Abstract
Introduction
Two-dimensional gel electrophoresis (2D-GE) is a powerful technology to compare complex protein mixtures; it has been applied to many fields of biomedical research and is widely used in biomarker discovery. In a 2D-GE gel thousands of proteins are separated in well defined spots; these protein spots can be revealed via a variety of staining techniques (Coomassie, Silver Stain, Sypro), 1 and captured by one or more digitized computer images per gel(CCD camera, laser scanner, and optical scanner). 2 The image capturing phase transforms the biological information of the 2D-GE gel into a quantitative computer-readable data set. Once all the studied gels have been collected and digitized the software-based image analysis can be started. Image analysis is crucial in extracting biologically relevant information from a two-dimensional gel electrophoresis experiment.
Despite the availability of several software applications to analyze 2D-GE images, there is no general consensus on 2D-GE data analysis protocol. Moreover several authors reported that the commercial packages are time consuming, can often miss values or give false positives, and induce variance in quantitative measures.3–9
The commercially available software perform the analysis workflow in two different ways. The classical package condensed the information onto spots. The spot detection is performed prior to matching and expression profile extraction. The second image analysis software group is based on the whole image information. These packages apply a warping procedure to remove running differences between gels, and the spot detection and protein expression profiles extraction occurred in a separated and independent step. 5 The emphasis in this analysis software has been on reducing the subjectivity of the image analysis. The fact that the alignment step is performed prior to the spot detection facilitates simultaneous spot detection on all gel images in an experiment and the resulting spot boundaries are identical on all gel images. 10 In Table 1 are collected the most popular commercial software for 2D-GE gel analysis.
The most popular commercial software for 2D-GE gel analysis.
Several research groups have developed freeware systems to handle certain key aspects of gel analysis, including archiving (SwissProt 2D), 11 comparison (Flicker), 12 interactive exploration (WebGel), 13 registration (bUnwarpJ and Sili2DGel),14,15 spot detection, 16 spot quantification precision and differential expression (Pinnacle). 17 However nobody has developed a complete package freely available and platform independent able to perform all the steps of a 2D-GE gel analysis experiment. 18
Leveraging also on these experiences we have developed an image analysis workflow based on the popular public domain image analysis software package ImageJ (http://rsb.info.nih.gov/ij/). ImageJ and its plug-in is easy-to-use software and can be used in routine applications. Our workflow has been developed according to the whole image information procedure. 19 It is based on six steps: aligning all the images, computing image fusion, creating a consensus spot pattern, propagating the consensus spot pattern to all gel images for quantification, and finally the statistic analysis.
In order to test our procedure, we performed a 2D-GE study of plasma from patients immediately after an acute myocardial event, comparing the results obtained using a widely diffused commercial package (Melanie; GeneBio, Geneva) to those obtained with our ImageJ-based procedure. We looked for biomarkers of pathology and/or treatment in acute myocardial infarction (AMI) patients treated with common anticoagulant protocols. The authors confirm that ethical approval was obtained for this research.
Material and Methods
2-DE Page
With this aim, we enrolled 9 patients admitted within 6 hours after the onset of chest pain symptoms, with myocardial infarction defined according to ESC/ACC criteria. All subjects signed informed consent forms prior to standard sample collection. 2D-GE was performed according to Maresca et al 20 and each sample was run in duplicate. For the first-dimension electrophoresis of plasma samples 200 μg (approximately 3 μl) were applied to 18-cm linear IPG strips 4–7 GE Healthcare (Uppsala, Sweden) and focused until 72000 V/hr were reached. Prior to SDS-PAGE, the IPG strips were equilibrated twice for 15 min in equilibration buffer (50 mM Tris-HCl pH 8.8, 6 M urea, 30% (v/v) glycerol, 2% (w/v) SDS and traces of bromophenol blue) containing 1% (w/v) DTT for the first equilibration step and 2.5% (w/v) iodoacetamide for the second step. SDS-PAGE was performed on 12.5% polyacrylamide gels according to Laemmli. 21 The run was carried out at 60 mA/gel at 16 °C and terminated when the dye front reached the lower end of the gel. Gels of plasma samples were visibly stained with Coomassie Blue, scanned using transmission mode to avoid saturation effects and saved in 16-bit TIFF format.
Image Analysis
Once all gels in the study had been collected and digitalized, they were analyzed using the ImageJ and some of its plugins or the commercial software Melanie.
We now go through the ImageJ-based procedure, and then we will compare the results obtained to the Melanie output. In Table 2 is collected the list of steps that describes how to perform the analysis using ImageJ and its plug-in.
List of steps that describes how to perform the analysis.
First, all images were warped by bUnwarpJ, 14 an algorithm for elastic and consistent image registration developed as an ImageJ plug-in. It performs a simultaneous registration of two images, allowing us to solve the problem of spatial distortions due to run-time differences and dye-front deformations.
We used the software to align all images in pairs, taking always the same image as reference and producing the corresponding warped images of the others. bUnwarpJ can be freely downloaded from http://biocomp.cnb.uam.es/~iarganda/bUnwarpJ/.
The reference image and warped images were subsequently displayed in a single stack image and summed to generate a fused image. We followed an image sum approach to retain as much information as possible from the original images.
Spot detection was performed on the fused image by the watershed plug-in written by Daniel Sage and freely downloaded from http://bigwww.epfl.ch/sage/soft/watershed/index.html. This plug-in is able to segment an image using the watershed algorithm by flooding directly on graylevel image. Of the several kind of outputs provided, we selected the binary output that allows us to apply the blob analyzer of ImageJ, so as to measure the catchment basins and save the blots, one for each protein spot, as a list of regions of interest (ROI). Each ROI corresponds exactly to a spot in the fused image. The list of ROI obtained by the spot detection procedure was our consensus spot pattern that is valid for the whole gel set of the experiment.
In other words, the list of ROI obtained is equivalent to the grid used in gene chip analysis, this grid was imposed on each of the aligned gel images so that a defined number of areas were quantified on every gel image of the experiment. The spot volume values extracted from each image were listed in a ImageJ “Results table”. The resulting table of “the whole image information procedure” did not have empty cells, while some commercial software, such as Melanie, are not able to eliminate all bias due to missing spot values. 22 All the data were analyzed by Calc (OpenOffice), a open source spreadsheet program downloadable from the web site http://www.openoffice.org/. For the normalization the volume of each spot on a given gel image was diveded by the total volume of all spots on that image. 23 The resulting table of our method did not have empty cells, while some commercial software, such as Melanie, are not able to eliminate all bias due to missing spot values. 22
Results
The warping step has produced a good alignment of all 2D-GE images. In Figure 1 is shown an example of warping step results. Figure 1A and B show two different 2D-GE images, in Figure 1C is shown the elastic registration obtained during the warping, and in Figure 1D the overlap of the two gels after the warping step. For the image fusion process all the warped images were used and the fused images do not show multiple spots thanks to the strength of the elastic alignment.
(A and B) shown two different 2D-GE images of two control subjects. In (C) is shown the elastic registration obtained during using bUnwarpJ plug-in. In Figure 1D is shown the overlap of the two gels after the warping step, in the red channel is shown the reference gel (Fig. 1A) and in the green channel is shown a warped gel.
Using the ImageJ procedure we were able to study 232 conserved spots, while with Melanie we analyzed a pattern of 205 matched spots. The spot detection and the matching were manually checked in both the procedures; Figure 2 shows the spots detected by the two procedure on one of the control subject gels (the reference gel used for the Melanie analysis).
2D-GE from the plasma of a control individual. (A) Spots detected and matched using Melanie. (B) Spots detected using the ImageJ procedure.
The scatter plot in Figure 3 shows that there is a linear relationship between the spot volumes evaluated by Melanie and the corresponding values obtained by the ImageJ procedure. In particular 42 spots, 33 more abundant spots (blue stars in Fig. 2) and the 9 differentially expressed spots (green stars in Fig. 2, see the next paragraph for the identification procedure) were considered for the comparison; the fact that the straight line in Figure 3 has a slope >1 means that volume values calculated for the same spot are on average larger by using the ImageJ procedure, which can be related to the slightly larger area segmented for each spot by ImageJ due to the fact that spots were segmented on the fused image (and not on every single gel, as Melanie does). Similar results were obtained for the spot list of every other gel as well as for the average volumes (data not shown).
Scatter plot of spot mean volumes as evaluated by Melanie and ImageJ.
The 9 differential spots (shown in Figure 4), whose mean normalized volume was significantly decreased in the myocardial infarction versus the control group, were all identified by
Profile of proteins differentially expressed in the plasma of myocardial infarction (MI) patients vs. control subjects. 7 spots were found significantly decreased in the plasma of the myocardial infarction patients by both methods (
By using the procedure described by Lemkin et al, 24 ie, by matching the spots of a gel with those of a reference map of human plasma (http://expasy.org/swiss-2dpage/viewer), we were able to tentatively identify 5 of the 7 significantly different spots as fibrinogen gamma chain fragments (with reference to Figure 4, spots 142, 143, 146, 147, and 148).
Discussion and Conclusions
Previous proteomic studies reported protein expression differences in plasma from patients during an acute coronary syndrome and from patients with moderate hypercholesterolemia, 25 and proteomic differences in the plasma of coronary ischemic patients resistant to aspirin as compared to aspirin-sensitive patients. 26 Interestingly, 3 of the very same gamma fibrinogen spots were reported as increased in untreated myocardial infarction, 25 but thrombolytic (fibrinolytic) therapy reduces the level of all fibrinogen chains (see for example 27 ). Since all the enrolled patients received an anticoagulant therapy, this decrease is possibly connected to the therapy, and thus may reflect modifications of the fibrin/fibrinogen balance in the patient blood.
Further experimental work in a larger patient cohort will be needed to confirm these data, and the study of the effects of anticoagulant therapy on hospitalized patients goes beyond the purpose of this work.
In conclusion, we have developed a free and easy alternative to a common commercial package for the segmentation and quantification of 2D gel spots; the procedure proved to be so effective, to confirm the results obtained by an established commercial solution.
We hope that the provided solution can help proteomic laboratories to quickly and inexpensively evaluate 2D-gel experimental results, without losing the required accuracy and providing a common reference for future analyses.
Disclosures
Author(s) have provided signed confirmations to the publisher of their compliance with all applicable legal and ethical obligations in respect to declaration of conflicts of interest, funding, authorship and contributorship, and compliance with ethical requirements in respect to treatment of human and animal test subjects. If this article contains identifiable human subject(s) author(s) were required to supply signed patient consent prior to publication. Author(s) have confirmed that the published article is unique and not under consideration nor published by any other publication and that they have consent to reproduce any copyrighted material. The peer reviewers declared no conflicts of interest.
Footnotes
Acknowledgements
/) in the frame of the regional law n.84-07/12/1993 (project ParIS–-Parkinson Informative System).