Selected Ion Extraction of Peptides with Heavy Isotopes and Hydrogen Loss Reduces the Type II Error in Plasma Proteomics

Abstract

Naturally occurring peptides display a wide mass distribution after ionization due to the presence of heavy isotopes of C, H, N, O, and S and hydrogen loss. There is a crucial need for sensitive methods that collect as much information as possible about all plasma peptide forms. Statistical analysis of the delta mass distribution of peptide precursors from MS/MS spectra that were matched to 63,077 peptide sequences by X!TANDEM revealed Gaussian peaks representing heavy isotopes and hydrogen loss at integer delta mass values of -3, -2, -1, 0, +1, +2, +3, +4, and +5 Da. Human plasma samples were precipitated in acetonitrile, and the resulting proteins were collected over a quaternary amine resin, eluted with NaCl, digested with trypsin, and analyzed by nano liquid chromatography-electrospray ionization-tandem mass spectrometry (LC-ESI-MS/MS) with an orbital ion trap (OIT). Fragment spectra (MS/MS) generated from the OIT data were fit to human fully tryptic peptides by X!TANDEM, which led to the identification of 3,888 protein gene symbols represented by three or more peptides (n ≥ 3). The peptide counts to plasma proteins from experimental MS/MS spectra were corrected against 29 blank LC-ESI-MS/MS spectra and 30 million random MS/MS control spectra to yield 2,784 true positive proteins (n ≥ 3; q ≤ 0.01). Peptides identified by fragmenting ions with Gaussian heavy isotopes and hydrogen loss that were matched to known plasma proteins, such as albumin (ALB), were shown to be true positives and agreed with the peptide sequences identified in the monoisotopic peak. Accepting the ions from the monoisotopic peak alone (±0.1 Da) yielded only 382 plasma proteins (n ≥ 3; type I error q ≤ 0.01; type II error ∼86%). In contrast, accepting all ions within ±0.1 Da around the hydrogen loss, monoisotopic, and heavy isotopic peaks led to the identification of 963 proteins (n ≥ 3; q ≤ 0.01; type II error ∼60%). Using the power of the OIT to resolve the Gaussian peaks from heavy isotopes and hydrogen loss resulted in the identification of three times more proteins with high confidence and a much lower type II error than analyzing peptides from the monoisotopic peak alone. The resolving power of the OIT may be exploited to increase observation frequencies and provide greater proteomic coverage and statistical power in comparative proteomics studies.

Figure 1

Scatter plots and distributions of the calculated peptide mass or delta mass values plotted against the log peptide intensity or log peptide p-value from all proteins (left) or true positive albumin (ALB) alone (right). A,E Scatter plot of precursor mass (MH) versus log intensity for all peptides; B,F, Scatter plot of MH versus log p-value for all peptides; C,G, Scatter plot of delta mass (Da) versus log p-value from all peptides; D,H, Histogram of the log MH versus density of all peptides.

Figure 2

Density and scatter plots of peptide delta mass versus log₁₀ precursor intensity for peptides from all plasma proteins. A, Density plot of the delta mass values (Da) from 67,898 MS/MS fit to plasma peptides; B, Scatter plot of peptide delta mass values (Da) versus the log peptide intensity values for the −2 to +2 Da precursor delta mass range for MS/MS fit to plasma peptides; C, Scatter plot of peptide delta mass values (Da) versus the log peptide intensity values for the −0.5 to +0.5 Da precursor delta mass range for MS/MS fit to plasma peptides; D, Density plot of the ion-extracted rearrangement and heavy isotope integer delta mass values (Da) ± 0.1 Da; E, Scatter plot of the delta mass values (Da) of the ions extracted from hydrogen loss and heavy isotope integer values ±0.1 Da.

Figure 3

Estimating the type I error rate from peptide count from the fit of MS/MS spectra using the exact p-value, FDR corrected q-value, and Monte Carlo simulation with 30 million random MS/MS spectra. Panels: A, The peptide-to-protein observation frequencies (counts) of the OIT from the fit of all experimental MS/MS spectra corrected by subtracting the peptides observed in blank injection MS/MS spectra; B, Plasma proteins with FDR q-values ≤0.01 identified in all MS/MS spectra after subtraction of peptides from blank injection MS/MS spectra at the peptide level; C, Peptide-to-protein observation frequencies (counts) from the OIT and the fit of all experimental MS/MS spectra corrected by subtracting the peptides observed in blank injections and 30,000,000 random MS/MS spectra; D, Plasma proteins with FDR q-values ≤0.01 identified in all MS/MS spectra after subtraction of peptides from blank injections and 30,000,000 random MS/MS spectra and Monte Carlo simulations.

Figure 4

Analysis of plasma albumin (ALB; a known true positive) and its most common true-positive peptide, EQLKAVMDDFAAFVEK. A, Tris-glycine 12.5% SDS-PAGE of a human plasma sample in a serial dilution. Plasma was diluted 50 times in the SDS-PAGE sample buffer prior to loading the indicated volume for electrophoretic separation. The arrow indicates the electrophoretic mobility of ALB, which is the predominant band found in large excess over all other plasma proteins; B, Density plot of the delta mass values from peptides matched to ALB from the fit of MS/MS to plasma peptides by X!TANDEM; C, Scatter plot of peptide delta mass values versus log peptide intensity for the peptides matched to albumin by X!TANDEM; D, density plot of the ions extracted from the Gaussian peaks at heavy isotopes and hydrogen loss ±0.1 Da that matched peptides from ALB; E, Scatter plot of the delta mass values of the heavy isotopes and hydrogen loss ±0.1 Da from ALB after ion extraction; F, The peptide delta mass density plot of the ALB peptide EQLKAVMDDFAAFVEK (Table 3).

Figure 5

Tandem mass spectrometry analysis of the most common true-positive peptide, EQLKAVMDDFAAFVEK from plasma albumin. Similar MS/MS fragmentation spectra that match to the same peptide from albumin was observed from hydrogen loss and heavy isotopes peaks at delta mass values of −3, −2, −1, 0, +1, +2 Da. Panels: A, the −3 Da delta mass peptide MS/MS spectra; B, the −2 Da delta mass peptide MS/MS spectra; C, the −1 Da delta mass peptide MS/MS spectra; D, the 0 Da delta mass peptide MS/MS spectra; E, the +1 Da delta mass peptide MS/MS spectra; and F, the +2 Da delta mass peptide MS/MS spectra. The output of the X!TANDEM algorithm is shown without any alteration.

Figure 6

Agreement between the monoisotopic mass from X!TANDEM and MaxQuant versus all data including heavy isotopes, hydrogen loss, and other precursor mass values. Panels: A, regression of all corrected X!TANDEM observation counts (n) onto those of the monoisotopic precursor mass [Multiple R-squared: 0.8074, F-statistic: 5189 on 1 and 1238 DF, p-value: < 2.2e^–16]; B, regression of all MaxQuant observation counts (n) onto all corrected X!TANDEM monoisotopic precursor mass observation counts [Multiple R-squared: 0.5358, F-statistic: 184.7 on 1 and 160 DF, p-value: < 2.2e^–16]; C, regression of all MaxQuant observation counts (n) onto those of the X!TANDEM monoisotopic precursor mass [Residual standard error: 0.5248 on 191 degrees of freedom Multiple R-squared: 0.603, F-statistic: 290.1 on 1 and 191 DF, p-value: < 2.2e^–16].