Using Data Independent Acquisition (DIA) to Model High-responding Peptides for Targeted Proteomics Experiments

Targeted mass spectrometry is an essential tool for detecting quantitative changes in low abundant proteins throughout the proteome. Although selected reaction monitoring (SRM) is the preferred method for quantifying peptides in complex samples, the process of designing SRM assays is laborious. Peptides have widely varying signal responses dictated by sequence-specific physiochemical properties; one major challenge is in selecting representative peptides to target as a proxy for protein abundance. Here we present PREGO, a software tool that predicts high-responding peptides for SRM experiments. PREGO predicts peptide responses with an artificial neural network trained using 11 minimally redundant, maximally relevant properties. Crucial to its success, PREGO is trained using fragment ion intensities of equimolar synthetic peptides extracted from data independent acquisition experiments. Because of similarities in instrumentation and the nature of data collection, relative peptide responses from data independent acquisition experiments are a suitable substitute for SRM experiments because they both make quantitative measurements from integrated fragment ion chromatograms. Using an SRM experiment containing 12,973 peptides from 724 synthetic proteins, PREGO exhibits a 40–85% improvement over previously published approaches at selecting high-responding peptides. These results also represent a dramatic improvement over the rules-based peptide selection approaches commonly used in the literature.Targeted proteomics using selected reaction monitoring (SRM)¹ and parallel reaction monitoring (PRM) is increasingly becoming the gold-standard method for peptide quantitation within complex biological matrices (1, 2). By focusing on monitoring only a handful of transitions (associated precursor and fragment ions) for targeted peptides, SRM experiments filter out background signals, which in turn increases the signal to noise ratio. SRM experiments are almost exclusively performed on triple-quadrupole instruments. These instruments can isolate single transitions as an ion beam and measure that beam with extremely sensitive ion-striking detectors. As a result, SRM experiments generally exhibit significantly more accurate quantitation when compared with similarly powered discovery based proteomics experiments, and frequently benefit from a much wider linear range of quantitation (3). SRM experiments often require less fractionation and can be run in shorter time on less expensive instrumentation. These factors allow researchers to greatly scale up the number of samples they can run, which in turn increases the power of their experiment.However, the process of developing an effective SRM assay is often cumbersome, as subtle differences in peptide sequence can have a profound impact on the physiochemical properties and subsquent SRM responses of a peptide. To successfully develop an SRM assay for a protein of interest, unique peptide sequences must be chosen that also produce a high SRM signal (e.g. high-responding peptides). Once identified, these high-responding peptides are often synthesized or purchased, and independently analyzed to determine the most sensitive transition pairs. Finally, the selected peptide and transition pairs must be tested in complex mixtures to screen for transitions with chemical noise interference and to validate the sensitivity of the assay within a particular sample matrix. Peptides and transitions that survive this lengthy screening process can then undergo absolute quantitation by calibrating the signal intensity against standards of known quantity.Although experimental methods have been developed to empirically determine a set of best responding peptides (4), these strategies can be time consuming and require analytical standards, which are currently unavailable for all proteins. More often than not, representative peptides are essentially chosen at random, using only a small number of criteria, such as having a reasonable length for detection in the mass spectrometer, a lack of methionine, and a preference for peptides containing proline (5). It is not uncommon for SRM assays to fail at the final validation steps simply because the peptides chosen in the first assay creation step happened to be unexpectedly poor responding peptides.In an effort to speed up the process of generating robust assays, several groups (6–9) have designed approaches to predict sets of proteotypic peptides using machine-learning algorithms. Proteotypic peptides are peptides commonly identified in shotgun proteomics experiments for a variety of reasons including high signal, low interference, and search engine compatible fragmentation. Enhanced Signature Peptide (ESP) Predictor (7) was the first successful modification of this prediction approach to use proteotypic peptides as a proxy for high-responding peptides for SRM-based quantitation. In brief, Fusaro et al. built a training data set from data-dependent acquired (DDA) yeast peptides and a proxy for their response was quantitated using extracted precursor ion chromatograms (XICs). The authors calculated 550 physiochemical properties for each peptide based on sequence alone and built a random forest classifier to differentiate between the high and low response groups. Other peptide prediction tools follow the same general methodology for developing training data sets. CONSeQuence (8) applies several machine learning strategies and a pared down list of 50 distinct peptide properties. Alternately, Peptide Prediction with Abundance (9) (PPA) uses a back-propagation neural network (10) trained with 15 distinct peptide properties selected from ESP Predictor''s 550. The authors of CONSeQuence and PPA found that their approaches outperformed the ESP Predictor on a variety of data sets.As with most machine learning-based tools, the generality of the training set to real-world data is key to the effectiveness of the resulting prediction tool. Although MS1 intensities extracted from DDA data can be useful for predicting high-responding peptides (11, 12), several factors make them less than ideal for generalizing to SRM and PRM experiments. In particular, DDA peptides must be identified before being quantified and key biochemical features beneficial for targeted analysis of transitions can reduce overall identification rates by producing fragment spectra that are difficult to interpret with typical search engines. By building training data sets on precursor intensities alone these tools ignore the fact that targeted assays actually use fragment ions for quantification. We propose that constructing training sets from DIA fragment intensities (13) will produce machine-learning tools that are more effective at modeling peptides that produce detectible transitions, rather than just proteotypic peptides.The use of digested proteins in training sets presents additional concerns. The observed variance in peptide intensities is confounded by variation in protein abundance. Converting peptide intensities to ranks can remove the dependence on varying protein levels at the cost of corrupting the training set with proteins that biochemically contain no high-responding peptides. PPA attempts to ease this concern by training with Intensity Based Absolute Quantitation values (14) for DDA peptides estimated from XICs. We hypothesize that constructing a training set from equimolar synthetic peptides removes most adverse effects of digestion from the training set, making it possible to construct a more generalizable tool.     

A Century of Change in Macrophyte Abundance and Composition in Response to Agricultural Eutrophication

Clear Lake, Iowa, USA is a shallow, agriculturally eutrophic lake that has changed drastically over the past century. Eight macrophyte surveys since 1896 were pooled and examined to characterize long-term impacts of eutrophication on macrophyte community composition and relative abundance. Surveys in 1981 and 2000 revealed few submergent and floating-leaved species and a dominance in emergent species (Scirpus, Typha). Over the past century, however, species richness has declined from a high of 30 species in 1951 to 12 found today, while the community composition has shifted from submergent-(99%) to emergent-dominated floras (84%). Potamogeton praelongus was the first emergent species to disappear but was followed by several other clear water Potamogeton species. Several floating leaved and emergent genera increased in relative abundance with eutrophication, notably Nuphar, Nymphaea, Phragmites, Polygonum, Sagittaria, Scirpus, and Typha. P. pectinatus was present over the entire century due to its tolerance of eutrophic conditions. Macrophyte growth was generally light-limited, with 93% of the variance in relative abundance of submergent species explained by changes in water transparency. Clear Lake exhibits signs of alternative stable states, oscillating between clear and turbid water, coupled with high and low submerged species relative abundance. The maximum macrophyte richness occurred as the lake oscillated between submergent- and emergent-dominated states. Changes in the water level have also impacted macrophyte growth since the area of the lake occupied by emergent macrophytes was negatively correlated with water level. Strongest correlations indicated that macrophytes respond to water level variations with a 2-year time-lag.     

Peptide-Centric Proteome Analysis: An Alternative Strategy for the Analysis of Tandem Mass Spectrometry Data

In mass spectrometry-based bottom-up proteomics, data-independent acquisition is an emerging technique because of its comprehensive and unbiased sampling of precursor ions. However, current data-independent acquisition methods use wide precursor isolation windows, resulting in cofragmentation and complex mixture spectra. Thus, conventional database searching tools that identify peptides by interpreting individual tandem MS spectra are inherently limited in analyzing data-independent acquisition data. Here we discuss an alternative approach, peptide-centric analysis, which tests directly for the presence and absence of query peptides. We discuss how peptide-centric analysis resolves some limitations of traditional spectrum-centric analysis, and we outline the unique characteristics of peptide-centric analysis in general.Tandem mass spectrometry has become the technology of choice for proteome characterization. In a typical bottom-up proteomic experiment, a mixture of proteins is proteolytically digested into peptides, separated by liquid chromatography, and analyzed using tandem mass spectrometry. The ultimate goal is to identify and quantify proteins by detecting and quantifying individual peptides, thereby shedding light on the underlying cellular mechanisms or phenotype. Several modes of data acquisition have been developed for bottom-up proteomics. The most commonly applied mode uses data-dependent acquisition (DDA)¹, in which tandem MS (MS/MS) spectra are acquired from the dissociation of precursor ions selected from an MS survey spectrum. Constrained by the speed of instrumentation, DDA can sample only a subset of precursor ions for MS/MS characterization, generally targeting the top-N most abundant ions detected in the most recent survey spectrum. In addition, DDA is typically coupled with a method referred to as “dynamic exclusion” (1) that attempts to prevent reselection of the same m/z for some specified period of time. These acquisition strategies greatly increase proteome coverage and extend the dynamic range of detection for shotgun proteomics. The resulting MS/MS spectra are typically analyzed using sequence database searching software such as SEQUEST, Mascot, X!Tandem, MaxQuant, Comet, MS-GF+, or OMSSA (2–8). Because these algorithms identify peptides by first associating each individual spectrum with a matching peptide sequence and then aggregating the thus matched spectra into a list of identified peptides, we refer to them as “spectrum-centric analyses.” In spectrum-centric analysis, spectra are most commonly interpreted using database searching, but can also be interpreted using de novo sequencing (9–11), or by searching against a spectrum library (12–14). For the past two decades, spectrum-centric analysis has been an essential driving force for the development of large-scale shotgun proteomics using DDA.DDA is a powerful and well-established technique for LC-MS/MS data acquisition. By targeting precursor ions observed in MS survey scans with highly selective MS/MS scans, DDA generates a large set of high quality MS/MS spectra, which can be automatically interpreted by database searching to identify thousands of proteins in a complex sample. When DDA was introduced, instrumentation was not fast enough to sample every observed precursor in the survey scan; thus, high-intensity precursors were preferentially targeted because they tend to generate higher quality MS/MS spectra that lead to peptide identification. Although this stochastic sampling approach results in a large amount of peptide identification in a single sample run, it comes at the cost of reproducibility of MS/MS acquisition between sample analyses and an inherent bias against low abundance analytes that are less likely to be sampled (15). On modern instrumentation, the speed of MS/MS acquisition has dramatically improved to the point where the majority of MS precursors that are not already in the dynamic exclusion list can be sampled for MS/MS analysis. However, even if every precursor observed in each survey MS scan is sampled, DDA is still biased against low abundance analytes that fall below the limit of detection in the MS analysis and will never be sampled. This bias is a practical limitation in the analysis of a complex mixture with high dynamic range in which many analytes will be below the limit of detection in MS analysis, but remain detectable by the more selective and more sensitive MS/MS analysis (16).DDA remains a powerful method for identifying a large number of proteins in a sample. However, because of the incomplete sampling, when a peptide is not identified in a conventional shotgun experiment using DDA, it is incorrect to conclude that the peptide is missing from the sample, or even below the limit of detection of MS/MS, because the peptide ions may have never been sampled for MS/MS analysis. To overcome such limitations, targeted acquisition approaches such as selected reaction monitoring (SRM, also commonly referred to as multiple reaction monitoring) are often the methods of choice. In targeted acquisition approaches, a set of predetermined precursor ions are systematically subjected to MS/MS characterization throughout the LC time domain. The collision energy for each targeted ion can be optimized for fragmentation efficiency. The resulting data are typically analyzed using “targeted analysis” (17–19), in which the software looks for the co-eluting patterns from a group of predetermined pairs of precursor–product ions (called transitions). With systematic MS/MS sampling and the combined specificity of chromatographic retention time, precursor ion mass, and the distribution of product ions, targeted acquisition allows highly sensitive and reproducible detection of the targeted analytes within a complex mixture.Modern targeted acquisition approaches are the gold standard for sensitively and reproducibly measuring hundreds of peptides in a single LC-MS/MS run (20–22). However, data acquired in this manner is only informative for the set of peptides targeted for analysis. Because of this narrow focus, iterative testing of different hypotheses (i.e. a different set of target peptides) also requires iterative acquisition of additional data. Moreover, assay development often requires retention time scheduling and/or refinement steps to find the optimal peptides and transitions for testing a particular hypothesis.With the existence of two complementary but distinct approaches—DDA for broad sample characterization and targeted acquisition for interrogation of a specific hypothesis—the natural question is if the benefits of both techniques may be combined in a single technique. A potential solution is an alternative mode of bottom-up proteomics referred to as data-independent acquisition (DIA) that has been described and realized with various implementations (16, 23–32). In DIA, the instrument acquires MS/MS spectra systematically and independently from the content of MS survey spectra. These DIA approaches differ from DDA methods, targeted acquisition methods, and from each other in MS/MS isolation window width, total range of precursor m/z covered, duration of completing one cycle of isolation scheme (called the cycle time), single or multiple isolation windows per MS/MS analysis, and instrument platform. Because of the benefits of systematic sampling of the precursor m/z range by MS/MS, data from a single DIA experiment can be useful for both peptide detection and quantification in a complex mixture. Similar to DDA approaches, DIA data is broadly informative because the MS/MS characterization is not specific to a predefined set of peptide targets. Similar to targeted approaches, MS/MS information of peptides across the entire LC time domain can be extracted from DIA data to test a particular hypothesis. As the acquisition speed of modern instrumentation continues to increase, DIA has become more popular because of its comprehensive and unbiased sampling.Although DIA resolves the problem of biased or incomplete MS/MS sampling, current DIA methods come with compromises (33), where the most common compromise is precursor selectivity. Constrained by the speed and accuracy of instrumentation, DIA methods typically use five- to 10-fold wider isolation windows compared with DDA to achieve the same breadth and depth of a single LC-MS/MS run. Because of this reduction in precursor selectivity, MS/MS spectra from DIA are noisier than DDA spectra. In particular, DIA by design generates mixture spectra, each containing product ions from multiple analytes with various abundance and different charge states. Fragmenting multiple analytes together also precludes DIA from tailoring collision energy for every analyte, a standard optimization in DDA and targeted acquisition.The low precursor selectivity and resulting complexity of DIA spectra severely challenges the performance of traditional spectrum-centric analysis, which generally assumes that the detected product ions were derived from a single, isolated precursor. The major challenges in interpreting mixture spectra lie in allowing for multiple contributing precursor ions, assessing the dynamic range of mixture peptides, distributing intensities of product ions shared by contributing peptides, and adjusting statistical confidence estimates. Because almost every spectrum is mixed in DIA data, it is poorly suited for analysis by classic spectrum-centric approaches initially designed for DDA data. Some sophisticated spectrum-centric approaches (34–39) address these challenges by deconvolving mixture spectra into pseudo spectra or by matching mixture spectra to combinations of product ions from multiple candidate peptides. However, identification of low abundance analytes by interpreting mixture spectra is inherently difficult because the MS/MS signals from low abundance analytes are naturally overwhelmed by the signals from high abundance ones.Recently, Gillet et al. demonstrated an alternative approach that analyzes DIA data in a targeted fashion (24), opening a new door for the investigation of tandem mass spectrometry data. Much like targeted analysis of transitions used in targeted acquisition methods, Gillet et al. use extracted ion chromatograms to detect and quantify query peptides. Similarly, Weisbrod et al. identify peptides by searching peptide fragmentation patterns against DIA data (25). Instead of interpreting individual spectra in a spectrum-centric fashion, these alternative approaches take each peptide of interest and ask: “Is this peptide detected in the data?” We refer to this approach as “peptide-centric analysis” in contrast with “spectrum-centric analysis.” In peptide-centric analysis, each peptide is detected by searching the MS and MS/MS data for signals selective for the query peptide. Peptide-centric analysis covers all methods that use peptides as an independent query unit, including but not limited to the targeted analysis. Peptide-centric analysis is intrinsically very different from spectrum-centric analysis (Fig. 1) and better suited for addressing many biological problems. This perspective discusses the analytical advantages of peptide-centric analysis and how they could translate to improvements in protein inference, and the analysis of DIA data.

Table I

Analytical comparison of spectrum-centric analysis versus peptide-centric analysis

Comparison between procedures using SDS for shotgun proteomic analyses of complex samples

Filter-aided sample preparation (FASP) and a new sample preparation method using a modified commercial SDS removal spin column are quantitatively compared in terms of their performance for shotgun proteomic experiments in three complex proteomic samples: a Saccharomyces cerevisiae lysate (insoluble fraction), a Caenorhabditis elegans lysate (soluble fraction), and a human embryonic kidney cell line (HEK293T). The characteristics and total number of peptides and proteins identified are compared between the two procedures. The SDS spin column procedure affords a conservative fourfold improvement in throughput, is more reproducible, less expensive (i.e. requires less materials), and identifies between 30 and 107% more peptides at q≤0.01, than the FASP procedure. The peptides identified by SDS spin column are more hydrophobic than species identified by the FASP procedure as indicated by the distribution of GRAVY scores. Ultimately, these improvements correlate to as great as a 50% increase in protein identifications with two or more peptides.