Multiple Reaction Monitoring-based, Multiplexed, Absolute Quantitation of 45 Proteins in Human Plasma^*

Reagents and Chemicals

All reagents were American Chemical Society (ACS) grade or higher. All solvents used, including water, were LC/MS grade.

Synthesis of Isotopically Labeled Tryptic Peptides

Isotopically labeled peptide standards were synthesized at a 5-μmol scale using Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry with a Protein Technologies Prelude peptide synthesizer (Tucson, AZ). Isotopically labeled amino acids, [¹³C₆]Lys (98% isotopic enrichment) and [¹³C₆]Arg (98% isotopic enrichment), were purchased from Cambridge Isotope Laboratories (Andover, MA) and were conjugated to TentaGel R resin by Rapp Polymere (Tubingen, Germany). Subsequent amino acid residues (100 mm) were double coupled using 20% piperidine as the deprotector and 1H-benzotriazolium 1-[bis(dimethylamino)methylene]-5-chloro-,hexafluorophosphate (1-),3-oxide (HCTU) as the activator. The cleavage was performed with 95:2.5:2.5 TFA:water:triisopropylsilane.

Purification of Synthetic Peptides

The cleaved peptides were removed from the synthesizer, and the TFA was evaporated under a stream of nitrogen. Ether was added to precipitate the peptides, and after centrifugation at 3000 × g for 5 min, the ether layer was decanted. Peptides were resolubilized in 0.1% TFA and purified by reversed-phase HPLC (Ultimate 3000, Dionex) while monitoring the peptide elution at 230 nm. The crude peptides were separated using a Vydac C₁₈ column (10 × 250 mm, 10-μm resin) with a linear gradient of 0.1% TFA in water (v/v) and 0.085% TFA in 50% acetonitrile (v/v) at a flow rate of 4 ml/min over 60 min. Fractions of interest were spotted onto stainless steel MALDI plates and measured by MALDI-TOF (Applied Biosystems/MDS Sciex, Concord, Ontario, Canada) mass spectrometry. Fractions containing greater than 80% of the target peptide by MALDI-TOF analysis were pooled and lyophilized.

Lyophilized peptides were resolubilized in 30% acetonitrile, 0.1% formic acid, and peptide concentrations were subjected to acid hydrolysis and amino acid analysis. Purity of the HPLC-purified peptides was determined by capillary zone electrophoresis (CZE) using a P/ACE^TM MDQ System (Beckman Coulter, Fullerton, CA) equipped with a UV detector monitored at 200 nm. All separations were performed at 25 °C in a bare fused silica capillary (Polymicro Technologies, Phoenix, AZ) of dimensions 50-μm inner diameter × 360-μm outer diameter, 60-cm total length, and 50-cm effective length. The applied voltage was 15 kV, and the background electrolyte was 50 mm sodium phosphate adjusted to pH 2.5 with phosphoric acid. The capillary was rinsed at 20 p.s.i. with 1 m NaOH for 1 min, Millipore-purified water for 1 min, 1 m HCl for 2 min, 0.1 m HCl for 3 min, and finally with background electrolyte (n-butyl glycidyl ether) for 3 min between each separation. Injection of 20 nl of each peptide solution (ranging from 4 to 10 pmol/μl in 30% (v/v) ACN, 0.1% FA) was accomplished by application of 0.5 p.s.i. for 20 s at the inlet end of the capillary. Purity was assessed by peak height for a given concentration of peptide injected relative to signal to noise of the background signal. The limit of quantitation was determined to be ∼150 nmol/μl. The absolute concentration of each synthetic SIS peptide was determined by amino acid analysis. These absolute concentrations were adjusted by the percent purity of each synthetic peptide as determined by CZE. This corrects for any possible contribution from incorrect or partial synthesis products.

MRM Q1/Q3 Ion Pair Selection by Nanoinfusion

Isotopically labeled peptides were diluted to 1 pmol/μl (1 mm) in 30% acetonitrile, 0.1% formic acid for infusion at a flow rate of 300 nl/min using a Harvard PicoPlus 11 syringe pump (Harvard Apparatus, Holliston, MA). Infused peptide solutions were analyzed by nanoelectrospray using a 4000 QTRAP hybrid triple quadrupole/linear ion trap mass spectrometer (Applied Biosystems/MDS Sciex) equipped with a nanospray ionization source. MS analysis was conducted in the positive ion mode with ion spray voltages in the 1800–2000-V range. The declustering potential was ramped (0–120 V in 2-V increments) during Q1 scans centered on 10-Da-wide mass ranges. MRM scans for optimization of MRM Q1/Q3 ion pairs were conducted with both Q1 and Q3 set to unit resolution (0.6–0.8-Da full width at half-height) while the collision energy was ramped (5–120 V in 2-V increments). An MS operating pressure of 3.5 × 10⁻⁵ torr was used during all MRM scans.

EDTA-plasma Collection

Plasma was collected from a healthy male donor who fasted for 16 h prior to blood collection. Blood was collected by venous puncture using a 21-gauge BD Vacutainer Multiple Sample Needle (reference number 367213, BD Biosciences). A total of 60 ml of blood was collected into 20 × 3.0-ml BD Vacutainer lavender tubes with 5.4 mg of K₂-EDTA (reference number 367856, BD Biosciences). Immediately following blood collection and mixing with the EDTA in the tubes, the samples were centrifuged at 1000 × g for 15 min at 22 °C to pellet the cells. Plasma was collected and centrifuged again to remove any remaining cells. The plasma was then divided into 1.0-ml aliquots in sterile cryotubes and immediately frozen at −80 °C for storage. Total elapsed time from collection to storage was 1 h. Plasma analyzed in the experiments described in this study had been stored at −80 °C for 4–8 weeks.

Preparation of Plasma Tryptic Digests and Addition of Internal Standard Peptides

Plasma tryptic digests were prepared by diluting 5 μl of whole plasma (∼350 μg of total protein) 1:10 with 25 mm ammonium bicarbonate prior to denaturation with 50 μl of sodium deoxycholate (10% (w/v) in 25 mm ammonium bicarbonate). Denatured plasma samples were reduced for 30 min at 60 °C with 5 mm tris(2-carboxyethyl)phosphine (net 22.5-fold excess over the ∼26 mm concentration of protein cysteine in plasma computed from known protein concentrations). Reduced plasma samples were alkylated for 30 min at 37 °C in the dark with 10 mm iodoacetamide (net 50-fold excess over plasma protein cysteine). Ammonium bicarbonate (25 mm) was added to the digests to reduce the sodium deoxycholate concentration to 1% (w/v) and to achieve a 500-μl digest volume containing a 1:100 dilution of plasma upon addition of trypsin. Modified, sequencing grade trypsin (Promega, Madison, WI) was added to the samples at a 20:1 substrate:enzyme ratio. Digestion was carried out for 16 h at 37 °C.

Samples were acidified by adding an equal volume of 1% (v/v) formic acid to stop digestion. For quantitation of unknown samples, a concentration-balanced mixture of 45 isotopically labeled internal standard peptides (0.1% (v/v) formic acid) was added to each plasma digest at a ratio of 1 volume of SIS peptide mixture to 4 volumes of acidified plasma trypsin digest. This came to 50 μl of SIS peptide mixture added per 1-μl equivalent of neat plasma.

Samples were desalted and concentrated prior to MS analysis by solid phase extraction using Waters Oasis reversed-phase 10-mg HLB cartridges (Waters, Milford, MA) following the manufacturer's recommended protocol. Samples were eluted with 200 μl of 50% (v/v) acetonitrile, 0.1% (v/v) formic acid. The eluted samples were frozen and lyophilized to dryness. Prior to LC-MRM/MS analysis, samples were reconstituted in 0.1% formic acid to a concentration of ∼1 μg/μl based on an initial plasma protein concentration of 70 mg/ml.

LC-MRM/MS Analysis of Plasma Digests

An Eksigent NanoLC-1Dplus HPLC was used for the injection of desalted plasma digest samples (1 μl) onto reversed-phase capillary columns (75 μm × 15 cm) packed in house using Magic C18AQ (5-μm diameter particles, 100-Å pore size; Michrom, Auburn, CA). A flow rate of 300 nl/min solvent A (2% acetonitrile, 0.1% formic acid) was used for 6 min. Separations were performed using a flow rate of 300 nl/min with a 32-min linear gradient from 0 to 23% solvent B (98% acetonitrile, 0.1% formic acid) followed by a 9-min linear gradient from 23 to 43% solvent B.

An Applied Biosystems/MDS Sciex 4000 QTRAP with a nanoelectrospray ionization source controlled by Analyst 1.5 software (Applied Biosystems) was used for all LC-MRM/MS analyses. All acquisition methods used the following parameters: 1900–2000-V ion spray voltage, a curtain gas setting of 25 p.s.i., a 200 °C interface heater temperature, a collision-activated dissociation pressure at 3.5 × 10⁻⁵ torr, and Q1 and Q3 set to unit resolution (0.6–0.8-Da full width at half-height).

Spray stability was improved and lifespan of the uncoated fused silica emitter tips (20-μm inner diameter, 10-μm tip; New Objective, Woburn, MA) was improved by use of 3–5-p.s.i. sheath gas and postcolumn, prespray addition of makeup solvent (80% (v/v) isopropanol) at a flow rate of 100 nl/min using a PicoPlus 11 syringe pump (Harvard Apparatus). MRM acquisition methods were constructed using 90 MRM ion pairs with peptide-specific tuned declustering potential (DP), collision energy (CE) voltages, and retention time constraints. A default collision cell exit potential of 23 V was used for all MRM ion pairs, and the scheduled MRM option was used for all data acquisition with a target scan time of 2 s and an 8-min MRM detection window.

MRM Data Analysis

All MRM data were processed using MultiQuant 1.0 (Applied Biosystems) with the MQL algorithm for peak integration. A 2-min retention time window with “report largest peak” enabled and a three-point smooth with a peak splitting factor of 2 was used. The default MultiQuant values for noise percentage and base-line subtraction window were used. All data were manually inspected to ensure correct peak detection and accurate integration. Linear regression of all calibration curves was performed using a standard 1/x (x = concentration ratio) weighting option to aid in covering a wide dynamic range (20). Raw data files for these analyses are available from the ProteomeCommons Tranche network (http://www.proteomecommons.org/data-downloader.jsp?fileName=O3Hbto5xoHSurD7i5FYQMe6IKixjYxsB8ao8W/gpnNmIxQX2hNxoHXJORjyDg9s5x1mVmm+7+J33b8FaEUBB7MuoGMAAAAAAAAAUXA==) using hash codes.

Selection and Synthesis of Isotopically Labeled Proteotypic Peptide Standards

Development of protein assays based on proteolytic peptide surrogates requires that the peptides selected are both detectable in every sample and reproducibly observed between sample preparations. Because of differing peptide ionization efficiencies and susceptibilities to CID, selection of a peptide to represent a protein in an MRM assay is a crucial step that dramatically affects the ultimate sensitivity and specificity of an assay. When selecting a peptide, several critical factors must be considered. 1) The peptide amino acid sequence must be unique to the target protein and contain no missed cleavage sites, 2) the peptide should be reproducibly observed in proteolytic digests, and 3) the peptide should not contain amino acids that are susceptible to chemical modification (Cys and Met).

When working with a species for which the genome has been sequenced and widely studied, as it has been for Homo sapiens, publicly accessible MS/MS spectra databases such as the Global Proteome Machine and Peptide Atlas are useful for determining which tryptic peptides of proteins are frequently observed by MS/MS analysis (21, 22). The concept of using “proteotypic” peptides for MS-based protein identification and quantitation is gaining wider acceptance, and bioinformatics tools are emerging to assist in their computational prediction (23–25). Additional experiments, however, are still required to verify the utility of these proteotypic peptides for use in MRM assays because public databases of MS/MS spectra often lack information about experimental methods and the MS instrumentation used to obtain these spectra.

We developed an experimental work flow to rapidly select proteotypic peptides that are suitable for use as internal standards with the intention of a creating an expandable mixture of SIS peptides for multiplexed absolute quantitation of proteins by LC-MRM/MS. Stable isotope-labeled derivatives of these peptides were synthesized and were used to develop MRM assays (Fig. 1). Our work flow contains checkpoints that quickly identify proteins with problematic proteotypic peptides (for example, the synthetic peptide is insoluble, the natural MRM signal does not cochromatograph with the SIS peptide, etc.), and these proteins are marked for selection of alternate proteotypic peptides.

For our initial studies, we selected 45 tryptic peptides from a list of moderate to high abundance plasma proteins shown previously to be reproducibly detectable by LC-MRM/MS analysis of whole plasma trypsin digests (14) (Table I). Nearly every one of these 45 proteins has been reported to be a potential marker of a clinically relevant disease (Table I) (18, 26–43). Thirty-one of these 45 proteins have been reported previously to be associated with cardiovascular disease (18).

Isotopically labeled forms (containing either a C-terminal [¹³C₆]Arg or [¹³C₆]Lys residue) of all 45 peptides were synthesized on a micromole scale. The C terminus of each peptide was selected for stable isotope-labeled amino acid incorporation to mass shift the entire y series fragment ions, which is the dominant ion series in CID fragmentation spectra of tryptic peptides (44), of each peptide. Selection of a y series ion as the Q3 fragment ion for each MRM Q1/Q3 ion pair thus ensures that both the Q1 and Q3 m/z values differ from those of the natural peptide. This maximizes the specificity of the MRM signal and eliminates potential cross-talk between the natural and SIS peptide MRM signals.

Proteotypic peptides selected by Anderson and Hunter (14) for apolipoprotein C-III, clusterin, and complement component C9 formed insoluble precipitates following cleavage from synthetic support resins. In cases of problematic peptides such as these, the Global Proteome Machine database of MS/MS spectra was used to select alternative peptides for MRM assay development (21). To enable determination of the absolute concentrations of the endogenous proteins by MRM, the concentrations of all of the purified synthetic peptides were determined by amino acid analysis (supplemental results). Postsynthesis peptide purity was assessed by both MALDI-TOF MS analysis and CZE. By CZE, the purity of the peptides was found to range from 77.4 to 98.7% with an average purity of 96.1% (supplemental results). The concentrations of the peptide stock solutions as determined by amino acid analysis were corrected using the percent purities determined by CZE analysis.

Optimized Selection of MRM Q1/Q3 Ion Pairs for Each Peptide

Consensus spectra collected in the Peptide Atlas and the Global Proteome Machine database can be useful for selecting MRM precursor and fragment ion pairs prior to SIS peptide synthesis; however, their utility in selecting the most sensitive MRM precursor/fragment ion pair for a peptide is limited. Creating the most sensitive MRM assay for a peptide requires judicious selection of both precursor ion charge state and fragment ions combined with empirical tuning of MS parameters. Public MS/MS data repositories were acquired with the intent of peptide sequence assignment. This requires fragmentation spectra that represent the entire peptide fragment ion series. The signal intensity of an MRM Q1/Q3 ion pair for a peptide is determined by the combined ionization efficiency of the precursor peptide and how readily it can be dissociated into a limited number of fragment ion species. Even sophisticated bioinformatics tools for MRM prediction are currently unable to predict optimal collision energy voltages without the use of simple formulas that directly correlate collision energy voltages with precursor ion m/z (24, 45).

Because the synthetic SIS peptides ionize and fragment identically to the natural peptides in the mass spectrometer, synthetic peptides can be used to optimize the MS analysis parameters. To ensure that the most sensitive combination of precursor and fragment ions are ultimately selected as the MRM ion pair for each peptide, peptide solutions were infused by nanoelectrospray ionization, and Q1 and MRM scans were used to optimize both the precursor and fragment ion signal intensities, respectively.

First the peptide ionization efficiency was optimized, and the most intense charge state was determined by ramping the DP voltages during the Q1 scans (Fig. 2, a and b). Tuned DP voltages ensure maximum efficiency of ion transfer of the precursor ion into the MS instrument and affect the maximum MRM signal that can be achieved with each peptide. Using the tuned Q1 parameters for each peptide, we next determined which were the dominant CID fragment ions that were generated from each peptide to create an MRM ion pair. To determine the most abundant fragment ions for each peptide, the infused peptides were analyzed using MRM ion pairs containing all possible [M + H]⁺ and [M + 2H]²⁺ b and y series fragment ions while ramping the collision energy (Fig. 2, c and d). Signal intensities from all MRM Q1/Q3 ion pairs for each peptide were ranked to ensure selection of the most intense precursor and fragment ion pair for MRM-based quantitation. This approach resulted in selection of CE voltages that maximize the generation of each fragment ion species (supplemental results). As a result of these studies, we found that doubly charged y ion fragments were the highest intensity fragment ions for 15 of the 45 peptides (Table I).

There are already several computational methods for the prediction of both proteotypic peptides and their corresponding Q1/Q3 ion pairs for MRM analysis (24, 25). The MIDAS^TM Workflow Designer (Applied Biosystems) software utility predicts the peptide charge state based on the number of basic residues in a peptide, and commonly uses y ion fragments with m/z values greater than the precursor ion m/z as Q3 masses for the peptide MRM ion pair (46). Collision energy voltages for peptide MRM ion pairs are often predicted using a generic formula (CE = 0.05 × (precursor m/z) + 5) that directly correlates peptide ion m/z and charge state with the predicted CE voltage (14). To assess the value of empirically selecting the most intense MRM Q1/Q3 ion pair for a peptide, signal intensities obtained with our optimized MRM pairs were compared with the signal intensities obtained with MIDAS-predicted MRM Q1/Q3 ion pairs. Although the same Q3 fragment ion was predicted by MIDAS for eight of the 45 peptides, the MIDAS-predicted CE and DP values resulted in lower observed signal intensities for these MRM ion pairs. On average, the signal intensity obtained with MRM parameters that were empirically determined by nanoinfusion analysis resulted in an 11.4-fold increase in signal intensity when compared with MRM parameters predicted by MIDAS (Table I).

Although we feel that optimizing the MRM CE voltages yields better results, we analyzed the linear correlation between precursor ion m/z and CE voltage for 45 unique peptides. We found that an improved formula for prediction of CE voltages suitable for MRM is defined as CE = 0.043 × (precursor ion m/z) + 2.25. This equation fits the data with a linear correlation coefficient (r) of 0.79.

Multiplexed LC-MRM/MS Analysis of Trypsin-digested Plasma

Identical MRM parameters can be used for the natural and heavy forms of each peptide taking into account Q1/Q3 mass differences due to stable isotope labels. To select the most sensitive MRM ion pair for each peptide, the three most intense MRM ion pairs determined by nanoinfusion analysis were tested under LC-MRM/MS conditions. LC-MRM/MS analysis of trypsin-digested plasma (1 μg) mixed with all 45 SIS peptides (100 fmol each) confirmed that the relative intensity of the top three MRM ion pairs identified by nanoinfusion was retained under LC-MRM/MS conditions (supplemental results). This analysis also permitted confirmation of the identity of ion signals observed for natural peptides.

Although MRM signals are highly analyte-specific, in a sample as complex as trypsin-digested plasma, multiple peaks generated by non-target ion species can often be observed for an MRM Q1/Q3 ion pair during an LC-MRM/MS analysis. The three most intense MRM ion pairs for each peptide were used to confirm the identity of all 45 natural peptide peaks by co-elution with their SIS peptide standard during an LC-MRM/MS analysis and allowed determination of their corresponding retention times (Fig. 3a). The benefit of the increased specificity due to SIS peptide co-elution was dramatically highlighted in the extracted ion chromatograms (XICs) of fibrinogen γ chain and gelsolin (Fig. 3a). Nonspecific signals were detected in the XICs of the MRM ion pairs for the natural forms of fibrinogen γ chain and gelsolin, and the two peaks observed for fibrinogen γ chain eluted within 30 s of each other, but only the early peak co-eluted with the authentic fibrinogen γ chain SIS peptide, allowing correct selection of the analyte peak. Analyte retention time certainty permits the use of retention time constraints as a third dimension of analyte specificity for each MRM assay.

Analyte-specific retention times (scheduled MRMs) for MRM ion pairs has only recently been made available to 4000 QTRAP users via a software and MS firmware upgrade. The ability to schedule MRMs based on retention time constraints results in a variable number of MRM Q1/Q3 ion pairs being scanned in each MS cycle, dramatically increases the maximum number of Q1/Q3 pairs that can be monitored per analysis, and also improves the ion statistics collected during the elution profile of a peptide (17).

Specificity of MRM Quantitation

Specificity and analytical precision are key performance characteristics of any assay. With the use of SIS peptides as internal standards, each MRM assay has three analyte-specific criteria: precursor ion m/z, fragment ion m/z, and retention time. Although we characterized top three MRM ion pairs for each peptide, to maximize the analyte capacity of multiplexed MRM quantitation, we focused on minimizing the number of MRM ion pairs in our analysis method. Careful selection of these criteria approaches absolute specificity of each MRM ion pair for the intended analyte.

Prior to selecting a single representative MRM ion pair for each peptide, the relative response of the three most intense MRM Q1/Q3 ion pairs for the natural and SIS peptides as determined by nanoinfusion analysis were compared by LC-MRM/MS analysis of both a plasma digest spiked with all 45 SIS peptides and a mixture of pure SIS peptides alone. This ensured selection of MRM ion pairs that are free of matrix interference from co-eluting ions (supplemental results). To rigorously characterize the linearity and sensitivity of these MRM assays, we chose to maximize the ion statistics collected per peptide by using a single MRM ion pair per peptide. The highest intensity MRM ion pair containing a Q3 y ion fragment free of matrix interference was selected as the ideal ion pair for each peptide (Table I, MRM ion pairs highlighted in bold). The final MRM analysis method was a 60-min analysis containing 90 MRM ion pairs with analyte-specific retention time limits (Fig. 3b) (17).

Creation of an SIS Mixture for Multiplexed MRM Analysis of 45 Peptides in Plasma

To achieve reproducible, multiplexed quantitation of 45 peptides in a single LC-MRM/MS analysis, a mixture of all 45 SIS peptides is needed that permits accurate, single step addition of the standard peptides to plasma digests. Quantitation with stable isotope dilution and LC-MRM/MS is linear across a broad 10³–10⁴ concentration range (13), but factors such as sample matrix and SIS peptide isotope impurity often reduce the observed linear dynamic range of individual peptides (15). Because the natural molar concentrations of the 45 plasma proteins to be quantified differ by >10⁴ (650,000 and 28 pmol/ml for albumin and L-selectin, respectively) (47, 48), an equimolar mixture of the SIS peptides would generate high ratios for certain analytes and low ratios for other. This would affect their quantitation accuracy.

Clinical guidelines for creating MS-based assays require that internal standards be added to samples at concentrations close to the mean concentration of the endogenous target analytes or at sufficient levels to permit reproducible measurement (20). Following these criteria, the concentration of each SIS peptide was titrated to determine a concentration that would result in a SIS peptide peak area within a factor of 10 of the endogenous peptide peak area when the SIS peptide was added to a standard plasma trypsin digest. As an example of this concentration balancing, the relative amounts of the highest and lowest abundance peptides, albumin and L-selectin, differed by more than a factor of 100 in the final concentration-balanced SIS peptide mixture. Peak area ratios between 0.1 and 10 (natural:heavy) were obtained for 41 of the 45 peptides when the balanced 45-peptide mixture was added to a trypsin-digested plasma sample and analyzed by LC-MRM/MS (Table II and Fig. 3a).

The volume of plasma that should be trypsin-digested and the corresponding amount of SIS peptide mixture required were considered when selecting a final concentration for the SIS peptide mixture. When handling high concentration protein samples like plasma, the error introduced during pipetting is a concern. As a result, 20–50 μl of a 10-fold dilution of plasma is routinely used for the tryptic digestion, although only 14 nl of neat plasma (∼1 μg of total protein) is required per LC-MRM/MS analysis. The final concentration of the balanced SIS peptide mixture was adjusted so that a 50-μl volume of the mixture was added to each microliter of neat plasma.

The SIS peptide mixture can be added to plasma samples either pre- or post-tryptic digestion. It is normally best to add internal standards as early in a sample preparation work flow as possible to compensate for any variation, like nonspecific degradation, during sample preparation. However, when the SIS peptides were added early in the work flow prior to digestion, we found that the peak area ratios (light:heavy) were elevated when compared with those obtained with postdigestion addition of SIS peptides (data not shown). The degree of the peak area ratio increase varied depending on the peptide and was not predictable. These ratio increases suggest that internal standard peptides are either being degraded or chemically modified over the course of the digestion. Depending on how rapidly peptides are liberated from their parent protein by trypsin, an internal standard peptide may be exposed to degradation effects to which the natural form is not subjected. To avoid this source of variation, in all subsequent experiments the SIS peptide mixture was added to the plasma tryptic digests postdigestion, i.e. during digest acidification and prior to desalting and concentration by solid phase extraction, so that any losses incurred during sample handling would still equally affect both natural and heavy peptides. Postdigestion addition of standards also requires less of the SIS peptide mixture because only as much sample as is required for LC-MRM/MS analysis needs to be prepared.

Reproducibility of MRM Quantitation with SIS Peptides

To characterize the effects on the analytical precision of the MRM measurements when the natural peptide peak areas were normalized against SIS peptide peak areas, we conducted replicate analyses of a standard plasma digest sample. This data set was used to evaluate how using peak area ratios, instead of analyte peak areas alone, affects measurement reproducibility (Fig. 4). Two aliquots from a plasma tryptic digest were spiked with either an equimolar mixture or a concentration-balanced mixture of all 45 SIS peptides. Each spiked plasma sample was analyzed by LC-MRM/MS 12 times with blank solvent injections between sample injections. Measurement of analyte peak areas over the course of all 24 LC-MRM/MS analyses resulted in >10% variation for all 45 analytes with peak area CVs of >20% for 39 of the 45 peptides (Fig. 4a).

Normalizing the natural peptide peak areas against the peak areas of co-eluting SIS peptides should improve the analytical precision by accounting for subtle interanalysis variations in ionization efficiency (20). By normalizing the natural peptide peak areas to the peak areas of the co-eluting reference SIS peptides, peak area differences between analyses can be effectively removed. Analytical precision was dramatically improved when peak area ratios were used to compare data between replicate LC-MRM/MS analyses (Fig. 4, b and c).

An unanticipated benefit of balancing the SIS peptide concentrations, which had been primarily performed to maximize the linear range of the assays, was a further reduction of analytical variation between analyses (Fig. 4c). When a concentration-balanced mixture of SIS peptides was spiked into plasma tryptic digests, MRM assays for 29 of the peptides had CVs of <4%, whereas only one MRM assay (for complement component C9) had a CV >10%. When equimolar amounts of each SIS peptide were added to the same plasma digest, only 13 MRM assays had CVs of <4%, whereas three assays had CVs of >10% with a maximum variation of 20.2% observed for vitamin D-binding protein. We attribute the additional reduction in percent CV that resulted from the use of concentration-balanced SIS peptides to the increased signal intensity and the improvement in signal quality for 18 of the SIS peptides whose concentrations had been increased to match the areas from the natural peptides. All subsequent MRM quantitation was conducted using the concentration-balanced 45-SIS peptide mixture.

To compare the analytical precision obtained with each of the top three most intense MRM ion pairs for each peptide, a plasma tryptic digest spiked with the concentration-balanced mixture of SIS peptides was analyzed 18 times by LC-MRM/MS. Although the lowest percent CVs did not always correlate with the highest intensity fragment ion, the reproducibility of the three MRM ion pairs did not vary greatly for each MRM peptide (Table I). For 41 of the MRM peptide assays, the difference in percent CV observed between MRM ion pairs was <5%. Increased variation correlated with decreasing fragment ion intensity. The largest range of variation between MRM ion pairs was <10% and correlated with low intensity fragment ions for lower concentration analytes (Table I).

Linear Response of MRM Quantitation

To ensure that analyte concentrations routinely encountered when quantifying unknown samples are measured within the linear response range for that analyte, the assays must be validated over a wider concentration range (49). Assay attributes such as linearity of response, limit of quantitation (LOQ), and interassay variation also need to be characterized to validate an assay (49). These parameters allow comparison with established assay technologies and are of particular interest when considering the potential translation of MRM-based protein assays into clinically relevant applications.

The standard method for characterizing assay linearity and LOQ would be to add standard proteins to a constant amount of blank sample matrix and internal standards. However, when assaying endogenous plasma proteins, preparation of a blank sample matrix devoid of all 45 endogenous proteins is not feasible. Similar difficulties were encountered when developing an MRM-based assay for alcohol dehydrogenase, which is endogenously expressed in liver (50). In such cases, a dilution series of pure internal standards is sometimes used to characterize assay linearity, but this likely over-represents the true linear range of the assay due to the absence of interfering components from the sample matrix (9, 51). We therefore used an alternative method, similar to that of Janecki et al. (50), to determine linear range. We diluted a tryptic digest of our standard plasma sample to generate a range of endogenous analyte concentrations spanning a 31,250-fold concentration range. These various analyte concentrations were combined with a constant amount of the concentration-balanced 45-SIS peptide mixture. Five replicate LC-MRM/MS analyses of each sample dilution were performed in the order from most dilute to most concentrated with two blank solvent injections between the different sample concentrations. To optimize ion statistics and sensitivity, a single MRM ion pair was used for each peptide (fragment ion noted in Table II).

Linear regression analysis was performed on the observed peak area ratios (natural:heavy) versus concentration ratios to generate calibration curves for each peptide (Fig. 5 and supplemental results). To calculate concentration ratios, we first needed to estimate the concentrations of all 45 natural peptides in our standard plasma sample because their actual concentrations are unknown. To accomplish this, a single concentration point was measured by LC-MRM/MS five times, and the concentration of each natural peptide was calculated from the average ratio compared with the area of its SIS peptide, which is at a known concentration (Table II). Then as we create a dilution series for the standard curves, we can measure how the observed peak area ratios change in response to changes in the peptide concentrations.

To avoid over-representing the degree of linearity in our calibration curves, we combined linear regression analysis with response factor plots to identify analyte concentrations that did not respond in a linear manner (49). Ideally response factor plots should have a near zero slope. Peak area ratios were sequentially excluded from the linear regression analysis until individual response factors for all data points included in the regression analysis fell within 80–120% of the average analyte response (slope) (Fig. 5b) (49). This approach ensures that the best linear fit is achieved, thereby improving quantitation accuracy and accurately representing the linear concentration range of each peptide assay.

Using this approach to define the assay linear range, 24 of the MRM peptide assays yielded linear responses over a >1000-fold concentration range (Table II). Using response factor plots to define the assay linear range also resulted in strong linear correlations (r > 0.99) for 42 of the 45 assays with average response factors (slopes) of 0.9–1.1 for 38 of the 45 assays (Table II). Biologically concentrations of these 45 plasma proteins typically vary by only a factor of 2–10 (48). Therefore, routine quantitation of these 45 proteins in unknown samples is performed using calibration curves that represent a narrower 30-fold concentration range. Linear concentration ranges over concentration ranges of >30 were obtained for 43 of the 45 peptide assays. By validating the linearity of these assays over a much wider concentration range, we can ensure that any biological concentrations encountered will fall well within the linear portion of the calibration curve.

Limit of Quantitation by MRM

The LOQ for an assay is an important measure of performance as it defines the lowest analyte quantity that can be accurately measured, and it is the true measure of the sensitivity of an assay. A signal-to-noise ratio of >10 is frequently used to define LOQ (15, 51). However, because MRM is an MS/MS scan mode, background noise is extremely low, and there is no practical way to accurately measure co-eluting noise in an MRM Q1/Q3 ion pair channel of interest. Therefore, it is advantageous to empirically determine the LOQ of an assay, defined as the lowest analyte concentration that can be measured with <20% CV (49). Five replicate LC-MRM/MS analyses of each analyte concentration in the dilution series used to define the assay linear range were used to define LOQs for each assay.

When considered together, assay linearity, linear range, and LOQ can help to identify problematic proteotypic peptides. One such problematic peptide is LLIYAVLPTGDVIGDSAK for α₂-macroglobulin. Although α₂-macroglobulin was measured with excellent analytical reproducibility (2.5% CV), the current assay had a poor linear dynamic range and an LOQ only 5-fold lower than the concentration normally encountered during routine quantitation of unknowns (Table II). No interference from co-eluting ion species was detected when the relative response of multiple MRM ion pairs was compared for analysis of this peptide in solvent alone versus in a sample matrix (supplemental results), and this peptide was well resolved as a symmetrical peak using our chromatographic conditions (Fig. 3a). These results suggest that an alternative proteotypic peptide should be selected for further development of an α₂-macroglobulin MRM assay.

Emphasizing the sensitivity of these MRM assays, LOQs for 27 of the peptides were found to be in the attomole range per analysis for a peptide on column (Table II). To place the LOQ of each peptide in context, it is important to take into account how the LOQ for an analyte relates to the natural biological concentration of the protein. Even a 4.3-fmol LOQ for albumin is still >2000 times lower than the natural mean concentration of albumin, a concentration difference far greater than what would be naturally encountered (48).

Reproducibility and Accuracy of MRM Quantitation

When characterizing assay reproducibility, it is important to determine the amount of technical variation introduced during sample preparation to estimate how well an assay can detect changes in analyte concentration. Analysis of replicate preparations of a standard sample can be used to assess the interassay reproducibility, or ruggedness, of an assay. An analytical assay can be further “stress-tested” by measuring the variation observed when these standard samples are prepared over the span of multiple days (49). To measure the interassay reproducibility of these 45 MRM peptide assays, three analytical runs containing 11 separate tryptic digests of a standard plasma sample were performed on different days. All samples were analyzed using the top three MRM ion pairs for each peptide (Table I) to determine whether reproducible concentration determination is achievable with a single MRM ion pair. Analyte response was measured for each day using one of the 11 plasma digests to create a six-point calibration curve that spanned a 30-fold range of concentration ratios. To create these calibration curves, the concentration of the SIS peptides was held constant while the amount of total digest was varied over a 30-fold concentration range. The highest and lowest concentration ratios used for these calibration curves contained 3-fold higher and 10-fold lower concentrations of natural peptide than used in unknown samples. Concentrations of the natural peptides in the standard sample were estimated using the calibration curve from the first analytical run as described above.

Analyte protein concentrations were calculated as the average of 30 measurements across all three analytical runs. Each analytical run consisted of a six-point calibration curve and 10 manually prepared plasma trypsin digests, and each run was performed and analyzed on a different day. When measured across three separate analytical runs, concentrations of 25 of the proteins had CVs of <10%, whereas 42 of the assays exhibited clinically useful CVs of <20% (Table II). These results demonstrate the highly reproducible nature of MRM protein quantitation and highlight its potential for clinical applications. Interassay variation should be further reduced by automation of sample preparation. Protein concentrations, determined from MRM assays based on a single peptide and multiple MRM Q1/Q3 ion pairs for each protein, were found to agree with previously reported clinical ranges for 23 of the proteins (Table II and supplemental results) (47, 48). An additional 16 measured protein concentrations were within a factor of 2 of literature values. Excellent agreement of concentrations calculated using different MRM ion pairs was obtained even when low mass a2 and b2 fragment ions, which should statistically be more susceptible to nonspecific interferences, were used. MRM ion pairs exhibiting interference were readily identified by changes in both their signal intensity relative to the other two MRM ion pairs and either an increase or a decrease in the calculated protein concentration (supplemental results). All protein concentrations were calculated with less than 30% technical variation except for vitamin D-binding protein (Table II). Closer examination of the vitamin D-binding protein results revealed that the signal obtained from the MRM ion pair used for quantitation (390.9/493.3) was attenuated in approximately half of the samples. The vitamin D-binding protein MRM ion pairs containing b3 and b5 ions did not exhibit this behavior and gave higher calculated concentrations that were just below clinically reported concentrations and only differed from each other by only 6.4% with CVs of 5.7 and 5.6%, respectively (supplemental results).

Fibrinogen α, β, and γ should be present at equimolar concentrations in plasma (48). Our MRM assays for fibrinogen α and β resulted in concentrations that fall within clinically reported ranges, whereas the concentration measured for fibrinogen γ was >10-fold lower than expected (Table II). This could be explained by incomplete tryptic digestion of fibrinogen γ. We are currently using our mixture of SIS peptides to characterize the kinetics of trypsin digestion for each of these 45 proteins. Preliminary data indicate that the levels of the fibrinogen α and β proteotypic peptides that we are measuring plateau within 2 h during trypsin digestion, whereas the fibrinogen γ peptide increases linearly in concentration over 22 h (data not shown). These results indicate that this particular fibrinogen γ peptide is located within a region of the protein that is resistant to tryptic digestion and that an alternate proteotypic peptide should be selected for development of a fibrinogen γ MRM assay. Such studies will assist in improving the overall accuracy of multiplexed protein quantitation achievable through the use of this balanced mixture of SIS peptides. Furthermore the speed of LC-MRM/MS analysis, combined with the application of this balanced mixture of SIS peptides, should be extremely valuable for revealing previously undetectable patterns of protein expression characteristic of specific diseases.

Conclusion

We have described the creation of an optimized mixture of 45 SIS peptides that we used to develop rapid, specific, and sensitive MRM assays for a panel of 45 human plasma proteins. Requiring minimal sample preparation, these MRM protein assays were reproducible and linear across a broad range of concentrations. This panel of isotopically labeled peptides should be easily expandable to include additional proteins. This would make this assay very useful for the discovery of biomarkers through expression profiling. A recent review of plasma protein concentrations by Hortin et al. (48) lists an additional 100 plasma proteins whose mean molar concentrations exceed that of L-selectin, which is the lowest concentration analyte of our 45-protein panel (14). These additional 100 plasma proteins should be directly detectable in tryptic plasma digests by LC-MRM/MS analysis and represent logical candidates for addition to our SIS peptide mixture. Ultimately this initial panel of characterized MRM assays for 45 human plasma proteins could be expanded into an assay capable of rapid, multiplexed profiling of more than 100 plasma proteins.

Accurate determination of analyte concentration is an essential characteristic of any assay for clinical use. Plasma protein concentrations from a single donor sample measured by MRM quantitation using a single tryptic peptide per protein resulted in 23 measured protein concentrations that were within reported clinical ranges (Table II) (48). Concentrations of an additional 16 proteins were within a factor of 2 of reported concentration ranges. Incomplete tryptic digestion of parent proteins can result in lower measured protein concentrations, whereas interference from co-eluting ion species can result in elevated measurements of protein concentration. Using our new MRM-based assay, accurate concentrations were rapidly obtained for 23 proteins using a single peptide plus an isotopically labeled internal standard peptide. However, to increase the accuracy of the concentration measurements of the remaining 22 proteins and to improve resistance of these assays to interferences, development of assays using additional proteotypic peptides for these proteins will likely be necessary.

When adapting these 45 MRM assays to third party laboratories, it would be prudent to verify that each peptide MRM assay is free of interferences by initially comparing the relative signal intensities of multiple MRM ion pairs for each peptide in addition to redetermination of both the analytical and technical reproducibility of each assay. Implementation of MRM-based protein assays for clinical applications and additional assay characteristics, such as interlaboratory accuracy and the effect of changing chromatographic conditions and MS instrumentation on reproducibility, will require validation and verification. Triple quadrupole LC-MS/MS instrumentation is already well established in hospital laboratories and is presently used to quantify several metabolites in clinical blood, serum, and urine samples (20). In addition to continued assay verification and validation, automated sample preparation work flows are needed to adapt these MRM-based protein assays to the clinical environment. An additional benefit of developing these MRM assays using trypsin-digested whole plasma samples is that sample preparation is relatively simple and highly amenable to automation using liquid-handling robotics.

Previous work by Kronkvist et al. (52) has shown that use of liquid-handling robotics to handle plasma samples and dispense internal standards can reduce relative standard deviations to as low as 0.2%, which could further reduce interassay variation for this panel of 45 MRM protein assays. Recent advances in ultrahigh pressure liquid chromatography could reduce analysis times by as much as 3-fold when combined with MRM analysis (53). MRM-based assays using isotopically labeled internal standard peptides permit absolute specificity with three analyte-specific criteria (retention time, precursor, and fragment ion m/z) that is not attainable with assays that depend on indirect colorimetric, fluorescent, or radioactivity readouts. In addition to multiplexed protein quantitation, MRM-based assays are also capable of quantifying non-synonymous small nucleotide polymorphisms, post-translational modifications, and alternate protein splice variants.