Linkage of MS data and antibodybased measurements

Another valuable use of these data for the PPP was to investigate relationships between the quantitative antibody-based measurements and the MS information derived from other work within the PPP. Based on the informatics integration methods (see Adamski et al., this issue), 9504unique IPI proteins were included in the combined data (see http://www.bioinformatics.med.umich.edu/app1/test/). The link between the MS data and the antibody-based measurements was made through IPI numbers. Two different search methods were used to find IPI numbers that corresponded to the analytes measured in the quantitative antibody-based assays (see Section 2), generating two lists of analyte-associated IPI numbers. Seventy IPI numbers that were common between these lists and the MS summary data were identified and are presented in Tab. 2. In four cases, two IPI numbers were associated with the same analyte name. Tab. 2 also gives the average concentration (the geometric mean over all samples, including the NIBSC sample, and all data sets) of each analyte, the number oflaboratories (out of 18) finding that IPI number, and the average number of peptides found for that IPI number. The relationships between the MS summary data and the average concentrations were examined (Fig. 6). Fig. 6A shows that individual laboratories made identifications in the 10-10 000 pg/mL range, with multiple laboratories finding the same IPI numbers above that range. Only single peptide identifications were made below

Fig. 5 Variation in the concentration of individual proteins across different preparation methods (see text for basis of selecting these proteins). Analyte and data set (in parentheses) are indicated in each plot. Replicate data from two to four different samples are plotted with respect to preparation method. Individual values for each sample are shown by the following: BD sample 1: open diamonds;BD sample 2: solid squares;BD

Fig. 5 Variation in the concentration of individual proteins across different preparation methods (see text for basis of selecting these proteins). Analyte and data set (in parentheses) are indicated in each plot. Replicate data from two to four different samples are plotted with respect to preparation method. Individual values for each sample are shown by the following: BD sample 1: open diamonds;BD sample 2: solid squares;BD

sample 3: solid triangles;and CAMS: solid circles. Averages of the replicate data are shown by a solid line for BDAA, a dotted line for BDAF, two dots and a dash for BDCA, and a dashed line for the CAMS specimen. Graph G includes ELISA data that has been normalized to the same scale as the microarray data, represented by darker versions of the corresponding lines for the microarray data.

Fig. 6 MS summary data with respect to concentrations measured by immunoassays and antibody microarrays. Concentration is pg/mL (log base 10). (A) Number of laboratories finding a given protein, (B) number of peptides for each protein identification.

around 200 pg/mL, with a steadily increasing average number of peptides above that (Fig. 6B). Both metrics increased steadily with concentration. The lack of data points in the 1-100 mg/mL range is primarily due to the low number of immunoassay and antibody microarray measurements in that range, as shown in Fig. 2.

Discussion

The analysis of the HUPO PPP reference specimens by antibody-based methods provided a useful complement to the other studies of the PPP. This work examined the use of immunoassays and antibody microarray methods to investigate the systematic variation of specific proteins between the PPP's reference specimens' sample preparation methods and to provide insights into the concentration-dependence of protein discovery by MS methods. The use of four distinct methods from four independent laboratories gave a broad view of the capabilities of antibody-based methods. Each of the four data sets had highly internally reproducible data, as shown by the high average correlations between replicate data, although the values did not always agree in the measurements of common analytes. The occasional lack of concordance between the sets underscores the importance of the use of common IRMs for cross validation and calibration between laboratories and methods. An international reference standard for 15 abundant serum proteins, CRM 470 [14], has been developed; its use has significantly reduced inter-laboratory variation in many protein assays in European quality assurance programs [22]. Of note, DB analyzers used standards, calibrators, and controls based on common IRMs that are generally applied in clinical chemistry. Antibody microarray measurements have not yet achieved the precision standard of clinical analyzers.

We investigated two aspects of the effect of sample preparation on protein concentration: systematic alterations of all proteins in general and consistent alterations in the concentrations of specific proteins. The most common general systematic alteration was a reduction of protein concentrations in the citrate-plasma preparation. This effect is attributable to the dilution of the plasma fraction of whole blood by the sodium citrate solution [23] and by the osmotic withdrawal of water from blood cells caused by the high salt concentration in the anticoagulant. When whole blood at a hematocrit of 0.4-0.5 is mixed with sodium citrate solution at a ratio of 9:1, the dilution of citrated plasma will be 15-19.5% (10% dilution from the citrate solution plus additional dilution from osmosis)

[23]. The concentration reduction in citrate-plasma was the most consistent in the DB data and was explainable by the dilution factor, with 14 of the 17 consistently reduced proteins lower than the serum preparation by less than 20%. We might note that most of the DB analyses were not approved for use with citrate-plasma. The other data sets showed less consistent alterations in the citrate-plasma concentrations, perhaps due to lower precision in the measurements or other sources of variation besides dilution, as discussed below. Of great importance for proteomics analyses, the dilution in citrate-plasma did not seem to affect protein identification in PPP analyses using various fractionation and MS methods, as the citrate-plasma specimens gave similar numbers of proteins identified relative to the other specimen types and similar detection of low-abundance immunoassayed proteins (see Simpson et al., and Omenn et al., this issue).

The preparation method that generally gave the highest protein concentrations varied among the four data sets. The GNF and MSI sets had higher protein abundances in the EDTA-plasma preparation, the DB set had higher abundances in the serum preparation, and the VARI set had highest values in the serum and heparin-plasma. The GNF and MSI sets focused on cytokine detection, and the relatively higher concentration of the cytokines in EDTA-plasma could indicate a protective effect of EDTA on cytokine stability, perhaps through EDTA's role as a protease inhibitor. The more abundant, common serum proteins measured in the other two sets could be less susceptible to protease activity and therefore not necessarily higher in the EDTA-plasma preparation. Other sources of variation in concentration could be the anticoagulant-induced release of certain analytes by lymphocytes, such as the release of tumor M2-PK in heparin-plasma but not in EDTA-plasma

[24], interference in certain assays by anticoagulants, or variability in protease activity or protein stability due to the presence or absence of certain anticoagulants.

The analysis of specific proteins showed that certain proteins were always highest or always lowest in certain preparation methods. The fact that some of these alterations were counter to the overall trends noted above shows that blood preparation methods can have variable effects on specific proteins or antibodies. Anticoagulants may in some cases specifically interact with certain proteins or specifically affect the stability of certain proteins. Such effects have been seen in previous studies. In one study, the levels of several hormones were either elevated or reduced between matched serum and EDTA-plasma and between matched serum and citrate-plasma samples [25]. Another study showed that parathyroid hormone is more stable in EDTA-plasma than in serum [26]. The levels of the cytokines IL-6, TNF-a, and leptin were found to be highly variable in citrate-anticoagulated- and heparin-anticoagulated-plasma but not in EDTA-anticoagulated-plasma or serum [27]. In some cases, an anticoagulant might actually bind to specific proteins. For example, EDTA binds to hemoglobin [28], which might be related to the observed consistent elevation of the hemoglobin measurements in the EDTA-plasma samples.

Based on the above observations, it is clear that comparisons between samples are only accurate when using samples that were collected with precisely the same method. Which preparation method to use in every case, however, is less obvious. No single preparation method is optimal for every analyte - the use of certain anticoagulants may interfere with some assays, and the activation of the clotting cascade may be detrimental for other assays. Therefore, the development of assays for individual proteins needs to be evaluated and optimized on a case-by-case basis. The information contained in Supplemental Tab. 3 could be used as a starting point for identifying potential anticoagulant-protein interactions that could affect an assay. Although assays for individual proteins must be individually optimized, it would be advantageous to use a single preparation method for proteomics methods and highly-multiplexed assays. Additional studies with an appropriate number ofsamples ofeach blood preparation method have to be performed to address the optimal blood preparation method for proteomics and highly-multiplexed studies, perhaps focusing on the consistency and stability of analytes rather than simply on concentration.

The final part ofthis study investigated the use ofthe antibody measurements to determine the concentration dependence of MS protein identification, using summary data from 18 different laboratories. A clear dependence on concentration was observed for both the number of laboratories finding certain proteins and the number of peptides found for each protein. It is encouraging that a precipitous decline in identifications at lower concentrations was not observed, but rather a steady decrease through most of the concentration range. Although the likelihood of identifying a protein and the quality of the identifications drop significantly for lower-abundance analytes, identifications were still made in the pg/mL range. Continued refinements and improvements in the technologies should make the identification of low-abundance proteins more common.

These studies demonstrate the benefits of high-throughput, high-precision, and high-sensitivity antibody-based analytical methods. We identified general and specific alterations in the protein concentrations that are related to the blood preparation method. In general, it appears that many cytokines are more stable in EDTA-plasma, specific interactions may occur in some cases with each anticoagulant, and a general dilution occurs with the use of citrate as an anticoagulant. The antibody-based methods also were useful for providing insights in the performance of MS-based protein identifications, showing that low concentration protein identifications are less frequent but still possible. In the continuing projects of the PPP, immunoassays and antibody microarrays will be useful in further studying these and other topics, such as characterizing the variation of many proteins in large populations of samples. Calibration using certified reference standards will be needed to reduce variation between laboratories and platforms.

F.V. thanks Herbert Schwarz, Harald Ackermann, and Lori Sokoll for excellent technical assistance and Carsten Schelp, Mary Lou Gantzer, Fritz Behrens, Manfred Lammers, Alex Rai, and Daniel Chan for helpful discussions. B.B.H. thanks John Marcus for excellent technical work. We thank Marcin Adamski and Raji Menonfor assistance with the PPP data sets and Akhilesh Pandey for helpful discussions on the IPI protein searches. B.B.H.

acknowledges support from the Michigan Proteome Consortium, the Michigan Economic Development Corporation (GR-356), and the Van Andel Research Institute. G.M. was supported by the NIH (grant 1P41RR018627-01) and the NSF (grant IIS-9988085). B.H.G. thanks Professor Peter Schultz and the Genomics Institute of the Novartis Research Foundation for continuing support. G.S.O. was supported by the Michigan Economic Development Corporation (GR-356). We acknowledge trans-NIH (NCIgrant supplement NCI-84982) and corporate support for the overall PPP (see www.hupo.org/ppp for detailed acknowledgements) .-MSI and DB generously contributed resources to the study.

Was this article helpful?

0 0

Post a comment