A Simplified Method for Computer Analysis of Autoradiograms from Two-dimensional Gels*

,

A simple method is described for computer analysis of a discrete number of spots on autoradiograms from two-dimensional gels. The method involves digitizing the density data on an autoradiograph with a rotating drum densitometer and displaying the data on a graphics computer terminal. The software allows the operator to select the boundaries of the spots to be analyzed from the terminal, then integrates the density of the spots and tabulates the data. Graphics options allow the operator to display a computer-generated image of the area of the film being analyzed. Accurate integration of weak or overlapping spots is accomplished by a nonlinear least squares fit of the density data to normal Gaussian curves in the x and y dimensions followed by analytical integration of the equations. Since the software is written in Fortran IV and the equipment required to run the programs is available in most computing centers, this technique should allow laboratories of modest resources to quantitate information from two-dimensional gels.
The technique of two-dimensional electrophoresis followed by autoradiography is often used to resolve and display complex mixtures of proteins in homogenates or subcellular fractions (1-3). It is widely appreciated that the full power of this technique will be reached only when the gel systems are standardized and computer analysis of the autoradiograms is used to compile standard coordinates and intensities for a large number of proteins (4). At least four laboratories have directed their efforts to this en.d and have reported methods for computerized analysis of the entire autoradiogram (3,5-7). Other workers have described computer programs for storing, transforming, or displaying the coordinates of proteins in the two-dimensional array (8,9).
For some applications, each of the published approaches to data analysis presents shortcomings. In our work, two-dimensional electrophoresis is used to resolve the proteins whose phosphorylation state is altered following treatment of ["PI P043--labeled, intact hepatocytes with hormones. Experience has shown that there are only 13 such proteins. While the computer software developed by Bossinger et at!. (5) or Garrels (3) would allow quantitation of the phosphorylation changes, in practice these programs are unnecessarily complex and costly in terms of computer time, required hardware, and ancillary services. For this reason, we have developed a simplified method for computer analysis of a small number of spots on autoradiograms from two-dimensional gels. The program displays a computer-generated image of any portion of an autoradiogram on the screen of a graphics computer terminal and allows the operator to select the regions of the xray film to be analyzed. This report describes the software and points out its advantages for quantifying the intensities of a small number of spots in a two-dimensional autoradiogram.
The program is written in Fortran IV and is designed to run on equipment usually available in a university computing center. For these reasons, the method should be of general utility to laboratories that use two-dimensional electrophoresis to separate a few proteins of interest from complex mixtures.

MATERIALS AND METHODS AND RESULTS'
Computer Display of the Autoradiogruph-PROG-2 is used to locate, integrate, and generate an image of the density data from the autoradiogram. Its commands and their functions are listed in Table 1 (see "Materials and Methods" in the Miniprint). The sections below give examples of the use of the important commands in PROG-2 with actual data. The first step in the quantitative analysis of an autoradiograph is to locate the coordinates of the data to be integrated. This task is accomplished by having the computer display the entire autoradiograph on the Tektronix graphics terminal and then selecting the appropriate x and y coordinates needed for higher resolution images. The process of defining and resolving a series of spots is presented in Fig. 1, A-D. Fig. 1A presents an autoradiograph prepared by a 4-day exposure of Dupont Chronex 4 film to 32P-labeled cytosolic proteins from glucagon-treated hepatocytes (12). The boxed area encompasses four major phosphoproteins ranging in M, = 80,000 to 56,000 (top to bottom) and isoelectric point from about 6.0 to 6.5 (left to right). Fig. 1B presents a computer generated image of the boxed section obtained by selecting x coordinates of 359-516 and y coordinates of 289-386 with the SPOT command. For this figure, the LEVELS command was used to set the plot level at 0.04 absorbance above the film background of 0.35 absorbance. Only dots were plotted in order to obtain a rapid display (10-13 s). Each dot in this image represents the optical density reading of a 0.2-mm square section of the x-ray film (pixel) with a density of 0.39 absor- Computer imaging of an autoradiograph. A, a twodimensional autoradiograph made from '"P-labeled cytosolic proteins isolated from glucagon-treated hepatocytes. The molecular weight dimension is vertical and the isoelectric focusing dimension is horizontal with the acidic end on the left. B, a computer-generated image of the boxed section of the autoradiograph. Each dot represents a 0.2mm square area of film (pixel). The computer plots dots for those pixels with optical densities 0.04 or more above film background of bance or greater. Pixels with densities below 0.39 absorbance are not plotted on the screen. Fig. 1C shows the use of the LEVELS command to enhance resolution and contrast of the same film area. To obtain this image, the plot level was raised to 0.09 absorbance above film background with dots plotted for pixels corresponding to densities between 0.44-0.63 absorbance, crosses plotted for pixels with densities between 0.63-1.02 absorbance, and stars plotted for pixels with densities above 1.02 absorbance. Proper use of the LEVELS command produces an image closely resembling the original autoradiograph and rapidly provides a three-dimensional impression of the density contours on a two-dimensional screen. Substantial image enhancement of weak spots can also be obtained with this command.  y coordinates were narrowed around Spot 3 and slightly different contouring was provided using the LEVELS command, as outlined above. Note how the crosses and stars add a definite impression of density contour. Integration of the density of the film is automatically provided for areas bounded by the square and the ellipse each time the SPOT command is used. The LEVELS command only affects the graphic output of the SPOT command; LEV-ELS does not affect the integration routines. Values printed with each image include: the square integral, its baseline, and the number of pixels summed; the elliptical integral, its baseline, and the number of pixels summed; the plot levels and the x, y coordinates (these details are not shown in Fig. 1). For Spot 3, these values were: square integral = 11,075 with a baseline of 0.35 absorbance and number of pixels = 1512; elliptical integral = 10,709 with a baseline of 0.35 absorbance and number of pixels = 1116. For well separated spots, the values of the square and elliptical integrals usually agree to within 0-5%; however, the elliptical integral more closely approximates the true shape of the spot. Moreover, if necessary, the density contributions of closely adjacent spots can be minimized by selection of the x and y coordinates to cause the neighboring spots to fall outside the area of the ellipse. For strong, well separated spots, the above method of integration provides acceptable accuracy. However, weak spots and closely adjacent spots require the more complex methods of analysis available with PROG-3 (see below).
At this point, Spot 3 has been located, its image displayed, and its density values integrated. The operator may now save the information in one of three ways by: ( a ) making a permanent copy of the image and integral information displayed on the screen using the Tektronix 4631 Hardcopy Unit; ( b ) storing the square and elliptical integral values, their baselines, and the number of pixels integrated in a running table with the SAVE TABLE command; or ( c ) transferring the entire matrix of optical density data bounded by the x and y coordinates to a new disk f i e with the SAVE command. The new file can then be retrieved and analyzed using the more complex techniques available with PROG-3 (see below). Options ( b ) and ( c ) allow the operator to identify the stored data with numbers and names, respectively. Any combination of the three options can be chosen for each spot.
Use of Complex Analysis to Integrate the Density of Weak or Overlapping Spots-The above discussion presents a straightforward method of integrating spot densities. For strong spots (integrated densities greater than 200), the routines available with PROG-2 are quite accurate. Most spots on a properly developed autoradiogram fall into this category. However, weak spots (integrated densities of 100-200) and closely adjacent spots (strong or weak) cannot be accurately analyzed by PROG-2. Weak spots are difficult to integrate with PROGS because small changes in the fiim background subtracted from the integral will markedly affect the final value. (See Figs. 3 Table 3 in the Miniprint for a description of this problem.) Overlapping spots are not accurately integrated with PROG-2 because the algorithms cannot discriminate between the density contributions of adjacent spots. A third program, PROG-3, has been developed to integrate weak or overlapping spots accurately. This program uses nonlinear least squares curve fitting techniques to generate normal Gaussian curves describing the density distribution of a spot in both the x and y dimensions and then analytically integrates the equations. (See "Materials and Methods" in the Miniprint for a full description of this pro-  Fig. 2A shows the image of Spot 4 generated with PROG-2. The integrated density of the elliptical integral from this analysis is 1301. The data in the area of the Tim bounded by the square is stored with the SAVE command, retrieved by PROG-3, and the density of the spot integrated. The value obtained is 1321, a result within 2% of that obtained with PROG-2. Fig. 2B shows the contour plot (one of PROG-3's plot routines) of the Gaussian curves fit to the data. The three ellipses represent constant optical densities of 0.40,0.54, and 0.86, outer to inner, respectively.  Table 3 in the Miniprint for a more complete description of this function of PROG-3.) The algorithms of PROG-3 can separate the density contributions from overlapping spots because they assume that the density distribution of a given spot follows a single Gaussian distribution in the x dimension and a separate single Gaussian distribution in they dimension. Thus, the algorithms will fit Gaussian curves to only one of a set of overlapping spots. The operator can direct the program to the spot of choice by proper selection of the spot boundaries when the data is stored for PROG-3 with the SAVE command. When the program is run, the algorithms ignore missing data or extraneous data from overlapping spots and provide the complete integral for a single Gaussian-shaped spot. Supporting data for this function of PROG-3 and other details about its uses can be found in the miniprint supplement.

DISCUSSION
While the technique of two-dimensional electrophoresis is widely used to resolve complex mixtures of proteins, quantitative analysis of the patterns obtained is not common. Among the main reasons for this situation is the large technical and financial investment needed for automatic analysis of the data in autoradiograms of two-dimensional gels (3)(4)(5)(6)(7). While this advanced technology is clearly required to be able to analyze and compare hundreds or thousands of spots, many laboratories use two-dimensional gels to separate a few proteins of interest from complex mixtures. The computer approaches described in this report offer many advantages that allow rapid, accurate quantitation of 10-50 spots in an autoradiogram. These advantages include simplicity, versatility, and low cost, yet the computer programs allow use of the full power of the two-dimensional technique. The obvious disadvantage of the approach is that the analysis is not automatic; operator intervention is required to select the area of the film to be analyzed. However, operator intervention also provides versatility that allows the program to be used for other purposes.
The main advantage of the software is that it is written in Fortran IV and is designed to run on a university computing center's main frame computer (a Control Data Corporation Cyber 730 in this case). Thus, the hardware required (disk drives, tape drives, central processor, and the graphics terminal) are owned and maintained by the computer center. The speed, memory capacity, and versatility of this type of system far exceed those of the mini-computer systems often used for analyzing two-dimensional gels (5,7). Moreover, this type of facility should be available in most university or research settings for a nominal usage fee. The high speed densitometer needed to scan the autoradiographs is perhaps less widely available. However, since the output of this unit can be stored on magnetic tape, it is quite practical to travel to an available, off-site densitometer and scan many autoradiographs in one session. The tapes can then be analyzed on-site over a longer period of time. It should be noted that, while the software described in this report was developed using one particular hardware configuration, the main advantages of the method are independent of the computer equipment used. If the three basic items of hardware are available, the programs should allow laboratories of modest resources to use the full power of the two-dimensional technique.
A second advantage of the software described in this report is that the algorithms used to integrate the density of the spots can be tailored to meet the complexity of the integration. For well separated spots of moderate to high density, the simple summing routines of PROG-2 provide rapid, accurate integration of the spot densities. Experience has shown that this mode of integration, combined with appropriate use of the BASE VALUE command to hold the subtracted background constant, can be used to integrate 80-90% of the spots in an autoradiogram. This mode of operation is very efficient because it uses little central processor time. Moreover, since the boundaries of the spots are selected interactively by the operator, the computer does not have to match or align the coordinates of spots in autoradiograms from different gels to compare two experiments. Thus, reproducibility of gel systems does not have to be as stringent as in totally computerized systems.
Weak or overlapping spots that require more complex analysis than is available with the SPOT command may be analyzed with PROG-3. This program contains algorithms that define the density distribution of a spot as Gaussian in the x and y dimensions and simultaneously estimate the appropriate background. In addition, since the equations do not require an entire spot for integration of density, PROG-3 is very useful for accurately integrating weak and/or overlapping spots (see "Results" for details). While the routines for PROG-3 could be used with all spots, they provide no improvement in accuracy for well separated spots of moderate to high density. Since PROG-3 requires more central processor time than PROG-2, its use is best reserved for weak or overlapping spots. The combined use of PROG-2 and PROG-3 can quickly and accurately process 10-50 spots of all types on a typical autoradiogram.
The final advantage of the software presented in this report is its versatility. Since the SPOT command of PROG-2 makes no assumptions about the area selected to be integrated, it can be used to integrate densitometric data from sources other than two-dimensional gels. As noted in the Miniprint, PROG-2 has been used to integrate direct po_sitive transparencies made from the stained proteins on two-dimensional gels, long thin areas (1 X 15 mm) in one-dimensional gels used in nucleic acid research and autoradiographs from two-dimensional peptide maps made on thin layer cellulose plates. Other uses are certainly possible.
The authors will provide documented copies of the program to interested parties on request.
Required Hardware: The ComputeT p r o g r a m dercrzbed for analyzing a u t oradlogramr from two-dimensional gels require access t o three a'a)or pieces of equlpmenr 1) a hlgh speed densitometer capable of scanning and recording the optxsl density of the Y-ray flln at raster sizes of 200 LI OT maller; 2) a large mainframe computer; 3) B graphlcs conpurer terminal. The actual equipment used ~n rhlr work included' 1) an O p r r o n~r P-1000 densltomter and its integral rape drlve; 21 the Control Data Corporation Cyber 730 CODputer a t the University of V i r g m~a Academic Computing Center; 3) a Tekfronix 4006 graphlcr terminal linked to a Tektronix 4611 Hardcopy Unit. The Hardcopy Unlt is n o t actually necessary but greatly facilitates storing the results of t h e analys~s. I t should be noted that neither the particular models of film scanner. computer O T graphlcs terminal nor the mode of data transfer a r e inm r f a n f . The S O~~Y~T F descrxbed could be adanted to run on a Variety of

RESULTS
developed to quantitate the effect of hormones on the phosphorylation of pp;g3in intact hepatocytes whose ATP pools had been pre-labeled with these ~$11: can be visualized via autoradiography of the q2P-labeled proteins used to integrate the density inforration of a piece of X-ray fily5produc d resolved in an scrylamide gel system (1s). However. the software can be by any source. including X-rays or other B emitters such as IC.

or ' H.
Validation of the Method: The programs described ~n this report were within limits. changer in the phorphorylario state of proteins in are consistency and linearity of the response of the X-ray film to changes in radioactivity. To test the linearity of some X-ray film to increasing amounts qf isotope eight nliquotr of 5 yl of H 0 containing 0-3400 cpm of Cion spreading) and exposed to 4 types of X-rmy film for 20 hours. scanning an autoradiograph 3 consecutive C~F S and analyzing the some sir spats in each scan. The six spots were chosen to represent P IS fold variation total integrated density. The differences in integrated density between replicate r c m r varied between 0.07-0.58 of the total spot density with an average of 0.38 ("-18).
The reproducibility of the OPtronlcs P-lo00 x a n n e r has been checked by provide I typical example of the use of PROG-2 to analyze the effect of Table 2 uses the densities of the spots labeled 1 -4 in Figure 1C to hormones on protein phosphorylation. The integrated densities were obtained for each spot using the SPOT cornand. The data was tabulated with the SAVE Spot 2 is not changed by treatment of the cell with horroner. There are many such protems in a typical autoradiographic pattern providing inportant internal controls to demonstrate that: contr01 and hormone treated cells are labeled to equal specific activities; equal amounts of protein are loaded Onto the two-dimensional gel system; and film and scanner responses are constant in each experiment. It is important to analyze several such proteins ~n each erperment to provide these arsurancer.

Spot Density Control
Spot Density 100 nu Gluclgon G1"crgo"lCo"trol Anal si5 of Weak S O~S ' If st all possible the analysxs of weak Spots should beyavoided by alFerinp the pTOteln loads.'the speclflc aCtlYltleS Of radioactive proteins, or the exposure timer to intensify weak S y t s on the autoradiographs. However, because of the short half life of [ PI some weak spots will probably require accurate analyszr. This slfuafion OCEYTS because low levels o f isotope decay before they darken The hlgher resolution X-ray analyrlr of weak spots. Flgure 3 shows an autoradiograph of P~labcled films. The followng sections demonstrate t h e utility of PROgiS in the proteins obralned from unstmulated hepatocytes. The boxed area in Figure 3A encornparres SIX phosphoproteins that valy 2 4 fold 1" >"regrated densify. Fzgvre 3B presents the computer-generated lnagc of the area. This ~e g l a n of film was chosen because Spots 1-4 represent a broad range of po$s~ble conflguraflons r a n g m g from a large strong spot. Spot 1. co a very weak spot, spot 4.

A AUTORADIOGRAPH
B COMPUTER IMAGE changer in the background level subtracted from the integral will msrkedly affect the final value (rea below). The magnitude of the effect of a change in background on the integrated value also depends on the spot shape. The by P msll change in background. while that for a more diffuse spot (for integral for a sharp strong spot (for example Spot 2) will be affected little example Spot 1) rill be more greatly affected. However no m a t t e r what the spot shape. the effect of B small change ~n background ;ill have the grc*te~t effect on weak spots.
The main problem with integrating spots such as Spot 4 is that small The effects of small changer in the film background an the integrated density o f strong and weak spots are clearly shown in Figure 4. The actual densities ef Spots 1-4 in Figure 3 were used for this calculation and the assumption is made that a11 spots have the same geometric shape for the rake of the illustration. The parition of the integrated densities of Spots 1  If PRCG-2 is used for analysis and background level is held constant at 10 fit C a u r s~a n curves to the data points do not require data from the entire spot. As long as the peak of rhe spot is in the field stored by the SAYF cormand. PROG.3 will renerate the correct value for the inteerated densitv.

Analysis of Overlapping Spots: The methods of calculation ured by PROC-3
This feiture a11ow~ t6e operator 10 calculate the integrals Gf overlappini spots. The use of PROG-3 in this manner can be demonstrated using Spot 2 from Table 3 Table 3 for the actus1 integrated valuer). q. The effect of raising film background on the integrated density