Parallel Reaction Monitoring for High Resolution and High Mass Accuracy Quantitative, Targeted Proteomics

Theoretical Comparison of SRM and PRM

The process of targeting a peptide with SRM involves two stages of quadrupole mass filtering with tight tolerances for both members of a precursor-product ion transition. Because all product ion transitions targeted for a given precursor peptide (usually 3 to 5) are required to simultaneously elute, the likelihood of mistaking a nontarget peptide or background ion for the targeted peptide is a rare occurrence; hence, SRM is considered to be a highly specific assay. The proposed PRM method involves only one stage of quadrupole mass filtering (of the precursor of interest) before mass analysis in an Orbitrap. The Orbitrap, however, by nature of its high resolution and high mass accuracy should more effectively separate ions of interest from background ions than the electron multiplier-based detection used in a QqQ. Thus, to motivate our experiments, we asked how PRM compares theoretically to SRM in terms of specificity. In other words, can the selectivity of Orbitrap HR/AM mass analysis make up for use of only one stage of mass filtering?

To answer this question, we digested the human proteome with trypsin in silico to yield over 20 million unique peptide sequences (see Experimental Procedures for details). For our calculations, we considered these peptides to be potential confounders in the measurement of the precursor and product ions of the 25 isotopically heavy-labeled peptides targeted in this study, as well as their 25 corresponding unlabeled peptides (50 total). For each unique confounder peptide, considered in charge states from 1 to 5, we further generated all possible b, y, a, b/y/a - water, and b/y/a - ammonia product ions, internal fragments, and immonium ions in charges ranging from 1 to one less the precursor charge state. We then asked how often, depending on the amount of evidence required by the assay and the mass analyzer employed, the numerous ions generated by the confounder population resulted in indistinguishable interference in the measurement of one of our target peptides. The results of these queries are summarized in Fig. 1C.

First, we consider a query requiring the least evidence of the targeted peptide, a SIM experiment. In SIM, an intact target peptide ion is isolated in Q1 and mass analyzed without further transformation. We begin our theoretical calculations with SIM to explore the effect of mass accuracy/resolution alone on selectivity. Assuming Q1 isolation widths of ±0.5 and 1 Th, and mass errors less than ±250 and 5 ppm, for the QqQ and QqOrbi, respectively, highly accurate mass analysis improves the chances of correctly identifying the target peptide (96 ± 3%) by exclusion of a large number of spurious peptides (7151 ± 1447, on average). Still, such an experiment, however, would only produce an unambiguous identification ∼1% of the time (Fig. 1C). Note, these calculations assume no upfront chromatographic separations and that all genome-predicted peptides are translated and detectable (i.e. the worst-case scenario).

Though it is obvious that high accuracy mass measurements increase specificity, it is less clear that this benefit persists in reaction monitoring experiments. To test this, we considered the number of intact peptides that could potentially be co-isolated with a given target peptide (assuming ±1 and ±0.5 Th Q1 isolation windows centered on the target peptide m/z for the QqOrbi and QqQ, respectively) and then generate at least one product ion (of any type) with a m/z within ±5 ppm (QqOrbi) or ±250 ppm (QqQ) of a y-ion transition from the target peptide. As shown in Fig. 1C, the specificity of a HR/AM analyzer substantially reduces the number of spurious peptide ions that can interfere in the correct identification of a target peptide by its y-ion transitions (shown as the rate of correct target identification, or true positive rate). The inclusion of 1, 2, or 3 y-ion transitions with high mass accuracy measurement results in approximately a (10 ± 14)-fold, (53 ± 35)-fold, and (112 ± 81)-fold greater likelihood of correctly identifying a target peptide compared with the inclusion of 1, 2, or 3 y-ion transitions at low resolution, respectively. Even if observation of all three y-ion transitions were required, as is typically the case in QqQ SRM assays, the likelihood of correctly identifying a target at low resolution and unit mass accuracy is still less than 1% for the 50 peptides considered here (0.6 ± 0.3%). Again, these calculations do not consider the significant benefit that is achieved by chromatographic separation and model a worst-case scenario. Even still, the detection of three y-ion transitions at high mass accuracy (i.e. <5 ppm mass error) provides nearly unambiguous target confirmation (1.6 ± 1.1 potential confounders) from the background of the entire human peptidome.

These theoretical calculations confirm our guiding supposition that PRM can provide greater routine specificity as compared with conventional SRM. Whether this result implies greater overall performance in targeted, quantitative proteomics studies compared with SRM, however, depends on several additional, but important, factors: whether both analyzers, one image current-based and the other electron multiplier-based, are capable of detecting the targeted species, and whether both analyzers can reproducibly and accurately measure the abundance of the targeted species. Given that PRM relies on an analyzer that is fundamentally less sensitive and slower than that used in SRM, investigation of these factors will reveal whether selectivity/specificity can overcome limitations in speed and sensitivity. In the following sections, we empirically investigate this issue through a systematic analysis of reproducibility, sensitivity, and linearity of both methods.

Detection Criteria for PRM and SRM

The detection criteria we developed for PRM incorporate the benefits of high selectivity and specificity HR/AM mass analysis, as discussed above. By making use of full mass range MS/MS spectra and the high specificity of product ions when measured with high mass accuracy, we developed an automated detection algorithm that assigned a spectral score to each PRM spectrum based on the presence of b- or y-ions within ±5 ppm of expected target-specific product ions using Eq. 1, and then generated an “extracted score chromatogram” (XSC) for that peptide. In the design of our spectral score, we chose to weight b-ions less than equivalently numbered y-ions because all of the peptides targeted in this study were isotopically heavy-labeled at the c-terminus and analyzed in a background of endogenous yeast peptides. Thus, y-ions, containing the heavy label, were more specific to our target peptides than b-ions and were weighted as such. If the target peptides of interest were not heavy-labeled, one might consider equally weighting b- and y-ions for scoring purposes. A positive detection event required that, in at least 2 of 3 replicates, the XSC met or exceeded a peptide specific threshold (equal to the length of the target peptide) within ±3σ min of the expected retention time (supplemental Fig. S1). Fig. 2 shows the application of this score to exemplary data from the peptide AETLVQAr (#17) under neat and matrix-containing conditions over all six peptide concentrations. The XSCs (gray) are overlaid with XICs (blue) generated using the summed intensity of all b- and y-ions present in each spectrum. Note that XSCs are used solely for establishing a detection event and not for quantification as the score is independent of ion intensity. Following a positive detection event, the target-specific ions that generated the detection event are extracted as an XIC for quantification. This peptide was detected at concentrations spanning four orders-of-magnitude (from 20 pm to 200 nm) without matrix, and over three orders-of-magnitude (from 200 pm to 200 nm) in the presence of yeast matrix.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Extracted PRM score chromatograms (XSC, ±5 ppm, gray) overlaid with XICs (±5 ppm, blue; 7-point boxcar smoothed) and single-scan PRM spectra for peptide AETLVQAr (#17) isolated at ±1 Th under neat (left) and matrix-containing (right) conditions from 2 pm to 200 nm. Red dotted line in XSC plots represents the score acceptance threshold (8) for this peptide. Product ions detected in each spectrum are highlighted and the spectral score is labeled. Yellow arrows in each XIC/XSC plot indicate the retention time at which the associated single-scan spectrum was acquired. Hashed area in XIC/XSC plots designates the retention time period during which peak elution was expected based on the ±3σ range around the average retention time observed in 200 nm analyses.

In Fig. 2, at high target peptide concentrations (e.g. 20–200 nm), the XSC often increases above threshold before the XIC shows any noticeable intensity change and stays above threshold after the XIC intensity has fallen. This characteristic is ascribable to the intensity-independence of the score, and suggests that high quality Orbitrap spectra, in terms of the presence of product ions with high mass accuracy, can be acquired independently of ion intensity above some intensity threshold. At low concentration, especially in the presence of matrix (see 200 pm in matrix), the XSC easily distinguishes nearby, spurious XIC signals from target-derived signals, demonstrating score specificity. The detection threshold was chosen to ensure that an accepted score (signifying target detection) would, at least, contain one y-ion with m/z greater than the precursor m/z. Although this threshold does not definitively ensure that a score above the threshold is entirely specific to only the target peptide, use of the threshold successfully separated spurious signals from target-specific signals, agreeing with or being more conservative than manual interpretation of the data in all cases. supplemental Fig. S2 plots the distribution of the spectral score for all possible combinations of b- and y-ions arising from an eight-amino-acid peptide, such as AETLVQAr, by the fraction of the score described by each ion. Example ion combinations are given at several scores below and above the threshold. supplemental Table S1A–D catalog the maximum XSC scores measured for each replicate, concentration, and PRM experiment type (supplemental Results and Discussion 2.2).

QqQ SRM experiments were performed to provide a standard for comparison of our QqOrbi PRM data. QqQ SRM data were analyzed manually based on detection criteria commonly used in the literature for SRM analyses (42). A positive peptide detection event occurred when eight of 10 criteria were met in a least two of three replicates. Originally, we required 10 of 10 criteria to be met in at least two of three replicates, but relaxed this criterion to improve the completeness of the data set and thus our ability to compare it to the PRM data set. The ten criteria were: 1) the appearance of three co-eluting transitions within ±2.5 min of the expected retention time (supplemental Fig. S1), 2–4) average transition m/z within ±500 ppm of the expected m/z, 5–7) average transition abundance within ±15% of the expected abundance, and 8–10) individual transition XICs with signal-to-noise meeting or exceeding three. Peptides, transitions, and other experimental parameters for SRM are listed in Table I.

Fig. 3A compares PRM with ±1 Th isolation to SRM for the detection of peptide GVSAFSTWEk (#1) at 200 pm and 2 nm (200 amol and 2 fmol, respectively, on column) in the presence of yeast matrix. In PRM, the signals present in the XIC during the entire duration for which this peptide was targeted are exclusively due to the targeted peptide, even at the lowest concentration detected (200 pm; XICs at lower concentrations had zero intensity). Numerous other y-ions (shown in gray) are also present at the 2 nm level, adding confidence to the detection and generating maximum spectral scores well above the threshold of 10 for this peptide. By contrast, transition XICs in the SRM data show numerous background signals throughout the duration over which the peptide was targeted (far right). At the 200 pm level, while numerous signals are present in the XICs for each SRM transition, the high level of chemical noise in all transition channels prevents confident determination of the presence of the target peptide. This example demonstrates the enhancement of sensitivity that can be achieved via PRM because of the high selectivity of HR/AM mass analysis, as well as the ease with which determinations of presence or absence of a peptide target can be made with HR/AM data.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

A, Comparison of QqOrbi PRM detection at ±1 Th with QqQ SRM for peptide GVSAFSTWEk (#1) at 200 pm and 2 nm in the presence of matrix. Transition XICs are shown for the entire duration over which the peptide was targeted, with a zoom-in on the relevant time period in the QqQ SRM case (from the region shaded in gray in the chromatograms at the far right). Additional y-type product ions present in the PRM data are shown in gray. XICs were extracted at ±5 and ±250 ppm for PRM and SRM, respectively. The maximum spectral score attained at each concentration in PRM is also labeled. B, Lowest concentration detected (as number of attomoles of peptide on column) for each peptide in neat and matrix-containing experiments for the 14 peptides targeted in both SRM and PRM.

PRM Measurement Precision

In quantitative studies, high measurement precision is critical to reliably distinguish differences between two analyses or samples. We assessed the degree of measurement precision for the PRM method, defined here as run-to-run area-under-the-curve (AUC) repeatability across technical replicates, for all concentrations and isolation widths. Overall, PRM exhibited high measurement precision with median percent relative standard deviation (%RSD) less than 10% in most cases (supplemental Table S2A). The main factors negatively affecting measurement precision (increasing %RSD) were low target peptide concentrations and individual peptide characteristics. Neither isolation width nor the presence of matrix resulted in statistically significant differences in precision (α = 0.05; NB, all discussion of mean %RSDs and statistical significance refer to log-transformed data; see supplemental Methods 1.2). The precision of measurements at the lowest concentration typically detected in neat PRM, 200 pm, was significantly lower than at all other (higher) concentrations. In the matrix-containing PRM data, however, no significant differences in the measurement precision of adjacent concentrations were observed (supplemental Table S2B). When grouped by peptide alone, only very hydrophilic, poorly retained peptides, #3 and #7, had significantly decreased measurement precision when compared with the other 23 peptides (supplemental Table S2C).

We find that PRM measurement precision is consistent with studies reporting run-to-run precision data in QqQ SRM experiments. For example, in the 2009 multi-laboratory study of SRM measurement repeatability and reproducibility for peptides spiked into plasma, Addona et al. (20) found both intra- and inter-laboratory precision to be similarly less than 15% RSD across the concentration range studied (1–500 nm). Likewise, precision improved for measurements at higher target concentrations. Kiyonami and colleagues (43) also reported similar precision metrics in their large-scale intelligent-SRM (iSRM) experiments, observing that 80% of the 757 peptides targeted in yeast exhibited less than 10% RSD. Compared with other studies using HR/AM mass spectrometers (quadrupole linear ion trap (QLT)-Orbitrap, QqTOF, and QLT-FT-ICR) for targeted, quantitative measurements (61⇓⇓⇓–65), QqOrbi PRM achieved, on average, two- to threefold better run-to-run measurement precision.

PRM Dynamic Range

On average, PRM experiments yielded quantitative information between 2–4 concentration orders-of-magnitude (93%), with the majority of experiments resulting in quantification across 3 orders-of-magnitude (54%, 0.2 to 200 nm; supplemental Table S3A, supplemental Fig. S3). The wider PRM isolation width, ±1 versus ±0.2 Th, resulted in improved dynamic range—neat, 10^3.3 versus 10^2.6, and matrix, 10^2.7 versus 10^2.2 (supplemental Table S3B). The presence of matrix in PRM experiments resulted in a modest depression of the quantifiable dynamic range, ∼0.5 orders-of-magnitude (matrix versus neat, 10^2.4 versus 10^3.0). Although tighter isolation widths slightly mitigated the effects of matrix-induced sensitivity depression, at the expense of overall sensitivity, this difference was not significant (supplemental Table S3A). These results indicate that, although an increase in selectivity due to gas-phase enrichment would be expected by tighter isolation widths, the concomitant decrease in ion transmission at very tight isolation widths (±0.2 Th), in conjunction with Orbitrap detection, results in decreased sensitivity as too few ions are present for the target signal to exceed the Orbitrap's thermal noise band. However, with a HR/AM analyzer, the wider isolation width, which provides greater ion transmission, but also higher levels of chemical noise, can be used without a decrease in performance because of the high selectivity of the mass analysis.

Linearity of PRM Measurement Response

The linearity of measurement response—or, measurement accuracy—was calculated as the percent RSD of response factors (AUC normalized by concentration) over the concentration range for which a peptide was detected. In general, linearity was not significantly influenced by isolation width or the presence of matrix: mean neat %RSDs of 37.1 and 36.4, and mean matrix %RSDs of 35.7 and 34.2, were observed in PRM experiments with ±0.2 and ±1 Th isolation widths, respectively (supplemental Tables S4A, S4B). If the lowest detected concentration, associated with lower overall measurement precision (vide supra), was excluded from each experiment, however, linearity improved and some significant differences emerged based on the presence of matrix (supplemental Tables S4C–S4F). Surprisingly, matrix-containing experiments were found to possess greater linearity than their neat counterparts in all experiments, with %RSDs of 23.2% for neat PRM and 12.0% for matrix PRM (supplemental Table S4D).

Because of the truncation of the measured dynamic range observed in the presence of matrix, we posited that greater linearity in matrix-containing experiments was an artifact of simply considering a smaller concentration range. Since lower precision on lower concentration measurements make it more challenging to arrive at an accurate average AUC from only three replicates, experiments quantifying over a wide dynamic range (e.g. neat experiments) could have decreased linearity due to low measurement accuracy at the lowest detected concentrations. To address this, we calculated an adjusted %RSD for each experiment and peptide that accounted for the dynamic range represented by a %RSD value (see supplemental Methods 1.2). The adjusted %RSD separates out contributions of dynamic range and concentration from the linearity estimate, thereby providing a more fair comparison between experiments where the dynamic range quantified is drastically different. These adjusted linearity values no longer demonstrate the surprising trend observed above when the lowest concentration was excluded (13.5% Adj. RSD neat versus 16.6% Adj. RSD matrix). Using the adjusted linearity metric, PRM was not significantly affected by the presence of matrix (or isolation width, as before) (supplemental Tables S4G, S4H).

QqQ SRM Measurement Precision, Dynamic Range, and Linearity

Like the PRM data described above, the QqQ SRM data, which queried a subset of 14 of the 25 peptides analyzed on the QqOrbi, also exhibited a high degree of run-to-run measurement precision, typically less than 15% RSD. Unlike in PRM, however, measurement precision, when all peptides and concentrations were considered together, was significantly greater in the presence of matrix (supplemental Table S5A). Lower peptide concentrations were again correlated with decreased measurement precision, though only significantly so under neat conditions (supplemental Table S5B). Peptides were quantified on average over concentration ranges of 10^3.3 (neat) and 10^2.2 (matrix), and the depression of dynamic range under matrix-containing conditions was statistically significant. Linearity under matrix conditions bested that under neat conditions with mean %RSDs of 16.7 and 42.1%, respectively. As in the QqOrbi, the loss of linearity was often the result of peptide detection at the lowest concentrations. Exclusion of these concentrations in %RSD calculations revealed no differences in linearity between matrix and neat conditions. Likewise, if only data covering the same dynamic range in both matrix-containing and neat conditions were considered, effectively truncating the neat data to the concentration range quantified in the matrix-containing data, linearity differences due to matrix interferences were not significant. Adjusted %RSDs, however, still exhibited significantly decreased linearity in the neat case. Mean adjusted %RSDs were 13.0 and 8.0% for neat and matrix data, respectively (supplemental Table S5C).

We postulate that the adjusted %RSD correction did not completely account for differences in linearity, as it did in the PRM data, because the SRM detection criteria were not sensitive enough to detect and exclude data at the lowest detected concentrations in neat experiments that were skewed by small amounts of chemical interference. In the matrix-containing SRM experiments, on the other hand, more abundant matrix interferences were easily detected and excluded by the detection criteria (supplemental Fig. S4). Detection criteria incorporating HR/AM data, however, more sensitively detected and excluded aberrant responses and thus, deviations in linearity were predictable and correctable (as adjusted %RSDs) based on the detected dynamic range. This same rationale can be applied to the discussion of measurement precision above: when all neat data were considered, precision was diminished relative to the matrix-containing data set because of the inclusion of background-skewed data not excluded by the detection criteria.

Empirical Comparison of QqQ SRM and QqOrbi PRM

To compare between QqQ SRM and QqOrbi PRM, we only considered the fourteen peptides analyzed in both data sets. Under neat conditions, run-to-run measurement precision (paired by peptide and concentration) was no different between QqQ SRM and QqOrbi PRM, regardless of the PRM isolation width employed. In the presence of matrix, however, SRM demonstrated significantly better measurement precision compared with both PRM data sets (5.6 versus 11.1% RSD, respectively; supplemental Tables S6A, S6B). Because neat SRM measurement precision, as discussed above, is believed to be artificially decreased because of the inclusion of chemical interference, it is likely that SRM exhibits superior measurement precision under neat conditions as well (supplemental Results and Discussion 2.3, Supplemental Table S6B).

SRM exhibited greater measurement precision likely because of the higher sampling rate of the QqQ (almost twice as many scans were acquired compared with the QqOrbi per 90 min chromatographic run). The ability of the QqQ to sample more points over a given chromatographic peak provided a more accurate determination of the peak AUC and, in turn, greater run-to-run repeatability. The difference in sampling rate between the two methods is due to the characteristics of the instruments used and, to some extent, necessary aspects of the experimental design. Because the QqQ is a “beam-type” instrument (as opposed to a scanning instrument like the QqOrbi), it operates at a duty cycle nearing 100%; this means that there is very little “down time” where the instrument is not acquiring data. The Orbitrap, conversely, as a scanning instrument, has inefficiencies inherent to its design. The Orbitrap transients employed here (required for MS/MS scans of resolution 17,500) were approximately twice the length of the 35 ms dwell times employed in the SRM method. Additionally, whereas the QqQ cycle time was fixed, the cycle time of the QqOrbi varied based on ion accumulation times, which were dynamically set based on ion flux. At low sample concentrations, QqOrbi injection times could reach as high as 120 ms to result in (at most) ∼40 ms of inter-scan “down time” where mass analysis was not occurring. Shorter transients and lower maximum allowable injection times could be used to match the cycle times of the two methods at the expense of overall performance: shorter injection times would decrease spectral quality/usability when ion flux is low, and lower mass analysis resolution would undermine the benefits that high selectivity brings to the PRM method.

Greater QqQ SRM measurement precision did not translate into greater sensitivity or linearity compared with PRM. Under neat conditions, SRM quantified a concentration range spanning 10^3.3 per peptide on average. This range is not statistically different from that achieved with PRM, 10^3.5, at the wider isolation width. In matrix-containing experiments, however, PRM quantified over a broader range at both isolation widths (10^2.4 and 10^2.9 at ±0.2 and ±1 Th, respectively) than did SRM (10^2.2), and significantly more at the ±1 Th isolation width (Table II). This result is illustrated on an individual peptide basis in Fig. 3B. For both neat and matrix containing experiments, the lowest concentration detected for each peptide studied in the SRM and PRM (±1 Th) experiments is plotted. SRM demonstrates a lower empirical detection limit than PRM in only one case (peptide NSWGTDWGEk, neat). As an aside, under matrix-containing conditions, SRM surprisingly performed no better QqOrbi SIM: SIM quantified on average over a range of 10^2.0 at an isolation width ±1 Th, statistically no different from SRM (see supplemental Results and Discussion 2.5–2.6, supplemental Table S6C).

Under neat conditions, PRM demonstrated significantly higher linearity over the quantified dynamic range compared with QqQ SRM (Table II). Given that PRM under matrix-containing conditions yielded quantitative data over a wider dynamic range than SRM, we calculated adjusted %RSDs, as before, to normalize the linearity metric for the dynamic range detected and permit a fair comparison of the data. With this consideration, PRM linearity was statistically no different from the linearity exhibited by SRM (Table II). Fig. 4 plots the SRM linearity data for each of the targeted peptides as a function of PRM linearity and is stratified by the presence of matrix and isolation width. This data presentation demonstrates the effect of using the adjusted linearity metric on mean linearity estimates (shown as vertical and horizontal lines), as well as the relative distributions of the linearity metrics across the data sets.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Comparison of linearity as %RSD and adjusted %RSD (supplemental Equation S1) for the 14 shared peptides targeted in QqOrbi PRM experiments (y axis) and QqQ SRM (x axis) under neat and matrix-containing conditions. Solid horizontal lines and dotted vertical lines represent the mean %RSD value, also labeled in the plot, of the associated data set. Data falling in the gray region demonstrate greater linearity in PRM experiments. Data in the white region demonstrate greater linearity in SRM experiments.

With these data, we find that a high resolution and accurate mass MS can acquire data of similar quality to that collected on a QqQ in a targeted, quantitative proteomics context with 1) minimal upfront development time, 2) straightforward data analysis, and 3) similar performance metrics. Our PRM methods were designed and optimized on a method-level, rather than on the peptide-level as with SRM methods (i.e. optimization of individual collision energies and transition sets for each target peptide). Thus, the PRM method has fewer parameters that require optimization (namely, only maximum injection time/AGC target, “global” collision energy, and, optionally, target scheduling) and can be performed with solely knowledge of the mass-to-charge ratios of the target peptides (and, optionally, the approximate retention time). These characteristics make PRM amenable to the “walk-up” instrument user who wishes to perform targeted, quantitative assays on a list of targets in a time-efficient manner—collection of MS/MS data and optimization of individual transition sets not required. We believe that the data presented here reflect the analytical performance a typical user might expect for PRM, however the analysis parameters used here may not represent the optimum conditions for all applications of PRM and would require application-specific investigation.

In addition to the minimal upfront method development needed to successfully perform a quantitative PRM assay, a notable aspect of using HR/AM data for targeted proteomics is the ease with which data can be interpreted and data analysis can be automated. Within the course of this study, we developed and automated a novel strategy for scoring and detecting low level analytes in complex, chimeric Orbitrap spectra that allowed sensitive detection of our target peptides without the expense of selectivity. The specificity of accurate mass measurements and the resulting paucity of marginal or ambiguous situations in the data (which are difficult to account for in automated methods) enabled this rather simple algorithm. On the other hand, even using well-established and optimized detection criteria from the literature for SRM, the SRM detection criteria were not sensitive enough to successfully detect and exclude data skewed by small amounts of chemical interference, leading to poor reproducibility and linearity. Furthermore, with the high specificity of a product ion measured at high mass accuracy for a target peptide (as demonstrated by our theoretical calculations) and the benefit of full mass-range mass analysis, the PRM paradigm lends itself well to future development of how targeted, quantitative proteomic data are analyzed. Through use of theoretical calculations, the high specificity of high mass accuracy data lays the groundwork for a generalized and statistically sound detection algorithm for reaction monitoring experiments that incorporates probability of correctness measures, as well as obviates the need for manual curation, ad hoc detection criteria (42), “decoy transitions” at the measurement-level (66), and the potentially high level of human intervention (and error) that comes with such detection strategies.

Last, our analysis of the analytical performance characteristics of PRM suggests that targeted HR/AM methods can rival the performance of QqQ SRM in terms of dynamic range, linearity, and, to a lesser extent, precision. Although SRM measurement precision was ∼twofold better (under matrix-containing experiments) likely because of differences in scan rate between the two mass analyzers, PRM yielded quantitative data over a wider dynamic range than SRM under matrix-containing conditions. High mass accuracy and high resolution underlie this result: when high levels of matrix background are present, a single stage of isolation in combination with highly resolved data is statistically significantly more sensitive than two stages of isolation in combination with low resolution data, the “gold-standard” method for quantitative proteomic analyses. Additionally, achievable linearity over the quantifiable dynamic range was found to be statistically the same between SRM and PRM. Thus, in answer to the query posed in our theoretical comparison of SRM and PRM, our experimental data suggest that high selectivity/specificity can overcome limitations in speed and sensitivity to reliably provide lower detection limits and higher accuracy measurements. We conclude that the proposed PRM analysis paradigm holds promise as a viable, and accessible, alternative and/or complement to SRM for the quantitative proteomics toolbox.

In support of our conclusions, a report by Weisbrod et al. (61), published during the preparation of this manuscript, described a method for data-independent discovery proteomics on an Oribtrap MS that reinforces the favorable performance of PRM relative to SRM for targeted proteomics. In their “Fourier Transform-All Reaction Monitoring” (FT-ARM) method performed on a QLT-Orbitrap MS, wide swaths of mass-to-charge space (i.e. 100 Th) were accumulated and isolated in the QLT using broadband waveforms, subjected in bulk to dissociation, and simultaneously mass analyzed to exploit the HR/AM properties of the Orbitrap. The authors noted the ability to quantify peptides with FT-ARM, using the product ions generated from in-bulk dissociation, with modest sensitivity, reproducibility, and precision despite quite significant interferences from other ions present in the 100 Th isolation swath. This further corroborates that HR/AM mass analysis enables real, quantitative information to be extracted from highly interference-riddled measurements with little upfront assay development or optimization. Because PRM involves significantly less co-isolated interference and uses a quadrupole mass filter-equipped platform better suited to accumulation and isolation of large quantities of ions (to improve sensitivity), these findings further support our conclusions that PRM yields high quality quantitative measurements, comparable to QqQ SRM, while simplifying method development.

Looking forward, the PRM paradigm also enables new modes of analysis that are currently unavailable on QqQ platforms. Because of the modular nature of hybrid HR/AM MS, these platforms provide an unprecedented amount of experimental flexibility. Possessing the capabilities of both high performance quantitative and high-throughput discovery proteomics instruments, one can envision mixed mode analysis types in which targeted and discovery experiments are performed simultaneously. For example, when not engaged in profiling an eluting target species, the instrument could be directed to perform conventional data-dependent or data-independent MS/MS, thereby maximizing instrument time, sample usage, and data density. Additionally, because of the high specificity of ions measured at high mass accuracy and consequently the reduced search space for potentially matching peptides, intelligent data acquisition strategies are possible that enable on-the-fly database searching, spectral matching, or spectral scoring (like that used here post-acquisition). Such strategies would enable the mass spectrometer to make decisions during acquisition, such as assessing whether a target peptide had been adequately identified, changing experimental parameters (CAD energies, isolation width, injection times, etc.) to improve performance for a particular target peptide, or determining its progress in a chromatographic run to make dynamic modifications to target peptide scheduling (67, 68).