Proteomics 2007, 7, 513–523
513
DOI 10.1002/pmic.200600648
RESEARCH ARTICLE
Improved comparative proteome analysis based on two-dimensional gel electrophoresis Murat Eravci1, 2, Sandra Fuxius1, 2, Oliver Broedel2, Stephanie Weist2, Selda Eravci2, Ulrich Mansmann3, Hartmut Schluter4, Joachim Tiemann4 and Andreas Baumgartner1, 2 1
Department of Radiology and Nuclear Medicine (Radiochemistry), Charité Universitätsmedizin, Campus Benjamin Franklin, Berlin, Germany 2 A1M Proteome Science, Berlin, Germany 3 Department of Medical Informatics, Biometry and Epidemiology, University of Munich, Germany 4 Department of Internal Medicine IV, Charité Universitätsmedizin, Campus Benjamin Franklin, Berlin, Germany
The purpose of this study was to test the extent to which differences in spot intensity can be reliably recognized between two groups of two-dimensional electrophoresis gels (pH 4–7, visualized with ruthenium fluorescent stain) each loaded with different amounts of protein from rat brain (power analysis). Initial experiments yielded only unsatisfactory results: 546 spots were matched from two groups of 6 gels each loaded with 200 mg and 250 mg protein, respectively. Only 72 spots were higher (p,0.05), while 58 spots were significantly lower in the 250-mg group. The construction of new apparatuses that allowed the simultaneous processing of 24 gels throughout all steps between rehydration and staining procedure considerably lowered the between-gel variation. This resulted in the detection of significant differences in spot intensities in 77–90% of all matched spots on gel groups with a 25% difference in protein load. This applied both when protein from 24 biological replicates was loaded onto two groups of 12 gels and when two pooled tissue samples were each loaded onto 6 gels. At a difference of 50% in protein load, more than 90% of all spots differed significantly between two experimental groups.
Received: August 24, 2006 Revised: November 2, 2006 Accepted: November 18, 2006
Keywords: Power analysis / Quantification / Rat brain / Two-dimensional gel electrophoresis
1
Introduction
The identification of differences in protein expression is one of the most important fields of application of proteome research. In recent years, enormous efforts have been invested in developing accurate methods for quantitative proteomics and considerable improvements have been achieved (reviewed in [1–5]).
Correspondence: Andreas Baumgartner, Department of Radiology and Nuclear Medicine (Radiochemistry), Charité Universitätsmedizin, Campus Benjamin Franklin, Hindenburgdamm 30, 12200 Berlin, Germany E-mail:
[email protected] Fax: 149-30-8445-2700
© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
2-DE is the key technology for studying the differences in protein expression level and PTM between various biological samples (reviewed in [6]). However, comparisons of spot intensities from 2-DE gels may be problematic due to the high gel-to-gel variance, the possible identification of several proteins from one spot, and labor-intensiveness [1– 3, 7–9]. With the technique of 2-D DIGE, two protein extracts can be prelabeled with two fluorescent cyanine dyes and then mixed and run on the same 2-D gel [10, 11]. This procedure avoids the problem of between-gel variability, but gel-to-gel variances of the ratios between the proteins from different samples may impair the reliable detection of small differences in spot intensities (e.g. [12], see also Discussion). To overcome the limitations inherent in 2-DE technology, gel-independent technologies have been developed that rely www.proteomics-journal.com
514
M. Eravci et al.
on the use of pairs of isotopically labeled reagents incorporated in vitro at either the peptide or the protein level, thus rendering possible comparative protein quantification by MS (reviewed in [4, 5]). In 1999, Gygi et al. [13] reported relative quantification of proteins after labeling of cysteine-containing peptide pairs with isotope-coded affinity tags (ICAT). Since then, the number of newly developed techniques for non-gel based quantification has been rapidly growing [14– 19], including methods for direct quantification of complex protein mixtures by different LC-MS runs without isotope labeling [20, 21]. However, all these elegant and promising new technologies are subject to a number of limitations (for reviews see [1–3, 22]). Where quantification is concerned, they are mostly restricted to comparing samples of n = 1 with each other, which makes it difficult to carry out appropriate statistical group comparisons. Due to the variability inherent in biological samples, only marked changes in protein expression, e.g. two-fold or higher, are usually considered relevant. Detection of less pronounced changes in protein expression may, however, be of major relevance. For example, the proteomic profiles of tissues derived from patients with Alzheimer’s disease as well as from those with cancer or any other diseases leading to progressive cell death are probably of greater evidential value in the early stages, when the expression of disease-specific proteins is only beginning to change, than in the late stages, when the levels of many nonspecific proteins are also dramatically altered due to progressive cell death. In pharmacological studies, manifold changes in protein expression following drug application may often reflect toxicological side effects rather than clarify the biochemical mechanisms underlying the action of the drug. There is therefore a need for the development of methods that allow the reliable statistical analysis of differences in protein expression of much lower than 100%. Many useful suggestions have been made as to how the reliability of quantitative comparisons carried out by conventional 2-DE technology could be improved, e.g. [23–26]. To our knowledge, however, hardly any study has yet evaluated the statistical power of the 2-DE methodology. Such a study should establish what percentage of all matched spots can be correctly classified as significantly different between groups when defined amounts of proteins are loaded onto the gels of the different groups (i.e. a 100% difference between, for example, the 250-mg and 500mg groups), a defined number of technical and/or biological replicates from a specific tissue are analyzed, and a defined methodology (i.e. pH range 4–7, labeling with CBB, Melanie software etc.) is applied. The need for such power studies in gel-based proteomics has now been stressed with increasing frequency [27]. We, therefore, set out to test the statistical power of our well-established, conventional 2-DE technique.
© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Proteomics 2007, 7, 513–523
2
Materials and methods
2.1 Materials 2.1.1 Reagents Pharmalytes, IPG buffers and IPG strips were purchased from GE Healthcare Biosciences (Freiburg, Germany) and electrophoresis reagents from Serva (Heidelberg, Germany), while all other reagents were acquired from Fluka (Buchs, Switzerland), Merck (Darmstadt, Germany), Roche Diagnostics (Mannheim, Germany) and Sigma Aldrich (Deisenhofen, Germany). 2.1.2 Purchased equipment The rehydration tray, Multiphor II unit and Ettan Dalt II System were purchased from GE Healthcare Biosciences, staining dishes from Tupperware (Frankfurt, Germany) and the Fuji FLA-3000 Fluorescence scanner from Raytest (Berlin, Germany). 2.1.3 Equipment developed by our own group (see Fig. 1) A rehydration tray for 24 IPG strips (18 cm) was constructed out of polycarbonate. The grooves of this rehydration chamber were only 19.5 cm in length, resulting in a smaller dead space and thus reduced crystallization of the rehydration solution containing the sample (Fig. 1a). An IPG strip tray for the simultaneous IEF of 24 IPG strips in one Multiphor II unit was made out of Tecapek with 24 milled grooves (Fig. 1b). The IPG strips are fixed by inserting the plastic ends of each strip into the insertion gap at the sides of each groove. Equilibration chambers for simultaneous equilibration of 24 IPG strips attached to the IPG strip tray were constructed in order to hold the IPG strip tray in a vertical position with sufficient volumes of each equilibration buffer (Fig. 1d). A second dimension electrophoresis unit which is the same as the Hoefer IsoDalt Unit with the exception that it processes 24 gels instead of 10 gels and is filled with 40 L of electrophoresis buffer instead of 20 L, was also constructed. This unit has 59 cm 6 32 cm large platinum-coated titanium electrodes instead of platinum wire electrodes in order to obtain a homogeneous electric field in the 40-L electrophoresis unit (Fig. 1e). A staining apparatus (Fig. 1g) and staining cassettes (Fig. 1f) for simultaneous fixation, staining and destaining of 24 2-D gels were also constructed. The staining cassettes hold the gels gently between two stainless steel nets during the whole staining procedure without the need for further handling of the gels. The dimensions of the staining cassettes allowed swelling and shrinking of the gels during the steps of our staining procedure. The staining device contains a motor-driven stainless steel frame with slots for 24 cassettes inside an www.proteomics-journal.com
Technology
Proteomics 2007, 7, 513–523
515
ous vortex stirring for 10 s. Thereafter the proteins were reduced by the addition of 5 mM tributylphosphine (TBP) and incubation for 60 min at room temperature. The alkylation was performed by the addition of 15 mM iodoacetamide and further incubation for 90 min at room temperature in the dark. Following reduction and alkylation the protein sample was centrifuged for 60 min at 20 000 6 g to pellet any debris. The protein concentration of the supernatant was determined according to Bradford [29]. The samples were diluted to the desired concentration with rehydration buffer (7 M urea, 2 M thiourea, 5% CHAPS, 10 mM DTT, 0.4% Pharmalyte 3–10, 0.5% Triton X-100, 0.025% bromophenol blue and 2% IPG-buffer). 2.3 2-DE - improved standard procedure (Experiments 3-13) 2.3.1 Application of the samples
Figure 1. Equipment constructed by our laboratory. (a) Rehydration tray for 24 IPG strips (18 cm), (b) IPG strip tray for 24 IPG strips, (c) IPG strip tray attached to the Multiphor II unit, (d) IPG strip tray in the equilibration chamber, (e) second dimension electrophoresis unit with platinum coated titanium electrode plates for processing 24 gels simultaneously, (f) stainless steel staining cassettes, (g) staining apparatus for simultaneous fixation, staining and destaining of 24 2-DE gels.
acrylic glass tank with a capacity of 40 L (Fig. 1g). Further information on all apparatuses constructed by our own group can be obtained from the authors on request. 2.1.4 Software BAS-Reader was from Raytest, Proteomweaver from Definiens (Munich, Germany), Delta 2D from Decodon (Greifswald, Germany), and SPSS for Windows was from SPSS (Munich, Germany). 2.2 Animal experiments and sample preparation Adult male Sprague-Dawley rats weighing approx. 300 g were decapitated without anesthesia. Various areas of the brain were dissected according to Glowinski and Iversen [28] and stored immediately at -807C. Protein samples were prepared from brain tissue, which was homogenized with solubilization buffer (9 M urea, 2.5 M thiourea, 5% CHAPS, 0.5% Pharmalyte 3–10, 0.1% Pefabloc SC and 150 U/mL Benzonase) in a motor-driven Potter-Elvehjem homogenizer, followed by five cycles of ultrasonication for 30 s and vigor© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
The IPG dry strips (pH 4–7 throughout) where applied with the gel side up to the rehydration chamber we had constructed ourselves. An aliquot of 340 mL of the sample was pipetted over the IPG strip, which was then swayed from side to side to achieve complete coverage of the gel surface. To prevent evaporation and crystallization of urea the slots were coated with IPG cover fluid. The rehydration was performed overnight. 2.3.2 IEF Up to 24 rehydrated IPG strips were transferred to the IPG strip tray we had developed ourselves. The IPG strip tray was placed in the Immobiline DryStrip tray on the Multiphor II unit (Fig. 1c). IEF was carried out for 65 kVh, using an EPS 3501 XL Power Supply (GE Healthcare Biosciences). Samples from various groups processed in a single experiment were processed in the same IEF run when equal amounts of proteins were loaded, and in different runs when the amount of protein differed between the groups (e.g. 200 mg vs. 250 mg,), unless otherwise stated. 2.3.3 Equilibration The IPG strip tray containing the 24 focused IPG strips was transferred on slow rotary shakers to the equilibration chambers we had constructed ourselves. The strips were first equilibrated for 15 min in equilibration solution (6 M urea, 250 mM Tris-HCl pH 6.8, 30% glycerol, 1% SDS) with 1% DTT. The IPG strip tray was then placed in a second chamber containing an equilibration buffer with 9% iodoacetamide for 15 min. 2.3.4 Second dimension electrophoresis SDS-PAGE was performed in a second dimension electrophoresis unit, which we had developed in our laboratory to process 24 gels simultaneously. SDS-PAGE was carried out, www.proteomics-journal.com
516
M. Eravci et al.
beginning with a low current of 20 mA/gel for 1 h, followed by 120 mA/gel overnight. This relatively high current was found to yield the best results when the chamber constructed by our own group was used. The SDS-PAGE was stopped when the bromophenol blue reached the bottom of the gel. 2.3.5 Fixation, staining, and destaining Thereafter the SDS-PAGE gels were fixed in the staining cassettes and transferred to the staining device. For all steps of the staining procedure, the cassettes remained in a slowly shaking frame inside the staining device. The staining device was filled and drained with all required solutions using vacuum pumps. Fixation was performed in a mixture of 30% ethanol and 10% acetic acid overnight. Staining was carried out with 1.5 mM ruthenium II tris(bathophenanthroline disulfonate) fluorescent stain [30] in 30% ethanol and 10% acetic acid overnight. Destaining of the background was performed in 20% ethanol overnight. 2.3.6 Fluorescence scanning The gels were scanned on a Fuji FLA-3000 fluorescence scanner using the scanning software BAS-Reader. The scans were performed at an excitation of 473 nm and an emission filter of 580 nm. The digital resolution was 16 bits/pixel, with a sensitivity of 100 mm and a pixel size of 100 mm. 2.3.7 Conventional 2-DE technique In Experiments 1 and 2, the GE Healthcare Biosciences rehydration tray was used for rehydration. IEF was performed on a DryStrip aligner in the Immobiline DryStrip tray (GE Healthcare Biosciences). The equilibration was also performed in the Immobiline DryStrip tray. Equilibrated strips were transferred to second dimension electrophoresis in the Ettan Dalt II System (GE Healthcare Biosciences). Fixation, staining and destaining were performed in polypropylene staining dishes (Tupperware). 2.4 Image analyses The software used for image analyses was either Delta 2D (Decodon) or Proteomeweaver (Definiens). With Delta 2D the spots of the 2-D patterns were warped between the gels with Delta 2D to get exactly the same spot positions for all gels. Then a fusion image was generated, on which the spots were first automatically detected and then edited manually. After successful validation of the spots, the resulting spot pattern was transferred to all other gels. The spots were quantified without normalization. With Proteomweaver, spots on 2-D images were detected automatically and edited manually on each image to eliminate the false positive and false negative signals from the automatic spot detection, using control tools such as 3-D visualization. After spot © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Proteomics 2007, 7, 513–523
detection, the spots were matched between all gels and quantified without normalization. It was necessary to exclude the normalization from the image analysis software because if normalization were automatically carried out this would also normalize the differences between two different protein loadings, which are the differences which were to be compared in this study (e.g. 200 mg vs. 250 mg). For normalization, the raw data generated by both programs was transferred to Microsoft Excel. Each spot volume was then adjusted to the absolute volume of all spots of the respective gel and group. The details of the experimental design are given in Section 3 and in Tables 1–6. 2.5 Statistical analyses Data were analyzed using the software SPSS for Windows from SPSS. Means 6 SD were used throughout. Deviations from normality were analyzed with the KolmogorovSmirnoff test. Between-group comparisons were performed using the Student’s t-test for independent samples (twotailed). The results of the Mann-Whitney U tests carried out in selected experiments are given for comparison. Pearson’s coefficients of correlation were calculated between all spots of different gels. The CV was calculated as follows for each experimental group: the mean of the SD calculated for all spots was expressed as a percentage of the mean value of all spot intensities. For example, when the mean 6 SD of all spots of an experimental group is 20 6 5, then the resulting CV is 25%. p–Values ,0.05 were considered significant throughout.
3
Results and discussion
Analysis of the data from several experimental groups with the Kolmogorov-Smirnoff test failed to reveal any relevant deviations from normality. For example, the test on the 200-mg dataset from Experiment 8 (Table 3) revealed a p-value of less than 0.05 for only 1 spot, 2 spots had p-values of less than 0.1 and 15 spots p-values of less than 0.3. In the 250-mg dataset, no spot yielded a p-value of less than 0.05 and only 18 spots had p-values of less than 0.3. Similar results were obtained for the other datasets (data not shown). All data were therefore analyzed by t-test. The Mann-Whitney U test, which is robust against deviations from normality, yielded comparable results (see Experiments 8–10, Tables 3 and 4). The results for all 13 experimental groups are listed in Tables 1–6. Table 1 shows the results of two experiments performed with commercially available standard equipment (see Section 2) before the analysis and elimination of different sources of gel-to-gel variation. In Experiment 1 only 130 (23.8%) of 546 matched spots showed significant differences between the two groups of gels loaded with 200 mg and 250 mg protein, respectively. We hypothesized that processing different amounts of protein in one IEF run impairs the www.proteomics-journal.com
Technology
Proteomics 2007, 7, 513–523
517
Table 1. Comparison of two experimental groups with a 25% difference in protein load (200 vs. 250 mg) – experiments before methodological improvements
No. of experiment
No. of gels compared
No. of spots matched
1
6 vs. 6 (in one IEF-run)
2
6 vs. 6 (in different IEF-runs)
p,0,05 (t-test)
CVa) (%)
Coefficient of correlationb) (range)
Brain area
Spot number
Percent (%)
546
130 (72:, 58;)
23.8
200 mg: 32.5 6 25.3 250 mg: 37.1 6 27.9
0.948–0.990
12 individual midbrains
724
394 (327:, 67;)
54.4
200 mg: 32.0 6 42.6 250 mg: 26.4 6 22.6
0.930–0.998
12 individual limbic forebrains
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities. b) Pearson’s coefficients of correlation were calculated between each gel and all other gels per group.
gel-to-gel reproducibility and thus repeated the experiment processing the 200-mg group and the 250-mg group in two IEF runs (Experiment 2). Thereafter 54.4% of the spots differed significantly. Most worrying, however, was the fact that in Experiment 1, 58 spots and in Experiment 2, 67 spots were significantly less intense in the 250-mg group than in the 200mg group. This indicates that quantification with conventional 2-DE techniques may yield unreliable results when the spot intensities differ by approx. 25% between two groups. Several studies have calculated coefficients of correlation between all spot intensities of two different gels in order to evaluate gel-to-gel variability, e.g., [31]. In Experiments 1 and 2, Pearson’s coefficients of correlation were calculated between each single gel and all the other gels, respectively. Calculations were done separately for the 200-mg and 250-mg groups. Table 1 shows that the coefficients ranged between 0.948 and 0.990 in Experiment 1 and between 0.930 and 0.998 in Experiment 2. These results reveal that even in experiments which are characterized by an unacceptably high gel-to-gel variability the between-gel coefficients of correlation all appeared to be very satisfactory. As these results seemed insufficient, we embarked on a thorough analysis of the factors causing gel-to-gel variability. At the same time we constructed new pieces of apparatus (Fig. 1) with which 24 gels can be processed simultaneously, for three reasons. First, simultaneous handling of all gels of one group during rehydration, in the IEF, during equilibration, in the second dimension and during the staining procedure may decrease gel-to-gel variance. Secondly, we were afraid that large numbers of gels (e.g. 12 or more) would need to be compared in order to achieve reliable statistical differentiation between groups whose protein loads differ only slightly (e.g. by 25%). Thirdly, the 2-DE methodology is time-consuming and thus simultaneous processing of 24 gels in parallel may considerably speed up this procedure. Table 2 lists some of the sources of between-gel variability. All experiments were performed with the new equipment unless otherwise stated. © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
In Experiment 3, 8 gels loaded with 200 mg protein from pooled midbrains were compared with each other. We had often observed that the numbers on the commercially used IPG strips (GE Healthcare) in one package were not consecutive, but interrupted by an interval of some times several hundred units. Other suppliers do not number the strips in their packages at all. The batches/lots of IPG strips are originally produced from several different gels. The strips from all of these “mastergels” are then grouped together to form commercially available packages. In this experiment, protein was loaded onto consecutively numbered IPG strips derived from a single mastergel. In the other group, protein was loaded onto IPG strips from different mastergels (but from the same lots/batches) as evidenced by large intervals of up to several hundred units between the numbers on the strips from one package. Table 2 shows that the CV was 13.4% when consecutively numbered strips were used, but 22.7% when strips from different mastergels were used. It would seem advisable for the producers of IPG strips not only to number these strips, but also to assure that a package contains only strips from a single mastergel. In Experiment 4, two groups with protein loads of 200 and 250 mg, respectively, were processed in the same IEF run. Significant changes were seen in only 32.2% of all spots, the intensities of 71 spots changed in the “wrong direction” and the CV were over 30%. In Experiment 5, 6 gels were processed gelside-down in the rehydration tray (conventionally) and compared to 6 gels processed gel-side up. Preliminary experiments in which the gels were processed gelside-up showed a more even reswelling over the whole strip length in all strips of a group (rendered visible by the bromophenol blue front in the IEF) than the conventional gel-side down rehydration. Table 2 reveals that performing the rehydration gelside-up considerably lowers the between-gel variance (CV of 10.1%, as compared to 17.0% in the gelside-down group). The results of Experiment 6 (Table 2) show that when the staining procedure was performed for all gels in parallel in our newly constructed www.proteomics-journal.com
518
M. Eravci et al.
Proteomics 2007, 7, 513–523
Table 2. Sources of variance impairing between-gel reproducibility
No. of Variable investi- Protein No. of No. of spots p,0.05 (t-test) experi- gated load (mg) gels matched Spot number Percent ment compared (%)
CVa) (%)
Brain area
Contribution to variance
3
IPG strips from 200 one vs. different master gels
8 vs. 8
813
n.d.
n.d.
One gel: 13.4 6 10.4 Midbrain, Different gels: 22.7 6 18.6 pool
At least 9%
4
Different protein 200 loads in one 250 IEF-run
12 12
568 568
183 (112: 71;)
32.2
200 mg: 250 mg:
30.4 6 22.8 24 individual 35.5 6 26.5 midbrains
5
Rehydration: 200 gel-side up vs. gel-side down
6 vs. 6
799
n.d.
n.d.
Upside: Downside:
10.1 6 8.9 Limbic fore- 7% 17.0 6 17.8 brain, pool
6
Staining: appa- 200 ratus for 24 gels vs. individual gel staining trays
6 vs. 6
1057
n.d.
n.d.
Apparatus: trays:
11.9 6 9.7 Cortex, 16.4 6 14.2 pool
7
Software: Delta 200 2D vs. Pro- 250 teomweaver 200 250
12 12
683 683
Delta 2D:
529
77.5
200 mg: 250 mg:
16.7 6 4.7 15.9 6 4.6
12 12
654 654
Proteomweaver: 402
61.4
200 mg: 250 mg:
21.2 6 5.8 20.5 6 6.4
5%
24 individual 5% midbrains
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities; n.d., not determined.
staining device, the CV was considerably better (11.9%) than it was when each individual gel was stained in polypropylene staining dishes (16.4%). The problem of software–induced variance in 2-DE image analysis has already been strengthened by several other study groups, e.g. [32]. Image analysis with the Proteomweaver software resulted in 654 matched spots compared to 683 after using Delta 2D software (Experiment 7). Table 2 also shows that analysis with Delta 2D yielded more spots correctly classified as significantly changed than Proteomweaver (77.5 and 61.4%, respectively) and the CV were also approx. 5% lower when Delta 2D was used. When spots were first identified on the fusion image created by the Delta 2D software, the number of spots generated was artificially high, usually between 1000 and 1300. It was possible to reduce this number to approx. 400–600 manually, by eliminating a large amount of obviously artificial spots, e.g. at the gel borders or in various smears. During manual correction of the fusion image up to more than 100 additional spots were identified by separating spots, which overlapped each other to various degrees and were not automatically separated. Thereafter, the spot image was automatically transferred to all gels and the quality of the matching between each gel and the fusion image was checked again. When the overall gel quality of an experiment was good and when the © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
warping was done carefully, the number of mismatches could be reduced to almost zero. This may be due to the precise warping procedure used in Delta 2D to align the spots between the gels of a single experiment properly. The most important software–derived source of error in spot quantification occurred when two or more spots were not clearly separated from each other, e.g. when one smaller spot appeared as a “shoulder” of a higher–intensity spot. The demarcations between such spots afforded by different gels may not be equal, resulting in unusually high CV. Thus, when such spots differ significantly between two “real” experimental groups (and not only in a power analysis), manual control is obligatory in order to establish whether the significant difference is reliable or may be due to a softwareand/or gel-derived variance in spot demarcation. Another source of variance occurred particularly in the range of lower spot intensities. The Pearson’s coefficients of correlation between all spot intensities and CV were always negative, but did not reach significance, as the variance increased only for a minority of spots with very low intensities. However, when the mean CV for all spots with intensities between 0.01 and 0.2 was calculated, for example for Experiment 8 (Table 3), it proved to be 35.6%, compared to the 14.3% calculated for all other spots with intensities between 0.02 and 24. This variance may be caused by some remaining background noise www.proteomics-journal.com
Technology
Proteomics 2007, 7, 513–523
519
Table 3. Comparison of two experimental groups with a 25% difference in protein load (200 mg vs. 250 mg) – experiments after methodological improvements
No. of experiment
8
No. of gels compared
No. of spots matched
p,0.05 Spot number
CVa) (%)
Brain area
24 individual midbrains
t-test
Mann-Whitney U
529 (77.5%)
495 (72.9%)
200 mg: 16.7 6 4.7 250 mg: 19.9 6 4.6
10 vs. 10 683 (last two figure rows omitted)
450 (66.3%)
428 (62.7%)
200 mg: 15.8 6 4.7 250 mg: 15.0 6 4.8
8 vs.8 683 (last four figure rows omitted)
379 (55.8%)
368 (53.9%)
200 mg: 15.4 6 5.1 250 mg: 14.6 6 5.0
6 vs. 6 683 (last six figure rows omitted)
300 (44.2%)
273 (40.2%)
200 mg: 14.5 6 5.3 250 mg: 14.1 6 5.1
12 vs. 12
683
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities.
even after background subtraction. This problem can be partly solved by creating a number of “false spots” (approx. 20) on the fusion image in “white spaces” between the identified spots, transferring them to all gel images and quantifying their intensities, which may then indicate the “real” lower limit of detection. Below this limit, all software– delineated spots are most likely artifacts. At least in our hands, image analysis with the Proteomweaver software resulted in a high number of mismatches, which had to be manually corrected on each gel. The software analysis of a given experiment thus took at least twice as long with Proteomweaver as it did with Delta 2D. In addition, the results were not as good as after analysis with Delta 2D and the technical support needs to be improved. In summary, the application of the above mentioned technical improvements results in a considerable decrease in the CV, i.e. in the gel-to-gel variance of each group (see Table 2, right column). In Experiments 3, 5 and 6 (Table 2) all samples were taken from the same tissue pool. The resulting CV of these groups, reflecting the “analytical variance”, ranged between 10.1 6 8.9% and 13.4 6 10.4%. This is considerably lower than the CV previously reported by other groups: 20–27% [33], 7–31% [25], 19–26% [24], 3–33% (mean 18.5% after analysis by Progenesis software) [34] and 32% [35]. Choe and Lee [36] found that 95 of all spots present on 3 out of 4 replicate gels exhibited a CV of less than 52%. Before the methodological improvements were introduced, the CV calculated for experiments using individual animals ranged between 26 and 37% (Table 1), reflecting the combined “analytical” and “biological” variances. These results are similar to the findings of Molloy et al. [24] who reported CV of 23–48% (depending on the sample type). The elimination of several factors responsible for gel-to-gel variability considerably lowered the CV to a range of between 9.5 and 19.9% (Tables 3–6, except for Experiment 12). In this © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
study, we compared only the “analytical” and combined “analytical and biological” variances of different experimental groups. More reliable information on the “true biological variance” could be obtained if the CV of both the “analytical” and the “analytical and biological” variances were directly compared in samples taken from the same experimental group. The above-mentioned improvements in gel quality are evident, however, only after software analysis. Figure 2 shows the proteome patterns of two different brain areas on two
Figure 2. 2-DE gels before and after methodological improvements. Limbic forebrain homogenate, 200 mg protein sample before (a) and after methodological improvements (b). Midbrain homogenate, 200 mg protein sample before (c) and after methodological improvements (d). Note that the four gels were loaded with protein from four different biological replicates and run in four different experiments.
www.proteomics-journal.com
520
M. Eravci et al.
Proteomics 2007, 7, 513–523
Evidently, pooling more than two replicates does not further enhance the rate of significant results. We then investigated the statistical power of our method following increases in the protein load. Table 5 shows that 795 spots were matched and 90.8% of them identified as significantly different when two groups of 12 gels each were loaded with 500 and 625 mg protein, respectively. These results are considerably better than those of Experiment 8 (listed again in Table 5 for comparison), in which only 200 and 250 mg, respectively, were loaded. However, when 800 mg and 1000 mg protein were loaded (Experiment 12), only 725 spots could be matched, and only 53% of them differed significantly between the two groups. This may be due to an increase in overlap between the spots when more protein is loaded, resulting in less reliable spot matching, which may in turn explain the high CV seen in this group. The results of Experiment 13 (Table 6) show that more than 90% of all spots differ significantly between groups of gels with 50% differences in their protein loads, even when only four gels per group were loaded with pools of two tissue samples each and compared with each other. Note that in Experiments 8–13, spots significantly differing between groups were always changed in the “right direction”, i.e. the intensities of the spots from gels with the higher protein loads were always greater than those of the gels loaded with the smaller amount of protein. In conclusion, the optimal experimental design for the analysis of rat brain tissue on 2-DE gels seems to be the investigation of 12 biological replicates per group. The protein extracted from two pooled replicates (preferably in the range of 500–600 mg) is loaded onto six gels per group. Such a design would allow the comparison of four experimental groups each containing six gels in one experiment. In addition, the processing of 24 gels in parallel throughout the whole procedure considerably reduces the amount of work required per experimental group.
gels produced before and after all methodological improvements, respectively. No major qualitative differences can be detected between the two groups of gels by visual inspection alone. We next conducted several experiments to investigate whether our improved methods were suitable for statistically differentiating a sufficient number of spots from groups of gels differing in their protein load by 25%. In Experiment 8 (Table 3) 12 gels loaded with 200 mg protein from 12 individual midbrains were compared with 12 gels loaded with 250 mg protein from 12 other midbrains. A total of 529 spots (77.5%) differed significantly between the two groups as evaluated by Student’s t-test. Furthermore, groups of 10, 8 and 6 gels each were created by omitting the remaining gels in each case and comparing these groups with each other (e.g. omitting the last two rows of figures corresponding to the last two gels of each group resulted in the comparison of ten gels with each other). The results are shown in Table 3. We next investigated whether the efficiency of the procedure could be improved by pooling two biological replicates and loading them onto 6 gels per group, again comparing 12 different biological replicates per group with each other (Table 4). The comparison of eight pools, each of two biological replicates, yielded significant differences in 80% of all matched spots, and the comparison of six pools (the last two gels of each group being omitted) yielded significant differences in 83.5%. These results were replicated in two further experiments (data not shown). Pooling three samples did not yield better results (data not shown). We next examined whether the number of gels could be further reduced by pooling four and five samples, respectively and processing them on groups of four gels each. The results are presented in Table 4 (Experiment 10) and show that only 43.4 and 60% of all matched spots were classified as significantly changed in groups 10a and 10b, respectively.
Table 4. Comparison of two experimental groups with a 25% difference in protein load (200 mg vs. 250 mg) – experiments with pooled tissue after methodological improvements
No. of experiment
No. of animals pooled
No. of spots matched
2
505
p,0.05 Spot number
CVa) (%)
Brain area
t-test
Mann-Whitney U
404 (80.0%)
428 (84.7%)
6 vs. 6 2 505 (last two rows of figures omitted)
422 (83.5%)
408 (80.8%)
a
4 vs. 4
4
482
209 (43.4%)
207 (42.9%)
200 mg: 16.7 6 17.8 32 individual septi 250 mg: 15.6 6 19.0
b
4 vs. 4
5
482
289 (60.0%)
293 (60.8%)
200 mg: 15.3 6 14.7 40 individual septi 250 mg: 15.3 6 18.1
9
10
No. of gels compared
8 vs. 8
200 mg: 250 mg: 200 mg: 250 mg:
12.4 6 11.3 32 individual 10.0 6 10.0 midbrains 12.1 6 11.1 9.5 6 9.2
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities.
© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com
Technology
Proteomics 2007, 7, 513–523
521
Table 5. Comparison of two experimental groups with a 25% difference in protein load – experiments with higher protein loads
No. of experiment
Protein load (mg)
No. of gels compared
No. of spots matched
p,0,05 (t-test) Spot number
Percent (%)
8 (For comparison)
200 250
12 12
683
529
77.5
11
500 625
12 12
795
722
90.8
12
800 1000
12 12
725
384
53.0
CVa) (%)
Brain area
200 mg: 16.7 6 4.7 250 mg: 15.9 6 4.6
The same 24 individual 500 mg: 14.2 6 11.6 midbrains for 625 mg: 14.2 6 14.9 all experiments
800 mg: 27.1 6 18.2 1000 mg: 29.1 6 20.7
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities.
Table 6. Comparison of two experimental groups with a 50% difference in protein load (200 mg vs. 300 mg) – experiment after methodological improvements
No. of experiment
13
No. of gels compared
No. of animals pooled
No. of spots matched
8 vs. 8
2
6 vs. 6 4 vs. 4
p,0.05 (t-test)
CVa) (%)
Brain area
32 individual midbrains
Spot number
Percent (%)
505
493
97.2%
200 mg: 12.4 6 11.1 300 mg: 12.9 6 15.6
2
505
485
96.0%
200 mg: 12.1 6 11.1 300 mg: 12.8 6 17.0
2
505
475
94.1%
200 mg: 12.6 6 10.3 300 mg: 13.0 6 15.5
a) The CV is the mean of the SD of all matched spots per group, expressed as a percentage of the mean value of all spot intensities.
In quantitative group comparisons performed by the gelbased DIGE technology, the gel-to-gel variance is abolished. However, the between-gel ratios of the proteins labeled with different CyDyes may be considerable [12]. In a power analysis based on their own work, Karp and Lilley [37] calculated that four DIGE gel replicates are required to reliably detect a 100% difference in protein amount in 80% of all spots. Moreover, no fewer than 18 gels are necessary to detect a 25% difference in spot intensities in 80% of the spots. However, improving the between-gel CV should also reduce the gel-togel variance of the CyDye ratios and therefore the number of replicate gels required for quantitative group comparisons by means of the DIGE technology. Although our results may represent some progress for gel-based quantitative proteomics, many problems are still waiting to be solved. Firstly, gel-based quantification is only possible within the well-known limits of the 2-DE methodology, e.g. hydrophobic and very low-abundant proteins are under-represented [7], for a review see [6]. Secondly, our experiments were performed exclusively with gels covering the pH 4–7 range. Preliminary experiments performed in the context of the experiments presented here showed that © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
for a reliable comparison of groups of gels with basic (e.g. pH 6–11) or narrow pH ranges a number of additional problems need to be solved. Optimal focusing and high quality 2-DE images are much more difficult to achieve than images in the pH 4–7 range. For example, horizontal smears (in particular on pH 7–11NL gels), irregular spot shapes (e.g. triangles and trapezoids on pH 6–9 gels) and geometric distortions may seriously impair between-gel reproducibility and reliable quantification. Extensive modifications of many different variables (e.g. kVh) have to be tested in order to achieve acceptable results. One particularly important variable is the mode of sample application, i.e. by cup-loading or by in-gel rehydration (for a review see [38]). Barry et al. [39] reported that the matching efficiency of cup-loading was consistently far better than that of rehydration methods when IPG strips in the range of pH 6–11 were used. We did not use cup-loading here, as we are not aware of studies showing that cup-loading results in better between-gel reproducibility in the pH 4–7 range. However, another source of between-gel variation may be sample loss, which can be considerable, in particular during rehydration and equilibration [40, 41], reviewed in [38]. www.proteomics-journal.com
522
M. Eravci et al.
Both Zuo and Speicher [40] and Zhou et al. [41] reported that sample loss was considerably lower after cup-loading than after in-gel rehydration when IPG strips covering the pH 3– 10 range were used. It would therefore seem worthwhile to conduct experiments comparing between-gel reproducibility after cup-loading vs. the in-gel technique also for gels with a pH range of 4–7 or “zoom-gels” with narrow pH ranges. In our study, we performed the equilibration simultaneously for all gels in a chamber constructed by our own group. Thus, the protein losses incurred during equilibration were most likely more “symmetrical” between samples than is commonly the case when each strip is equilibrated individually. Thirdly, it is now well known that on 2-DE gels one spot may represent several different proteins, e.g. [8, 9], even when IPG gels covering only one pH range are used [7]. This may result in ambiguous or even false protein identification by MS. This problem could theoretically be best approached by labeling proteins with different isotopes post-2-DE. For example, post-2-DE labeling of peptides from spots with the highest and lowest intensities, respectively, as evaluated after fluorescent labeling, would allow for unambiguous identification by LC-MS. However, to our knowledge such a method has not yet been developed. Detection of quantitative differences in spot intensities not detected by fluorescent labeling is, however, only possible when samples are prelabeled with different isotopes and all spots analyzed by MS [18, 19]. We therefore conclude that gel-based quantitative comparisons still have limitations, some of which may be overcome by combining them with MS-based methods. Finally, it is clear that our data are specifically derived from the analysis of homogenates from different regions of rat brain. The results may be better if less complex tissue (such as cell cultures) is analyzed or worse when the biological variance of the proteome pattern is increased, e.g. after pharmacological treatments or in diseased tissues. Considering the enormous complexity of the brain, subcellular proteomics is considered a most promising approach [42]. Therefore, the between–gel CV of gels loaded with protein from subcellular fractions of brain tissue, such as synaptosomes, nuclei or myelin, should now also be investigated. We intentionally included different areas of rat brain in our analyses in order to investigate whether the reliability of between-group comparisons is comparable between brain areas. This was the case. However, this approach may have compromised the interpretation of our results to a certain extent. For example, individual midbrains were used when we compared groups consisting of two pools each (Table 4, Experiment 9). In contrast, individual septi were used when we compared groups consisting of four and five pooled animals, respectively (Table 5, Experiment 10). The CV were much higher when four or five biological replicates were pooled and smaller numbers of gels were compared with each other (Table 5). In our design, we cannot completely be sure whether this result was attributable to the different number of pooled replicates or whether it was, at least in part, because a different brain area was being analyzed. © 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Proteomics 2007, 7, 513–523
4
Concluding remarks
In recent years, gel-based, comparative proteomic studies have contributed a rapidly increasing number of significant new findings on the pathophysiology of neuropsychiatric disorders, e.g. [43, 44], and promising projects such as the HUPO brain proteome project will further enhance our understanding of brain physiology and pathophysiology, e. g. [45, 46]. Our initial results from comparative proteomic analysis of brain tissue show that it is advisable for each laboratory to conduct its own power analyses, as already emphasized by others [27]. Furthermore, following the elimination of several factors responsible for gel-to-gel variability and the construction of new apparatuses allowing the processing of 24 gels in parallel quantitative proteomic analysis throughout the whole 2-DE procedure the between-gel CV improved considerably. This allowed the detection of significant differences in spot intensity of between 77 and 90% of all matched spots on gel groups with a 25% difference in protein load. The methodological work presented here may therefore improve the reliability of the of the quantitative proteomic analysis of brain tissue.
The authors would like to thank Thomas Forster, Dieter Breuer and their colleagues from the technical workshop of the Charité University Medical Center for their invaluable help and support.
5
References
[1] Moritz, B., Meyer, H. E., Proteomics 2003, 3, 2208–2220. [2] Righetti, P. G., Campostrini, N., Pascali, J., Hamdan, M., Astner, H., Eur. J. Mass Spectrom. 2004, 10, 335–348. [3] Righetti, P. G., Castagna, A., Antonucci, F., Piubelli, C. et al., J. Chromatogr. A 2004, 1051, 3–17. [4] Ong, S. E., Mann, M., Nat. Chem. Biol. 2005, 1, 252–262. [5] Domon, B., Aebersold, R., Science 2006, 312, 212–217. [6] Görg, A., Weiss, W., Dunn, M., Proteomics 2003, 4, 3665– 3685. [7] Gygi, S. P., Corthals, G. L., Zhang, Y., Rochon, Y., Aebersold, R., Proc. Natl. Acad. Sci. USA. 2000, 97, 9390–9395. [8] Zhu, W., Venable, J., Giometti, C. S., Khare, T. et al., Electrophoresis 2005, 26, 4495–4507. [9] Campostrini, N., Areces, L. B., Rappsilber, J., Pietrogrande, M. C. et al., Proteomics 2005, 5, 2385–2395. [10] Ünlü, M., Morgan, M. E., Minden, J. S., Electrophoresis 1997, 18, 2071–2077. [11] Tonge, R., Shaw, M. M., Middleton, B., Rowlinson, R. et al., Proteomics 2001, 1, 377–396. [12] Karp, N. A., Kreil, D. P., Lilley, K. S., Proteomics 2004, 4, 1421– 1432. [13] Gygi, S. P., Rist, B., Gerber, S. A., Turecek, F. et al., Nat. Biotechnol. 1999, 17, 994–999.
www.proteomics-journal.com
Proteomics 2007, 7, 513–523
Technology
523
[14] Washburn, M. P., Ulaszek, R. R., Yates, J. R. 3rd, Anal. Chem. 2003, 75, 5054–5061.
[30] Rabilloud, T., Strub, J.-M., Luche, S., Girardet, J. L. et al., Proteome 2000, 1, 1–14.
[15] Ross, P. L., Huang, Y. N., Marchese, J. N., Williamson, B. et al., Mol. Cell. Proteomics 2004, 3, 1154–1169.
[31] Challapalli, K. K., Zabel, C., Schuchhardt, J., Kaindl, A. M. et al., Electrophoresis 2004, 25, 3040–3047.
[16] Kuhn, K., Prinz, T., Schäfer, J., Baumann, C. et al., Proteomics 2005, 5, 2364–2368.
[32] Wheelock, A. M., Buckpitt, A. R., Electrophoresis 2005, 26, 4508–4520.
[17] Pasquarello, C., Sanchez, J.-C., Hochstrasser, D. F., Corthals, G. L., Rapid Commun. Mass Spectrom. 2004, 18, 117–127.
[33] Blomberg, A., Blomberg, L., Norbeck, J., Fey, S. J. et al., Electrophoresis 1995, 16, 1935–1945.
[18] Smolka, M., Zhou, H., Aebersold, R., Mol. Cell. Proteomics 2002, 1, 19–29.
[34] Nishihara, J. C., Champion, K. M., Electrophoresis 2002, 23, 2203–2215.
[19] Schmidt, A., Kellermann, J.,Lottspeich, F., Proteomics 2005, 5, 4–15.
[35] Mahon, P., Dupree, P., Electrophoresis 2001, 22, 2075–2085.
[20] Wang, G., Wu, W. W., Zeng, W., Chou, C.–L., Shen, R.-F., J. Proteome Res. 2006, 5, 1214–1223.
[37] Karp, N. A., Lilley, K. S., Proteomics 2005, 5, 3105–3115.
[21] Old, W. M., Meyer-Arendt, K., Aveline-Wolf, L., Pierce, K. G. et al., Mol. Cell. Proteomics 2005, 4, 1487–1507. [22] Julka, S., Regnier, F., J. Proteome Res. 2004, 3, 350–363. [23] Marengo, E., Robotti, E., Gianotti, V., Righetti, P. G. et al., Electrophoresis 2003, 24, 225–236. [24] Molloy, M., Brzenzinski, E. E., Hang, J., McDowell, M. T., VanBogelen, R., Proteomics 2003, 3, 1912–1919. [25] Hunt, S. M. N., Thomas, M. R., Sebastian, L. T., Pedersen, S. K. et al., J. Proteome Res. 2005, 4, 809–819. [26] Marengo, E., Robotti, E., Antonucci, F., Cecconi, D.et al., Proteomics 2005, 5, 654–666. [27] Wilkins, M. R., Appel, R. D., Van Eyk, J. E., Chung, M. C.et al., Proteomics 2006, 6, 4–8. [28] Glowinski, J., Iversen, L. L., J. Neurochem. 1966, 13, 655– 669. [29] Bradford, M. M., Anal. Biochem. 1979, 72, 248–254.
© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[36] Choe, L. H., Lee, K. H., Electrophoresis 2003, 24, 3500–3507. [38] Görg, A., Obermaier, C., Boguth, G., Harder, A. et al., Electrophoresis 2000, 21, 1037–1053. [39] Barry, R. C., Alsaker, B. L., Robison-Cox, J. F., Dratz, E. A., Electrophoresis 2003, 24, 3390–3404. [40] Zuo, X., Speicher, D. W.Electrophoresis 2000, 21, 3035–3047. [41] Zhou, S., Bailey, M. J., Dunn, M. J., Preedy, V. R., Emery, D. P. W., Proteomics 2005, 5, 2736–2747. [42] Tribl, F., Marcus, K., Bringmann, G., Meyer, H. E. et al., J. Neural Transm. 2006, 113, 1041–1054. [43] Beasley, C. L., Pennington, K., Behan, A., Wait, R. et al., Proteomics 2006, 6, 3414–3425. [44] Piubelli, C., Fiorini, M., Zanusso, G., Milli, A. et al., Proteomics 2006, 6, S256–S261. [45] Hamacher, M., Marcus, K., van Hall, A., Meyer, H. E., Stephan, C., J. Neural Transm. 2006, 113, 963–971. [46] Schulenborg, T., Schmidt, A., van Hall, A., Meyer, H. E. et al., J. Neural Transm. 2006, 113, 1055–1073.
www.proteomics-journal.com
