| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Molecular & Cellular Proteomics 6:1059-1072, 2007.
© 2007 by The American Society for Biochemistry and Molecular Biology, Inc.
,
From the Laboratory of Thrombosis Research, Department of Clinical Research, University of Bern and ¶ Institute of Clinical Chemistry, University Hospital and University Children's Hospital, 3010 Bern, Switzerland
 |
ABSTRACT |
|---|
 TOP ABSTRACT EXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
All the chemical and enzymatic reactions do have shortcomings however. They rely on the assumption that all proteins and their isoforms and all amino acid side chains of one type, e.g. free sulfhydryl group of cysteines, have the same chemical properties and react equally well with any chemical reagent. This is clearly not the case as every protein chemist can confirm. Further disadvantages are high costs and the need to perform pairwise analyses that prevents retrospective or prospective studies with clinical samples. New promising approaches exploit the wealth of data produced by LC-MS/MS of protein digests, such as chromatographic peptide peak intensity (10–12), spectrum sampling (SpS) (12, 13), protein abundance index (PAI) (14, 15), or probabilistic peptide scores (PMSS) (16). PMSS is an extension of spectrum sampling. It adds up all the probabilistic peptide scores, as calculated by the OLAV/PHENYX peptide identification engine, that identify the same protein instead of counting just the number of spectra. It can be assumed that peptide concentrations in two different but compositionally similar samples will be the same when the protein concentrations before digestion were the same. In data-dependent LC-MS/MS it can therefore be inferred that a similar number of fragment spectra on these protein-specific peptides will be conducted. Accordingly the number of spectra identifying one protein will become smaller if the concentration in one sample is lower than in the other one. SpS and PMSS are based on this principle, and Colinge et al. (16) have shown that the latter approach compares favorably with the spectral counting approach. Recent publications have presented evidence that integrated ion chromatographic peak areas may be more robust than SpS for comparative protein quantitation especially if only one or two peptides of one particular protein can be identified (12). However, integrated peak areas can be influenced by different effects including ion suppression, by limited ion trapping capacity of mass spectrometers, or simply by the parameters applied to create extracted ion chromatograms, e.g. m/z tolerance, background subtraction, etc. Fundamental to all these approaches are correct peptide identifications.
Absolute protein quantitation in complex protein mixtures can be determined by the staining intensity of a protein spot on 2DE relative to the total intensity on the gel. However, this approach relies on complete separation of all proteins without any loss, or equal losses between repetitions, in the first dimensional gel and an equal staining efficiency for all proteins. Unfortunately these prerequisites are difficult, if not impossible, to achieve with 2DE. Alternatively isotope-labeled synthetic peptides have been used as internal standards with LC-MS/MS (17). Such an approach is costly when many proteins are to be investigated, and difficulties arise in spiking appropriate amounts of synthetic peptides to avoid suppression effects. Mann and co-workers (14, 15) have shown recently that SpS can be used for absolute protein abundance estimation called PAI or emPAI. PAI is calculated by the number of observed peptides (SpS) divided by the number of observable peptides (defined as theoretically possible peptides produced by enzymatic proteolysis of a protein). For absolute protein quantity PAI is divided by the sum of all PAIs. The emPAI is equal to 10PAI – 1. The authors found that PAI values correlated better with logarithmic protein concentrations for unknown reasons.
Probabilistic identification scores add some kind of quality measure to each spectrum count because the more ions are available for fragmentation the better the resulting fragment spectrum quality can be, resulting in an increased identification score. Therefore, we pursued the possibility to use probabilistic identification scores for protein quantitation of high density lipoproteins.
Lipoproteins shuttle hydrophobic metabolites, especially lipids, from sites of absorption and production to sites of storage and excretion through the blood system of mammals. Derangements in lipid metabolism can have different causes, and elevated lipid levels in blood are a high risk factor for development of cardiovascular disease, especially if hypercholesterolemia is associated with high LDL cholesterol (LDL-C) and low HDL cholesterol (HDL-C) concentrations. It has been recognized that reduced or elevated levels of certain apolipoproteins can cause life-threatening health conditions, including an increased risk to develop atherosclerosis. For example, apoE–/– mice that are deficient in apolipoprotein E, a ligand of lipoprotein receptors, develop atherosclerotic lesions resembling those observed in humans (18). Elevated HDL-C is considered as antiatherogenic due to its reverse cholesterol transport function. It has been postulated that a specific HDL subclass enriched in apoA-I (LpA-I) is the main antiatherogenic factor. However, contradictory literature about this issue exists as reviewed by Alaupovic (19). The different results published might be explained by the use of different antibody-based immunological techniques with different antibodies and standards to measure apolipoprotein concentrations. Possibly due to such controversies, in clinical chemistry laboratories lipoproteins are still characterized only through lipid measurements and not through their protein composition.
Here we present results of a pilot study on protein compositional analysis of total HDL based on antibody-free mass spectrometry technology. The data presented in this study indicate the possibility that such an approach has the potential to clarify existing controversial results, reveals new facts about lipid metabolism mechanisms, and might help to define atherosclerotic risk factors based on HDL protein composition rather than cholesterol levels alone.
|
EXPERIMENTAL PROCEDURES |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
Standard Protein Mixtures—
Bovine serum albumin (ALBU_BOVIN, P02769, molecular mass = 66,433 Da) was from Merck. Bovine fetuin (FETUA_BOVIN, P12763, molecular mass = 36,353 Da) and chicken lysozyme (LYSC_CHICK, P00698, molecular mass = 14,313 Da) were from Calbiochem (VWR International). Recombinant human growth hormone (SOMA_HUMAN, P01241, molecular mass = 22,129 Da) from Novo Nordisk Pharma was generously provided by P. Mullis (Department of Clinical Research, University of Bern, Switzerland). Chicken actin (ACTS_CHICK, P68139, molecular mass = 41,817 Da), jack bean concanavalin A (CONA_CANEN, P02866, molecular mass = 25,616 Da), bovine insulin (INS_BOVIN, P01317, molecular mass = 5,739.6 Da), bovine ß-lactoglobulin 
- and ß-form (LACB_BOVIN, P02754, molecular mass = 18,281 Da), horse holomyoglobin (MYG_HORSE, P68082, molecular mass = 17,568 Da), and rabbit parvalbumin (PRVA_RABIT, P02624, molecular mass = 11,934 Da) were from Sigma. A stock solution of each standard protein was prepared in 20 mM Tris/HCl, pH 8.0, 50 mM NaCl (except SOMA that came as a 10 mg/ml solution), and aliquots were stored at –40 °C. The exact protein concentration of each solution was determined by quantitative amino acid analysis at the Analytical Research and Analysis Facility of the University of Bern (Johann Schaller, Department of Chemistry and Biochemistry, University of Bern, Bern, Switzerland). Defined amounts of these proteins were mixed to create five different mixtures with ALBU, CONA, FETUA, LACB, LYSC, MYG, and THYG building a stable matrix with ACTS, PRVA, INS, and SOMA added in at different ratios as given in Table I. PRVA was contaminated with 17.2% (w/w) triose-phosphate isomerase (TPIS_RABIT, P00939, molecular mass = 26,625 Da).
|
|
Gel Electrophoresis and Protein Identification—
Desalted HDL preparations were analyzed by 1DE and/or 2DE, protein bands from 1DE or protein spots from 2DE were cut from gels, and proteins were digested with trypsin and identified by LC-MS/MS as described previously (20).
Shotgun LC-MS/MS—
Protein samples were diluted to 1.0 mg/ml with 50 mM ammonium bicarbonate. An aliquot of 5 µl was denatured by addition of 1 µl of 0.3 M DTT in water and 6 µl of ACN followed by incubation for 30 min at 40 °C. Samples were diluted with 18 µl of 50 mM ammonium bicarbonate containing 5 mM CaCl2, and digestion of proteins was started by addition of 2 µl of trypsin solution (50 ng/µl) and incubation at 
30 °C on an orbital thermoshaker 5437 (Eppendorf, Dübendorf, Switzerland). After 6 h a new trypsin aliquot was added to complete the digestion overnight. Trypsin activity was quenched by dilution with 0.1% formic acid to a final protein concentration of 33 ng/µl. When not stated differently, 500 ng of protein digest was loaded onto a self-made microbore column (0.15-mm inner diameter x 60-mm length) at a flow rate of 4 µl/min solvent A (0.1% FA in water/ACN (98:2)). Columns were packed with GROM-SIL 300 Octyl-6 MB, 5 µm, reversed phase material (Grom GmbH, Rottenburg-Haiflingen, Germany). Columns were developed by a biphasic ACN gradient of 0–5% solvent B (0.1% FA in water/ACN (4.9:95)) in 1 min followed by 5–40% solvent B in 60 min at a flow rate of 3 µl/min. The column effluent was directly coupled to an Esquire3000+ ion trap mass spectrometer from Bruker Daltonics (Bremen, Germany) via a capillary ESI source operated at 3,700 volts. The gain control (ion charge control) was set to 30,000 with a maximum accumulation time of 200 ms. CID was triggered on the two most abundant, not singly charged peptide ions in the m/z range of 360–1400. Precursors were set in an exclusion list for 0.5 min. CID spectra interpretation was performed with PHENYX on the Vital-IT server operated by GeneBio (Geneva, Switzerland) against the latest release of the UniProt-Swiss-Prot protein database allowing variable modifications of Met oxidation, Asn/Gln deamidation, and formylation of free amino groups. Parent and fragment mass tolerances were set to 2.0 and 0.8 Da, respectively. Up to three missed cleavages and half-tryptic peptides were allowed. All peptide identifications on proteins known to be present in the analyzed sample reaching a p value of 
0.00001 were accepted for protein quantitation calculations. We have shown previously that peptide identifications with higher p values are essentially false positive identifications (21). Identification results were extracted into Microsoft Excel by means of an in-house developed perl script. Each sample was analyzed at least four times. Protein identification of the four LC-MS/MS runs of the synthetic protein mixture 1 was additionally performed with SEQUEST and PeptideProphet validation with the Trans-Proteomic Pipeline software suite version 2.8 (D. Goodlett laboratory, University of Washington, Seattle, WA) and MASCOT (Matrix Science) by using the same search parameters.
Absolute Protein Quantitation, PMSSI—
The total protein score was calculated by summing all peptide Z-scores belonging to the same protein (PMSS). PMSS values were corrected by multiplication by the molecular weight of the native protein (considering no post-translational modifications) as given on the EXPASY website (www.expasy.org) and division by the number of theoretically observable peptides, which was defined as peptides following the trypsin cleavage rules (C-terminal to lysine and arginine, not if proline at P1' position) with zero or one missed cleavage and having a molecular mass between 720 and 3,000 Da (equals the typical mass range of peptide identifications achieved on our setup).
These criteria diverge from what we used as an allowance criteria of positive identification during database search (which was optimized for the most complete list of peptide identifications). We chose these criteria for observable peptides nevertheless because >91% of all peptide identifications were on fully trypsin-specific peptides with zero or one missed cleavage (protein-dependent). Most of the remaining identifications were based on missed cleavages due to in-source fragmentation of fully tryptic peptides. Some chymotryptic cleavages were also observed. These nonspecific trypsin identifications are not predictable; hence they cannot be used to define a generalized criterion for observable peptides. However, these deviations did not contribute significantly to the final results.
The weight fraction percentage (% w/w) or peptide match score summation index (PMSSI) of each protein was calculated by dividing the corrected PMSS value of each individual protein by the sum of all corrected PMSS values essentially as published by Ishihama et al. (15). Peptide ion peak areas were extracted with a visual basic script run on the Bruker DataAnalysis software, version 3.1, with background noise subtraction.
Determination of ApoA-I by Capillary Electrophoresis with UV Detection at 200 nm (CE-UV)—
Samples were solubilized in 2% SDS, and phenylalanine as internal standard was added (1 mg/ml final concentration). After incubation in a boiling water bath for 3 min, the samples were diluted five times in electrophoresis buffer: 50 mM sodium borate, pH 9.1, 0.2% SDS, 5% methanol. Capillary electrophoresis was performed in an open capillary with inner diameter of 50 µm and length of 50 cm at 30 kV and 30 °C for 20 min on a 270HT CE system (Applied Biosystems). UV detection occurred on line at a wavelength of 200 nm. Absolute apoA-I quantitation was performed by comparing the area ratios of the apoA-I and the phenylalanine (internal standard) peak using a linear calibration curve. Purity in percent (%) was calculated from the ratio of the peak areas between apoA-I and the total area of all other protein peaks. All samples were analyzed twice.
Statistical Testing—
The non-parametrical, one-way analysis of variance test by Kruskal-Wallis using 
2 statistics was used to test whether protein concentrations between the samples of the hypercholesterolemia group and the normolipidemia group were statistically different at a significance level of 95%. Statistical testing was performed on the means of all LC-MS/MS experiments and multiple HDL isolations recorded for each blood sample. Non-parametrical statistics was used because the measured values from the limited population size in this study were not normally distributed as determined by a graphical quantile-quantile plot.
|
RESULTS |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
0.8 probability, whereas there was no improvement by applying more stringent thresholds on MASCOT search results by either using an ion score of >24 or an expect value of 1, respectively (the latter result is not shown). PHENYX Z-score correlation remained unchanged (R2 = 0.995) after applying a more stringent acceptance criteria in the form of a p value threshold of 10–8. This result indicated that true probabilistic Z-scores, as calculated by PHENYX, are as good to represent protein abundance as SpS and correlated better than MASCOT ion or PeptideProphet probability scores when only first rank identification was used as the sole acceptance criteria for identification.
|
|
Despite certain signal variations, calculated ratios between peak areas of one representative peptide per protein in mixture 1 and the other four mixtures correlated with the ratios calculated from the theoretical concentrations (R2 = 0.83) (Fig. 2, panel E). This correlation was actually excellent in the range between 0.2 and 3, corroborating the conclusions by others that peak areas can be used for comparative proteomics. We were able to show that within this ratio range the peak area ratios correlated well with PMSS ratio calculations too (R2 = 0.89) (Fig. 2, panel F, and Supplemental Table 1). In summary, peak area ratios are a good tool to validate Z-score-based protein ratios; the latter is quite accurate when dealing with protein concentration differences not bigger than 3–5. At higher ratios, the score-based calculation tends to underestimate the real differences, and proteins with only one or two peptide identifications cannot accurately be quantitated. The underestimation of elevated concentration differences can be explained by a 10-fold smaller dynamic range of peptide identification scores with finite numbers spanning at most 2 orders of magnitude compared with integrated LC-MS peak areas, which can cover 3 orders of magnitude on an ion trap mass spectrometer.
The PMSSI-calculated absolute protein quantities did compare well with the theoretical values as demonstrated in Fig. 3 and Table I. Exceptions were MYG and SOMA, which were determined at higher than theoretical concentrations, and CONA and PRVA, which were detected with lower concentrations than expected. Most tryptic peptides of MYG and SOMA were detected in their +2 and +3 charge state; hence each peptide could be identified twice resulting in higher scores compared with proteins from which peptides were detected most often only with one charge state. CONA and PRVA were only partially digested because only a few peptides could be detected sporadically. Incomplete digestion of a protein is a common fact in protein chemistry and will result not only in a decreased estimation of the concentration of the affected protein but ultimately to an overestimation of the concentration of all other proteins in the sample. However, this error source will apply to all other shotgun LC-MS/MS approaches too. Further validation of the score-based absolute protein quantitation was achieved with HDL samples as outlined further below.
|
1.12 kg/liter. We also tested the possibility of isolating HDL by immunoaffinity column chromatography with a mouse anti-human apoA-I antibody coupled to Sepharose beads (R. James, University of Geneva, Switzerland). However, such preparations seemed to be composed of a specific subpopulation differing clearly in protein composition from any other HDL class isolated with our earlier method (20). Additionally the immunoaffinity preparations were contaminated with abundant plasma proteins including human immunoglobulins and apolipoprotein B among others (results not shown). We determined the average HDL yield in the isolated HDL density class to be 37 ± 7% based on the measurement of the HDL-C concentration of 12 different HDL preparations before and after isolation (result not shown). Losses of HDL were indicated by a second Sudan Black-stained ring with a density intermediate between HDL and LDL. Analysis by SDS-PAGE and shotgun LC-MS/MS of this second ring revealed a lipoprotein population that was enriched in apoA-I, C apolipoproteins, and apoE; depleted in apoA-II, apoD, and serum amyloid protein A4 (SAA4); and slightly contaminated with apoB (<1% by weight), respectively (not shown). This additional HDL population corresponded therefore to HDL of lower density and was not included in this study together with HDL of very high density.
Further Validation of Absolute Protein Quantitation by PMSSI—
ApoA-I is the main constituent of HDL. We used CE to separate apoA-I from all other proteins present in HDL and for detection of protein peaks at 200 nm. The apoA-I purity measurements of two different HDL preparations and four different KBr density gradient ultracentrifugation preparations of reconstituted HDL (rHDL) were compared with PMSSI values. As illustrated in Fig. 4, the apoA-I concentrations measured by these two independent methods correlated very well (correlation coefficient of 0.94). Furthermore we analyzed by PMSSI and gel densitometry six different mixtures composed of progressively diluted rHDL (regarded as an apoA-I standard with a purity of 93%) in a matrix of six proteins, ALBU, FETUA, LACB, LYSC, SOMA, and THYG (Supplemental Fig. 2). ApoA-I concentrations as well as the concentrations of the matrix proteins, estimated by PMSSI, linearly correlated with the corresponding theoretical concentrations (Supplemental Fig. 2), whereas gel densitometry estimations were consistently much lower than expected. There are different possible causes for this discrepancy. Determination of apoA-I from purified HDL by gel image densitometry proved to be difficult because of the following. First, apoA-I co-migrated consistently with all other proteins on 1DE or 2DE as revealed by co-identification of apoA-I in all other spots or bands (20). Second, the low molecular weight proteins apoA-II and apoC-I were not stained representatively on gels. Third, apoA-I bands/spots were very intense making accurate band/spot density detection difficult (Supplemental Fig. 2). Fourth, although apoA-I densitometry data from samples run on one gel were comparable with each other, they were not when compared between different gels (Supplemental Fig. 3). This can be explained by non-reproducible polyacrylamide polymerization effects between gels resulting in minor changes of protein migration and possible staining inhomogeneities between gels, although the same staining procedure was carried out each time. Fifth, apoA-I staining intensities were close to saturation for SYPRO RubyTM when 15 µg of protein was loaded per lane. Loading of less protein could circumvent this problem, although the detection of less abundant protein species would be compromised resulting in an overestimation of apoA-I concentration.
|
With gel electrophoresis, we could detect and identify a total of 16 different proteins in isolated HDL. The nine most abundant ones were reproducibly identified and quantified by shotgun LC-MS/MS (Fig. 5 and Supplemental Figs. 3 and 4). The relative S.D. values within the four LC-MS/MS repetitions were exponentially and inversely related to the actual abundance of a given protein. Relative S.D. medians ranged from 83 to 34 to 4% for protein abundances of 0.9% (apoC-II) to 1.7% (apoE) to 62% (apoA-I), respectively. All measured protein concentrations were plotted in Fig. 5 with each panel representing one protein with the HC group against the N group. Kruskal-Wallis non-parametric statistical tests revealed a statistically highly significant (p < 0.01) reduced concentration for apoA-I in the HC group and increased concentrations for apoC-I, apoC-III, and apoE, respectively. As there were only few samples tested in this pilot study, these differences cannot be regarded as representative for the entire population; however, they were in agreement with the clinical lipid measurements (see "Discussion"). Relative albumin concentrations did not differ between the groups. This observation corroborates our previous notion that albumin most likely associates stoichiometrically with HDL particles (20). If albumin were dispersed after ultracentrifugation in a gradient-like manner with highest concentration at the bottom and lowest concentration at the top of centrifugation tubes, the albumin relative to HDL protein concentration would not be the same. This was obviously not the case.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
However, the method depends on sample complexity, instrument configuration, and LC separation performance. Automatic gain control or ion charge control, as it is called by Bruker, maximizes the trap filling of a selected precursor ion. In theory, fragment spectral quality should be relatively independent of changing precursor ion intensity over the course of a chromatographic peak unless the precursor ion is of low abundance making maximal trap filling impossible within the set maximum ion accumulation time. As a result, fragment spectra of low abundance precursors have generally lower intensities (smaller signal-to-noise ratios) that might result in lower identification scores as opposed to high abundance precursors with good signal-to-noise fragment spectra. The ion trap was programmed to select the two most intense, not singly charged precursors during any full MS scan for subsequent fragmentation (signals one and two). These two precursors were then placed in an exclusion list for 30 s. In the meantime, MS/MS was performed on the two next most intense peaks and so on. However, if all available MS signals had already been submitted for MS/MS before the 30-s exclusion period of signals one and two was over, the instrument returned to them for lack of other suitable precursors. Additionally the third isotopic peak of a high abundance precursor can also be chosen for fragmentation due to its high signal intensity. Therefore, the PMSSI approach is based on the fact that despite the dynamic exclusion list the precursor ion of a particular peptide, being more abundant in sample A than B, is selected, on average, more in sample A than B. Our results did suggest that this was indeed the case if the complexity of the peptide population desorbing from the reversed-phase column is low enough during the entire LC run. Thus, the reversed-phase column development must be adapted to the complexity and the amount of protein digest loaded as well as the scanning speed of the instrument. Indeed it was possible to show with two different samples, each composed of not more than 16 proteins, that a loading of 500 ng of protein digest and an ACN gradient of 60 min was required for best performance (Supplemental Fig. 1). A shortened 40-min gradient resulted in missing a substantial amount of the low abundance peptides that resulted in a changed protein composition biased in favor of the more abundant proteins. A prolongation of the gradient to 80 min had no influence on the outcome compared with the 60-min gradient (results not shown). In conclusion, low abundance precursor ions, derived from low abundance proteins, are stochastically underrepresented by the SpS and PMSSI methods; hence the accuracy of their quantitation suffers as seen by the high relative S.D. values (Supplemental Table 2).
As illustrated in Figs. 3 and 4, PMSSI worked reliably with the exception of proteins that were (a) hydrolyzed by trypsin to well ionizing peptides always forming a doubly and triply charged ion within the m/z range of 500–1,000 during the electrospray ionization process (for example, MYG and SOMA) or (b) incompletely digested proteins like PRVA and CONA. It needs to be mentioned that these deviations will also influence the results of isotope-labeled samples. The comparative labeling strategy has its advantage in the fact that two (with the iTRAQTM currently up to four) samples are compared in a single LC-MS/MS run. However, all labeling strategies rely on chemical reactions that are difficult to equalize with different samples, and it is always a pure comparative approach for only few samples at a time. The PMSSI approach does not rely on any chemical reaction, and it is possible to measure as many replicates as needed, dispersed over extended periods of time, as long as standard operating protocols are followed meticulously (Fig. 6).
We chose a well established HDL isolation protocol based on floatation in a density gradient. This protocol does not allow the separation of HDL subpopulations that differ in apolipoprotein composition, like LpA-I (lipoprotein enriched in apoA-I) or LpA-I/A-II (lipoprotein enriched in apoA-I and apoA-II) (19). However, it allowed the reproducible isolation of HDL particles with an average density of 1.12 kg/liter that were free of any significant contamination by low density lipoproteins or other plasma proteins. In the existing literature, we could not find any quantitative numbers for total HDL protein in plasma because HDL is usually measured in the form of its cholesterol content. The overall recovery of about 1 g/liter HDL protein from normolipidemic plasma is what can be expected with this method.2 The calculated apoA-I levels ranged from 221 to 684 mg/liter, which corresponded to about 
to 
of what is published elsewhere (19). The apparently missing apoA-I is explained in several ways. First, HDL subclasses of higher and lower density were not collected for further analysis in this study as mentioned under "Results." Second, Asztalos et al. (22) reported specific loss of 10% apoA-I associated with small HDL particles during ultracentrifugation caused by denaturing forces and conditions used with this approach. Correspondingly our HDL yields, determined by HDL-C measurements, confirmed the loss of HDL protein in the order of 60%. Third, we used UV absorption measurements at 280 nm for absolute protein quantification. Although this method is rugged and easy to use, protein-specific extinction coefficients differ according to the relative abundance of tyrosine and tryptophan residues in the corresponding amino acid sequence. Despite this shortcoming, UV 280-nm measurements allowed for a robust comparison between different HDL samples.
Another explanation for discrepancies of apoA-I yields is due to the use of antibodies in most published results. Antibodies do have different avidities and have to recognize apolipoproteins that are part of a multimolecular protein-lipid complex, resulting in a possible steric hindrance of antibody-antigen recognition. This possibility was supported by our observation that immunoaffinity-purified HDL differed significantly from different HDL density classes. In contrast, a mass spectrometry-based method, like PMSSI, is independent of quaternary structures because proteins and protein-lipid complexes are dissociated under denaturing conditions followed by proteolysis of proteins into many peptides. Peptides are then separated on a reversed-phase column enabling the mass spectrometer to detect multiple surrogates of many antigens within the course of one analysis. Although the ionization process is biased by the chemical structure of peptides (see overestimation of MYG and SOMA in Fig. 3), this bias can be considered to be randomly distributed over all tryptic peptides of a proteome and should therefore have only a minor effect.
In the following we give possible interpretations of the observed protein compositions in relation to HDL metabolism. However, we would like to emphasize that only a distinct HDL subpopulation of one defined density range was analyzed. Together with the fact that we analyzed only a small number of samples in this preliminary study, the clinical significance of the given interpretations cannot be warranted completely.
Lipid-free apoA-I, small pre-ß HDL, and discoidal HDL are important acceptors for cholesterol and after remodeling through the action of LCAT and other plasma enzymes become spherical HDL. LCAT is activated by apoA-I and apoC-I and esterifies cholesterol molecules of HDL using phosphatidylcholine as fatty acyl donor. Therefore, apoA-I can be considered as the rate-limiting factor for HDL-C (23). Indeed apoA-I mass correlated positively with HDL-C (Fig. 6, top left panel) and total HDL protein (not shown). Additionally the relative apoC-I concentrations in hypercholesterolemia HDL appeared to be increased (Fig. 5), which might indicate a certain compensatory function for LCAT activation by apoC-I in the case of reduced apoA-I levels. Such a compensation mechanism is not obvious in the apoC-I plot of absolute protein concentrations of Fig. 6 (top left). However, if HDL-C were solely dependent on apoA-I in a linear relationship, hypercholesterolemia samples would have HDL-C levels that were significantly lower than the actual measured levels. For instance, C4 with an apoA-I concentration of 344 mg/liter would have a theoretical HDL-C level of 0.66 instead of the measured 1.07 mM when correlated with N1 for instance. The absolute HDL-associated apoC-I concentration of C4 (29.3 mg/liter) was equal to the absolute HDL-associated apoC-I concentrations of normolipidemia samples (mean ± S.D. of 23.9 ± 10.7 mg/liter), corroborating a positive effect of apoC-I on HDL-C in hypercholesterolemia.
As mentioned above, lipid-poor HDL precursors are important acceptors of cellular cholesterol and contribute to lipid metabolism and reverse cholesterol transport. HDL precursors are dissociated from chylomicrons and very low density lipoprotein during lipoprotein lipase-mediated hydrolysis of triglycerides. ApoC-III is an inhibitor of lipoprotein lipase, and increased levels of apoC-III will cause an accumulation of triglycerides in serum and, due to reduced availability of HDL precursors, a reduction in HDL-C. This relationship is indicated in the Fig. 6, top right panel. C2 and C3 did not fit into such a model, indicating that lipid metabolism is a multifactorial mechanism. A defect in lipid clearance might actually explain the two "outliers." ApoE-specific receptors mediate the uptake of apoA-I and apoB lipoproteins. A plot with HDL-associated apoE concentration in relation to total serum cholesterol and the combined apoC-I and apoC-III concentrations as the third dimension (Fig. 6, lower right panel) revealed that normolipidemia samples were aligned at a cholesterol concentration of about 5 mM and had a heterogeneous distribution of apoE (0–22.3 mg/liter) as well as apoC-I + apoC-III (59.5–125.0 mg/liter), respectively. The hypercholesterolemia samples had similar traits with 8.8–28.9 mg/liter apoE and 59.9–172.1 mg/liter for the C apolipoproteins; however, they segregated into two populations, one containing C1, C2, and C3 and the other containing C6 and C7 with C4 and C5 being at the crossroad. C1 to C5 had increased apoE and cholesterol concentrations and low to average C apolipoproteins, whereas C6 and C7 had average apoE but most extreme apoC. There are three common apoE isoforms, apoE-II (Cys112/Cys158), apoE-III (Cys112/Arg158), and apoE-IV (Arg112/Arg158), that strongly affect their binding properties to the LDL receptor and their intracellular recycling (for a review, see Ref. 24). Subjects homozygous for apoE-II have elevated triglyceride, cholesterol, and extracellular apoE. All these characteristics would fit to the traits of C2 and C3. However, we could not detect increased amounts of the more acidic apoE-II isoform by 2DE nor the tryptic peptide 158CLAVYQAGAR167, indicative for the apoE-II form, by LC-MS/MS analyzes. From this, it must be concluded that the familial hypercholesterolemia of subjects C1, C2, and C3 is caused by a lipoprotein receptor defect, and genetic tests are still ongoing to reveal its origin.
We identified serum amyloid A4 as a constitutive member of total HDL, and with samples C2 and C5 we also detected trace amounts of SAA (accession number P02735). Serum amyloid A proteins are acute phase response proteins that are increased during inflammation and are chronically elevated in metabolic syndrome and type 2 diabetes. In vitro experiments had shown that SAA is able to replace apoA-I and apoA-II from HDL, but in vivo results in mice did not confirm this (25, 26). Our results indicate that serum amyloid A4 is constitutively associated with HDL where it is maintained at a constant ratio relative to apoA-I as long as HDL-C is at physiologically normal levels of >1 mM (Fig. 6, lower left panel). In samples with reduced HDL-C, the relative contribution of SAA4 to total HDL protein and apoA-I appeared to increase, whereas relative concentrations of apoA-II remained unaffected. Serum amyloid can form dense but large HDL particles independently of the availability of apoA-I in apoA-I knock-out mice and cell culture (27–29). Our finding suggests that SAA4 protein may take over some functionality of apoA-I-enriched lipoproteins but most likely does not contribute to reverse cholesterol transport.
Derangements in lipid metabolism are clinically indicated by the measurement of total serum cholesterol, triglyceride, HDL-C, and LDL-C. Hypercholesterolemia is defined as a severe form with high LDL-C and low HDL-C that is associated with an increased risk to develop atherosclerotic disease and a milder form with high LDL-C and high HDL-C. Our quantitative apolipoprotein results of a certain HDL density class did highlight different causes for hypercholesterolemia. HDL samples of the three related subjects C1, C2, and C3, who do have relatives with atherosclerotic incidences, had a protein pattern that clearly differentiated them from all other samples. Subjects C4 and C5 had less extreme lipid levels and could be categorized by our protein quantitation as having a milder form of hypercholesterolemia. The two sisters, C6 and C7, were differentiated from the other hypercholesterolemia subjects by high HDL-C. In accordance, the HDL apolipoprotein pattern was comparable with that of the normolipidemia group except for their extremely high levels of C apolipoproteins.
|
CONCLUDING REMARKS |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Published, MCP Papers in Press, March 5, 2007, DOI 10.1074/mcp.M600326-MCP200
1 The abbreviations used are: 2DE, two-dimensional gel electrophoresis; 1DE, one-dimensional gel electrophoresis (SDS-PAGE); ACTS, actin; ALBU, serum albumin; apoX, apolipoprotein X (X = A-I, A-II, C-I, C-II, C-III, etc.); CE-UV, capillary electrophoresis with UV detection at 200 nm; CONA, concanavalin A; emPAI, exponentially modified protein abundance index; FA, formic acid; FETUA, fetuin; HC, hypercholesterolemia; HDL, high density lipoprotein; HDL-C, HDL cholesterol; INS, insulin; LACB, ß-lactoglobulin; LCAT, lecithin-cholesterol acyltransferase; LDL, low density lipoprotein; LDL-C, LDL cholesterol; LYSC, lysozyme; MYG, myoglobin; N, normolipidemia; PAI, protein abundance index; PMSS, peptide match score summation; PMSSI, peptide match score summation index; PRVA, parvalbumin; rHDL, reconstituted HDL; SAA, serum amyloid protein A; SOMA, growth hormone (somatotropin); SpS, spectrum sampling; THYG, thyroglobulin. 
2 R. Bolli, personal communication.
* This work was supported in part by ZLB Behring. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
S The on-line version of this article (available at http://www.mcponline.org) contains supplemental material.
To whom correspondence should be addressed: Laboratory of Thrombosis Research, Dept. of Clinical Research, Freiburgstrasse, 3010 Bern, Switzerland. Fax: 41-31-632-2683; E-mail: manfred.heller{at}dkf.unibe.ch
|
REFERENCES |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONCONCLUDING REMARKSREFERENCES |
|---|
