Created: 1st September 1999, last updated: 12th November 1999, © 1999 ABRF
P. Hunziker,a T. T. Andersen,b Y. Bao,c S. A. Cohen,d N. D. Denslow,e J. D. Hulmes,f A. M. Mahrenholz,g K. Mann,h K. M. Schegg,i K. A. West,j and J. W. Crabbj
aBiochemisches Institut, University of Zurich, Zurich, Switzerland; bDepartment of Biochemistry and Molecular Biology, Albany Medical College, Albany, NY 12208; cDepartment of Microbiology, University of Virginia Medical School, Charlottesville, VA 22908; dWaters Corp., Milford, MA 01757; eDepartment of Biochemistry and Molecular Biology, University of Florida, Gainesville, FL 32610; fImClone Systems Inc, New York, NY 10014; gDepartment of Biochemistry and Molecular Biology, Indiana University School of Medicine, Indianapolis, IN 46202; hMax-Planck-Institut Biochemie, 82152 Martinsried, Germany; iDepartment of Biochemistry, University of Nevada, Reno, NV 89557; and jThe Eye Institute, Cleveland Clinic Foundation, Cleveland, OH 44195.
The ABRF amino acid analysis study evaluated the general utility of amino acid analysis (AAA) for identification of proteins after denaturing gel electrophoresis and electroblotting to polyvinylidene difluoride (PVDF) membrane. Thirty-eight participating laboratories analyzed a known control (ovalbumin, 5 µg applied to the gel) and either lysozyme or bovine serum albumin as unknown samples (1-, 5-, and 10-µg amounts applied to the gel). Analyses of the unknowns yielded average compositional errors of approximately 30%, 19%, and 18%, respectively, from the low, intermediate, and higher sample amounts; the ovalbumin control exhibited an approximately 17% average error. Compositional data were submitted to the ExPASy and PROPSEARCH Internet sites for protein identification. Without search parameter adjustments or restrictions, both computer programs provided identification of about 20%, 66%, and 74% of the data from the 1-, 5-, and 10-µg gel samples, respectively. Deleting problematic data (Gly, Met, and Pro) did not always facilitate protein identification. Incorporating control results into the ExPASy search increased identifications 2% to 10%, and restricting search parameters by species, isoelectric pH, and molecular weight increased identifications by more than 80%. Average amounts analyzed for correct identifications were approximately 0.4 µg, 1.8 µg, and 2.9 µg for the 1-, 5-, and 10-µg gel samples, respectively. The results support the efficacy of AAA in the low microgram and nanogram range for the identification of PVDF-immobilized proteins from two-dimensional gels. (J Biomol Tech 1999;10:129-136)
Key words: amino acid analysis, polyvinylidene difluoride (PVDF) membrane, protein identification.
Address correspondence and reprint requests to P. Hunziker, Biochemisches Institut, Universitat Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland (email: Peter.Hunziker@access.unizh.ch).
During the past decade, the Association of Biomolecular Resource Facilities (ABRF) has evaluated the efficacy of amino acid analysis (AAA) on polyvinylidene difluoride (PVDF) membrane three times--first in 1990, again in 1995, and most recently in 1997. Both of the first two studies1,2 demonstrated generally higher errors of analysis (21% to 27% average error) relative to nonimmobilized samples. The 1995 study also revealed significantly high background on blank PVDF membrane,2 suggesting that the higher error in analyses on PVDF may in part result from background contamination from the membrane. This report describes the 1997 ABRF AAA study that was designed to probe whether PVDF-electroblotted samples exhibit less error than PVDF-absorbed samples and to evaluate resource laboratory capability for identifying proteins on PVDF using AAA. Our results complement those from the 1996 ABRF AAA study,3 which demonstrated that the most AAA data from 5 µg of a lyophilized unknown provided identification of the protein through the Internet computer query programs PROPSEARCH4 and ExPASy.5 The 1997 ABRF AAA study supports AAA coupled with PROPSEARCH and ExPASy as a practical approach to the identification of PVDF immobilized proteins in the low microgram and nanogram range.
Chick ovalbumin and chick egg white lysozyme were purchased from Sigma (www.sigma.com); bovine serum albumin (BSA) was obtained from the National Bureau of Standards. Electrophoresis sample solutions were prepared containing ovalbumin at 0.25 mg/mL plus lysozyme and BSA at 0.05 mg/mL, 0.25 mg/mL, or 0.5 mg/mL. Equal-volume aliquots (20 µL) of each of the three sample solutions were applied to BioRad Ready Gels (www.bio-rad.com) (12% Tris glycine gels with 4% stacking gels), resulting in 1-µg, 5-µg, and 10-µg amounts applied of each unknown plus 5 µg of the ovalbumin control. Electrophoresis was performed according to Laemmli6 using a buffer composed of 25 mM Tris, 192 mM glycine, and 0.1% sodium dodecyl sulfate (SDS, pH 8.3). The samples were run for 15 minutes at 50 V to allow the sample to enter the gel and then for 1 hour at 120 V to separate the proteins. Samples were electrotransferred to PVDF (Immobilon PSQ, Millipore, www.millipore.com) in a buffer system of 10 mM MES (pH 6), 20% methanol, 0.01% SDS for 4 hours at 50 V. The resulting blots were stained with Coomassie blue, destained, and dried.
Each participant received a piece of PVDF containing an unknown protein (BSA or lysozyme) electroblotted from the 1-, 5-, and 10-µg gel samples plus the ovalbumin control. A photocopy of a two-dimensional (2D) gel containing the unknown proteins and appropriate markers was included so that the molecular weight and isoelectric point of the unknowns could be estimated. Each participant was asked to analyze all four bands plus an unstained section of the blot as a blank and to submit their data to the ExPASy (http://www.expasy.ch/ch2d/aacompi.html), PROPSEARCH (http://www.embl-heidelberg.de/aaa.html), or both Internet sites to identify their unknown protein. Participating laboratories also submitted their raw data (total picomoles of each amino acid for each sample analyzed) to the ABRF Amino Acid Analysis Research Committee through an independent collaborator who removed identifiers to keep the data anonymous. The Committee calculated compositional error and amount of protein analyzed as described elsewhere1-3,7-13 and submitted the data to the PROPSEARCH and ExPASy sites for protein identification. For both sites, searches were performed without and with filtering of potential matches by species (ExPASy site only), molecular weight, and isoelectric point.
Thirty-eight laboratories returned 40 data sets to the Amino Acid Analysis Committee (two laboratories returned two data sets each from two different AAA methods). However, not all laboratories analyzed all levels of the unknown nor the ovalbumin control. Thirty-one data sets were received for 1 µg of unknown, 35 for 5 µg of unknown, 37 for 10 µg of unknown, and 38 for the 5-µg ovalbumin control. Twenty laboratories submitted data obtained from a blank portion of the PVDF. About 56% of the data was obtained from precolumn, as opposed to postcolumn, AAA methods. This level of usage is comparable to that seen in the 1996 ABRF AAA study3 but represents a slight increase over that seen for precolumn AAA techniques in other recent ABRF studies.2,12 The most popular AAA methodologies in 1997 were the postcolumn Ninhydrin method (~37% of the data), the precolumn PTC method (~29%), and the AQC method (~22%), the use of which was increased relative to other ABRF studies.1-3,7-12,14 Other AAA instrumentation used by one or two participants included precolumn o-phthalaldehyde (OPA), precolumn OPA/9-fluorenylmethoxycarbonyl (FMOC), postcolumn Fluram, and postcolumn OPA.
The average compositional error of each data set is shown in Figure 1. The lowest amounts (1-µg gel samples) were the most challenging, yielding an average error of 30.3 ± 14.0% (n = 31). The average compositional error for the 5-µg unknowns (19.2 ± 10.2%, n = 35) was comparable to that obtained with the 10-µg unknowns (18.5 ± 11.0%, n = 37) and to the 5-µg ovalbumin control (17.2 ± 7.1%, n = 38). As can be seen in Figure 1 and Table 1, the accuracy of the results were independent of instrumentation and method of analysis. This observation supports the conclusion of previous ABRF studies that the methodology used for AAA is far less important than the skill and care of the operator. Interestingly, a slightly greater accuracy was obtained from analysis of BSA (20.0 ± 12.1%, n = 52) than lysozyme (24.1 ± 13.6% error, n = 50). This is a sample-related issue, because the groups receiving the two unknowns obtained the same accuracy from analyses of the ovalbumin control (16.9% and 17.2% error, respectively). Figure 2 shows that the most error-prone amino acids in this study were Gly, at least for the 1-µg samples, and Met and Pro, which in lysozyme are rare residues. These amino acids are also subject to exaggerated errors from destruction or co-eluting by-products.
FIGURE 1. Compositional accuracy. Average percent error is shown for all submitted data from amino acid analysis (AAA) on PVDF membrane of the 1-, 5-, and 10-µg unknown and 5-µg ovalbumin gel samples. Shading of the bars indicates the AAA methodology used for each analysis. OPA, o-phthalaldehyde; AQC, 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate; PTC, phenylthiocarbamoyl; FMOC, 9-fluorenylmethoxycarbonyl.
TABLE 1
Average Percent Error Values for the Major Chemistries
|
|
||||||||
| Chemistry Used |
|
|
|
|
||||
|
|
||||||||
| PTC | 31.8 ± 14.9 (9) | 19.9 ± 6.2 (12) | 15.6 ± 4.6 (9) | 19.4 ± 5.8 (12) | ||||
| Ninhydrin | 29.1 ± 6.8 (10) | 19.9 ± 11.4 (11) | 19.2 ± 10.8 (15) | 16.0 ± 5.5 (12) | ||||
| AQC | 31.8 ± 21.6 (8) | 19.3 ± 15.4 (8) | 21.5 ± 17.2 (8) | 16.8 ± 11.2 (8) | ||||
| All Chemistries | ||||||||
| Average | 30.3 ± 14.0 (31) | 19.2 ± 10.2 (35) | 18.5 ± 11.0 (37) | 17.2 ± 7.1 (38) | ||||
| Median | 27 | 17.5 | 16.3 | 16.9 | ||||
|
|
||||||||
The values shown are average percent error. The major chemistries shown represent 35 of 38 sites reporting. Numbers in parentheses indicate the number of facilities using that chemistry for the sample indicated.
FIGURE 2. Problematic residues. Error for each amino acid was averaged over all analyses, revealing Gly, Met, and Pro as the most problematic residues in this study. Black bars, data for 1-µg unknown; cross-hatched bars, data for 5-µg unknown; gray bars, data for 10-µg unknown; white bars, data for 5-µg ovalbumin.
The level of amino acids in PVDF blanks ranged from 0.012 to 0.4 µg total amino acids (one data set excluded), with an average of 0.13 µg (n = 19); this was more than 10% of the predicted sample mass at the lowest level. Lower background levels were reported by laboratories exhibiting lower error, suggesting that sample handling downstream of the electroblotting procedure also contributed to background contamination. However, low background was not a sufficient condition for high accuracy; several sites with relatively high error reported low background levels.
Overall, 75% of the participating laboratories correctly identified the unknown proteins without regard to species from their own PROPSEARCH or ExPASy queries; 48% of the participants identified the protein and the correct species. A summary of protein identifications using these two search programs along with average error and amount analyzed from all the submitted data sets is presented in Table 2. Without regard to species and without adjustments or restrictions, both computer programs provided correct identification for about 20% of the data sets from the 1-µg unknowns, about 66% from the 5-µg unknowns, about 74% from the 10-µg unknowns, and about 28% of the ovalbumin control data. Identification of unknowns correlated with the accuracy of the AAA but with a rather wide window of permissible error. For example, the data resulting in correct identifications using PROPSEARCH had scores of 2.32 or lower in all cases and most of these data exhibited average compositional errors of 20% or less (Fig. 3). However, six data sets with 20% to 30% average errors were also identified correctly (Fig. 3). When using the ExPASy program with ovalbumin control data, successful identifications increased for all levels (Table 2) but most significantly for the 1-µg gel samples.
TABLE 2
Correlation of Identification With Percent Error
|
|
||||||||||||||||||
|
|
||||||||||||||||||
|
|
||||||||||||||||||
| Propsearch | Without Control | With Control | ||||||||||||||||
|
|
|
|
||||||||||||||||
Sample |
% Correct (n) |
Avg Error (%) |
Avg µg Analyzed |
% Correct (n) |
Avg Error (%) |
Avg µg Analyzed |
% Correct (n) |
Avg Error (%) |
Avg µg Analyzed |
|||||||||
|
|
||||||||||||||||||
| 1 µg | 18.8 (6) | 17.1 ± 6.5 | 0.43 ± 0.11 | 21.9 (7) | 19.4 ± 8.5 | 0.45 ± 0.11 | 31.3 (10) | 21.1 ± 7.5 | 0.40 ± 0.12 | |||||||||
| 5 µg | 62.9 (22) | 14.4 ± 5.8 | 1.83 ± 1.25 | 68.6 (24) | 14.5 ± 5.5 | 1.79 ± 1.21 | 71.4 (25) | 15.5 ± 5.7 | 1.63 ± 1.23 | |||||||||
| 10 µg | 75.7 (28) | 13.4 ± 6.2 | 2.99 ± 1.75 | 73.0 (27) | 14.5 ± 6.5 | 2.88 ± 1.77 | 78.4 (29) | 17.0 ± 10.6 | 2.81 ± 1.73 | |||||||||
| Ovalbumin | 27.0 (10) | 10.8 ± 3.5 | 2.34 ± 1.44 | 29.7 (11) | 10.2 ± 3.2 | 2.18 ± 1.50 | ||||||||||||
|
|
||||||||||||||||||
Percent correct refers to the correct identification of the protein without regard to species. Average error refers to the average percent error of those analyses that yielded correct identification. Micrograms analyzed refers to the average amount of protein found by amino acid analysis, and n is the number of data sets.
FIGURE 3. Correlation of PROPSEARCH scores with compositional accuracy. PROPSEARCH scores describe the fit of the experimentally determined amino acid composition with that of a known protein; scores of 1.0 correspond to 99% reliability of a correct identification.4 The correlation of PROPSEARCH assigned scores with average compositional percent error is shown for the unknowns analyzed at all levels. Circles, data identifying the correct protein and species as the number 1 choice; squares, data identifying the correct protein but not the species as the number 1 choice; triangles, data identifying an incorrect protein as the number 1 choice.
As shown in Figure 4, when potential matches using ExPASy were filtered with regard to species and over a limited range of isoelectric points and masses, the number of identifications increased even more dramatically. With restrictions for species, isoelectric point, and molecular weight, correct identifications were obtained from more than 60% of the data from the 1-µg gel samples and more than 80% of the data from the 5-µg and 10-µg unknown and control samples. ExPASy searches were also performed after omitting the high error results associated with Gly, Met, and Pro (Fig. 2). For all samples, improved score numbers were obtained when Met was omitted. The deletion of Gly significantly improved the score numbers only at the 1-µg level (Table 3). In all other cases, neither deleting all three amino acids together or individually reliably enhanced protein identification.
FIGURE 4. Extra search parameters enhance identifications by amino acid analysis. Results from ExPASy performed using extra search parameters are shown for the 1-, 5-, and 10-µg unknowns and 5-µg ovalbumin electrophoresis samples. White bars, no search restrictions employed; black bars, search restricted to a known species; gray bars, search restricted to a known species within a defined molecular weight and isoelectric point range. A correct identification was defined as the correct protein (but not necessarily species) selected as the number 1 ranked protein.
TABLE 3
ExPASy Score Changes on Omitting High-Error Amino Acids From Composition
Data
|
|
||||||
Sample |
Lower Score |
No Change |
Higher Score |
|||
|
|
||||||
| 10 µg | ||||||
| Without Gly | 35 | 19 | 46 | |||
| Without Met | 46 | 46 | 8 | |||
| Without Pro | 27 | 30 | 43 | |||
| Without Gly, Met, and Pro | 41 | 18 | 41 | |||
| 5 µg | ||||||
| Without Gly | 40 | 20 | 40 | |||
| Without Met | 46 | 37 | 17 | |||
| Without Pro | 17 | 26 | 57 | |||
| Without Gly, Met, and Pro | 46 | 14 | 40 | |||
| 1 µg | ||||||
| Without Gly | 75 | 6 | 19 | |||
| Without Met | 41 | 34 | 25 | |||
| Without Pro | 38 | 14 | 50 | |||
| Without Gly, Met, and Pro | 78 | 3 | 19 | |||
| Ovalbumin | ||||||
| Without Gly | 47 | 11 | 42 | |||
| Without Met | 84 | 8 | 8 | |||
| Without Pro | 29 | 16 | 55 | |||
| Without Gly, Met, and Pro | 84 | 8 | 8 | |||
|
|
||||||
Lowering the distance score indicates a better fit and is an improvement, while a higher score indicates increased distance from theory. Results are expressed as a percentage of the total for the group.
The amounts analyzed were significantly lower than the amounts applied to the gels (Table 2). For example, for the data used for the identifications in Figure 4 (with restrictions for species, molecular weight and isoelectric point), the corresponding amounts recovered by AAA were 0.4 ± 0.1 µg, 1.9 ± 1.1 µg, and 3.2 ± 1.6 µg from the 1-, 5-, and 10-µg gel samples, respectively. Average compositional error associated with these analyses was 23.4 ± 8.0% (1-µg gel samples), 16.3 ± 7.0% (5-µg gel samples), and 15.9 ± 8.3% (10-µg gel samples).
The 1997 ABRF AAA study sought to examine whether PVDF-electroblotted samples exhibit less error and background contamination than PVDF-absorbed samples and to probe the general utility of AAA for identifying proteins on PVDF. A marginal improvement in compositional accuracy was observed in analyses of PVDF-electroblotted samples compared with PVDF-absorbed samples. An approximately 19% average error was observed overall in 1997 from the 5- to 10-µg gel samples compared with an approximately 21% average error from the 5 to 10 µg of protein absorbed to PVDF in the 1995 study (9). The average accuracy of ABRF AAA studies on PVDF1,2 remains inferior to comparable analyses of nonimmobilized samples. For example, an 11.9% average error (n = 71) was obtained from analyses of 5 µg of lyophilized protein in the 1996 ABRF AAA study.3 To further explore possible ways to reduce background contamination, blank pieces of PVDF were analyzed in this study after electroblotting. Although some laboratories reported very little contamination, others found blank PVDF containing up to 400 ng of amino acids. Higher background was observed in the 1995 ABRF AAA study, which reported two to three times more contamination associated with some solvent-washed PVDF blanks. Nevertheless, differences in size of the PVDF pieces distributed in 1995 and 1997 preclude a definitive conclusion that electroblotted PVDF contains less background. The current results emphasize the difficulty associated with obtaining high accuracy amino acid analyses on PVDF and reinforce the necessity for meticulously careful sample handling techniques. Although accuracy enhances the usefulness of compositional data for protein identifications, the results also show that the available computer query programs are relatively tolerant to error.
Successful protein identifications in the 1997 study increased with the accuracy of the data, with the greatest number of identifications from compositions exhibiting 5% to 15% error, as found in the 1996 AAA study.3 Nevertheless, 20% to 30% average compositional error in some cases did not prohibit correct identifications. Protein identification from AAA data was facilitated by taking advantage of select search parameters. For example, ExPASy allows entry of data from a known protein that has been analyzed in parallel with the unknown. This entry has the effect of correcting for systematic error. When data for unknowns were submitted to ExPASy with the amino acid values of a known protein, the total number of correct identifications increased about 10% for the lowest level analyses. Such control data appear particularly appropriate when submitting a number of data sets in a batch for computer searches. Filtering of potential matches by species, molecular weight, and isoelectric point should be included in searches whenever possible, because they enhance the probability of a correct identification. These results demonstrated a dramatic increase in all identifications when these filters were incorporated into the search parameters. However, deleting data for problematic amino acids was of little benefit for most of the samples and does not appear to be routinely warranted. Only the omission of Gly for the 1-µg unknowns and of Met for ovalbumin significantly improved the search results.
Most participating resource laboratories correctly identified the unknown proteins in this study from their own efforts and the easy-to-use PROPSEARCH and ExPASy programs. These results exemplify the potential power of the technology. As proteomics research expands and the use of 2D gels grows, AAA on PVDF offers a reasonably sensitive approach to protein identification with high-throughput capacity.15 In this regard, applications of the technology can be useful for independent identifications and for confirming tentative identifications obtained from mass spectrometry, Edman degradation, and Western analysis.
We would like to thank Dr. Umit Yüksel (Crylolife, Inc.) for tabulating the data and maintaining the anonymity of the participants and Dr. M. R. Wilkins for assistance with batch submission of searches. This work was supported in part by DOE grant DE-FG02-95ER61839 to J. W. Crabb, on behalf of the ABRF.
1. Tarr GE, Paxton RJ, Pan YCE, Paxton RJ. In Techniques in Protein Chemistry II. San Diego: Academic Press, 1991:139-150.
2. Mahrenholz AM, Denslow ND, Andersen TT, et al. In Marshak DR (ed): Techniques in Protein Chemistry VII. San Diego: Academic Press, 1996:323-330.
3. Schegg KM, Denslow ND, Andersen TT, et al. In Marshak DR (ed): Techniques in Protein Chemistry VIII. San Diego: Academic Press, 1997:207-216.
4. Hobohm U, Houthaeve T, Sander C. Amino acid analysis and protein database compositional search as a rapid and inexpensive method to identify proteins. Anal Biochem 1994:222:202-209.
5. Wilkins MR, Pasquali C, Appel RD, et al. From proteins to proteomes: large scale protein identification by two-dimensional electrophoresis and amino acid analysis. Biotechnology 1996:14:61-65.
6. Laemmli UK. Cleavage of structural proteins during the assembly of the head of bacteriophage T4. Nature 1970;227:680-685.
7. Crabb JW, Ericsson LH, Atherton D, Smith AJ, Kutny R. In Villafranca JJ (ed): Current Research Protein Chemistry. San Diego: Academic Press, 1990:49-61.
8. Ericsson LH, Atherton D, Kutny R, Smith AJ, Crabb JW. In Jörnvall H, Höög JJ, Gustavsson AM (eds): Methods of Protein Sequence Analysis 1990. Basel: Birkhauser Verlag, 1991:143-150.
9. Strydom DJ, Tarr GE, Pan YCE, Paxton RJ. In Angeletti RH (ed): Techniques in Protein Chemistry III. San Diego: Academic Press, 1992:261-274.
10. Strydom DJ, Andersen TT, Apostol I, Fox JW, Crabb JW. In Angeletti RH (ed): Techniques in Protein Chemistry IV. San Diego: Academic Press, 1993:279-288.
11. Yüksel KÜ, Andersen TT, Apostol I, et al. In Crabb JW (ed): Techniques in Protein Chemistry V. San Diego: Academic Press, 1994:231-240.
12. Yüksel KÜ, Andersen TT, Apostol I, Fox JW, Paxton RJ, Strydom DJ. In Crabb JW (ed): Techniques in Protein Chemistry VI. San Diego: Academic Press, 1995:185-192.
13. Crabb JW, West KA, Dodson WS, Hulmes JD. In Coligan JE, Ploegh HL, Smith JA, Speicher DW (eds): Current Protocols In Protein Science, unit 11.9, suppl 7. New York: John Wiley & Sons, 1997:11.9.1-11.9.42.
14. Niece RL, Williams KR, Wadsworth CL, et al. In Hugli TE (ed): Techniques in Protein Chemistry. San Diego: Academic Press, 1989:89-101.
15. Yan JX, Wilkins MR, Ou K, et al. Large-scale amino-acid analysis for proteome studies. J Chromatogr A 1996;736:291-302.