Maximized quantitative phosphoproteomics allows high confidence dissection of the DNA damage signaling network

Abstract

The maintenance of genomic stability relies on DNA damage sensor kinases that detect DNA lesions and phosphorylate an extensive network of substrates. The Mec1/ATR kinase is one of the primary sensor kinases responsible for orchestrating DNA damage responses. Despite the importance of Mec1/ATR, the current network of its identified substrates remains incomplete due, in part, to limitations in mass spectrometry-based quantitative phosphoproteomics. Phosphoproteomics suffers from lack of redundancy and statistical power for generating high confidence datasets, since information about phosphopeptide identity, site-localization, and quantitation must often be gleaned from a single peptide-spectrum match (PSM). Here we carefully analyzed the isotope label swapping strategy for phosphoproteomics, using data consistency among reciprocal labeling experiments as a central filtering rule for maximizing phosphopeptide identification and quantitation. We demonstrate that the approach allows drastic reduction of false positive quantitations and identifications even from phosphopeptides with a low number of spectral matches. Application of this approach identifies new Mec1/ATR-dependent signaling events, expanding our understanding of the DNA damage signaling network. Overall, the proposed quantitative phosphoproteomic approach should be generally applicable for investigating kinase signaling networks with high confidence and depth.

Figure 1

Modeling outcomes of SILAC reciprocal labeling as a means to reduce technical error and/or variation (EV). (A) Workflow showing reciprocal labeling scheme with a forward experiment (Experiment A, left) and a reverse experiment (Experiment B, right) with anticipated outcomes and proposed causes of EVs (middle). (B) Anticipated distribution of false positives in a comparison of two identical samples if error and variation occurred randomly and independent of isotopic labeling. (C) Anticipated distribution of false positives in a comparison of two identical samples if error and variation were unidirectionally biased (i.e. similar ratio in both a forward and reverse experiment).

Figure 2

Reciprocal labeling in an isogenic yeast cell line reveals extensive error and variation that is unidirectionally biased. (A) Histograms for SILAC ratios of two independent phosphoproteome experiments comparing isogenic wild-type S. cerevisiae as depicted in Fig. 1A. EVs are colored in orange. (B) Scatterplot comparing experimental data from the two SILAC experiments shown in (A). EVs are colored in orange. (C) Histogram showing unidirectional bias of error and variation toward quadrants 2 and 4 in plot from (B).

Figure 3

Data filtering based on quantitation consistency drastically reduces error and variation. (A) Scatterplot from Fig. 2B. indicating the “Quadrant” filtering scheme (gray data points removed in Q2 and Q4) and additional filtering based on the requirement for at least 2 PSMs per experiment for each data point (plot on the right). (B) Histogram showing EVs in Q1 and Q3 (orange data points) from A as a percentage of total dataset using either 1 PSM or 2 PSM filtering. (C) Scatterplot from Fig. 2B. indicating the “Bow-tie” filtering scheme (gray data points removed) and additional filtering based on the requirement for at least 2 PSMs per experiment for each data point (plot on the right). For Bow-tie filtering, in addition to removing EVs in Q2 and Q4, data points in Q1 and Q3 were required to be within an interval of correlation correspondent to fourfold of the log2 scale. (D) Histogram showing highlighted points in quadrants 1 and 3 from (C) as a percentage of total dataset using either 1 PSM or 2 PSM filtering.

Figure 4

“Bow-Tie” approach efficiently reduces decoy peptide identifications. (A) Scatterplot from Fig. 2B. indicating hits from a decoy database. Decoy peptides are displayed in red. As in Fig. 3, unfiltered EVs in Q1 and Q3 are displayed in orange. (B) Histogram displaying number of decoy peptide hits in Q1 and Q3 with quadrant filtering applied. (C) Scatterplot from Fig. 2B. indicating hits from a decoy database and employment of Bow-tie filtering. Decoy peptides are displayed in red. As in Fig. 3, unfiltered EVs in Q1 and Q3 after Bow-tie filtering are displayed in orange. (D) Histogram displaying number of decoy peptide hits in Q1 and Q3 with Bow-tie filtering applied.

Figure 5

Quantitative phosphoproteomic analysis of Mec1-dependent signaling. (A) Scatterplot (with Bow-tie filter applied and PTMProphet score ≥ 0.9; only data points within Bow-tie filter displayed) of forward and reciprocal SILAC experiment comparing phosphoproteome of rad9Δ cells to phosphoproteome of rad9Δ mec1Δ cells. Cells were treated with 0.02% MMS for 2hrs. (B) Histogram depicting distribution of phosphorylation sites in Q1 and Q3 compared to control experiments. (C) Estimation of false discovery rate (FDR) in quantitative analysis for experiment in 5A. FDR for quadrants 1 and 3 is estimated based on error and variation in wild-type control experiment (Fig. 2). See “Materials and methods” for more details.

Figure 6

Expanding the Mec1 signaling network. (A) Scatterplot (with Bow-tie filter applied and PTMProphet score ≥ 0.9; only data points within Bow-tie filter displayed) of forward and reciprocal SILAC experiment comparing phosphoproteome of rad9Δ cells to phosphoproteome of rad9Δ mec1Δ cells. S/T-Q consensus motif is highlighted in green. Cells were treated with 0.02% MMS for 2 h. (B) Histogram of proportion of S/T-Q phospho-motif by quadrant in the Mec1 experiment from (A). (C) Pie chart showing proportion of nuclear proteins (GO Cellular Location^,) in Mec1-dependent (log2 ratio > 1.0) S/T-Q sites from the Mec1 experiment in (A). (D) Uniprot keyword enrichment analysis performed on proteins containing Mec1-dependent S/T-Q phosphorylation from (A). (E) String analysis of proteins with Mec1-dependent phosphorylation in the S/T-Q consensus revealed a sub-network of proteins involved in DNA repair via homologous recombination (HR). Image adapted from https://string-db.org/. (F) String analysis of proteins with Mec1-dependent phosphorylation in the S/T-Q consensus revealed a sub-network of proteins related to the nucleolus. Image adapted from https://string-db.org/. (G) Simplified model depicting major processes directly regulated (green arrows) by Mec1. Results suggest that the Yak1 represents a novel kinase under direct control by Mec1.