A Dual Workflow to Improve the Proteomic Coverage in Plasma Using Data-Independent Acquisition-MS

Abstract

Plasma is one of the most important and common matrices for clinical chemistry and proteomic analyses. Data-independent acquisition (DIA) mass spectrometry has enabled the simultaneous quantitative analysis of hundreds of proteins in plasma samples in support population and disease studies. Depletion of the highest abundant proteins is a common tool to increase plasma proteome coverage, but this strategy can result in the nonspecific depletion of protein subsets with which proteins targeted for depletion interact, adversely affecting their analysis. Our work using an antibody-based depletion column revealed significant complementarity not only in the identification of the proteins derived from depleted and undepleted plasma, but importantly also in the extent to which different proteins can be reproducibly quantified in each fraction. We systematically defined four major quantitative parameters of increasing stringency in both the depleted plasma fraction and in undepleted plasma for 757 observed plasma proteins: Linearity cutoff r² > 0.8; lower limit of quantification (LLOQ); measurement range; limit of detection (LOD). We applied the results of our study to build a web-based tool, PlasmaPilot, that can serve as a protocol decision tree to determine whether the analysis of a specific protein warrants IgY14 mediated depletion.

Graphical Abstract

INTRODUCTION

The plasma component of blood is the primary biofluid for diagnostic and prognostic clinical analyses.¹ Every tissue ultimately interacts with plasma, which is readily available and sampled. Thus, by detecting and quantitatively analyzing the appropriate target, plasma may serve as an individualized window into physiology and a person’s health/disease status. Plasma is a complex biochemical matrix whose proteome spans a dynamic range in excess of 10 orders of magnitude in which the 10 most abundant proteins account for 90% of its total protein mass.² This broad and lopsided dynamic range of proteins within plasma can render the physiological window opaque to discovery-based analyses of less abundant proteins, which in turn can potentially hinder new candidate biomarker identification.³

Depletion strategies have been important components of discovery workflows because they address the mass imbalance and broad dynamic range inherent to plasma.^4,5 Selective depletion of highly abundant proteins can be accomplished by chemical⁶ and capture-based methods.⁷ Chemical depletion involves the precipitation of all proteins except albumin,⁷ while immunoaffinity-based methods generally employ immobilized antibodies targeted toward a series of high abundant plasma proteins.⁸ Although the number of proteins targeted by immunoaffinity-based methods can differ, these strategies almost invariably target albumin, IgG, α1-Antitrypsin, IgA, IgM, transferrin, haptoglobin, α2-macroglobulin, fibrinogen, complement C3, α1-acid glycoprotein (orosomucoid), HDL (mainly apolipoproteins A-I and A-II), and/or LDL (mainly apolipoprotein B).⁹

Regardless of the depletion strategy, the remaining effluent, the depleted proteins are typically highly diluted and must be further prepared prior to downstream MS-based proteomic analysis. The absence of depletion inherently simplifies sample preparation and decreases the introduction of variance and error, but limits dynamic range.¹⁰ Furthermore, depletion can therefore affect the quantitative profiles of both high abundant proteins and any bound less-abundant binding partners.^11–13 Nevertheless, relative to the analysis of undepleted plasma samples, depletion serves as an additional lens through which large numbers of less abundant plasma proteins can be revealed by MS.

Data independent acquisition mass spectrometry (DIA-MS) can simultaneously analyze hundreds of proteins in plasma samples with a common goal of establishing DIA-MS-based discovery workflows that have specificity, quantitative accuracy and reproducibility.^14,15 While the value of depletion to proteomic coverage of low abundant targets extends to DIA-MS, its inclusion or exclusion yields samples with distinct biochemical matrices and altered protein compositions. In contrast to targeted mass spectrometry approaches, like multiple reaction monitoring (MRM also known as selected reaction monitoring) that target a constrained number of peptides for the accurate quantitation of a specific protein,¹⁶ DIA-MS workflows cannot be simultaneously tailored to suit all plasma protein subsets. Furthermore, the cumulative effects of depletion efficiency, capture reagent specificity, nonspecific binding, and indiscriminate protein loss are assumed to be unequal across the proteome and several fundamental questions remain unaddressed. To what extent does plasma depletion influence downstream protein quantitation and for which proteins is it beneficial or a hindrance? Are there any advantages of analyzing both undepleted plasma alongside plasma depleted of its most abundant proteins?

Our work sought to address these questions by systematically optimizing and defining the preparative parameters that maximize proteome coverage and quantitative accuracy from depleted and undepleted plasma aliquots, respectively. We incorporated these optimized protocols to establish a parallel dual workflow strategy in which samples are divided into two aliquots: the first aliquot was shunted to IgY14-based depletion (which targets the top 14 most abundant plasma proteins for depletion⁸) and desalted after trypsin digestion with the goal of analyzing less-abundant candidates that minimally interact with high abundance matrix proteins; the second aliquot was undepleted to preserve the quantitative accuracy among candidates that more extensively interact with high-abundance matrix proteins (such as albumin). Both IgY14-depleted and undepleted aliquots were digested and desalted prior to being subject to DIA-MS analysis. Finally, we systematically defined four major quantitative parameters of each observed protein (linearity cutoff r² > 0.8; LLOQ, HLOQ-quantitative range; LLOD) and built a protocol decision tree encompassing 757 plasma proteins. Our optimized dual workflow and its ensuing protocol decision tree can serve as an important guide for the targeted quantitative analysis of specific protein subsets, or of broader discovery-based proteomic studies.

MATERIALS AND METHODS

All aqueous solutions were prepared using HPLC-grade water (Fisher Scientific) unless otherwise specified.

IgY14 Depletion

All human plasma (Bioreclaimation IVT (lot# BRH1120184)) was depleted using IgY14 column gradient chromatography (Seppro IgY14 LC-10, Sigma-Aldrich) with a flow rate of 0.5 mL/min using the ProteomeLab PF2D (Beckman Coulter) coupled with a fraction collection module and an online 280 nm UV detector.^17,18 The liquid chromatography (LC) program is summarized in Supplementary Table S1. All mobile phases were freshly prepared and sterile-filtered using a 0.45 μm filtration flask (Cat#430770 Corning). Mobile Phases were prepared fresh according to the Seppro IgY14 technical instructions. Plasma was spin-filtered by centrifugation (Spin-X 0.45 μm 2.0 mL, Cat38163, COSTAR). The filtrate was diluted 6-fold with dilution buffer (Seppro IgY14 kit) containing exogenously spiked β-Galactosidase (0.1 μg/μL). Upon priming the LC lines and rinsing the column with mobile phase, 600, 900, 1200, or 1500 μL of filtered and diluted plasma (corresponding to 100, 150, 200, and 250 μL of plasma equivalent, respectively) was loaded onto the IgY14 column sequentially for depletion optimization. 7.5–10 mL of flow-through was collected between 10–30 min and checked to ensure a pH range of 6–7. 15–25 mL of IgY14-depleted eluate was collected at 35–60 min (pH2–3), and neutralized with 2 mL of neutralization buffer (Seppro IgY14 kit).

Sample Preparation for LC-MS/MS

The following solutions were divided into single-use aliquots after preparation, stored at −80 °C, and thawed as individual aliquots on an as-needed-basis immediately prior to use: Reaction buffer consisting of 20% (w/v) octyl-beta-glucopyranoside (OGS, Sigma-Aldrich O9882–5G); 50 mM Tris (2-carboxyethyl) phosphine (TCEP, ThermoFisher Scientific PI20491); 5 mg/mL β-Galactosidase (β-Gal, Sigma-Aldrich G3153–5MG). A stock solution of 10% (v/v) formic acid was prepared in an amber glass container by diluting a 1 mL ampule of formic acid (Fisher Scientific, A117–10 × 1AMP) in 9 mL of LC-MS grade water (Fisher Scientific W6212,). The following solutions were freshly prepared: 100 mM Tris, 4 mM CaCl₂ pH8.5; 200 mM Iodoacetamide (Sigma-Aldrich, I6125–10G). Internal standard solution consisted of 2 μg/μL β-gal protein dissolved in water. Trypsin solution was freshly prepared by dissolving N-tosyl-l-phenylalanine chloromethyl ketone (TPCK)-treated trypsin (Sciex, 4445250) in 0.1% (v/v) formic acid at 2 μg/μL.

For each reaction, 400 μg of protein were mixed with 27.5 μL reaction buffer, 5 μL 20% (w/v) OGS, and 5 μL internal standard solution, depending on the experiment. Sample reduction was performed by adding 5 μL of 50 mM TCEP to each sample, followed by incubation at 60 °C for 60 min. Alkylation: 2.5 μL of 200 mM Iodoacetamide was added to each sample, followed by incubation for 10 min at room temperature. After adding 60 μL 100 mM Tris, 4 mM CaCl₂ pH8.5, TPCK-treated trypsin was added to achieve a trypsin:protein ratio of 1:20, and incubated at 37 °C. Tryptic digests were acidified to a final concentration of 0.1% formic acid (FA). Peptide desalting was performed using solid phase extraction (SPE) using 96-well Oasis HLB plates with a dedicated vacuum manifold (Waters WAT058951). The HLB sorbent was activated with methanol and washed 3 times with 1 mL 0.1% FA. Each tryptic digest was diluted with 300 μL of 0.1% FA, supplemented with 400 μL of 4% phosphoric acid, and slowly loaded (~2 drops/s) onto the sorbent. Loaded peptides were washed 3 times with 1 mL 0.1% FA, and eluted with two successive 0.5 mL of 50% ACN in 0.1% FA. Eluted peptides were dried down using a speedvac and stored dry at −80 °C in anticipation of LC-MS/MS analysis.

LC-MS Analysis

LC-MS/MS analysis for all DDA-MS experiments was performed using an Ultimate 3000 nano LC (Thermo Scientific) connected to an Orbitrap Elite mass spectrometer (Thermo Scientific) equipped with an EasySpray ion source. Peptides were first loaded onto a trap column (PepMap100 C18, 5 μm, 100 Å, 300 μm i.d. × 5 mm, Thermo) followed by separation on a PepMap RSLC C18 column (2 μm, 100 Å, 75 μm i.d. × 25 mm, Thermo) using a flow rate of 300 nL/min with a linear gradient of 5–35% B for 90 min, 35–95% B for 3 min, holding at 95% B for 7 min and re-equilibrating at 5% B for 25 min at 400 nL/min (mobile phase A was 0.1% formic acid in water and mobile phase B was 0.1% formic acid in acetonitrile). The nanosource capillary temperature was set to 275 °C and the spray voltage was set to 2.2 kV. MS¹ scans were acquired in the Orbitrap Elite at a resolution of 60 000 fwhm (400–1700 m/z) with an AGC target of 1 × 10⁶ ions over a maximum of 100 ms. MS² spectra were acquired for the top 15 ions from each MS1 scan in CID mode in the ion trap and a target setting of 1 × 10⁴ ions, a max accumulation time of 50 ms, and an isolation width of 2 Da. The normalized collision energy was set to 35% and one microscan was acquired for each spectrum. Monoisotopic precursor selection was enabled and only MS¹ signals exceeding 500 counts triggered the MS2 scans, with +1 and unassigned charge states not being selected for MS² analysis. Dynamic exclusion was enabled with a repeat count of 2, repeat duration of 30 s, and exclusion duration of 90 s.

All samples acquired by DDA-MS were searched using Sequest (Thermo Fisher Scientific, San Jose, CA, USA; version 1.0) configured with the human canonical proteome database (UP_Human_Canonical_Reviewed_March16) downloaded from Uniprot, containing 49 374 protein entries, including decoys. The database was additionally supplemented with iRT sequences and β-gal peptide sequences for use with QC applications.¹⁹ Search parameters assumed strict tryptic digestion of peptides, a fragment ion mass tolerance of 1.00 Da, and a parent ion tolerance of 50 ppm. Carbamidomethyl of cysteine was specified as a fixed modification, while deamidation of asparagine and glutamine as well as methionine oxidation were specified as variable modifications. Validation of MS/MS based peptide and protein identification was performed using Scaffold software (version Scaffold_4.5.3, Proteome Software Inc., Portland, OR). Peptide identifications were accepted if they could be established at greater than 95.0% probability by the Peptide Prophet algorithm²⁰ with Scaffold delta-mass correction. Protein identifications were accepted if they achieved than 99.0% probability with a minimum of 2 identified proteotypic peptides. Protein probabilities were assigned by the Protein Prophet algorithm.²¹ Proteins containing indistinguishable peptides based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Proteins sharing significant peptide evidence were grouped into clusters. Proteins were annotated with GO terms from gene_association.goa_human (downloaded Mar 18, 2015).²²

MRM experiments were performed using a Prominence LC system (Shimadzu) coupled to a QTRAP 6500 mass spectrometer (SCIEX). Peptides were reconstituted to a final concentration of 2 μg/μL in 0.1% FA containing ¹⁵N-labeled synthetic peptides from β-galactosidase (New England Peptide) as an internal standard. Five μL of reconstituted peptides was loaded onto an XBridge Peptide BEH C₁₈ analytical column (100 mm, 2.1 mm i.d., 3.5 μm particle size), protected by a VanGuard XBridge BEH C₁₈ guard column (5 mm, 2.1 mm i.d., 3.5 μm particle size) (Waters). The column temperature was set to 36 °C. Mobile phase A consisted of 2% (v/v) acetonitrile and 0.1% (v/v) formic acid in water, whereas mobile phase B consisted of 5% water (v/v) and 0.1% (v/v) formic acid in acetonitrile. Flow rate was maintained at 0.250 mL/min. The initial gradient spanned from 2 to 5% mobile phase B for 5 min, increased to 20% over the next 10 min interval, increased to 45% over the next 4 min, and finally increased to 90% over the remaining 0.5 min and maintained for 3 min. Column equilibration consisted of 2% mobile phase B for 3 min. Total LC time was 26 min. MRM was performed in unit resolution configuration under positive electrospray ionization. Ion spray voltage and temperature were set at 5KV and 500 °C, respectively. Curtain gas, GS1, and GS2 were set at 35, 60, and 55 AU respectively. A scheduled MRM transition list (Supplementary Table S1) was created with a 90 s window around the expected retention time, with a 1 s cycle time and a defined unit resolution for both Q1 and Q3. Data acquisition and analysis for MRM was performed with Analyst 1.6.2 and MultiQuant 3.0 software (SCIEX).

DIA-MS was performed using an Eksigent NanoLC 415 System coupled to a TripleTOF 6600 mass spectrometer (both Sciex). Dried desalted peptides were reconstituted to a final concentration of 1 μg/μL (unless otherwise specified) in 0.1% FA containing 1:40 internal retention time standards (Biognosys, Zürich, Switzerland). LC separation was performed by injecting 4 μL (~4 μg protein) onto a ChromXP C₁₈ column (150 × 0.3 mm, Sciex) at a flow rate of 5 μL/min, using 0.1% FA in water and 0.1% FA in acetonitrile as mobile phase A and B, respectively (Optima LC-MS grade, Fisher Scientific). Mobile phase B was increased from 3% to 35% in 60 min, then from 35% to 85% in 3 min, and finally held at 85% for 5 min before re-equilibration at 3% mobile phase B for 7 min, for a 75 min total LC run time. Peptide profiles were obtained by DIA using sequential windowed acquisition of all theoretical fragment ion mass spectra (SWATH) consisting of 100 Q1 variable isolation windows across a mass range from 400 to 1250 Da (30 ms dwell time per window). A full MS1 scan (250 ms dwell time) was acquired before each SWATH cycle. Ion spray voltage was set at 5500 V and ion spray temperature at 100 °C. Curtain gas was set as 25 AU, GS1 at 15 AU, and GS2 at 20 AU, and the declustering potential was 80 V.

Data Analysis for DIA-MS.

SWATH-MS.wiff files from the DIA-MS runs were first converted to profile mzML using ProteoWizard²³ v.3.0.6002. The whole process of SWATH-targeted data analysis was carried out using OpenSWATH²⁴ v.2.0.0 running on an internal computing cluster. OpenSWATH utilizes a target-decoy scoring system (PyProphet v.0.13.3) such as mProphet to estimate the identification of FDR. The best scoring classifier that was built from the sample of most protein identifications was utilized in this study. On the basis of the final spectral library used in this study, which encompasses 1667 unique Swiss-Prot proteins represented by 43 899 peptides and 353 160 transitions,²⁵ OpenSWATH first identified the peak groups from all individual SWATH maps at a global peptide FDR of 1% and aligned them between SWATH maps based on the clustering behaviors of retention time in each run with a nonlinear alignment algorithm.²⁶ For this analysis, the MS runs were realigned to each other using locally weighted scatterplot smoothing method and the peak group clustering was performed using “LocalMST” method. Specifically, only those peptide peak groups that deviate within 3 standard deviations from the retention time were reported and considered for alignment with the max FDR quality of 5% (quality cutoff to still consider a feature for alignment). Next, to obtain a high-quality quantitative data at the protein level, we discarded those proteins whose peptides were shared between multiple different proteins (nonproteotypic peptides).²⁷ Data preprocessing and statistical analysis of MS runs into quantitative data was performed using mapDIA²⁸ v2.4.1. The fragment-level intensities were normalized based on MS² TIC (total ion current) from each DIA MS run to remove systematic bias between MS runs. This is followed by outlier removal and peptide/fragment selection that preserve the major quantitative patterns across all samples for each protein. The selected fragments and peptides are used in the final model-based statistical significance analysis of protein-level differential expression between specified groups of samples. Quantitative peptide and protein level summary outputs were then used for all downstream biological analysis.

Additional Downstream Analyses.

To assess data quality and assay characterization, we calculated Calibration Curves²⁹ measuring the analytical response against concentrations of all given protein analytes in our assay derived from loading serially diluted samples. This was accomplished using a Linear OLS (Ordinary Least Squares) Regression fit. These analyses were used to compute Limits of Detection (LOD), Limits of Quantification (LOQ), Precision and Reproducibility using % CV, and Accuracy using % error to present a comparative set of protein measurements for characterization of assay performance and precision.

RESULTS AND DISCUSSION

Depletion Optimization

We sought to determine optimal IgY14-column loading by performing depletion on 100, 150, 200, and 250 μL of plasma each containing a uniform concentration of exogenously spiked β-Galactosidase (0.1 μg/μL). We performed nano-LC-MS/MS with DDA on the ensuing four IgY14-depleted fractions in parallel with an undepleted plasma sample, and quantified β-Galactosidase, ApoE, and Albumin. The recovery of these three proteins, assessed by their spectral counts in IgY14-depleted plasma, served as representative indices for proteins not targeted for depletion (β-Galactosidase), endogenous proteins that may be subject to nonspecific secondary depletion (ApoE), and proteins targeted for depletion whose presence indicates of IgY14-column overloading (Albumin).

Figure 1A indicates that Albumin was detected in the IgY14-depleted fraction when a plasma volume of 250 μL was applied to the IgY14 column, indicating that the depletion column was saturated between 200 and 250 μL. By contrast, we detected β-Galactosidase to a similar extent in all of our IgY14-depleted and undepleted samples suggesting that recovery for proteins unrelated to any targeted proteins was consistent across all tested plasma loads. Finally, while ApoE is not specifically targeted for depletion by the IgY14 column, but in our experience ApoE can be susceptible to partial depletion. Thus, we reasoned that ApoE could serve as an index for nonspecific depletion that could arise from lower-affinity interactions when the column is inadequately loaded and targeted proteins are insufficiently present to occupy their binding sites. We observed consistent recovery of ApoE across all IgY14 loading conditions (Figure 1A). On aggregate, these results indicated that the optimal volume of plasma for this IgY14-depletion column was 200 μL. Along with Albumin, ApoE, and β-Galactosidase, this preliminary DDA-based analysis also yielded a quantitative comparison of a total of 130 proteins across the undepleted and the four IgY14-depleted plasma conditions. Among these proteins depicted in Figure 1B, we observed 66 proteins with a significantly increased quantitative signal in all of the IgY14-depleted samples and 29 proteins with a significantly decreased signal (Figure 1B). These results not only confirm that the IgY14 column yields depletion of proteins beyond the 14 specifically targeted proteins, but also that the analysis of both IgY14-depleted and undepleted plasma fractions can substantially improve coverage owing to the complementarity of their intrinsic proteomes.

Figure 1.

(A) Quantitative comparison three representative indices of proteins not targeted for depletion (β-Galactosidase), endogenous proteins that may be subject to nonspecific secondary depletion (ApoE), and proteins targeted for depletion whose presence indicates of IgY14-column overloading (Albumin), in both IgY14-depleted and undepleted plasma (n = 3, mean ± standard deviation). (B) Heatmap and two-way clustering of the proteins identified in IgY14-depleted and undepleted plasma samples. 100, 150, 200, and 250 μL of plasma containing β-Galactosidase were subjected to IgY14 depletion, independently digested, and analyzed by LC-MS/MS with DDA mode. Quantitation is based on total spectra normalized to the β-Galactosidase signal using Scaffold.

Given that several high-abundant plasma proteins targeted by the IgY14 column have well established high-affinity protein–protein interactions, our finding that auxiliary depletion occurs is not unsurprising. Indeed, we speculate that additional proteins, undetected in this experiment, are also susceptible to this type of secondary depletion. To this end, we performed string network analysis on the 29 proteins whose decreased signal indicated that they had undergone primary or secondary depletion (Supplementary Figure S1). This secondary analysis indicated that almost all IgY14-depleted proteins are known to interact within a distinct network, and we therefore infer that nontargeted proteins with significant direct or indirect binding affinities to targeted high-abundance proteins are favored for secondary depletion.

Digestion Optimization

In order to determine the optimal digestion time for the MS-based analysis of plasma proteins, we performed LC-MS/MS with Data Independent Acquisition on undepleted plasma subjected to tryptic digested under uniform conditions for 2, 4, 6, and 10 h. Since peptides are the direct measure, we used peptides as our primary metric to assess technical optimization. Optimal digestion conditions differ from protein to protein, and partially digested peptides resulting from suboptimal digestion conditions can adversely affect quantification. We performed a spectrum-centric search using DIA-umpire in order to assess the extent of incomplete tryptic digestion as well as nonspecific proteolytic peptides. As expected, the 2 h digestion resulted in a greater proportion of partially digested miscleaved tryptic peptides relative to the other time points, which was partially offset by a reciprocal decrease in nonspecific proteolytic peptides. The 2 h digestion also yielded the largest number of canonical tryptic peptides (Figure 2). These DIA experiments are meant to ultimately provide quantitative results more comprehensively across proteomes. This goal is best served by maximizing the pool of candidate peptides on which to draw for quantitation. For this reason and for the purposes of this study, we assessed optimal digestion time by the total number of identified peptides as well as their reproducibility across 3 biological replicates per digestion time, each of which was injected in triplicate for LC-MS/MS analysis (Figure 2). For the purposes of our study, bioreplicates were samples drawn from a single pool of plasma, prepared and analyzed in parallel. Our results indicate that the number of peptides is both highest and most consistent across the three bioreplicates with a 2 h digestion time.

Figure 2.

Comparison of the peptides identified in plasma samples over varying digestion times. An EDTA plasma stock was aliquoted and digested under uniform conditions for 2, 4, 6, and 10 h. Each digestion time was performed with 3 individual bioreplicates each of which was injected 3 times for LC-MS/MS analysis using Data Independent Acquisition. Peptide identities were derived using openSWATH with FDR < 5%. (A) The number of peptides identified for each of the three bioreplicates (graphed as red, blue, and green points to represent individual injections) across each digestion time. Box plots depict median with interquartile ranges. (B) The extent to which peptide identities overlap between bioreplicates for each of the digestion times. (C) The extent to which canonical tryptic peptides are generated relative to nonspecific proteolytic peptides and miscleaved tryptic peptides over the four digestion times. This result is based on a thorough search of DIA-umpire generated MS² spectra by ProteinPilot and includes all nontryptic digestion possibilities. The percentage of every group of digestion completeness is calculated by the number of those peptides vs the total number of identified peptides.

Although the number of identified peptides was maximal at 2 h, shorter digestion times can suffer from inconsistencies in the identities of the peptides that are produced. This is because insufficient digestion times produce partial digestion of random protein subsets by virtue of the stochasticity with which the digestion enzymes interact with proteins and protein fragments at a molecular level. We plotted individual Venn diagrams for each digestion time to depict the extent to which peptides identified in every technical injection within a biological replicate, overlap between bioreplicates (Figure 2B). The 2 h digestion time had the highest number of peptides as well as the most consistency as defined by the number of peptides that overlapped between the three bioreplicates. This compared favorably with all other time points, indicating that a digestion of 2 h had both the most- and the most consistently identified peptides. While our study was not designed to determine why peptide identities by DIA decreased with digestion times beyond 2 h, there are several possibilities. We speculate that while increased digestion times may decrease the number of partially digested peptides and improve the ability to quantitate these peptide subsets, it may also result in the degradation of some more labile plasma peptide subsets, as previously reported.^30,31 Thus, plasma protein subsets that did not meet our quantitative criteria may be quantifiable under digestion conditions that are more precisely tailored, but the potential trade-off may be that other proteins may be less consistently detected.

Plasma Sample Desalting Strategies

The salinity of both undepleted and IgY14-depleted plasma samples requires a desalting strategy prior to LC-MS/MS analysis. Undepleted plasma samples have a well-defined desalting protocol that involves protein digestion followed by Solid Phase Extraction using HLB (Waters) elution plates. Although the IgY14-depletion protocol yields higher volume samples with increased salinity (150 mM NaCl), Solid Phase Extraction subsequent to digestion remains a viable option for desalting and sample concentration with a plate-based format that enables digestion and desalting of 96 samples in approximately 1.5 days.

We additionally considered two alternative strategies for desalting IgY14 samples (Supplementary Figure S2). The first alternative involved protein desalting by size exclusion chromatography (Zeba spin columns) ahead of digestion, followed by solid phase extraction of the peptides once the digestion is complete. Although this alternative approach decreases sample salinity ahead of digestion, it effectively yields more dilute samples (<30 μg/mL) with large volumes (3 mL) that require enrichment by solid phase extraction ahead of LC-MS/MS analysis. From a practical perspective, this strategy has several centrifugation steps that are incompatible with automation and requires approximately 3 days to digest and desalt 96 samples. The second alternative involves offline C₁₈-reverse phase LC-based protein desalting followed by elution into a trypsin-compatible ammonium bicarbonate buffer. Although this strategy does not require any additional steps ahead of LC-MS/MS analysis, it is incapable of parallel or multiplexed sample processing and thus requires in excess of 14 days to desalt and digest 96 samples. Given the obvious throughput advantages, we similarly desalted both IgY14-depleted and undepleted plasma samples at the peptide level after digestion with solid phase extraction using HLB (Waters).

We evaluated the reproducibility and recovery of this HLB desalting strategy by performing MRM-MS, to quantify a known concentration of β-galactosidase exogenously spiked into 3 plasma aliquots which were digested and either subject to HLB-plate-based desalting, or directly injected onto the LC-MS system which contained a divert valve to facilitate on-column desalting. This experiment yielded an average recovery of 72.7% with a percent coefficient of variance (% CV) of 2.2% (N = 3). We determined that owing to the low coefficient of variance, these results were sufficient to apply solid-phase HLB-based desalting to larger sample cohorts.

A Protocol Decision Tree for Plasma Protein Analysis

We applied DIA-MS to the optimized preparative parameters in order to assess the influence of IgY14-depletion protocol on plasma proteome analysis. Specifically, we sought to assess the extent to which plasma depletion influences protein quantitation, to define those proteins whose quantitation is benefited and/or hindered by depletion, and to address the potential advantages of concurrently analyzing both undepleted and IgY14-depleted fractions of the same sample.

In the first series of experiments, we analyzed the number of proteins and peptides detected while increasing the amount of sample loaded onto the analytical column. Beyond the metrics of number of identified and quantified peptides, higher numbers of peptides loaded onto columns can have a practical consequence of increased instrument down-time due to a higher frequency of cleaning to preclude MS instrument performance being adversely affected. For this reason, lower sample loading could be advantageous provided analytical performance remains uncompromised. However, owing to the dynamic range of plasma proteins, individual proteins exhibit unique linear ranges. By extension, the identification and the reproducible quantitation of specific proteins may require different on-column protein loads. As expected, we are able to detect more peptides with improved quantitative confidence as sample loading is increased. Our results for identification indicated that both peptide and protein numbers similarly plateau when the column was loaded above 500 ng of plasma digest (Figure 3A and B). Interestingly, this pattern was mirrored in both undepleted and IgY14-depleted plasma samples. Ultimately, we consistently observed more proteins and peptides in the IgY14-depleted plasma samples than in the undepleted samples. A more in-depth analysis revealed that proteins identified in IgY14-depleted plasma are also derived from an average of 6 peptides/protein compared to 4 peptides/protein in the undepleted plasma, which can be beneficial from the perspective of selecting and validating quantitative and qualitative peptides on which to build MRM-type analyses. Having identified greater numbers of proteins in the IgY14-depleted fraction, we sought to determine the extent to which the observed proteins were shared between the depleted and undepleted fractions. While some proteins (and peptides) were shared, IgY14-depleted plasma reveals a proteome which is largely complementary to that revealed in undepleted plasma (Figure 3C).

Figure 3.

Numbers of proteins (A) and peptides (B) identified by DIA-MS with increased sample loading on column. The digest of both IgY14-depleted and undepleted plasma were loaded LC-MS from 63 to 2000 ng; three technical replicates were injected at every concentration. (C) Proportional Venn diagrams representing proteins and peptides uniquely and commonly identified in undepleted and IgY14-depleted plasma samples with increasing peptide load on-column.

We sought to determine if the complementarity between undepleted and IgY14-depleted plasma observed in Figure 3C extended beyond identification to include quantitation. To address this question, we subjected the entire data set to independent two-way clustering analysis. This analysis yielded a heatmap that reveals a distinct grouping pattern (Figure 4). First, the clustering analysis correctly grouped IgY14-depleted and undepleted samples along the y-axis with increasing on-column loads. Second, 2-way clustering resulted in distinct “quartiles” that partition according to the linear dynamic ranges of proteins. Specifically, the upper left quartile is mainly populated with proteins whose identification and quantitation is largely dependent on depletion; these likely represent mid- to low-abundance proteins whose detection is suppressed or obscured by the dynamic range inherent to plasma. By extension, the lower left quartile indicates that these same proteins are largely absent or lack any discernible quantitative range in undepleted plasma samples. The two right-sided quartiles reciprocally show proteins whose quantification is by-and-large impaired by depletion; these likely represent high abundant proteins that are directly or secondarily affected by depletion. Thus, 2-way clustering reveals that the complementarity between IgY14-depleted and undepleted workflows applies to both protein identification and quantification. Furthermore, implicit in this complementarity is the finding that a simultaneous analysis of both IgY14-depleted and undepleted fractions is inevitably greater than the analysis of one fraction performed in isolation of the other, regardless of the desired level of analytical stringency.

Figure 4.

Heatmap of identified proteins (x-axis) in IgY14-depleted and undepleted plasma with increasing plasma-digests loaded on-column (63–2000 ng, three replicates each, y-axis). Ordering along both x- and y-axes was independently defined by 2-way clustering (Jmp software). Blue-to-red color scale represents absence (deep blue), to high-abundance (deep red) protein-level quantitation derived from signal intensities of MS² peak areas.

Our results indicate that the parallel analysis of IgY14-depleted and undepleted fractions from a given plasma sample significantly improves identification and quantitation of DIA-MS-based plasma proteome analysis. Our results can also help distinguish proteomic subsets whose qualitative and/or quantitative analysis requires depletion from those that do not. Additionally, both DIA-MS and MRM utilize transitions and MS2 chromatographic peak areas for the quantitation of analytes. On account of this shared fundamental principle, our DIA-derived comprehensive plasma proteome results can serve as a valuable tool to guide sample preparation workflows ahead of MRM assay development targeted to individual proteins. Our characterization of the dual workflow was designed not only to ascertain the extent to which each of the observed proteins could be identified and quantified, but also to define the parameters of that quantification—namely, linearity and linear dynamic range. In this way, these results can serve as a decision-tree for plasma protein analysis (Figure 5), which progresses from left-to-right toward increasing levels of quantitative stringency. Thus, if the experimental goal is limited to protein identification, the combined analysis of IgY14-depleted and undepleted fractions (yields 757 identified proteins, 619 of which are quantifiable by DIA-MS according to our criteria for quantitation of a defined LOQ and observed in a minimum of 2 replicates with a CV < 20%; upon performing linear ordinary least squares regression fit on the quantified proteins, 588 of these had a quantifiable linear range as defined by an RSQ > 0.8 across a minimum of 3 points. The range between their LLOQ (lower limit of quantitation) and ULOQ (upper limit of quantitation) was defined as their linear dynamic range.

Figure 5.

PlasmaPilot Protocol decision tree and web-tool based on our dual workflow results. The combined analysis of IgY14-depleted and undepleted plasma fractions yields a total of 757 unique identified plasma proteins. (Top) Proteins are partitioned according to the workflow (depletion, undepletion, or both) required to achieve identification, quantification, and quantification with good linearity. Right-most panel partitions proteins with good linearity according to the workflow that yields the broadest linear range. ¹Identifiable proteins were detected by DIA-MS with an FDR < 1% as calculated by OpenSWATH. ²Quantifiable proteins were detected by DIA-MS with a defined LOQ and observed in a minimum of 2 replicates with a CV < 20%. ³Quantified proteins with good linearity, having an RSQ > 0.8 across a minimum of 3 concentration points. ⁴Quantified proteins with good linearity are segregated according to the workflow that yields the widest linear range. ⁵Proteins that are equivalent between depletion and undepletion strategies are numerated in the middle “both” column. ⁶72 proteins can similarly be quantified by either the IgY14-depleted or the undepleted workflows, one of these proteins (Q04917) shows linear ranges of similar magnitudes spanning different quantitative ranges for IgY14-depleted and undepleted workflows. (Bottom) The PlasmaPilot web interface, which provides a prospective sample preparation guide for the quantitative MS-based analysis of these plasma proteins. Bottom left shows proteins of interest as a data table with a recommendation pie chart, and bottom right shows peptide coverage—all based on the level of quantitative stringency selected by the user. The web application hosted on Heroku, built using AngularJS and Bootstrap for the front-end and NodeJS for the back-end can be accessed at http://plasmapilot.herokuapp.com. Data are stored on the Heroku Postgres database and visualized using the Bar Chart and Pie Chart components from the DevExtreme HTML5 JavaScript Charts Library. PlasmaPilot is used via the home page, where protein(s) of interest are queried in the text box using UniProt IDs or gene names (separated by new line or comma). After selecting the desired analytical stringency (Identifiable, Quantifiable, Linearity), the query is submitted. As part of the analysis results, we provide the observed sample preparation results along with recommendations based on the mode selected. Recommendations are based on the degree of detection achieved in IgY14-depleted and undepleted samples. Query proteins are visualized in the Pie chart for a better understanding of their distribution to infer recommended sample preparation method. Each queried protein can be investigated further to assess protein coverage and to determine which peptides were detected for future targeted proteomic efforts.

In contrast to discovery based DIA-analysis, our protocol decision tree can help determine the necessity of depletion for the targeted assay development of a given protein subset. In this light, IgY14 depletion workflow is only warranted if a protein subset of interest resides among the 301 proteins uniquely identified in IgY14-depleted plasma. Alternatively, if the experimental goal expands to include quantitation of a target that resides among the 285 proteins that can be identified in both IgY14-depleted and undepleted fractions, 65 of these partitions to IgY14-depleted and 14 to the undepleted fractions. Furthermore, as expected, some proteins that can be reliably identified will not meet the criteria for quantitation. A further 40 proteins that are quantifiable in both workflows similarly partition if the targeted analysis requires quantitation with good linearity (9 partition to the undepleted plasma; 31 partition to IgY14-depleted plasma). A final partitioning takes place as proteins are ranked according to the magnitude of their linear dynamic range. This final level of stringency in our study leaves only 72 proteins with similar linear dynamic ranges between IgY14-depleted and undepleted fractions, and for practical purposes, 71 of these proteins would be best analyzed in the absence of depletion. Interestingly, a single remaining protein does not clearly partition; its linear dynamic range is similar in magnitude but spans an orthogonal quantitative range in IgY14-depleted or in undepleted fractions. Thus, as quantitative stringency increases, one subset of proteins inevitably fails to meet the analytical requirements, and a growing fraction of those proteins shared by both workflows partition to one workflow or the other. Our protocol decision tree may be valuable for investigators beyond our research group, but manually searching a list for a protein or proteins of interest can be cumbersome. For this reason, we built a freely available web-based search tool in which a user can query multiple Uniprot ID’s, which highlights protein coverage under our experimental conditions, and which can be used to determine if IgY14 depletion is required for identification, simple- or accurate-quantitation, and quantitation with a broad linear range. We have built this tool, which we call “PlasmaPilot” as a web-based graphical user interface, implemented and hosted on Heroku, providing a user-friendly way to query and visualize our plasma data at http://plasmapilot.herokuapp.com/.

In conclusion, the combination of both IgY14-depleted and undepleted plasma samples for DIA-MS results in improved plasma proteome coverage and quantitation. Twinning IgY14-depleted and undepleted workflows not only enabled us to define those proteins that were uniquely identified in a given preparation, but also informed which preparative strategy was quantitatively superior for those proteins that were consistently identified.

ABBREVIATIONS

DIA-MS

data independent acquisition mass spectrometry

MRM

multiple reaction monitoring

LC

liquid chromatography

SPE

solid phase extraction

LLOQ

lower limit of quantification

HLOQ

higher limit of quantification

LOD

limit of detection

LDL

low-density lipoprotein

HDL

high-density lipoprotein

TCEP

Tris (2-carboxyethyl) phosphine

OGS

octyl-beta-glucopyranoside

TPCK

N-tosyl-l-phenylalanine chloromethyl ketone