Personal-computer-based vision software enhances electrophoresis
For decades, scientists have used electrophoresis?the migration of charged molecules in solution in response to an electric field?in biochemical research. But, with the development and application of image-analysis software tools, the technique is now opening up an entirefield of scientific and medical research.
By John Haystead, Contributing Editor
For decades, scientists have used electrophoresis—the migration of charged molecules in solution in response to an electric field—in biochemical research. But, with the development and application of image-analysis software tools, the technique is now opening up an entirefield of scientific and medical research.
"One reason for the recent interest in electrophoretic separation is the emerging field of "proteomics," says Andy Borthwick, senior applications scientist at Nonlinear Dynamics (Newcastle upon Tyne, England), a developer of software for electrophoresis. "Proteins are required for the structure, function, and regulation of the body's cells, tissues, and organs, and proteomics—the study of proteins and their biological functions—has become an important science in modern drug discovery."
Vision technology and image-analysis software are giving powerful new capabilities to a standard biotechnology research tool.
Proteomics researchers use electrophoresis to separate and identify proteins in a mixture. The rate of each protein molecule's electrophoretic migration depends on the strength of the field, net charge, size, and shape of the molecules and also on the ionic strength, viscosity, and temperature of the medium in which the molecules are moving. More advanced 2-D gel-electrophoresis techniques improve on this process by displacing the proteins twice (at right angles to one another) as opposed to a single dimension. Because the resulting samples are separated over a larger area, the resolution of each individual component, or spot on the gel, is increased.
Kendrick Laboratories (Madison, WI) is one service laboratory specializing in 2-D gel electrophoresis on proteins for a wide variety of medical, biotechnology, and pharmaceutical researchers. "While it's possible to perform individual 2-D gel analyses visually, the detection limits of the computerized system are far superior to the human eye," says Melody Kyper, senior biochemist at Kendrick. "If I can see a difference with the naked eye, it probably represents a ten- or hundredfold difference," she says.
FIGURE 1. Early interpretation of protein spot data between gels was limited to categorizing those spots that either showed significant changes in their amounts or were present or absent in one of the two gels being compared. As a result, analysis was at best basic and qualitative in nature. New software tools such as Phoretix 2-D gel-analysis system are now providing researchers with detailed qualitative data. A panel on the lefthand side of the Phoretix display screen, known as the navigator, serves as the main operating center for the software. Incorporating tools to adjust image parameters, the navigator remains on screen throughout the complete image-analysis process.
Kendrick uses a Molecular Dynamics (Sunnyvale, CA) Personal Densitometer SI (PDSI) 375 scanner to capture images for later analysis. The scanner accepts gels directly and also is used to scan x-ray films. Although the scanner generates 12-bit data images, Kendrick converts these to 8-bit data. The scanner is also capable of generating 50-µm pixel sizes, but Kyper says they prefer to use 100-µm resolution. "We find the 50-µm pixel images too big to handle efficiently, and 100 µm is more than adequate for image analysis."
According to Nancy Kendrick, president of Kendrick Laboratories, the Molecular Dynamics scanner was chosen because of the linearity of its laser optics. "For accurate quantitative measurements, it's critical that the signal of the scanner be directly proportional to the amount of stain on the gel." Kendrick says that although a number of diode-array scanners were considered for this application, they were not as linear as the laser scanner. Kendrick checks the linearity of its system once a month using optical-density (OD) disks of known values.
"One of the advantages of traditional line-scanning densitometers was that they were often built with exceedingly good optics and had dynamic ranges of up to 4 OD units," says Borthwick. "But for most electrophoresis applications, such a range is unnecessary because stains cause bands and spots to have an optical density of less than 1.5 OD units, and the darkening of x-ray film in autoradiography provides a range of 2 OD units for most films, with 3 OD units achieved rarely," he says. The Molecular Dynamics laser densitometer has an OD range greater than 3 OD units. The scanner is linked to the lab's Pentium 300/196MB computer via a SCSI interface. Images are then viewed on a Philips/Magnavox 107S 17-in. flat square, high-contrast CRT with 1024 x 768 resolution.
To perform image analysis, Kendrick uses Phoretix image-analysis tools from Nonlinear Dynamics. These provide quantitative results as opposed to strictly qualitative interpretations. Kyper also points out that while it's relatively easy to track simple experiments involving one test sample and one control sample, when an experiment involves a great many gels or images, computerized collection, analysis, and reporting tools are invaluable. "Without such tools, the level of meaningful data achievable would be limited."
In operation, Phoretix gel-analysis software automatically detects and identifies all of the spots on a gel, assigns coordinates and volume measurements, and rapidly and quantitatively compares different test samples against each other and against a control image (see Fig. 1).
The Phoretix software analyzes gray-scale tif images of the electrophoresis gels. These images may be 1- or 2-D electrophoresis samples, but the majority of proteomic applications involve 2-D gels. Stained gels are first passed through a scanner or densitometer or captured by a CCD camera capable of capturing the depth of intensity of each spot of stained protein. Images of radioactively labeled gels are captured on x-ray film. "Most scanners are capable of capturing images in 8-, 12- or 16-bit gray scales, and the software can handle all of these formats," says Steve Mallam, software team leader of the 2-D project. The Phoretix software includes interfaces to the various capture devices for setting up and calibrating image-intensity collection parameters.
The Phoretix Windows-based software is written in Microsoft Visual C++. Mallam also uses Microsoft Visual SourceSafe for source-code control and Visual Intercept from Elsinore Technologies (Raleigh, NC) for defect tracking.
Though the Phoretix tool doesn't implement third-party image-processing packages, it does incorporate several third-party program-control libraries such as the Rouge Wave/Stingray Objective Grid and Objective Toolkit libraries from Rogue Wave Software (Boulder, CO). Objective Grid allows developers to bind to any data source, including ODBC, DAO, and ADO. Supporting Microsoft Excel-like formulas, the object-oriented cell architecture allows users to add any custom cell type or embed an OCX or ActiveX control. The Objective Toolkit provides simple GUI controls, customizable toolbars and menus, docking windows, image classes, MDI alternatives, shortcut bar classes, tabbed controls and windows, tree control and tree view, user interface extensions, view classes, and utility classes.
Although the program runs on standard personal computers with a minimum Pentium 200-MHz processor and 32 Mbytes of RAM, Mallam recommends at least 64 Mbytes on machines running Windows 98 or 2000. "Obviously, users working with 16-bit gray-scale images will require a higher-throughput machine for the same level of processing efficiency as those working with lower-resolution images."
The first step in image analysis is to define an area of interest in the image that suitably reflects the requirements of the particular experiment. Usually researchers are interested in only a small portion of the total image, and as the raw image data are not modified by the software, with all processing and analysis handled as attached or associated files, original images can be analyzed multiple times and in different fashions. In all cases, users want to exclude edges, because most "noise" associated with electrophoretic imaging will appear near the edges of the gel where dye has collected.
FIGURE 2. The spot-detection wizard allows users to easily select the best sensitivity and spot-detection parameters for their particular gel images and experimental requirements.
To meet the needs of both experienced and relatively novice users of the tool set, the Phoretix software includes a "spot-detection wizard" (see Fig. 2). The wizard presents a nine-panel window from which you can visually select the best approach to defining spots according to spot size, operator size (size of the overlay placed around a detected spot), acceptable noise, and background sensitivity.
After all the spots of interest have been automatically detected, users can further manually edit the images, such as splitting apart spots that are very close together. Alternatively, they can adjust the threshold by which the software automatically performs this function. In some cases, however, users prefer to establish higher-than-necessary sensitivity-detection parameters and then manually remove any unwanted spots that appear.
The next step in the process is background substraction, which automatically applies a threshold across the entire image to compensate for any ununiform staining. Because there are a number of modes available for doing this, the user can view and compare the results of each approach on the screen (see Fig. 3).
Although there are many approaches to matching images from different electrophoresis samples, the most common starting point is to select one gel to be used as a base reference or control. Phoretix compiles an index for all of the spots on this control image, which is then used to automatically correlate the spots on each subsequent image. The algorithms that perform this matching first identify a common focal point in the two images and then work outward until the images are fully correlated.
FIGURE 3. Background subtraction is one of the most crucial steps in accurately quantifying spot material. Phoretix includes three automatic methods and one manual method, all of which treat each spot independently to account for variations in background across the gel. Verification of the background subtraction can be done by using the background only image display (left), which removes all spot material and provides an instant visual overview of how well the background subtraction has been done.
The software can be set to perform this image matching automatically, or the user can assist the software by manually identifying the same spot on two images as a common reference point. The user overlays the reference gel spots and links various identifiable spots on the distorted gel to similar spots on the reference gel. As each of these "seeds" is added, regional warping of the overlay aligns the two images. Once satisfied with the positioning of the seeds, the warping is removed and the pattern-matching process proceeds. The process doesn't alter either the reference or the image being matched to it.
Once image matching is complete, multiple postanalysis techniques can be applied to the data to suit the users specific needs and requirements. These include calibration of images according to molecular weight, charge separation, and differential analysis of multiple gels. For example, some users may want to create an average image to represent all of the gels collected from an experiment. Users can change the parameters defining these average images by adjusting the amount of experimental error or standard deviation accepted in each spot, making the representative groups as tight or as broad as they wish. All of the postanalysis application packages are selectable from on-screen menus.
Melody Kyper, senior biochemist at Kendrick Laboratories, performs PC-based analysis of gels scanned using a Personal Densitometer SI (PDSI) 375 from Molecular Dynamics.
Raleigh, NC 27616
Madison, WI 53713
Redmond, WA 98052
Sunnyvale, CA 94086-4520
Newcastle upon Tyne,
NE1 3JA England
North America Headquarters
New York, NY 10020
Rogue Wave Software
Boulder, CO 80301