Proteomic image analysis

While lying on a rock at the edge of a lake, you stare blankly at the sky and the myriad stars that play from horizon to horizon. As your eyes slowly adjust to the darkness, you catch the translucent wisp of light that flows across the sky. Slowly you begin to pick out the patterns made familiar by decades of skygazing. The large box with the three-starred handle, Ursa major (the big dipper). The five-starred zigzag that forms Andromeda. The four corner stars that taper midway to a three-starred belt, Orion.

Somewhere else, another lone figure sits staring blankly at a light box or video terminal, eyes adjusting tentatively to the glow. This time, it is points of black on white, not the other way around, that greet the viewer. And although his eyes do not seek Orion’s form in the spray of proteinaceous spots and smudges, the searcher does look for patterns in the display before him.

Proteomics and 2DE

Figure 1. The proteomic workhorse. The traditional approach to protein detection and identification involved the use of two-dimensional electrophoresis (2DE), in which proteins are separated by their isoelectric point and relative mass, followed by single-spot analysis via mass spectrometry.

Perhaps the strongest, if much maligned, workhorse of proteomic research is the analysis of protein expression profiles using two-dimensional electrophoresis (2DE). When 2DE is coupled with mass spectrometry (MS) for determining the identity of the individual protein spots that splay across the gel, no technology has done more to give researchers a foothold in identifying new drug targets or metabolic changes caused by illness (Figure 1).

In “Pathways to the proteome: From 2DE to HPLC”, in this issue of Modern Drug Discovery, Mark Lesney details some recent advances using 2DE,the technological limitations, and how researchers are mimicking the technology by switching from a solid polyacrylamide matrix to liquid chromatography. In brief, 2DE is theseparation of proteins extracted from a cell or tissue sample by placing an electric potential across a matrix of cross-linked polyacrylamide. The proteins separate according to their charge. This is the first dimension.

The gel is then laid across the top of a second polyacrylamide gel that denatures or unfolds the proteins in the presence of an ionic detergent. As the proteins move through the second gel, smaller proteins move faster than larger ones, and thus the protein population is separated on the basis of molecular weight. In some experiments, before the gels are run, the proteins are tagged with radioactive or fluorescent markers to allow later detection. In other cases, the proteins are chemically stained after separation. Either way, staining or marking the proteins results in a series of spots and smears that represent a portion of the proteins in the sample.

By examining extracts from tissues or cells at different stages of development, under different growth conditions, in the presence of drugs, or upon infection by a pathogen, researchers can look for changes in the spotting pattern of the cells. These changes include

• movements of individual spots, perhaps caused by alterations in the posttranslational modification(e.g., phosphorylation) of the protein, or
• the disappearance of a spot or the creation of a new one, signaling the shutting off or turning on of the expression of a given protein, respectively.

Gels do not last forever, and it is essential that the data from 2DE be stored in a format that is easily maintained.

Image analysis
Although there are innumerable methods for creating an image of a gel—includingflatbed scanners, fluoro- and phosphoimagers, and densitometers—the ability to use data relies on image analysis software packages, such as those listed in Table 1. The first generation of imaging systems, developed in the early 1980s,were decidedly user-unfriendly, relying on mainframe hardware and barely programmable software, and their operation was largely the purview of computer scientists. By the late 1980s, however, with the development of Unix operating systems and improvements in graphical user interface (GUI) technologies, later image analysis systems were more easily adapted to laboratories,and the variety of available software expanded.

But it has been in the past decade, with the expanded use of personal computers, further GUI refinements, and the emergence of the Web, that this software has hit its stride, becoming a useful tool for even the least-computer-savvy researcher. Image analysis software is used to log information, tell a robotic arm the location of a particular spot when excising it from a gel for further analysis by MS,or perform comparative analysis between gels that are generated in-house or available on Web-based databases.

The first step in gel analysis is to detect and quantify a single spot.At this point, the user must still establish background thresholds. This user interaction is especially critical when trying to identify spots of low intensity.Many programs then filter the image and raise the contrast to further highlight the smaller spots. A spot identifier isolates individual spots on the gel, typically by analyzing the intensities of individual and groups of pixels, and gives them codes.

In some cases, the program compares the location of a given spot for a given sample, an indication of the protein’s relative mass and isoelectric point, and discerns the true identity of the spot (e.g., H. sapiens metalloprotease-1) on the basis of a search of an internal or Web-based database. Large Scale Biology Corp. (Vacaville, CA) has developed two proprietary databases, called the Molecular Anatomy of Pathology and the Molecular Effects of Drugs. These databases contain a series of 2DE gels that describe the effects of disease processes and drugs on human protein profiles. Similarly, the Human Proteome Organisation (www.hupo.org) is developing the Human Protein Index, the protein version of the Human Genome Project.

Examples of Web-based databases are the one hosted by Julio E. Celis at the Danish Centre for Human Genome Research at the University of Aarhus (http://biobase.dk/cgi-bin/celis) and the Swiss Institute of Bioinformatics’ (Geneva) Swiss-2DPAGE (www.expasy.ch/ch2d/). The Danish project offers a database of 2DE gels for human as well as animal and plant systems, and looks at disease states such as skin disorders and cancer. The Swiss project offers a variety of information besides simply a gel database, including advice on setting up your own 2DE database.

Comparative analysis
If you only wanted to look at one gel, life would be simple. But proteomics is a comparative science. Thus, it is necessary to identify protein spot patterns between gels.

In matching gel images, the program performs automatically what the human eye does naturally: It looks for pairs of features that have the same or similar spatial distributions. If there is a doublet in the top right corner near a smeared singlet in gel 1, the program searches for a similar doublet–singlet combination in gel 2. This process is then repeated thousands of times until the program has accounted for all of the spots that it can identify. If it tries to compare several gels, one gel is established as the reference gel, and all spot combinations on the other gels are compared with it. One merely has to imagine the problem of quickly and accurately identifying and correlating 10,000 spots or more on two gels to see why computational advances have proven so critical to proteomics.

In examining the effect(s) of a drug on a cell or trying to diagnose a disease condition from a tissue sample, proteomic researchers compare what amount to before-and-after gels. To accommodate these comparisons, some programs have borrowed a tool used to explore the cosmos. To locate and identify comets and asteroids, astronomers compare two or more photographs of a specific region of space, perhaps taken hours or days apart. By flicking back and forth between the photos, the eye picks up subtle changes in the position of points of light, an indication that some heavenly orb is moving relative to the stars behind it. Many 2DE analysis packages do the same thing computationally, switching back and forth between the images of two gels such that individual spots move or appear anddisappear relative to the spots around them. The researcher can then highlight these changing spots manually on-screen. Alternatively, the program can perform this function automatically, which is useful when trying to compare gels generated under different conditions or in different locations.

Multidimensional mayhem
The real power of these programs is their abilityto overcome the technical glitches and human errors inherent in the use of 2DE. Anyone who has run a polyacrylamide gel is familiar with the problems of reproducibility caused by imperfections in the gel matrix.The slightest polymerization failure can lead to a severely warped gel, where, for example, all of the spots 5 cm from the top of the gel move to the right by 1 cm. This problem wreaks havoc with image analysis. Originally, the only hope of rescuing the image was to repeat the experiment until the appearance of the data improved. With analysis software, researchers can examine several gels of the same sample and average the locations of individual defects, ostensibly creating a “perfect” gel in the process. Alternatively, many of the newer programs allow for localized perturbations by keeping the larger picture in mind.

The one caution for this type of manipulation, however, is that perturbations from an event such as protein phosphorylation may be minor. Thus, it is critical that the “fudge factor” introduced by the program to accommodate gel imperfections not obliterate any actual differences between samples. This is where it becomes critical to consider any data manipulations with regard to the other spots on the gel. Ideally, when a cell reacts to a drug treatment, changes will occur in only a few spots and other protein expression levels will be maintained. One might compare these proteinaceous standard bearers (e.g., the housekeeping proteins that maintain a cell’s internal environment) to Polaris, the star around which the northern heavens rotate.

Given the human predilection for pattern recognition—our desire to “spot” the difference, as it were—it is not surprising that 2DE image analysis has come to play such a large part in improving our understanding of how we, as metabolic machines, function.

Further reading

• Appel, R. D.; et al. Electrophoresis 1997, 18, 2724–2734.
• Binz, P.-A.; et al. Anal. Chem. 1999, 71, 4981–4988.
• Celis, J. E.; et al. FEBS Lett. 2000, 480, 2–16.
• Cordwell, S. J.; Nouwens, A. S.; Walsh, B. J. Proteomics 2001, 1, 461–472.
• Pandley, A.; Mann, M. Nature 2000, 405, 837–846.
• Patton, W. F. Electrophoresis 2000, 21, 1123–1144.