Improved Method for Determining Absolute Phosphorylation Stoichiometry Using Bayesian Statistics and Isobaric Labeling

Abstract

Phosphorylation stoichiometry, or occupancy, is one element of phosphoproteomics that can add useful biological context (Gerber et al. Proc. Natl. Acad. Sci. U. S. A. 2003, 100, 6940-5). We previously developed a method to assess phosphorylation stoichiometry on a proteome-wide scale (Wu et al. Nat. Methods 2011, 8, 677-83). The stoichiometry calculation relies on identifying and measuring the levels of each nonphosphorylated counterpart peptide with and without phosphatase treatment. The method, however, is problematic in that low stoichiometry phosphopeptides can return negative stoichiometry values if measurement error is larger than the percent stoichiometry. Here, we have improved the stoichiometry method through the use of isobaric labeling with 10-plex TMT reagents. In this way, five phosphatase treated and five untreated samples are compared simultaneously so that each stoichiometry is represented by five ratio measurements with no missing values. We applied the method to determine basal stoichiometries of HCT116 cells growing in culture. With this method, we analyzed five biological replicates simultaneously with no need for phosphopeptide enrichment. Additionally, we developed a Bayesian model to estimate phosphorylation stoichiometry as a parameter confined to an interval between 0 and 1 implemented as an R/Stan script. Consequently, both point and interval estimates are consistent with the plausible range of values for stoichiometry. Finally, we report absolute stoichiometry measurements with credible intervals for 6772 phosphopeptides containing at least a single phosphorylation site.

Keywords: Bayesian modeling; SPS-MS3; TMT; error intervals; global proteome; human cell lines; mass spectrometry; phosphatase; phosphorylation; stoichiometry.

Figure 1.

(A) Workflow for phosphorylation stoichiometry experiment. Briefly, reduced and alkylated cell lysate from five biological replicates of HCT-116 cells were separately digested with trypsin, and each sample was split into two aliquots for TMT-10 labeling. One labeled aliquot from each sample was subjected to phosphatase treatment while its sister aliquot underwent a mock treatment. All 10 aliquots were combined for Sep-Pak cleanup and subjected to reversed phase HPLC and then analyzed by SPS-MS3 on a Thermo Orbitrap Fusion Lumos. Stoichiometries were then calculated for each peptide and assigned to phosphopeptides from a previous independent phosphopeptide identification experiment. (B) Sample calculation of how stoichiometry is calculated for an observed peptide from our experiment. The stoichiometry for each sample is calculated. In the example shown, the stoichiometry is calculated for the red sample. An equivalent formula is to use the ratio of treated to untreated to calculate the stoichiometry: $1-\frac{1}{T:U}$.

Figure 2.

(A) Workflow for independent phosphopeptide identification experiment. Fe-IMAC enrichment was performed on the digested cell lysate from HCT-116 cells. The phosphopeptide enriched digest was then TMT-labeled to account for chemical changes caused by TMT-labeling and subjected to fractionation by reverse-phase HPLC. Fractions were analyzed by high resolution MS2 analysis. Resulting phosphopeptide identifications were localized to sites using a modified A-score to generate the known phosphorylation sites library used in Figure 1A. (B) Summary of phosphopeptides identified during this experiment. Pie chart breaks down the phosphopeptides by type: acidic, basic, proline-directed, and other. Sites were assigned a type based on a previously described algorithm.

Figure 3.

Summary of phosphorylation stoichiometry experiment results. A total of 124 419 peptides corresponding to 8351 proteins were identified. A total of 6772 unique peptides (2556 proteins) were matched to phosphorylation sites identified in our phosphopeptide enrichment experiment.

Figure 4.

(A) Example TMT-data for peptides known to harbor phosphorylation sites. Stoichiometries were calculated for each sample (red, yellow, green, blue, and purple), the average and standard deviation are reported. Solid colors represent channels where the aliquot was treated with phosphatase while the striped colors represent channels where the aliquot was mock treated. (B) Table displaying the individual sample phosphorylation stoichiometries calculated for each peptide in panel A). Red characters represent the expected phosphorylation site. All sites chosen were identified as regulatory phosphorylation events through targeted studies based on the phosphositeplus.org database.

Figure 5.

Histograms of the phosphorylation stoichiometry for each estimation method. (A) Histogram when no correction is performed. (B) Histogram where each negative stoichiometry is replaced with 0. (C) Histogram when stoichiometry is estimated using the Bayesian modeling approach. The red dashed line represents 0%.

Figure 6.

Peptides were rank ordered (lower values first) by their estimated stoichiometry. 80% confidence intervals (red bars) and 95% confidence intervals (black bars) were drawn around each point. The y-axis represents the phosphorylation stoichiometry as a fraction instead of a percent. Resulting caterpillar plots are shown for each method. (A) Standard method with no corrections performed. (B) All negative stoichiometry calculations were replaced with 0. (C) Stoichiometry values were estimated using our Bayesian model.