| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Molecular & Cellular Proteomics 5:573-588, 2006.
© 2006 by The American Society for Biochemistry and Molecular Biology, Inc.
,
From the The Plasma Proteome Institute, Washington, D. C. 20009-3450 and ¶ Applied Biosystems, Foster City, California 94404
 |
ABSTRACT |
|---|
 TOP ABSTRACT EXPERIMENTAL PROCEDURESRESULTSDISCUSSIONREFERENCES |
|---|
Candidate-based specific assays rely on the specificity of capture or detection methods to select a specific molecule as analyte. Capture reagents such as antibodies can provide extreme specificity (particularly when two different antibodies are used as in a sandwich immunoassay) and form the basis of most existing clinical protein assays. There is intense interest in miniaturizing sets of such assays (5, 6) in array formats (on planar substrates, beads, etc.), although significant problems remain in the production of suitable antibodies and in the simultaneous optimization of multiple assays in one fluid volume.
Mass spectrometry provides an alternative assay approach, relying on the discriminating power of mass analyzers to select a specific analyte and on ion current measurements for quantitation. In the field of analytical chemistry, many small molecule analytes (e.g. drug metabolites (7), hormones (8), protein degradation products (9), and pesticides (10)) are routinely measured using this approach at high throughput with great precision (CV < 5%). Most such assays use electrospray ionization followed by two stages of mass selection: a first stage (MS1) selecting the mass of the intact analyte (parent ion) and, after fragmentation of the parent by collision with gas atoms, a second stage (MS2) selecting a specific fragment of the parent, collectively generating a selected reaction monitoring (plural MRM) assay. The two mass filters produce a very specific and sensitive response for the selected analyte that can be used to detect and integrate a peak in a simple one-dimensional chromatographic separation of the sample. In principle, this MS-based approach can provide absolute structural specificity for the analyte, and in combination with appropriate stable isotope-labeled internal standards (SISs), it can provide absolute quantitation of analyte concentration. These measurements have been multiplexed to provide 30 or more specific assays in one run (11). Such methods are slowly gaining acceptance in the clinical laboratory for the routine measurement of endogenous metabolites (e.g. in screening newborns for a panel of inborn errors of metabolism (12)) and some drugs (e.g. immunosuppressants (13)).
Recently the MRM assay approach has been applied to the measurement of specific peptides in complex mixtures such as tryptic digests of plasma (3). In this case, a specific tryptic peptide can be selected as a stoichiometric representative of the protein from which it is cleaved and quantitated against a spiked internal standard (a synthetic stable isotope-labeled peptide (14)) to yield a measure of protein concentration. In principle, such an assay requires only knowledge of the masses of the selected peptide and its fragment ions and an ability to make the stable isotope-labeled version. C-reactive protein (3), apolipoprotein A-I (15), human growth hormone (16), and prostate-specific antigen (17) have been measured in plasma or serum using this approach. Because the sensitivity of these assays is limited by mass spectrometer dynamic range and by the capacity and resolution of the assisting chromatography separation(s), hybrid methods have also been developed coupling MRM assays with enrichment of proteins by immunodepletion and size exclusion chromatography (18) or enrichment of peptides by antibody capture (SISCAPA (19)). In essence the latter approach uses the mass spectrometer as a "second antibody" that has absolute structural specificity. SISCAPA has been shown to extend the sensitivity of a peptide assay by at least 2 orders of magnitude (19) and with further development appears capable of extending the MRM method to cover the full known dynamic range of plasma (i.e. to the pg/ml level).
There is compelling evidence that high and medium abundance plasma proteins have value as clinical biomarkers and thus that there may be applications for specific MRM assays even without antibody enrichment. One may therefore ask how many plasma proteins can be measured by quantitating their peptides in a plasma digest, and how precise could these measurements be? If the measurement strategy proves to be robust, could it be carried out using existing high throughput LC-MS/MS platforms? To address these questions we generated and tested MRM assays based on peptides from a variety of high to medium abundance plasma proteins to see how many could be measured effectively by LC-MS/MS with and without subtraction of the most abundant proteins. An understanding of the performance of MS/MS in this application could enable routine and relatively inexpensive measurement of classical plasma proteins and also provide a foundation for use of MS/MS in more sensitive anti-peptide antibody-enhanced SISCAPA assays for low abundance proteins.
|
EXPERIMENTAL PROCEDURES |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONREFERENCES |
|---|
In our initial attempt to generate MRMs by purely in silico methods, a set of 177 proteins and protein forms was assembled that are demonstrated or have potential to be plasma markers of some aspect of cardiovascular disease (20), and a subset of 62 proteins was selected for which an estimate of normal plasma abundance was available. Predicted tryptic peptides for each of these were generated along with relevant Swiss-Prot annotations and a series of computed physicochemical parameters: e.g. amino acid composition, peptide mass, Hopp-Woods hydrophilicity (21), and predicted retention time in reversed-phase (C18) chromatography (22). An index of the likelihood of experimental detection was derived from a data set reported by Adkins et al. (23) by counting the number of separate "hits" for the peptide in the data set divided by the number of hits for the most frequently detected peptide from the same protein. An overall index of peptide quality was generated according to a formula that gave positive weights to Pro, KP, RP, and DP content and negative weights to Cys, Trp, Met, chymotrypsin sites, certain Swiss-Prot features (carbohydrate attachment, modified residues, sequence conflicts, or genetic variants), and mass less than 800 or greater than 2000. The 3619 tryptic peptides predicted for the 62 protein marker candidates (6–497 peptides per target) ranged in length from 1 to 285 amino acids. Within the useful range of 8–24 amino acids, 721 peptides had a C-terminal Lys and 690 had a C-terminal Arg. In this report, peptides from 30 of these target proteins ending in C-terminal Lys were selected for further study.
We also selected peptides based on two types of direct proteomic survey experiments. In the first case we carried out classical LC-MS/MS analysis of plasma digests in which the major ions observed were subjected to MS/MS using the ion trap capabilities of the 4000 Q TRAP instrument. The identified peptides showing the best signal intensity and chromatographic peak shape for a given parent protein were selected. In addition, we used the Global Proteome Machine database of Craig et al. (24) to select peptides from target proteins that were frequently detected (multiple experiments).
Finally we used an adaptation of the MIDAS workflow, described previously for discovering post-translational modifications (25, 26), to look for measurable tryptic peptides from a variety of plasma proteins. In this approach, the protein sequence is digested in silico, likely y-ion fragments are predicted, and theoretical MRMs are generated for all the peptides in an acceptable size window. These MRMs are then used as a survey scan in a data-dependent experiment to detect specific peptide peaks, and each resulting MRM peak is examined by full scan MS/MS to obtain sequence verification of the hypothesized peptide. To verify peptide specificity in designated protein targets, selected peptides were searched with BLASTP for exact matches against the genome-derived human, mouse, and rat Ensembl peptides using Ensembl Blastview (www.ensembl.org/Homo_sapiens/blastview).
"Random" MRMs—
Two approaches were used to generate pseudorandom MRMs. In the first case we used 100 MS1 values distributed randomly (by the Excel RAND function) between 408.5 and 1290.2 (the maximum and minimum of an early set of real MRMs we tested) paired with MS2 values chosen randomly between this MS1 and the maximum of the real MRMs (1495.6), thus mimicking the properties of our real MRMs (which are generally +2 charge state peptides and +1 charge fragments). In a second set we paired 131 MS1 values chosen randomly from among MS1 values in a large table of real MRMs with MS2 values chosen randomly from the real MS2 values of the same list, imposing only the constraint that each MS2 had to be between 1 and 2 times the paired MS1 mass (to approximate our selection criteria for real MRMs). Peaks detected in these MRMs were examined by triggering MS/MS (the MIDAS workflow).
Reagents—
The following chemicals were used: trypsin (Promega), sodium dodecyl sulfate (Bio-Rad), iodoacetamide (Sigma), formic acid (Sigma), tris-(2-carboxyethyl)phosphine (Sigma), and acetonitrile (Burdick and Jackson).
Plasma Depletion and Digestion—
All experiments were performed on aliquots of a single human plasma sample from a normal volunteer. The six highest abundance proteins were depleted from plasma using the multiple affinity removal system ("MARS"; Agilent Technologies spin columns) according to the manufacturer’s protocol. Depleted sample was then exchanged into 50 mM ammonium bicarbonate using a VivaSpin concentrator (5000 molecular weight cutoff, VivaScience). Undepleted plasma was also desalted before digestion.
Both depleted or undepleted plasma samples were denatured and reduced by incubating proteins in 0.05% SDS and 5 mM tris-(2-carboxyethyl)phosphine at 60 °C for 15 min. The sample was then adjusted to 10 mM with iodoacetamide and incubated for 15 min at 25 °C in the dark. Trypsin was added in one aliquot (protease:protein ratio of 1:20) and incubated for 5 h at 37 °C.
Labeled Peptide Internal Standards: polySIS—
A series of SIS peptides was added to samples in selected experiments by spiking with the tryptic digest of a "polySIS" polyprotein.2 Briefly this protein was produced by cell-free transcription and translation of a synthetic gene coding for 30 concatenated tryptic peptide sequences (derived from 30 plasma proteins) in the presence of U-13C6,U-15N2-labeled lysine (a total mass increment of 8 amu compared with the natural peptide). The 30 peptides were selected based on our initial in silico MRM design approaches and are thus not fully optimized using experimental data. Of these peptides, 13 were used in the present studies (the remainder were not reproducibly detected in digested plasma with peak area >1E+04). The positioning of the label atoms at the extreme C terminus of each peptide has the effect that all fragments that contain the C terminus (i.e. the y-ions) will show the mass shift due to the label, whereas all the fragments that contain the N terminus (and hence have lost one of more C-terminal residues: the b-series ions) will have the same masses as the corresponding fragments from the natural (sample-derived) target protein. These features (shifted y-ions and normal b-ions) provide a simplification in interpreting the fragmentation patterns of the SIS peptides in comparison with the similar QCAT concept described recently by Beynon et al. (27). To determine the absolute concentration of polySIS protein, an aliquot was diluted with 1 M urea, 0.05% SDS, and 50 mM Tris, pH 8 and subjected to N-terminal Edman sequencing, yielding an initial concentration of 5 ± 1 pmol/µl. A tryptic digest of the polySIS protein was spiked into whole and depleted human plasma digests at the final concentrations shown in Table I.
|
|
RESULTS |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONREFERENCES |
|---|
Because different peptides from a single protein can vary widely in detectability by ESI-MS, we attempted to improve upon the in silico approach to MRM design using experimental data from a conventional peptide survey scan of a human plasma digest and applying the selection criteria to peptides with demonstrated detectability. Using a 3-h LC gradient, MS/MS scans were collected for the major doubly or triply charged ions across the separation using information-dependent data acquisition, and a second run was performed using time-filtered exclusions of the peptide ions detected in the first run. The combined results identified 54 plasma proteins ranging in abundance from albumin down to fibronectin (normal plasma concentration of 
300 µg/ml). This experimental MS/MS data provided explicit information for peptide selection, charge state, and most abundant y-ion m/z value under the specific conditions used (i.e. electrospray ionization with collisional peptide fragmentation), allowing improved design of MRMs. When these MRMs were then used to analyze the same sample in a subsequent run, triggering MS/MS scans at any MRM signal, most of the peptides were detected as peaks in the chromatogram and identified by database search as expected. In most of these MRM chromatograms, only a single peak was detected.
Because peptide detection sensitivity using MRM is expected to be greater than that achieved in a full scan MS survey approach, a comprehensive de novo MRM design method was explored for those proteins not detected in the above survey experiment. Using a novel software tool, a large set of MRMs was generated for each of a series of target proteins by selecting all predicted tryptic peptides in a useful size range together with multiple high mass y-ion fragments of each (the "MIDAS" workflow (25, 26)). These MRMs were then tested in LC-MS/MS runs of the unfractionated plasma digest, grouped in panels that included all the predicted tryptic peptides of one or two proteins at a time (50–100 MRMs per run), with MS/MS scans triggered on any peaks observed. Of 12 proteins examined, nine produced at least one usable MRM (signal-to-noise (S/N) ratio >20).
MRM results from the above approaches were pooled, and a set of optimized MRMs was assembled that covered a total of 60 peptides representing 53 proteins in human plasma (Table II; seven proteins were represented by two peptides). This set includes 18 peptides selected by the in silico approach (indicated by an X in the SIS column of Table II): eight of the initial 30 in silico peptides were eliminated as likely to be too low abundance for detection, and better alternative peptides were selected from experimental data for four others. For all but one of the peptides we elected to measure two fragments (i.e. using two MRMs per peptide), yielding 119 MRMs. Finally we included MRMs for 18 stable isotope-labeled internal standard ("SIS") versions of target peptides (i.e. the tryptic digest of the polySIS protein) spiked into the digest plasma samples. The resulting set of 137 MRMs was measured in all the replicate runs described below using an 18-ms dwell time per MRM and a resulting cycle time of 3 s between measurements.
|
|
700 ng of total protein assuming an average plasma concentration of 70 mg/ml (28)) after depletion of the most abundant proteins using an Agilent MARS spin column (84% of protein mass should be removed based on a calculation using normal abundances). This loading, comprising an estimated 110 ng of total peptides, proved to be a loading compatible with very good nanoflow chromatography of the MRM peptides. Experiments B and E used higher loadings to explore the trade-off between peak stability (chromatographic quality, adversely affected by increased load) and S/N ratio (improved by increased analyte quantity). We concluded that the 110 ng loading was optimal. Chromatographic elution times were quite reproducible, showing average CVs of 2% (experiment D) and 2.5% (experiment E).
Fig. 2A shows histograms of the coefficients of variation for the five experiments (individual values for each MRM are contained in Table II). In analyses of depleted plasma, more than 60% of the MRMs show within-run CVs less than 10%, and almost half have CVs below 5%. A number of these MRMs (e.g. 
1-antichymotrypsin, apolipoprotein E, hemopexin, heparin cofactor II, plasminogen, prothrombin, fibrinogen 
chain, complement C4, and factor B) showed an average within-run CV of 3–4% across three experiments, precision equivalent to that of good clinical immunoassays. Analyses of whole (undepleted) plasma digests showed generally higher CVs (20–50% of MRMs had CV <10%). It is important to note that these reproducibility measures are computed on raw peak areas without correction using internal standards. Four of the measured proteins were expected to be removed by the depletion process used (Agilent MARS spin column). In comparing average peak areas obtained in analyses of digests of whole and depleted samples, we found substantial reductions in albumin (1.3E+08 reduced to 1E+04), transferrin (1.5E+05 reduced to 5E+03), and haptoglobin (4.6E+06 reduced to 1E+05). 1-Antitrypsin was not detected reliably in any of the runs (presumably a bad peptide choice for MRM assay), and so its removal could not be confirmed. We did not incorporate MRMs to measure the two immunoglobulins subtracted by the MARS column.
|
|
10, which is consistent with the quantitative requirement of a S/N ratio of 10 for a reported lower limit of quantitation. The highest peak areas measured (albumin peptides in whole plasma digest samples) are above 1E+08, demonstrating a maximal working dynamic range of >4 orders of magnitude above this cutoff.
|
|
0.67 µg/ml (29) or 20.3 pmol/ml. Fibronectin is a 260-kDa protein present in plasma at a normal concentration of 300 µg/ml (30) or 1200 pmol/ml. Given that an amount of digest corresponding to 0.01 µl of plasma was loaded on column in experiment D, these peptides would be expected to be present on column at 200 and 12,000 amol, respectively. In the case of L-selectin we had spiked a SIS peptide at 2.0 fmol and thus could determine that the natural (sample-derived) peptide was present at 0.1 times the amount of SIS (single point quantitation), yielding a measured 200 amol and implied plasma concentration of 0.6–0.67 µg/ml in good agreement with expectation. CVs for fibronectin in experiments D and E were 4 and 4%, respectively, and for L-selectin were 22 and 11%, respectively, presumably reflecting the fact that L-selectin was near the lower limit (1E+04) for high quality detection.
Six of the 53 selected target proteins were not reliably observed, however. Three are probably explained by low normal abundances: we did not obtain a reproducible signal for the selected peptides from coagulation factor V, vitamin K-dependent protein C, or C4b-binding protein. There were also instances in which peptides from more abundant proteins were not reliably detected: the inter- trypsin inhibitor light chain (despite the fact that a peptide from the heavy chain of this protein gave a good quality MRM), apolipoprotein C-II, and 1-antitrypsin. In these latter cases, the problem is most likely poor choice of peptides: numerous alternative peptides exist for both the inter- trypsin inhibitor light chain and 1-antitrypsin, but for small proteins, such as apolipoprotein C-II, there may be no better alternative, and additional enrichment of these peptides will be necessary.
In general, depletion of the most abundant proteins using the Agilent MARS column improved the performance of MRMs for non-subtracted proteins substantially. This effect appeared to be due to improved detection sensitivity and improved chromatographic peak shape (achieved by decreasing the total peptide loaded by 6-fold), both of which contribute to improved MRM peak signal-to-noise and lower CVs. Fig. 5 illustrates the benefit of depletion in removing the albumin peptide (major peak in Fig. 5A) and thus boosting the minor peaks in the depleted sample (Fig. 5B). At very high loading of undepleted plasma digest, we noticed large shifts in peak retention times, but at loadings in the region of our nominal load the effect of high abundance peptides on MRM retention times was minor.
|
2.5 orders is usable (i.e. beginning at peak area >1E+04, the approximate cutoff below which CVs become large). Duplicate digests show excellent comparability (R2 = 0.995 and 0.998 for F1_1 versus F1_2 and F2_1 versus F2_2, respectively). Duplicate depletions (which necessarily include the effects of different digests as well) are only slightly worse (e.g. R2 = 0.989 and 0.991 for F2_1 versus F1_1 or F2_2 versus F1_2, respectively).
|
|
, fibronectin, haptoglobin ß, and inter- trypsin inhibitor heavy chain) occur in rat homologs (all the other human sequences did not occur in the other species’ Ensembl peptides). Of the final 47 MRMs, 12 were contributed by the in silico approach leading to the 30 polySIS peptides, and one (hemopexin) was contributed by an earlier in silico effort (19): these are indicated in Table II by an X in the SIS column for the associated Lys-labeled internal standard. A total of eight in silico selections were replaced by better performing peptides from the same target protein as a result of experimental testing (four before and four after selection of the 137 MRMs), two subsequently failed and have not yet been replaced, and eight were dropped before testing because of expected insufficient abundance. Thus 13 in silico selections survived, whereas 10 were replaced in testing.
Of the six failures (see above), we expect three proteins to be within the concentration range that should be measurable and are selecting substitute peptides for these. The distribution of CVs for the 47 best MRMs is shown in Fig. 2B: in experiment D, 40 of these had CVs below 10%, and 19 had CVs below 5%.
|
DISCUSSION |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONREFERENCES |
|---|
Our initial attempts to select usable tryptic peptides by purely in silico means were reasonably successful as approximately half of the peptides chosen (13 of 23) produced acceptable MS signals in plasma. Although the prediction of ionization properties of tryptic peptides can be expected to improve substantially in the future, we used experimental MS/MS data, in combination with computational methods, to select more successful target peptides. Two experimental methods proved particularly useful. High abundance peptides were detected in conventional LC-MS/MS data-dependent full scan MS experiments in which a subset of high signal peptides seen in MS1 are subjected to MS/MS. Lower abundance peptides were detected by constructing lists of candidate MRMs to all appropriately sized predicted tryptic peptides from a target protein and then characterizing any detected MRM peaks by MS/MS (the MIDAS workflow in which MRM methods are designed using a specifically designed script within the Analyst software). Because MRMs are typically more sensitive than full scan survey MS for detection of very low abundance components, the MIDAS approach allowed us to find successful MRMs for more lower abundance peptides and represents the preferred strategy for MRM design going forward. This process was facilitated by the combination of high sensitivity triple quadrupole MRM and ion trap MS/MS scan capabilities on the hybrid triple quadrupole linear ion trap 4000 Q TRAP mass spectrometer. A general process for MRM design using these sources of information is shown in Fig. 8 in which bold arrows indicate the preferred approach. The final step in the process, selection among alternative candidate MRMs for a given protein based on measured CVs in replicate runs, is recommended because of the significant differences in CV observed among MRMs with similar signal strength: CV appears to be a characteristic of a peptide to some extent separate from ion current.
|
10%) of MRMs indicates that chromatographic elution time is an important additional factor in providing the absolute analyte specificity desired in these assays. Following more extensive experience with specific MRMs and any variation that may occur in them due to shifts in peak elution times, it may be possible to establish a set of plasma MRMs that are truly free of secondary peaks and that could therefore potentially be measured using short LC separations (constrained only by increases in ion suppression effects as peptides crowd together).
The unambiguous measurement of a peptide derived from L-selectin (normal concentration of 0.67) suggests that proteins present in normal plasma at 1 µg/ml are measurable by MRM in depleted and undepleted samples with minimal upfront sample fractionation. Different tryptic peptides from the same protein can produce ion currents differing by factors of at least 1E+03 in LC-MS/MS experiments, excluding the peptides that are not detected at all. This variation is due to multiple factors, including propensity to ionize in the electrospray source, coincidence in elution time with other easily ionizing peptides, efficiency of release during tryptic digestion, and presence of unrecognized post-translational modifications arising due to biology or during sample preparation. Although we do not yet know the frequency of high efficiency peptides among the tryptic products of each protein, we believe that numerous additional plasma proteins can be measured down to 1 µg/ml using MRMs and estimate that 50–100 proteins may be added to our list as we extend this effort to additional candidates from the plasma component database (2, 20).
Given that albumin, the most abundant protein in plasma, is measurable by MRM in undepleted plasma in the same experiment that detects L-selectin, the dynamic range of quantitatively measurable proteins appears to be 1 µg/ml to 55 mg/ml or 5E+04. This dynamic range of 4–5 orders of magnitude is consistent with typical quantitation experiments performed using MRM on this instrument and covers almost half the logwise dynamic range of currently known and measured proteins in human plasma (1). The dynamic range of peak area measurements appears to be similar, extending from 1.3E+08 ± 5% for undepleted albumin to less than 1E+04 (the approximate cutoff below which CVs become >10%). In this context it is important to note that measurement of proteins over this dynamic range does not require measuring peptides at such a broad range of signal intensities because it is easily possible to select lower signal peptides from high abundance proteins (and high signal peptides from low abundance proteins) to diminish the dynamic range requirement at the peptide level. As long as stable isotope-labeled internal standards are used, the quantitative information contained in Nat:SIS ratios should be unaffected by peptide choice.
The reproducibility of peptide MRM measurements was impressive in many cases. For the three experiments analyzing depleted plasma digests, an average of 78% of the 47 best MRMs had within-run CVs of 10% or less, and 36% had CVs less than 5%. A number of MRMs gave low CVs across five experiments (i.e. in depleted and whole plasma digests at widely varying total loads) including peptides from hemopexin (CVs ranging between 2 and 6%), vitronectin (4–7%), kininogen (5–6%), complement factor B (3–6%), complement C4 (3–5%), apolipoprotein B100 (5–7%), antithrombin III (5–7%), and 1-antichymotrypsin (2–7%). These precision values are comparable to many small molecule MRM measurements and approach the results of conventional immunoassays used in clinical diagnostics (32). They demonstrate clearly that peptide MRM measurements can be used in high precision quantitative assays. Based on reproducibility at this level, it appears likely that useful comparative data across samples might be obtained by comparing simple peak area measurements (i.e. without SIS standardization) and that relatively small quantitative differences could be detected (differences of 20% would be >3 standard deviations from the mean for many of the best MRMs and thus very easily detected). The observation that individual peptides showed consistently higher or lower CVs suggests that quantitative precision depends on characteristics of the peptide and perhaps its environment of co-eluting peptides. Therefore it will be useful in future work to select MRM peptides based on CV across replicate runs in addition to the characteristics normally considered (e.g. ion current and chromatographic peak shape).
MRMs designed for absolute quantitation using a set of 13 SIS peptides (with sequences identical to 13 of the selected tryptic peptides) performed well. When spiked at known loading these provided internal standards for quantitation and resulted in a measured value for L-selectin in almost exact agreement with the literature value for normal subjects (0.67 µg/ml). This result is encouraging but not definitive as we did not measure target protein concentrations in our sample by independent methods, and the individual SIS peptides (derived by digestion of a quantitated precursor protein) were not individually quantitated. Thus not all of our endogenous: standard ratio measurements agreed so well with literature values. In addition, single point concentration curves as used here do not provide the same accuracy in quantitation as multiple point curves. Nevertheless the extreme reproducibility (R2 = 0.991) of the measured ratios (and implied absolute peptide concentrations) between experiments D and E (Fig. 4) across almost 3 orders of magnitude in peak area and 3-fold difference in total peptide load are striking. To effectively standardize peptides (and proteins) over wider ranges, a more sophisticated strategy will be utilized using smaller equimolar groups of peptides selected to represent each decade of peptide concentration (i.e. one polySIS product or a set of equimolar SIS peptides for each decade of the abundance scale). In addition, the completeness of polySIS digestion, and thus the relative stoichiometry of the resulting SIS peptides, must be rigorously characterized.
Depletion of the highest abundance plasma proteins using an immobilized antibody column proved to be very useful in measuring our MRMs. We expected to be able to load a digest of 6 times as much depleted plasma as undepleted plasma because these two samples would contain approximately equal amounts of total peptides, and this was confirmed. We also confirmed that the reference loading chosen (110 ng of peptides derived from digestion of 10 nl of depleted plasma) was near but not above the limit for good quality LC peak shape and reproducible chromatography in a nanoflow system. However, we had few if any MRMs that were detectable in the depleted sample but not the undepleted sample. The major difference emerged instead in the lower CVs of peaks measured in the depleted sample due presumably to the decreased level of competing peptides and consequent higher signal to noise. In future extensions of this method to lower abundance proteins, it is likely that depletion will expose numerous targets not otherwise detectable. Depletion significantly reduced the levels of albumin, transferrin, and haptoglobin as expected.
Highly reproducible depletion and tryptic digestion are a necessity as steps in routine sample preparation for MRM analysis. In a small study (experiment F) designed to look at these effects, R2 values 
0.995 were obtained for peak areas in duplicate digests, and R2 values 0.989 were obtained for duplicate depletion and digestion in the same experimental set. It thus appears that both depletion and digestion can be carried out with sufficient reproducibility to provide useful measurements and that MRM assays provide an ideal method for assessing sample preparation reproducibility.
The multiplexing capability of LC-QqQ-MS platforms for measuring peptides in complex digests is substantial, providing an opportunity to measure large panels of proteins accurately in each run. Based on the performance of the present set of 137 MRMs, which were all monitored continuously across the entire LC gradient as 18-ms sequential measurements, it is clear that 100–200 MRMs might be used routinely to measure peptides in long LC gradients. However, given reproducible chromatographic elution times, it is possible with existing systems to measure each MRM only during a short time window when the peak is expected to occur (e.g. a window of 10% of total run length given an average 2–2.5% CV in peak elution time measured in our experiments D and E). Once this approach is implemented, based on extensive knowledge of elution time and column reproducibility, and provided that MRMs are selected that do not cluster too much in elution time, substantially more MRMs could be used in a single LC MRM experiment.
An additional important consideration for throughput of MRM measurements is the duration of the chromatography run. In our replicate experiments D and E, a 30-min gradient was used that led to a total cycle time (including intersample wash) of 75 min. It would clearly be advantageous if this could be reduced, allowing more samples to be run per day. The extreme analyte specificity indicated by the low density of peaks in MRM space suggests that most of our MRMs should perform well with less benefit from chromatographic separation, and the ability to focus the MRM measurements in discrete time windows may allow more MRMs to be brought closer together in elution time without sacrificing the required multiple measurements across each peak. Hence we expect that substantial improvements in run time should be possible, probably in conjunction with a shift to higher flow rate (e.g. capillary flow) systems providing increased robustness in routine operation. Higher flow rate LC systems have significantly lower (10x) absolute sensitivity than nanoflow, but in the current application it would be easily possible to load 10x more sample (i.e. 100 nl) as sample at these loadings is rarely limiting.
Considered in a broad context, MRM assays appear to have several advantages in addition to those described above. 1) The instrumentation used to measure peptide MRMs is very similar to that used in existing robust platforms for high throughput quantitative measurement of drug metabolites in plasma, for the detection of inborn errors of metabolism in newborns and for pesticide analysis. A large base of compatible instrumentation and expertise thus already exists. 2) Because the target proteins are detected via tryptic peptide surrogates after sample denaturation and digestion, peptide MRM results should be insensitive to alterations in protein folding and intersubunit associations (factors that can negatively affect immunoassays). 3) MRM results are obtained by integrating discrete, anticipated peaks in simple ion chromatograms using well established and widely used commercial software, and thus the method is not dependent on the much more complex data processing infrastructure required to handle the peak matching and spectral identification challenges of a discovery proteomic pipeline. 4) It is probable, although not yet demonstrated, that the cost per high quality measurement will be lower than for immunoassays (both to create the assay to begin with and to apply it to large sample sets).
Important limitations of peptide MRMs must also be recognized. 1) Although the method is applicable to post-translationally modified peptides, the post-translational modification must be known or specifically hypothesized, and its properties must be designed into the assay up front (25) (and normally cannot be discovered de novo by our approach). 2) Some proteins may not readily produce usable peptides. The protein may be too short and hence produce few candidates to begin with or may, like many immunoglobulin sequences, be too variable. These cases will require further exploration (alternative proteolytic enzymes or sample derivatization procedures). 3) Genetic variants altering a single amino acid in the selected peptide will prevent its determination by the wild-type MRM; hence these must be recognized in advance and designed specifically. Provided these issues are taken into account, none appears to prevent productive use of peptide MRMs for measuring abundances of plasma proteins.
Based on the results presented here, numerous clinically important plasma proteins can be measured by peptide MRMs with precision comparable to current clinical immunoassays. The panel of assays presented should have general use as the nucleus of candidate-based biomarker validation approach and should lead ultimately toward a comprehensive assay platform for all human proteins in plasma.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Published, MCP Papers in Press, December 6, 2005, DOI 10.1074/mcp.M500331-MCP200
1 The abbreviations used are: CV, coefficient of variation; QqQ, triple quadrupole; MRM, multiple reaction monitoring; S/N, signal-to-noise; Nat, natural sample-derived peptide; SIS, stable isotope-labeled internal standard; polySIS, polyprotein SIS; SISCAPA, stable isotope standards and capture by anti-peptide antibodies; MIDAS, multiple reaction monitoring-initiated detection and sequencing; MARS, multiple affinity removal system. 
2 L. Anderson and C. L. Hunter, manuscript in preparation.
* 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.
To whom correspondence should be addressed: The Plasma Proteome Inst., P. O. Box 53450, Washington, D. C. 20009-3450. Tel.: 301-728-1451; Fax: 202-234-9175; E-mail: leighanderson{at}plasmaproteome.org
|
REFERENCES |
|---|
|
TOPABSTRACTEXPERIMENTAL PROCEDURESRESULTSDISCUSSIONREFERENCES |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
