the diagonal are overexpressed in the sample represented on the x-axis. The overall expression level for any gene can be determined by measuring the absolute distance from the origin. Figure 5-3 depicts a microarray scatterplot.
Figure 5-3 Microarray scatterplot for two samples. For each point, the position relative to the main diagonal indicates the relative expression levels for test and control sequences. Points that appear above the diagonal are overexpressed in the sample plotted along the y-axis. Likewise, points that appear below the diagonal represent genes that are overexpressed in the sample plotted on the x-axis. Messages that are identically expressed in both samples are plotted directly on the diagonal. The overall expression level for any gene is determined by measuring the absolute distance from the origin. (Adapted from: Jeffrey Augen, "Bioinformatics and Data Mining in Support of Drug Discover," Handbook of Anticancer Drug Development. D. Budman, A. Calvert, E. Rowinsky, eds. Lippincott Williams and Wilkins. 2003)