Global and site-specific effect of phosphorylation on protein turnover

Summary

To date, the effects of specific modification types and sites on protein lifetime have not been systematically illustrated. Here, we describe a proteomic method, DeltaSILAC, to quantitatively assess the impact of site-specific phosphorylation on the turnover of thousands of proteins in live cells. Based on the accurate and reproducible mass spectrometry, a pulse labeling approach using stable isotope-labeled amino acids in cells (pSILAC), phosphoproteomics, and a unique peptide-level matching strategy, our DeltaSILAC profiling revealed a global, unexpected delaying effect of many phosphosites on protein turnover. We further found that phosphorylated sites accelerating protein turnover are functionally selected for cell fitness, enriched in Cyclin-dependent kinase substrates, and evolutionarily conserved; whereas the Glutamic acids surrounding phosphosites significantly delay protein turnover. Our method represents a generalizable approach and provides a rich resource for prioritizing the effects of phosphorylation sites on protein expression lifetime in the context of cell signaling and disease biology.

Graphical Abstract

eTOC Blurb

By developing and applying a large-scale mass spectrometry-based method, Wu and Ba et al. quantitatively assessed the stabilizing and destabilizing effects of phosphorylation on protein expression. The findings revealed that phosphorylation impacts protein degradation in a site-specific manner, which is associated with surrounding amino acids, local structure, and phosphoprotein functions.

Introduction

Biological signaling networks receive and transduce signals based on both amplitude and duration (Kholodenko, 2006). As the primary functional biomolecule involved in most cellular processes, proteins have been characterized by various properties, such as structure, abundance, localization, stability, and turnover. Protein post-translational modifications (PTMs) can alter the above properties, leading to diverse functions (Smith and Kelleher, 2018; Smith et al., 2013). Phosphorylation is a critical PTM, which has been demonstrated to be essential for signaling transduction (Olsen et al., 2006), mediating protein-protein interaction (Betts et al., 2017), and altering the three-dimensional protein structure, thermal stability (Huang et al., 2019), and subcellular localization (Krahmer et al., 2018) – frequently in a modification site-specific manner. However, the impact of phosphorylation sites on protein turnover has so far not been assessed at the proteome-scale.

In order to adapt to temporal environmental changes, cells have to utilize kinases and phosphatases to effectively and instantly catalyze phosphate-transfer between substrates. Nevertheless, the long-term regulation of turnover and decay of those constitutively phosphorylated proteins is also critical for the cellular systems. We aim to focus on this important regulatory mechanism in the present study. Such a “phosphate-transfer independent” mechanism was shown to enable the rewiring of cellular biochemical states after adaptation (Hunter, 2007; Nguyen et al., 2013) and the establishment of cell fitness against intrinsic genetic alterations (Lahiry et al., 2010; Ochoa et al., 2020). Specifically, the functional crosstalk between phosphorylation and ubiquitination was discovered to substantially downregulate the levels of key phosphoproteins in a variety of pathways such as EGFR/MAPK signaling (Nguyen et al., 2013; Suizu et al., 2009) and cell-cycle control (Skowyra et al., 1997; Verma et al., 1997). However, high-throughput discovery tools are currently lacking to illustrate how phosphorylated proteins are degraded by proteostasis and proteolysis pathways.

In the last decade, mass spectrometry (MS) based proteomics has greatly facilitated the analysis of proteins and their PTMs (Aebersold and Mann, 2016; Mann and Jensen, 2003). MS-based mapping of phosphorylation and ubiquitylation sites previously suggested that distinct phosphorylation sites often co-occur with ubiquitylation (Swaney et al., 2013). This analysis, however, relied on the MS detection frequency of PTM sites after an individual or tandem PTM enrichment (Swaney et al., 2013), which does not consider the relative stoichiometry of both PTMs. Moreover, such qualitative results are likely affected by the abundance of bulk proteins and the sensitivity of MS analyzers. As an arising MS-based technique, Data-independent acquisition mass spectrometry (DIA-MS) (Aebersold and Mann, 2016; Gillet et al., 2012; Venable et al., 2004) usually generates continuous, high-resolution MS2 peak profiles along with the liquid chromatography (LC) separation which can be used for the simultaneous identification and quantification of thousands of peptides, including their PTMs (Bekker-Jensen et al., 2020; Meyer et al., 2017; Rosenberger et al., 2017a). Because of its high quantification accuracy and consistency between samples, DIA-MS facilitates pulse experiments of stable isotope-labeled amino acids in cells (pulse SILAC, or pSILAC), usually involving multiple labeling time points (Jovanovic et al., 2015; Liu et al., 2016; Ong et al., 2002; Schwanhausser et al., 2011). Promisingly, with an optimized computational framework (Salovska et al., 2020), we have demonstrated that pSILAC-DIA enables the measurement of the turnover rates of proteins and their alternative splicing isoforms with high efficiency, reproducibility, and accuracy, and compares favorably to alternative approaches (Liu et al., 2017; Liu et al., 2019; Salovska et al., 2020).

Based on the above rationale, in this work, we developed a proteomic method, namely DeltaSILAC (delta determination of turnover rate for modified proteins by SILAC), that directly interrogates the impact of phosphosites on protein lifetime in comparison to the respective unphosphorylated forms. Using the datasets generated from growing cancer cells, we present a large-scale, unbiased assessment of human phosphoproteome turnover regulation.

Results

Establishing DeltaSILAC to quantify phosphoproteome turnover.

We first describe the rationale of our DeltaSILAC method (Figure 1). In a typical pSILAC experiment, the growing cells are cultured in the heavy SILAC medium during a predetermined time course. The cells are maintained at steady-state, where the degraded and synthesized protein copies are balanced (Claydon and Beynon, 2012; Welle et al., 2016). Such a concentration balance is assumed to apply to all the proteins (specific circadian regulations may be the exceptions). Accordingly, the protein-specific turnover rate can be determined by MS measurement, e.g., through modeling the rate of loss of the light isotope (k_Loss) that follows the an exponential decay (Pratt et al., 2002) (See Figure 1 and Methods). Herein, we suggest that global modified proteoforms can also conceivably be expected to form a similar concentration balance in steady-state cells – otherwise the cellular system would be perturbed. This is analogous to e.g. western blot (WB) experiments measuring phosphorylated proteins in unperturbed growing cells, in which most WB results should remain the same upon different harvest time points. Indeed, we compared the abundances of both proteome and phosphoproteome and found them to be globally stable over the pSILAC labeling period in our experiment (see below, Figure S1A–B), Thus, we used the pSILAC strategy to derive protein turnover rates and lifetime for both modified or unmodified forms.

Figure 1. Development of DeltaSILAC to quantify protein turnover with site-specific phosphorylation.

(A) Experimental workflow illustrating the DeltaSILAC method. The experimental method is comprised of pSILAC labeling, phosphoproteomics, and DIA mass spectrometry.

(B) Data analysis strategy illustrating the peptide sequence-level matching used in DeltaSILAC.

(C-D) Venn diagram of overlap between unmodified and phosphorylated proteins and peptides at expression (C) and turnover (D) levels.

(E) Circos plot of relative HeLa_7 (CCL2)/ HeLa_8 (Kyoto) ratios in six layers, including expression and turnover, as depicted. Fold changes from high to low are shown in red to green. The data are phosphoproteome centric, i.e., data matched to available phosphoproteomic identifications.

(F) Spearman correlation analysis between six layers using HeLa_7/ HeLa_8 ratio data. Spearman’s rho is shown, with positive correlations visualized in blue color and negative correlations in red color.

(G) The scatterplot indicating HeLa_7/ HeLa_8 fold-change ratios of k_Loss estimates between matched unmodified and phosphopeptides.

To demonstrate the DeltaSILAC workflow using a well-defined model system, we investigated a panel of 12 HeLa cell strains initially collected from different research laboratories (Liu et al., 2019). A considerable heterogeneity of gene copy number alteration (CNA) was previously documented between these HeLa cells (especially between HeLa CCL2 and Kyoto strains), which leads to a systematic rewiring of mRNA, protein, and bulk-protein degradation levels (Liu et al., 2019). This prior dataset provided the basis to assess the roots of phosphoprotein abundance variability. In the present study, using DIA-MS we confidently quantified 24,119 ± 446 Class-I phosphopeptides (i.e., those with a confident phosphosite localization) (Bekker-Jensen et al., 2020; Olsen et al., 2006) across 12 HeLa cell lines (Figure S2 & Table S1). We found that phosphoprotein abundance is globally more variable than the corresponding mRNA and protein levels between cell strains (Figure S2D–F). While we acquired phosphoproteomic profiles for all HeLa strains covered by the original study (Liu et al., 2019) (Figure S2), we herein chose HeLa_7 and HeLa_8 as representative HeLa CCL2 and Kyoto strains for phosphoproteome turnover measurement.

To systematically determine the effect of phosphosites on protein turnover, we deployed a pSILAC experiment at 0, 1, 4, 8, and 12 hours for both HeLa_7 and HeLa_8 cells (Figure 1). The labeling interval is similar to previous studies (Salovska et al., 2020; Schwanhausser et al., 2011). The cell lysates of different time points were then trypsin proteolyzed. Each peptide aliquot was split, with 5% of peptides used for direct proteomic measurement and 95% for phosphoproteomics. An established DIA-MS method (Mehnert et al., 2019) was employed to measure both proteomes and phosphoproteomes during labeling (Figure 1A–B). We confidently identified 7,583 and 7,550 proteins for HeLa_7 and HeLa_8 from single-shot measurements, assigned by 87,266 and 86,881 peptides (both peptide- and protein- FDR were < 1%). For 5,456 and 5,608 proteins (detected by 47,831 and 50,978 peptides), we computed turnover rates (k_Loss), which can be mathematically transformed to an estimation of the protein half-life time T_1/2 (i.e., T_50% values (Zecha et al., 2018) in this report, see Methods and Figure 1C–D & Figure S1C). Moreover, among ~24,000 phosphopeptides detected in HeLa_7 and HeLa_8, 13,781 and 15,777 phosphopeptides (that is, 12,134 and 13,078 phosphosites) were quantified with a T_1/2. We found that T_1/2 correlation between HeLa_7 and HeLa_8 for phosphopeptides (R=0.90) was similar to the T_1/2 correlation for proteins (R=0.86, Figure S1D–G), suggesting that the quantitative performance of pSILAC-DIA applied to total proteome analyses (Liu et al., 2017; Liu et al., 2019; Salovska et al., 2020) can be extended to phosphoproteomics. Altogether, this dataset (Table S2) presents a pioneer systematic analysis of phosphoproteome turnover.

We then benchmarked the phosphopeptide dynamics by the comparison to other omics layers. The genome-wide heterogeneity between HeLa_7 and HeLa_8 reported previously (Liu et al., 2019) was assessed at the levels of gene copies, mRNA, bulk-protein expressions, phosphopeptide abundances, and turnover rates of all peptides and phosphopeptides, as well as the correlations between these levels, revealing the gene dosage impact across layers (Figure 1E). We found that the change of phosphopeptide abundance follows mRNA and protein levels (R=0.40 and 0.46 respectively), while the turnover regulation of phosphopeptide does not (Figure 1F). Additionally, the correlation between turnover ratios for phosphopeptides and unmodified peptide counterparts are moderate (R=0.39; Figure 1G), similar to the correlation between abundance ratios (R=0.46). This result, therefore, highlights the vital need to analyze phosphopeptides for both abundance and turnover independently. Although the gene dosage compensation mechanism at the protein turnover level was confirmed by a positive mRNA~peptide k_Loss correlation (R=0.12), we found the mRNA~phosphopeptide k_Loss correlation to be weaker (R=0.06, p>0.05, after relevance-correction against peptide k_Loss). This implicates that broad CNA may have a limited impact on phosphoproteome turnover overall. Therefore, we analyzed the datasets of HeLa_7 and HeLa_8 independently (i.e., not relying on the phosphoproteomic turnover difference between the two cells) in the following analysis.

To summarize, with DeltaSILAC, we measured lifetime (in terms of T_1/2) for both modified and unmodified versions of the same peptide sequences from the same protein (Figure 1D).

Site-specific impact on protein turnover by phosphorylation.

To better understand the effects of site-specific protein phosphorylation on protein turnover, we herein used the concept of “phosphomodiform” introduced by Huang et al. for measuring protein thermal stability changes induced by differential phophosites (Huang et al., 2019). In essence, due to the existence of numerous proteoforms in live cells, “phosphomodiform” characterization investigates the site-specific functions of phosphorylation by bottom-up proteomics, e.g., phosphosite abundance profiling employed by most previous phosphoproteomics studies, the thermo stability analysis in Huang et al (Huang et al., 2019), or turnover measurement employed in this report.

To pinpoint phosphorylation influence on phosphomodiform lifetime, we applied a paired strategy – we subtracted the T_1/2 of the backbone sequence-matching non-phosphorylated peptides (np-peptide) from the T_1/2 of the respective phosphopeptides (p-peptide), resulting in a delta value (ΔT_1/2). After filtering (see Methods), 1,919 and 2,147 such pairs of np- and p- peptides in HeLa_7 and HeLa_8 were quantified. We found that for those proteins identified with multiple phosphorylation sites, different sites could lead to a variable ΔT_1/2 (Figure 2). For example, the pT490 increased T_1/2 of AHNAK by ~7.5 hours (i.e., ΔT_1/2 = 7.71 and 7.58 hours in HeLa_7 and HeLa_8) whereas pT4100 shortened T_1/2 by > 15 hours (i.e., ΔT_1/2 = −15.61 and −16.86 hours; Figure 2A). For MAP4 certain phosphosites such as pS507 and pT521 could greatly enlarge T_1/2 by > 40 hours, having a much higher turnover modulation effect than other phosphomodiforms from the same protein (Figure 2B). In comparison, for MARCKS, phosphorylation on most sites detected on average similarly increased protein half-life of about 5 hours (Figure 2C). Interestingly, all the 19 phosphomodiforms of SF3B1 measured in our results barely changed their lifetime as compared to the non-phosphorylated counterparts, indicating that turnover can be robust despite variable modifications for this protein (ΔT_1/2 =0.72 ± 1.96 hours, Figure 2D–E). Altogether, different phosphosites on different proteins may lead to a wide-range of effects on their lifetime.

Figure 2. Regulation of protein turnover by phosphorylation in DeltaSILAC data.

(A-D) The averaged ΔT_1/2 values (N=3 biological replicates) determined by an independent analysis of HeLa_7 and HeLa_8 cells for protein examples of multiple phosphorylation sites. These examples include (A) neuroblast differentiation-associated protein AHNAK (AHNAK), (B) Microtubule-associated protein 4 (MAP4), (C) Myristoylated alanine-rich C-kinase substrate (MARCKS), and (D) Splicing factor 3B subunit 1 (SF3B1).

(E) Demonstration of localization of nine phosphorylated sites of SF3B1 in its known structure.

(F) The Heavy-to-Light (H/L) ratios determined for S79 and S86 phosphorylated form (p-peptide) and their shared un-phosphorylated form (np-peptide, LASVPAGGAVAVSAAPGS₇₉AAPAAGS₈₆APAAAEEK) in 60S acidic ribosomal protein P2 (RPLP2). Each displayed datapoint denotes the H/L ratios obtained at a different time point along the pSILAC time course in Figure 1. Data are represented as mean ± SD.

(G) The DIA MS/MS peak groups during LC elution containing unique fragment ions for the peptide carrying phosphorylated S79 (pS79). Note that the heavy (above middle zero lines) to light peaks (below zero line) are scaled, respectively.

(H) The DIA MS/MS peak groups during LC elution containing unique fragment ions for the peptide carrying phosphorylated S86 (pS86).

(I) The MS1 peak groups from the same DIA run during the same elution region (corresponding to a peptide containing single phosphorylation).

(J-M) Individual examples of H/L ratios determined for paired p-peptide and np-peptide for (J) Protein FAM207A, (K) Glycylpeptide N-tetradecanoyltransferase 1 (NMT1), (L) Prelamin-A/C (LMNA), and (M) Peroxiredoxin-6 (PRDX6). Each displayed datapoint denotes a different time point along the pSILAC time course. Data are represented as mean ± SD.

(N) Data variability of ΔT_1/2 values summarized as standard deviation (S.D.) for different phosphosites within the same protein. All proteins were classified by the number of phosphosites measured by DeltaSILAC. The red numbers denote the median values in each group.

Furthermore, we found that DeltaSILAC was able to impressively discriminate the turnover discrepancy between phosphomodiforms carrying closely located phosphosites, even those identified on the same peptide backbone. The sites S79 and S86 in RPLP2 present such an example, as they share the same np-peptide (Figure 2F). The unique MS2 signatures carrying either pS79 or pS86 were extracted from the DIA-MS dataset. Moreover, the heavy-to-light (H/L) ratios of the two sites during pSILAC labeling were confidently resolved to respective LC peak groups, which have only a < 1 min retention time (RT) interval (Figure 2G–H), whereas a T_1/2 difference of 9.66 hours between pS79 and pS86 (+12 vs. +2.34 hours) was estimated. Because the two phosphopeptides have the identical MS1 m/z (Figure 2I) and the small RT difference between them (which can sometimes be even smaller for other phosphopeptide positional isomers), they would present a challenge for MS1-based alignment across multiple pSILAC samples (Cox and Mann, 2008; Lim et al., 2019; Searle et al., 2019).

To gauge the precision of ΔT_1/2, we checked the H/L ratios of np-peptides and p-peptides during pSILAC labeling. We found our sequence-matching strategy necessary and useful because even np-peptides can show variable labeling rates, such as npT34 and npS203 of FAM207A in Figure 2J. The extended protein lifetime by phosphosites was credibly inferred from H/L ratios measured by DIA-MS during the pSILAC time course, such as pS47 of NMT1 and pS212 & pS107 of LMNA (Figure 2K–L). Other sites instead increased the phosphomodiform turnover, such as pT177 in PRDX6 (Figure 2M). The determined np-peptide and p-peptide T_1/2 were highly stable across replicates, validating the reproducibility of our experimental and computational pipeline in DeltaSILAC and indicating that even small changes in protein half-life (for example, ~1 hour) are detectable with statistical cutoffs (Figure S3A–B). Therefore, DeltaSILAC reliably uncovered a 4-6 hours variability of half-life time, on average, between differential phosphomodiforms of the same protein (Figure 2N).

Altogether, our analysis demonstrates that DeltaSILAC can accurately assess the effects of phosphorylation on protein turnover in a site-specific manner.

Phosphorylation delays turnover for many proteins in growing cells.

To understand the global effect of phosphorylation on protein turnover, we analyzed T_1/2 in three comparative scenarios. 1. Total comparison: T_1/2 of all phosphopeptides identified versus T_1/2 of all bulk proteins identified (Figure 3A–B, left panel). This “unmatched” comparison was conducted regardless of the mapping of np- or p- peptides. 2. Protein-matched comparison: T_1/2 of phosphopeptides whose corresponding proteins were identified versus T_1/2 of the corresponding proteins at the bulk-level (Figure 3A–B, middle panel). This comparison thus excluded ~10,000 phosphopeptides and ~3,500 proteins that were only detected by either phosphoproteomics or proteomics (Figure 1D). 3. Peptide-matched comparison: T_1/2 of p-peptides versus T_1/2 of np-peptides (Figure 3A–B, right panel). This comparison only accepts T_1/2 data with a pair of np- and p- peptides. Comparison 1 suggests that the total phosphoproteome and the proteome have comparable T_1/2 (median is 16.8 vs. 16.3 hours for HeLa_7 and 15.8 vs. 16.1 hours for HeLa_8). The slightly reduced T_1/2 in HeLa_8 from HeLa_7 is expected because HeLa Kyoto has a shorter doubling time than HeLa CCL2, as reported (Liu et al., 2019). Intriguingly and significantly, a trend of decreasing protein turnover conferred by phosphorylation was demonstrated by both protein- and peptide- matched strategies: the T_1/2 median increase by phosphorylation was determined to be 0.70 and 1.10 hours for HeLa_7 and HeLa_8 respectively according to the protein-matched comparison (Comparison 2), and 2.53 and 2.55 hours according to peptide-matched comparison (Comparison 3) (Figure 3A–B). The volcano plots suggest that, based on the peptide-matched comparison, the T_1/2 is significantly upregulated for 623 and 679 phosphopeptides but only downregulated for 148 and 156 in both cell lines (P < 0.01, T_1/2 change > 1.5 hours; Figure 3C–D). The different outcome from Comparison 1 might be explained by the phosphoproteomic enrichment step. This step covered more low-abundant proteins, which in turn have a shorter T_1/2 than high-abundant proteins detected by total proteomics (Claydon and Beynon, 2012; Liu et al., 2017; Liu et al., 2019) (Figure S3C). The above results further emphasize the importance of using sequence-matched controls for analyzing the turnover of phosphomodiforms.

Figure 3. Phosphorylation increases lifetime for the many proteins in growing HeLa_7 and HeLa_8 cells.

(A-B) Distribution of T_1/2 values in total proteome versus phosphopeptides, matched proteins versus phosphopeptides, and matched peptides versus phosphopeptides in HeLa_7 (A) and HeLa_8 (B) cells. P values were calculated by Wilcoxon sum test. The borders of the box represent the 25th and 75th percentile, the bar within the box represents the median, and whiskers represent the range.

(C-D) Volcano plots for ΔT_1/2 of 3 vs. 3 biological replicates in HeLa_7 (C) and HeLa_8 (D) cells. P values were calculated by Student’s T-tests. The color dots denote the significantly altered gene expressions (P < 0.01; ΔT_1/2 > 1.5 hours).

To verify the stabilizing effect of specific phosphorylation on the respective proteins, we transfected HeLa CCL2 cells with tagged wild-type (WT), “phosphomimetic” (S to D), and “phosphodead” (S to A) versions for three phosphosites: S191 at Catenin beta-1 (CTNNB1), S110 at Synaptosomal-associated protein 23 (SNAP23), and S118 at Transcriptional repressor protein YY1, all of which were determined by DeltaSILAC to prolong the protein lifetime when phosphorylated. After transfection, cells were cultured for 24 hours to allow the expression of the transgenes and subsequently chased in the presence of cycloheximide for measuring different protein degradation rates. Using western blots (Figure 4A–D), we confirmed that the phosphomimetic CTNNB1-S191D and SNAP23-S110D remarkably stabilized the proteins when compared to the exogenous expression of wild-types. In the meanwhile, the phosphodead CTNNB1-S191A clearly destabilized the protein, whereas SNAP23-S110A rendered a similar but modest destabilizing effect (Figure 4B and 4D). Notably, this result is consistent with the quantitative DeltaSILAC readouts that upon phosphorylation CTNNB1-S191 induced a larger ΔT_1/2 than SNAP23-S110 (+9.85 hours vs. +2.81 hours) for respective proteins (Table S2). In addition, using an orthogonal imaging approach for determining the stability of YY1 with a similar strategy to what previously described (Mandad et al., 2018), we generated three synthetic genes based on YY1 (WT, S118A and S118D) and inserted them in a vector that allowed the co-expression with a marker for normalization (mNeonGreen, Figure 4E–F). Again, YY1-S118D was significantly stabilized compared to the wild-type protein, where the phosphodead mutant YY1-S118A facilitated protein turnover (Figure 4F).

Figure 4. Phosphosites of CTNNB1-S191, SNAP23-S110, and YY1-S118 stabilizing the corresponding protein expressions.

CTNNB1, Catenin beta-1; SNAP23, Synaptosomal-associated protein 23; YY1, Transcriptional repressor protein YY1.

(A-D) Western blotting verifying the protein stabilization effects of phosphosites CTNNB1-S191 and SNAP23-S110. HeLa CCL2 cells were transfected with Flag-tagged wild-type (WT), “phosphomimetic” (S to D) and “phosphodead” (S to A) versions of CTNNB1-S191 and SNAP23-S110 respectively and treated with Cycloheximide for up to 8 hours. Western blotting imagines (A and C) and the resultant quantitative visualization using Image J (B and D) were shown.

(E) Imaging approach for measuring protein stabilization induced by YY1-S118. The turnover sensors (1) are composed of the wild-type sequence of YY1 fused to the ALFA-tag epitope for quantitative imaging (Mandad et al., 2018). A P2A self-cleaving peptide followed by the green fluorescent protein mNeonGreen was included for normalization. The representative microscope images were shown for cells after 2 hours Cycloheximide treatment. Scale bar: 10 μm.

(F) Quantitative results of high-content imaging-based verification on YY1-S118 (t_1/2 WT: 1.0 [95% CI: 0.9–1.3]; t_1/2 S118A: 0.7 [95% CI: 0.7–0.8]; t_1/2 S118D: 3.3 [95% CI: 2.9–3.8]). Each dot represents the average of three separate experiments with SEM (n = 3).

Taken together, the matching comparisons and individual verification experiments somewhat surprisingly elaborate that many proteins tend to increase rather than decrease T_1/2 when phosphorylated. The peptide-matching strategy (Comparison 3), generating more significant ΔT_1/2 values, is thus preferably used in the following analyses, whereas the protein-matching strategy (Comparison 2) was also occasionally adopted because it generates 2.5 times more valid ΔT_1/2 values for phosphosites (6,834 for HeLa_7 and 7,654 for HeLa_8).

Structural features associated with protein turnover altered by site-specific phosphorylation.

The effects of phosphosites on protein turnover revealed by DeltaSILAC prompted us to investigate whether certain local structural environments and other properties of particular phosphosites are associated with ΔT_1/2. By comparing site-specific T_1/2 to a published dataset reporting phosphomodiform melting temperature (T_m) (Huang et al., 2019), we found a small but significant correlation of R=0.20 (P<2.2E-16) across all phosphosites. This relationship holds positive even after stringent correction using bulk-protein abundance which positively correlates with both T_1/2 and T_m (Figure S3C–F). This result indicates that phosphorylation sites bringing more thermal stability to proteins could also increase protein lifetime (Figure 5A) and might be suggestive of a coupling mechanism between structural stability and expression stability for achieving cellular proteostasis. Furthermore, using the DIA intensity ratio of p-peptide vs. bulk protein (Phos-/ Prot) and p-peptide vs. np-peptide (Phos-/ Pept) as the proxy of phosphorylation occupancy, we found a significant albeit weak association between the high phosphorylation occupancy and the long cellular lifetime for different phosphomodiforms (Figure S3G–J), which aligns well with the results in Figure 3.

Figure 5. Global relationships between local structural features and phosphorylation-altered protein turnover.

(A) The melting temperature (T_m, °C) of matched site-specific phosphomodiforms is distributed into T_1/2 quintiles from small to large in HeLa_7 and HeLa_8 cells (***P < 0.001, *P < 0.05, Wilcoxon sum test).

(B) The GRAVY score of phosphopeptides (based on their naked sequences) is distributed into ΔT_1/2 quintiles from small to large percentage in HeLa_7 and HeLa_8 cells (**P < 0.01, *P < 0.05, Wilcoxon sum test).

(C) Comparisons between the most detected protein domains and ΔT_1/2 values (to bulk proteins) for measured phosphosites in HeLa_7 cells.

(D) Distribution of ΔT_1/2 values for phosphopeptides with indicated phospho-amino acids and combinations thereof in HeLa_7 cells (***P < 0.001, *P < 0.01, Wilcoxon sum test).

(E-G) Comparisons of ΔT_1/2 values and predict transmembrane topologies (E), solvent accessibility (F), and predicted secondary structure elements (G) in HeLa_7 cells (*P < 0.05, Wilcoxon sum test).

All corresponding results in HeLa_8 can be found in Figure S4.

Next, to facilitate the relative analysis of ΔT_1/2, we divided ΔT_1/2 into quintile segments, with Q1 (0-20%) representing phosphosites reducing lifetime (i.e., faster turnover), Q2-Q4 (20-80%) representing intermediately regulated cases, and Q5 (80-100%) representing those dramatically slowed down turnover with phosphorylation (Figure S3K–N). First, we found that the grand average of hydropathy (GRAVY) score of all np-peptide sequences negatively correlates to ΔT_1/2, suggesting the higher peptide-hydrophobicity tends to stabilize the existence of phosphomodiform (P= 0.0029 and 0.0014 between Q1 and Q5 in HeLa_7 and HeLa_8, Wilcoxon sum test; Figure 5B). Second, different protein domains did not show a general effect on ΔT_1/2 (Figure 5C & Figure S4A–D). One of the exceptions is the increased lifetime for phosphorylated MARCKS (P= 0.006 and 0.034), an intrinsically disordered, alanine-rich protein whose phosphorylation translocates the protein from plasma membrane to cytoplasm through conformational changes (Bubb et al., 1999). Another exception is the shortening lifetime of phosphorylated KI67/Chmadrin repeat (P= 4.3E-09 and 1.5E-04) with presumable functions in the cell cycle process (Schluter et al., 1993). Third, global ΔT_1/2 values among the phosphorylated serine (S), threonine (T), and tyrosine (Y) sites revealed longer lifetime for pS than pT containing peptides (median is 2.43 vs. 1.48 hours of ΔT_1/2, P=0.0011, for HeLa_7). On average, pY only yields the a ΔT_1/2 of 0.07 hours in our results. Notably, the combination of two phosphorylated amino acids seemingly further increased protein the lifetime further: ΔT_1/2 is 6.87 hours for two pS containing peptides and 3.70 for those with one pS and one pT (Figure 5D), higher than single phosphosite containing peptides. This result reinforces the delaying effect of individual phosphorylation on protein turnover globally. Identical observations were obtained from HeLa_8 independently (Figure S4E–F). Finally, phosphosites without transmembrane topologies (non-TM) demonstrated smaller ΔT_1/2 than those with transmembrane domains (e.g., helices, Figure 5E, and Figure S4G). Nevertheless, phosphosites predicted in a buried region with low solvent accessibility and in helical, or β-sheet demonstrated domains exerted smaller half-life changes than those in exposed and loop area, in agreement to with the thermal stability data reported previously (Huang et al., 2019) (Figure 5F–G and Figure S4H–I).

In summary, we discovered that local protein structural properties in the vicinity of specific phosphorylation sites influence protein turnover in response to phosphorylation.

Functional insights for altered protein turnover induced by phosphorylation.

Next, we studied the relationship between phosphosite functions and turnover alteration (Figure 6 and Figure S5). Ochoa et al. recently conducted a meta-analysis of phosphoproteomic datasets and extracted 59 features annotating phosphosite functions. They further integrated these features to a single score that prioritizes phosphosites relevant for cell fitness (Ochoa et al., 2020). Correlating ΔT_1/2 to this score reveals a strong, negative trend (Figure 6A), reassuringly indicating that those phosphosites of a high fitness score are likely to have a faster turnover (in Segment Q1), and underscoring the importance of previous studies on the crosstalk between phosphorylation and ubiquitination and phosphodegrons (Holt, 2012; Swaney et al., 2013). In-depth correlation analysis suggests that ΔT_1/2 has a high evolutionary relevance. As compared to Q1-4, Q5 sites exhibit a much higher variant tolerance (SIFT score) that summarizes the lower conservation for phosphosite residue to alanine mutations or any mutations (Ng and Henikoff, 2003; Ochoa et al., 2020) (Figure 6B & Figure S5A–B). Thus, phosphosites, mainly those slowing down protein turnover, tend to be less conservative during evolution. Moreover, the kinase and kinase-substrate annotation suggested that phosphosites in kinases, in general, regulated their phosphomodiform lifetime as phosphosites in other proteins (see also Figure 3). Deviating from this norm are a few kinases showing faster turnover in steady-state cells, such as Rho-associated coiled-coil containing kinases, ROCK1, and ROCK2 (Figure 6C and S6A–D). Kinase-substrate mapping further revealed that the phosphosites activated by Cyclin-dependent kinase 1 (CDK1) have a drastically short lifetime (Figure 6D & Figure S5C), in agreement with the knowledge that the phosphorylation coupling degradation is essential in cell division (Holt, 2012). Next, to obtain an unbiased functional view for site-specific ΔT_1/2, we performed a biological process (BP) enrichment analysis in each segment from Q1 to Q5 (Figure 6E for HeLa_7 & Figure S6E for both cells). A few BPs are universally enriched by the phosphoproteome (throughout Q1-Q5), such as cell-cell adhesion, RNA splicing, and RNA binding. Specific BPs essential for translational control such as translational initiation (P=0.035) and cell cycle (P=0.008) are enriched in Q1, demonstrating their shorter lifetime upon phosphorylation. The SH3 domain was preferably enriched in Q2-4, indicating the robustness of these phosphoproteins’ turnover, on average. Q5 enriched distinctive BPs such as translocation (P=1.93E-5), glucose transport (P=4.1E-4), and tRNA transport from the nucleus (P=6E-4), suggesting a stabilization of the expression of proteins participating in cellular transport by phosphorylation. In addition, distributing ΔT_1/2 values to their subcellular protein locations (Thul et al., 2017) only revealed an exceptional case of Nuclear Speckles, where the phosphoproteins have a relatively faster turnover (Figure S6F).

Figure 6. Functional features associated with phosphorylation-altered protein turnover.

(A) The Functional scores of phosphosites are distributed into ΔT_1/2 quintiles (Q1 to Q5: small to large) in HeLa_7 and HeLa_8 cells (**P < 0.01, *P < 0.05, Wilcoxon sum test).

(B) The Sift scores of the alanine mutant for phosphosites are distributed into ΔT_1/2 quintiles (Q1-Q5: from small to large) in HeLa_7 and HeLa_8 cells (***P < 0.001, **P < 0.01, Wilcoxon sum test).

(C) Depiction of human kinome indicating all kinases that harbor phosphorylation site with less (green) or greater (brown) phosphomodiform T_1/2 value compared to bulk proteins in HeLa_7 cells. A large node represents a greater difference. For kinase with multiple phosphosites, only the largest change (maximum absolute ΔT_1/2 value) was shown.

(D) Comparisons between the detected kinase substrate motifs and ΔT_1/2 values (to bulk proteins) for identified phosphosites in HeLa_7 cells.

(E) Functional processes enrichment analysis for all five quintiles of proteins according to the ranked ΔT_1/2 value of phosphomodiforms compared to the matched proteins in HeLa_7 cells.

(F) Sequence analysis of the ±14 amino acids around the phosphosite with the smallest and longest ΔT_1/2 values (Q1 vs. Q5, where Q1 has a faster turnover for p-peptides than np-peptides, and Q5 has a much slower turnover for p-peptides than np-peptides) in HeLa_7 and HeLa_8 cells. The percentage of significant residues (P < 0.05) were shown.

(G) Distribution of ΔT_1/2 values of phosphopeptides with different numbers of Glutamic acids (Es) around the phosphosite (±14 amino acids) in HeLa_7 cells (***P < 0.001, Wilcoxon sum test).

(H) Distribution of ΔT_1/2 values (to bulk proteins) with different percentages of modifiable ubiquitination sites (the number of ubiquitination sites over the total number of Lys) around the phosphosite (±14 amino acids) in HeLa_7 cells (*P < 0.05, one-sided Wilcoxon sum test).

All corresponding results in HeLa_8 can be found in Figure S5–6.

Besides the biological function annotation, we assessed the frequency of amino acids (A.A.) surrounding phosphosites, with the expectation to identify more common principles that couple phosphorylation with degradation kinetics (Figure 6F). The central A.A. position in this analysis confirmed that pT and pY outperform pS in accelerating protein turnover (see also Figure 5D). Interestingly, we did not detect apparent enrichment of lysine (K) residues around phosphosites in Q1, although lysine residues can be potentially ubiquitylated. Instead, we discovered a prevalent enrichment of glutamic acid (E) in Q5. Indeed, more Es within ±14 A.A. residues remarkably increase the phosphosite expression stability, which was not previously reported (Figure 6G and Figure S5D). Finally, we mapped the phosphosites with their ΔT_1/2 to a dataset identifying ubiquitylation peptides containing K-ε-diglycine (van der Wal et al., 2018) in human cells (diGly, Figure 6H and Figure S5E). We found that ΔT_1/2 indeed shrinks with an increasing percentage (from 0-60%) of modifiable Ks (Martin-Perez and Villen, 2017; Swaney et al., 2013) around the phosphosites. However, the highest density of K(diGly) seemed to conversely impair the degradation effect (i.e., ΔT_1/2 is larger for 80-100% range, Figure 6H), suggesting that the density of K(diGly) alone without the ubiquitin stoichiometry data is not enough to denote the protein degradation extent.

In summary, the biological annotation based on DeltaSILAC data recognizes functional relevance of site-specific effect of phosphorylation on protein stability.

Regulation of protein turnover due to phosphorylation can be largely conserved across cells.

Above, we analyzed cervical cancer cells HeLa_7 and HeLa_8 by DeltaSILAC. We next sought to assess the generalizability of the major observations in two colorectal cancer (CRC) cell lines, SW948 and RKO, by conducting DeltaSILAC measurement. We performed absolute and relative correlation analysis for T_1/2 and ΔT_1/2 between all four cell lines (Figure 7A–C). The correlations of proteome-wide, as well as phosphoproteome-wide T_1/2 between two CRC cells and two HeLa strains, were higher than correlations across the tissue types of the cells (R=0.82-0.83 for CRC cells, R=0.87-0.89 for HeLa cells, in contrast to R=0.53-0.59 between them). The relative measure ΔT_1/2, resulted in almost comparable correlations (R=0.76 for CRC cells, R=0.83 for HeLa cells, and R=0.31-0.49 between them). Also, we were able to broadly reproduce observations made in HeLa cells, such as the enrichment of Es in Q5 (Figure 7D) and the globally decreased protein turnover by phosphorylation in the CRC cells (Figure S7A). In addition to analyzing the cultured human cancer cell lines, we also applied DeltaSILAC to measure the phosphoproteome turnover of the synchronized PC12 cells (see Methods), a rat pheochromocytoma cell line that was used in enormous pharmacological and cell signaling transduction studies. Similar conclusions were made in PC12 cells (Figure S7B–C). Thus, ΔT_1/2 demonstrates a promising, non-stochastic parameter to be measured for different cell lines.

Figure 7. Phosphorylation conferred protein turnover regulation across HeLa_7, HeLa_8, SW480, and RKO cells.

(A-C) Spearman’s correlation among four cell lines of T_1/2 values for all peptides (A), phosphopeptides (B), and ΔT_1/2 values (C).

(D) Sequence analysis of the ±14 amino acids around the phosphosite with the fastest and slowest ΔT_1/2 values (Q1 vs. Q5) in SW948 and RKO cells. The percentage of significant residues (P < 0.05) were shown.

(E) A proposed model coupling protein turnover to phosphorylation with different surrounding features.

Discussion

Here we present a proteomic method, DeltaSILAC, that quantitatively measures the site-specific impact on protein turnover rendered by PTM events on thousands of proteins. Phosphorylation was previously profoundly studied with respect to the instant “phosphate-transfer” activities by kinases and phosphatases. However, many druggable phosphorylation sites are constitutive, stabilized, or rewired in the disease status. We make use of the fact that in steady-state growing cells, the abundance of almost all the proteins, as well as their modified versions, achieve a balance between synthesis and degradation (Claydon and Beynon, 2012; Liu et al., 2016; Schwanhausser et al., 2011). Therefore, a pSILAC experiment provides an excellent opportunity to study the de facto cellular outcome of this balance for both proteins and modiforms. DeltaSILAC essentially quantifies overall phosphoproteome degradation control such as the ubiquitin-proteasome pathway and lysosomal proteolysis irrespective of “phosphate-transfer” kinetics. This is because, in DeltaSILAC, the counterpart non-modified peptide (np-peptide, or at least the deriving protein) for each Class-I phosphopeptide has to be identified and quantified with a half-life, simulating a virtual, non-modified peptide reference that is also individually site-specific. Thus, it should be stressed, that the bona fide turnover change might be smaller than ΔT_1/2 appeared, due to the potential mutually causal relationship between lifetimes of phosphomodiforms and their non-modified counterparts. Nevertheless, we believe DeltaSILAC presents a significant methodological advance, in contrast to traditional qualitative LC-MS/MS studies (Swaney et al., 2013), and avoids quantitative bias due to PTM enrichment efficiency, basic protein abundance, instrument sensitivity drift between injections, and the possibility of other co-existing PTMs. In DeltaSILAC strategy, these traits may only affect peptide detection possibility, but do not lead to inaccurate turnover measurements.

Of note, the high reproducibility of DIA-MS favorably supports pSILAC experiments. The single-shot DIA-MS already achieved substantial coverage on both proteome and phosphoproteome, shortening the overall instrument time needed for a global analysis. Importantly, as illustrated in Figure 2F–I, the MS2-level ion signatures unique to PTM can be extracted from DIA, together with high-resolution LC separation, to reach unprecedented analytical specificity for each PTM sites without the need of peak alignment (Salovska et al., 2020). Such specificity may be challenging to achieve for MS1 alignment algorithms (Cox and Mann, 2008; Lim et al., 2019) or for Tandem mass tag (TMT)-based multiplexing (Welle et al., 2016), especially when the elution peaks of phosphopeptide positional isomers significantly overlap. Another prominent advantage of DeltaSILAC is its robustness because the H/L ratios of the matched PTM and non-PTM peptides are comparable in each sample and across experiments within the same cell line. This relative comparison avoids the need of determining the precise cell doubling time for calculating the absolute protein lifetime in conventional pSILAC experiments.

Our dataset combining phosphoproteomics and turnover estimation revealed novel and unexpected biological insights. First, the majority of phosphorylation sites were found to be increased in their lifetime, at least in the all investigated cancer cells under stable growing status. Although this phenomenon has to be measured in other in-vivo systems or animal models, our data strongly demonstrates that the impact of phosphosites on protein lifetime is rather heterogenous. The delayed turnover of phosphomodiforms, on first sight, seems to be contradictory to the functional crosstalk events discovered between phosphorylation and ubiquitylation (Hunter, 2007; Nguyen et al., 2013; Suizu et al., 2009). Nevertheless, further scrutiny suggests that the phosphosites of faster turnover (i.e., those in Q1) are indeed functionally more essential for cell fitness, more evolutionarily conserved, and enriched for CDK1 substrate sites, all consistent with previous findings (Hunter, 2007; Nguyen et al., 2013; Skowyra et al., 1997; Verma et al., 1997). Also, the phosphosites CTNNB1-S191, SNAP23-S110, and YY1-S118 that were previously shown to be important in cancer and other processes (Riman et al., 2012; Verweij et al., 2018; Xu et al., 2020) were all verified by exogenous mutational experiments (Figure 4) to stabilize the respective proteins, underscoring the reliability and utility of DeltaSILAC. Thus, the discovery that phosphorylation often negatively acts on protein turnover might be an observation that was underrepresented in previous studies due to the lack of unbiased quantitative measurements. Second, local protein structural feature analysis and biological annotation illustrate that the turnover-delaying phosphomodiforms may involve multiple-phosphorylated sites, frequently in the loop and exposed regions, and might significantly regulate cellular transport and various processes. For example, the nuclear pore complexes (NPC) were previously identified to maintain over a cell’s life through a slow but finite exchange (Toyama et al., 2013). According to our data, with a few sites being phosphorylated these NPC proteins might resist degradation even more, reinforcing their possible roles in cellular aging (Toyama et al., 2013). Third, our motif analysis in all the cell line datasets indicates that the phospho-T may have an average shorter lifetime than phospho-S at the cellular steady state (Figure 5D), prompting the need for future studies on differential kinetics of S and T phosphorylation (Hein et al., 2017). Last but not least, the dataset suggests the higher frequency of glutamic acid is remarkably linked to longer phosphomodiform lifetime. The above observations, although preliminary, open an attractive avenue for perturbating the activity of crucial phosphorylation sites that are known to be responsible for drug response and disease development, for the possibility of phenotype management (Figure S7D–F). The targeted management could be directed by the prediction rules (Holt, 2012) (Figure 7E) exemplified above. Generating data inventories of turnover features by DeltaSILAC measurements on more samples and relevant models is thus appealing in the future.

As for potential future applications, DeltaSILAC could be quickly adapted to study protein modifications other than phosphorylation. However, the pSILAC design will limit this approach in cell line models or other systems where the isotopic labeling is possible (Fornasiero et al., 2018; Hidalgo San Jose and Signer, 2019; Kruger et al., 2008). Also, mechanistically, more time points of pSILAC labeling and subsequent biochemical experiments are required to deeply understand more detailed or alternative mechanisms underlying the altered turnover rates. Examples of interesting questions are. e.g., whether the newly synthesized proteins are less likely to be phosphorylated or modified? Do modified and non-modified proteins follow different exponential decay kinetics (McShane et al., 2016)? How does the DeltaSILAC result change between different subcellular compartments?

In conclusion, our study offers a highly powerful workflow for timing the turnover of thousands of site-specific modifications, providing complementary knowledge to the current understanding of functional protein PTMs.

STAR☆METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Yansheng Liu (yansheng.liu@yale.edu).

Materials Availability

Reagents generated in this study will be available upon request.

Data and Code Availability

The datasets generated during this study are available at ProteomeXchange Consortium with the dataset identifier PXD017496.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell culture.

Twelve of 14 HeLa strains used in the multi-omics study were used in this report (Liu et al., 2019), analyzing phosphoproteomic variability. The collection includes the original six HeLa cells variants subtype CCL2 (2, 6, 7, 12, 13, and 14) and six HeLa cells variants subtype Kyoto (1, 3, 4, 8, 9, and 10). The original HeLa_11 (deviating genome (Liu et al., 2019)) and HeLa_5 (CCL2.2) were not used in this study. All HeLa cells were gifts from multiple laboratories, as documented previously. HeLa _7 (CCL-2) and HeLa_8 (Kyoto) were original from ATCC. The human colon cancer cell lines, SW948 (ATCC, CCL-237) and RKO (ATCC, CCL-2577), were kindly provided by Dr. Prem Subramaniam at Columbia University. All cells were tested and confirmed to be negative for mycoplasma. All cells were cultured for up to five additional passages for aliquoting and protein harvest. The cell culture protocol was detailed previously (Liu et al., 2019). Cells were routinely cultured in 5% CO₂, and 37 °C in DMEM Medium supplemented with 10% FBS, Sigma Aldrich, together with a penicillin/streptomycin solution (Gibco). PC12 cells (ATCC, CRL-1721) were cultured in T75 flasks at 37 °C, humidified 5% CO₂ in complete RPMI-1640 medium (Life Technologies, #11875093) and then primed by 16 hour serum-free conditioning as for signaling transduction studies (Kiyatkin et al., 2020).

METHOD DETAILS

Pulsed SILAC experiment.

For the two HeLa cell lines (HeLa_7, HeLa_8), SILAC DMEM medium (Thermo #88364) lacking L-arginine and L-lysine was firstly supplemented with 10% dialyzed FBS (Thermo Fisher, # 26400044) and the same penicillin/streptomycin mix. For both SW948, RKO and PC12 cell lines, the SILAC RPMI-1640 media lacking L-Arginine, L-Lysine (Thermo Scientific, # 88365) was used instead, with the same basic configuration. The Heavy L-Arginine-HCl (13C6, 15N4, purity >98%, #CCN250P1), and L-Lysine-2HCl (13C6, 15N2, purity >98%, # CCN1800P1) were purchased from Cortecnet and spiked into the culturing medium in the same manner as described previously for DMEM or RPMI medium (Li et al., 2019; Liu et al., 2019). Before SILAC labeling, cells were seeded (at 40-50% confluency for CRC cells and 50-60% for HeLa and PC12 cells) and incubated in normal light DMEM/RPMI medium for 24 hours, at 5% CO₂ 37°C, overnight. For pSILAC labeling of both HeLa cells, five-time points, including 0 hours and four labeling points 1, 4, 8, and 12 hours were applied, with three biological replicates (as individual dishes) per time point per cell line. This means a total of fifteen 10-cm dishes per cell line were prepared before labeling. For all cell lines, each replicate sample per time point yielded corresponding protein amount, sufficiently supporting 700 μg of protein mixture (see below) processed for proteomic and phosphoproteomic analysis.

DeltaSILAC proteomic sample preparation.

Label-free and labeled cells of HeLa cells, including HeLa_7 and HeLa_8, SW948, RKO, were harvested and digested, mainly as previously described (Collins et al., 2017; Li et al., 2019). Cells were washed three times by precooled PBS, harvested, and snap-frozen by liquid nitrogen. The cell pellets were immediately lysed by adding 10 M urea containing complete protease inhibitor cocktail (Roche) and Halt™ Phosphatase Inhibitor (Thermo) and stored in −80°C for further analysis. After harvesting all dishes, samples of HeLa_7 and HeLa_8, as well as samples from SW948 and RKO, were processed for tryptic digestion. The cell pellets were ultrasonically lysed by sonication at 4 °C for 2 min using a VialTweeter device (Hielscher-Ultrasound Technology) and then centrifuged at 18,000 × g for 1 hour to remove the insoluble material. A total of 700 μg supernatant proteins (determined by BioRad Bradford assay) were transferred to clean Eppendorf tubes. The supernatant protein mixtures were reduced by 10 mM tris-(2-carboxyethyl)-phosphine (TCEP) for 1 hour at 37 °C and 20 mM iodoacetamide (IAA) in the dark for 45 min at room temperature. Then five volumes of precooled precipitation solution containing 50% acetone, 50% ethanol, and 0.1% acetic acid were added to the protein mixture and kept at −20 °C overnight. The mixture was centrifuged at 18,000×g for 40 min. The precipitated proteins were washed with 100% acetone and 70% ethanol with centrifugation at 18,000×g, 4°C for 40 min, respectively. 300 μT of 100 mM NH₄HCO₃ was added to all samples, which were digested with sequencing grade porcine trypsin (Promega) at a ratio of 1:20 overnight at 37 °C. After digestion, the peptide mixture was acidified with formic acid and then desalted with a C18 column (MarocoSpin Columns, NEST Group INC). The amount of the final peptides was determined by Nanodrop (Thermo Scientific). About 5% of the total peptide digests were kept for total proteomic analysis.

DeltaSILAC phosphoproteomic sample preparation.

From the same peptide digest above, 95% of peptides per sample was used for phosphoproteomic analysis. The phosphopeptide enrichment was performed using the High-Select™ Fe-NTA kit (Thermo Scientific, A32992) according to the kit instruction, as described previously (Gao et al., 2019). Briefly, the resins of one spin column in the kit were divided into five equal aliquots, each used for one sample. The peptide-resin mixture was incubated for 30 min at room temperature and then transferred into the filter tip (TF-20-L-R-S, Axygen). The supernatant was removed after centrifugation. Then the resins adsorbed with phosphopeptides were washed sequentially with 200 μL× 3 washing buffer (80% ACN, 0.1% TFA) and 200 μL×3 H₂O to remove nonspecifically adsorbed peptides. The phosphopeptides were eluted off the resins by 100 μL×2 elution buffer (50% ACN, 5% NH₃•H₂O). All centrifugation steps above were conducted at 500 g, 30sec. The eluates were collected for speed-vac and dried for mass spectrometry analysis.

DIA mass spectrometry.

For each proteomic and phosphoproteomic sample generated by DeltaSILAC, DIA-MS analysis was performed on 1 μg of peptides, as described previously (Li et al., 2019; Mehnert et al., 2019).

LC separation was performed on EASY-nLC 1200 systems (Thermo Scientific, San Jose, CA) using a self-packed analytical PicoFrit column (New Objective, Woburn, MA, USA) (75 μm × 50 cm length) using C18 material of ReproSil-Pur 120A C18-Q 1.9 μm (Dr. Maisch GmbH, Ammerbuch, Germany). A 120-min measurement with buffer B (80% acetonitrile containing 0.1% formic acid) from 5% to 37% and corresponding buffer A (0.1% formic acid in H₂O) during the gradient was used to elute peptides from the LC. The flow rate was kept at 300 nL/min with the temperature-controlled at 60 °C using a column oven (PRSO-V1, Sonation GmbH, Biberach, Germany).

The Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Scientific) instrument coupled to a nanoelectrospray ion source (NanoFlex, Thermo Scientific) was calibrated using Tune (version 3.0) instrument control software. Spray voltage was set to 2,000 V and heating capillary temperature at 275 °C. All the DIA-MS methods consisted of one MS1 scan and 33 or 40 MS2 scans of variable isolated windows (Mehnert et al., 2019), with1 m/z overlapping between windows. The MS1 scan range is 350 – 1650 m/z, and the MS1 resolution is 120,000 at m/z 200. The MS1 full scan AGC target value was set to be 2.0E5 (2.0E6 for label-free samples), and the maximum injection time was 100 ms. The MS2 resolution was set to 30,000 at m/z 200 with the MS2 scan range 200 – 1800 m/z, and the normalized HCD collision energy was 28%. The MS2 AGC was set to be 5.0E5 (1.5E6 for label-free samples), and the maximum injection time was 50 ms. The default peptide charge state was set to 2. Both MS1 and MS2 spectra were recorded in profile mode. Detailed MS settings can be inspected through raw files provided at ProteomeXchange.

Western blotting-based measurement of protein stabilization.

Full length CTNNB1/SNAP23 wild-type (pCMV Flag-tagged Human CTNNB1/SNAP23) plasmid or the relevant CTNNB1-S191A/SNAP23-S110A and CTNNB1-S191D/SNAP23-S110D mutant were cloned into the pCMV-MCS-Flag vector to generate corresponding expression plasmids. Plasmids were transfected with Lipofectamine 3000 (Thermo Fisher Scientific, L3000015). The following antibodies were used: Monoclonal ANTI-FLAG® M2 antibody produced in mouse (Sigma, F3165) and α Tubulin antibody (B-5-1-2) (Santa Cruz Biotechnology, sc-23948). To study protein degradation, Cycloheximide (Targetmol, T1225) was used at 100 μg/mL for up to 8 hours. Cells were lysed in EBC lysis buffer (50 mM Tris HCl, pH 8.0, 120 mM NaCl, 0.5% Nonidet P-40) supplemented with protease inhibitors (Selleck Chemicals) and phosphatase inhibitors (Selleck Chemicals). Equal amounts of protein (10 μg per lane) derived from each sample were separated on 10% SDS-PAGE gels and transferred to polyvinylidene difluoride (PVDF). The membranes were blocked for 1 hour with 5% nonfat dried milk and subsequently incubated with primary antibodies overnight at 4 °C. In the following morning, the membranes were incubated with the secondary antibodies for 1 hour at room temperature. The bound antibodies were visualized using SuperSignal West Femto chemiluminescence reagent (Pierce) in SmartChemi 610 (Beijing sage creation). The images were analyzed with Image J software.

Fluorescent imaging-based measurement of protein stabilization.

Synthetic genes were designed in silico and were ordered at GenScript (USA). The human coding sequence of the YY1 transcription factor (NM_003403) or its mutants (S118A. S118D) were inserted in a custom-designed plasmid containing the coding sequence of the 2X-ALFA-Tag (Gotzke et al., 2019), followed by the coding sequence of the P2A self-cleaving peptide (Szymczak-Workman et al., 2012) and the sequence of the human-optimized mNeonGreen (Shaner et al., 2013) used for identifying positive cells. Sequences were sub-cloned in pcDNA3.1(+) plasmids (Thermo Fisher Scientific, cat. num. V79020). All final plasmids were confirmed by sequencing with a forward CMV primer (CGCAAATGGGCGGTAGGCGTG) and a reverse BGH primer (TAGAAGGCACAGTCGAGG). For imaging experiments cells were plated at a concentration of 100’000 cells/well on SensoPlate 24-well glass-bottom plates (Greiner Bio-One; cat. num. 662892) coated with 0.1 mg/ml poly-L-lysine. Cells were transfected in solution with Lipofectamine 2000 (Thermo Fisher Scientific). Before transfection, the quantity of DNA was measured in triplicate at the NanoDrop 2000 and confirmed by densitometric analysis on quantitative DNA gel electrophoresis. For each well the same quantity of DNA (500 ± 5 ng) was incubated with 1.0 μl of Lipofectamine and used for transfections as suggested by the manufacturer. After 24 hours of expression, cells were chased in the presence of 5 μM of Cycloheximide, fixed in 4% buffered PFA, quenched, stained with DAPI, and imaged in PBS at the Cytation 5 cell imaging multi-mode reader (BioTek).

QUANTIFICATION AND STATISTICAL ANALYSIS

Mass spectrometry data analyses.

DIA-MS data analyses were performed using Spectronaut v13 (Bruderer et al., 2015; Bruderer et al., 2017). For both proteomic and phosphoproteomic measurements, the hybrid assay libraries were respectively generated, which were based on both DIA measurements as well as data-dependent acquisitions (DDA) on relevant samples, as well as the optimized pSILAC-DIA workflow (Salovska et al., 2020). To generate the library for pSILAC DIA-MS datasets, the default settings for Pulsar search of Spectronaut was used with modification in the Labeling setting: a) “Labelling Applied” option was enabled, b) SILAC labels (“Arg10” and “Lys8”) were specified in the second channel. c) The complete H/L labeling of the whole library was ensured by selecting the “In-Silico Generate Missing Channels” option in the Workflow settings. d) Importantly, for phosphoproteomic datasets, the possibility of Phosphorylation at S/T/Y was enabled during database searching (as a variable modification), together with Oxidation at methionine was set as variable modification, whereas carbamidomethylation at cysteine was set as a fixed modification. The final spectral libraries (HeLa proteome library containing 179,199 peptide precursor assays for 8,040 protein groups, HeLa phosphoproteome library containing 139, 117 peptide precursor assays for 7,758 protein groups, CRC proteome library containing 78,763 peptide precursor assays for 6,593 protein groups, and CRC phosphoproteome library containing 56,056 peptide precursor assays for 4,859 protein groups) were all made public through PRIDE (see below). Otherwise, for targeted peptide identification and protein quantification in label-free samples, Spectronaut settings were kept as Default.

For the targeted data extraction and subsequent identification and quantification for pSILAC datasets, the Inverted Spike-In (ISW) workflow was used, as described previously (Salovska et al., 2020). This means the “Spike-In” workflow was selected in Multi-Channel Workflow Definition, and both “Inverted” and “Reference-based Identification” options were enabled. Both peptide and protein FDR cutoff (Qvalue) were controlled at 1%, and the data matrix was strictly filtered by Qvalue. The “minor peptide group” was defined as Modified Sequences. In particular, the PTM localization option in Spectronaut v13 was enabled to locate phosphorylation sites (Bekker-Jensen et al., 2020; Rosenberger et al., 2017b), with the probability score cutoff >0.75 (Bekker-Jensen et al., 2020), resulting Class-I peptides (Olsen et al., 2006) to be identified and quantified. All the other settings in Spectronaut were kept as Default.

pSILAC based calculation for turnover analysis.

In a pSILAC experiment analyzing protein turnover, because the growing cells are respectively maintained in a steady state (Claydon and Beynon, 2012), it is assumed that the degraded and synthesized protein copies are balanced (Welle et al., 2016). Accordingly, almost all (if not all) the phosphomodiforms, including alternative splicing isoforms (Zecha et al., 2018) and proteins with PTMs, should achieve the concentration balance between synthesis and degradation in such a state (i.e., without any perturbation). This principle essentially enables the protein turnover calculation by monitoring the intensities of light and heavy peptides across several time points.

To fit the model of protein turnover estimation, we used a similar approach as was employed in our previous studies (Liu et al., 2017; Liu et al., 2019; Pratt et al., 2002). Below, we describe such an approach in the present study.

1. We quantified peptide precursor intensities for light and heavy signals from the above Spectronaut results.

2. We calculated the rate of loss of the light isotope (k_Loss) by modeling the relative isotope abundance (RIA, analogous to Pratt et al. (Pratt et al., 2002)). RIA is determined by the signal intensity in the light channel divided by the sum of light and heavy intensities i.e., RIA = L/(H + L), onto an exponential decay model assuming a null heavy intensity (RIA = 1) at time 0; i.e., RIA(t) = e^−kLoss×t.

3. We used nonlinear least-squares estimation to perform the fit. An average of the peptide precursor k_Loss values was performed to calculate the k_Loss values for all unique peptide sequence precursors, which were then aggregated as below.

4. We then summarized k_Loss to evaluate the turnover rate for each bulk-protein as previously described, using all peptides weighted for a particular bulk protein (Liu et al., 2017; Liu et al., 2019), but also for each peptide and especially phosphopeptide in this study. In particular, for k_Loss determination, we applied a strict filter strategy as previous studies (Zecha et al., 2018) by only accepting those peptide H/L ratios showing monotone increasing pattern during pSILAC labeling process and by only accepting those turnover rates with the CV of log2(k_Loss) (Claydon and Beynon, 2012) <20% across three biological replicates.

5. To calculate the half-life T_1/2 (Pratt et al., 2002; Rost et al., 2016; Zecha et al., 2018) for each peptide and phosphopeptide, we used T_1/2 = Ln(2)/k_Loss so that the k_Loss rate can be converted to a time domain. It should be noted that DeltaSILAC determination is performed per cell line. Thus, the relative doubling time difference between cell lines does not impact the calculation of our T_1/2 values, which are essentially identical to the previously reported T_50% values (Zecha et al., 2018).

6. Finally, to pinpoint the turnover influence of each PTM, we subtracted T_1/2 of backbone sequence-matching non-phosphorylated peptide (np-peptide) from the T_1/2 of counterpart phosphopeptide (p-peptide), resulting in the value of ΔT_1/2.

   This means ΔT_1/2 = T_{1/2 p-peptide} − T_{1/2 np-peptide} was applied.

The whole process of calculating T_1/2 is also illustrated in Figure S1C.

Image analysis.

Image analysis (Figure 4 E–F) was performed in ImageJ (Schneider et al., 2012). Briefly, transfected cells (mNeonGreen positive) were selected in the green channel by applying an empirically derived threshold, which was identical for all the images from the same time point analyzed. The threshold procedure generated region-of-interest masks for all the cells. The fluorescence intensity in the channel corresponding to the FluoTag®-X2 580 red nanobody (rhodamine channel) was then determined in the masks (typically 3000–4000 masks per time point and experiment, for a total of >10,000 cells analyzed for each experiment). These values were then averaged for each of the experiments and the standard error of them mean (SEM) was calculated across three separate experiments (n = 3).

Bioinformatics analyses.

Circos-0.69-9 (http://circos.ca/) (Krzywinski et al., 2009)was used for the circle visualization (Figure 1E). Spearman’s correlation coefficients were calculated using R (functions cor() or cor.test() to infer statistical significance). The colored scatterplots from blue-to-yellow were visualized by the “heatscatter” function in R package “LSD” using a two-dimensional Kernel Density Estimation (Figure 1G and 6A–C). The correlation matrix was created by the “corrgram” function in R package “corrgram” (Figure 1F). Partial correlation was analyzed by the “pcor” function in R package “ppcor” to inspect relationship between two continuous variables whilst controlling for or correct against the effect of another continuous variable. The melting temperature (Tm, °C) value (Figure 5A) for each phosphosite was taken from the reported dataset(Huang et al., 2019). The flanking amino acid sequences (±14 amino acids) of phosphorylation sites were retrieved by motifeR (https://www.omicsolution.org/wukong/motifeR/) (Wang et al., 2019). The Calculate the grand average of hydropathy (GRAVY) value for protein sequences, defined by the sum of hydropathy values of all amino acids divided by the protein length, was computed by GRAVY Calculator (http://www.gravy-calculator.de/index.php). The flanking amino acid sequences (±14 amino acids) of phosphorylation sites that can be unambiguously assigned to specific serine, threonine, or tyrosine residues were used for GRAVY calculation (Figure 5B). Secondary structure element, solvent accessibility, and transmembrane topology were predicted using Predict_Property standalone package (v1.01)(Wang et al., 2016) with the protein FASTA sequences (Figure 5E–G). The functional score can reflect the importance of phosphosite for organismal fitness (Ochoa et al., 2020). The sift score predicts the functional impact of missense variants based on sequence homology and the physicochemical properties of the amino acids (Vaser et al., 2016). Lower scores represent deleterious variants. For every phosphorylated protein site, the functional score (Figure 6A), sift scores of alanine mutant (Figure 6B), the average sift scores of all variants, protein domains (Figure 6C), and kinase substrate motifs (Figure 6D) were retrieved from the previous report (Ochoa et al., 2020). Kinase family tree (Figure 6C) was depicted by Coral (http://phanstiel-lab.med.unc.edu/CORAL/)(Metz et al., 2018). Functional annotation was carried out in David Functional Annotation Tool v6.8 (https://david.ncifcrf.gov/summary.jsp) (Huang da et al., 2009) with all detected proteins in this study as background (Figure 5E). The compound responses were enriched by phosphorylation site-specific functional enrichment through ssGSEA2.0/PTM-SEA(Krug et al., 2019). Sequence analysis (Figure 6F and 7D) was conducted and visualized by IceLogo (https://iomics.ugent.be/icelogoserver/)(Colaert et al., 2009). The modifiable ubiquitination sites in the flanking sequence (±14 amino acids) of the phosphosite were identified according to the reported dataset(van der Wal et al., 2018), and the percentage of modifiable ubiquitination sites was calculated as the ratio of the number of ubiquitination sites over the total number of Lys (Figure 6H). The cellular compartment location of proteins with phosphorylation was annotated by a subcellular map of the human proteome (Thul et al., 2017). All boxplots were generated using the R package “ggplot2”. The bold line within box indicates median value; box borders represent the first and third quartile, and whiskers and grey panels represent the minimum and maximum value within 1.5 times of interquartile range. Outliers are depicted using hollow dots. The heatmap was created using the R package “pheatmap”. Statistical analyses were conducted using Wilcoxon sum test. For volcano plots, P values were calculated by Student’s T-tests. Statistical significance is indicated in the figure legends.

Supplementary Material

2

Table S1. Phosphoproteomic profiling across 12 HeLa cell lines (Related to Figure 1)

For the 12 HeLa cell lines included in Liu Y. 2019 (see Methods), the raw DIA-MS signals for distinctive phosphorylated peptides were shown. The _a, _b, _c denotes three dish replicates for the whole process.

3

Table S2. The k_Loss, T_1/2, and turnover data calculated for HeLa_7 and HeLa_8 (Related to Figure 2)

The k_Loss data and T_1/2 are reported at the protein (Prot-), individual peptide (Pep-) and individual phosphopeptide (Phos-) levels.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Monoclonal ANTI-FLAG® M2 antibody	Sigma	Cat# F3165
α-Tubulin antibody (B-5-1-2)	Santa Cruz	Cat# sc-23948
Chemicals, Peptides, and Recombinant Proteins
DMEM high glucose	Gibco	Cat# 10564011
RPMI 1640	Gibco	Cat# 72400047
Fetal bovine serum (FBS)	Sigma-Aldrich	Cat# F8318
DMEM for SILAC	Thermo Fisher Scientific	Cat# 88364
PRMI 1640 Medium for SILAC	Thermo Fisher Scientific	Cat# 88365
Dialyzed fetal bovine serum (FBS)	Thermo Fisher Scientific	Cat# 26400044
Penicillin/streptomycin solution	Gibco	Cat# 15140122
Heavy L-Arginine-HCl (13C6 for SILAC)	Cortecnet	Cat# CCN250P1
Heavy L-Lysine-2HCl (13C6,15N2 for SILAC)	Cortecnet	Cat# CCN1800P1
Trypsin-EDTA (0.25%)	Gibco	Cat# 25200072
Urea	Sigma-Aldrich	Cat# U5128
Cocktail	Roche	Cat# 4693116001
Halt™ Phosphatase inhibitor	Thermo Fisher Scientific	Cat# 78428
Tris-(2-carboxyethyl)-phosphine (TCEP)	Thermo Fisher Scientific	Cat# T2556
Iodoacetamide (IAA)	Sigma-Aldrich	Cat# I1149
Acetone	Thermo Fisher Scientific	Cat# 650501
Ethanol	Thermo Fisher Scientific	Cat# 459828
Acetic acid	Thermo Fisher Scientific	Cat# 71251
NH4HCO3	Thermo Fisher Scientific	Cat# 09830
Trypsin	Promega	Cat# V5111
Formic acid	Thermo Fisher Scientific	Cat# 1283200
Acetonitrile (ACN)	Thermo Fisher Scientific	Cat# TS-51101
Trifluoroacetic acid (TFA)	Thermo Fisher Scientific	Cat# 400445
Ammonium hydroxide	Sigma-Aldrich	Cat# 320145
Lipofectamine 3000	Thermo Fisher Scientific	Cat# L3000015
Lipofectamine 2000	Thermo Fisher Scientific	Cat# 11668019
Cycloheximide	Targetmol	Cat# T1225
Critical Commercial Assays
BioRad Bradford protein assay kit	BioRad	Cat# 5000201
High-Select™ Fe-NTA kit	Thermo Fisher Scientific	Cat# A32992
Deposited Data
Mass spectrometry data	ProteomeXchange Consortium	PXD017496
Experimental Models: Cell Lines
HeLa cell lines (1-14)	Liu et al (2019)	N/A
HeLa_7	ATCC	CCL-2
HeLa_8	ATCC	Kyoto
SW948	ATCC	CCL-237
RKO	ATCC	CRL-2577
PC12	ATCC	CRL-1721
Software and Algorithms
Spectronaut v13	Biognosys, Inc	N/A
GraphPad Prism 8	Graphpad software, Inc	N/A
ImageJ	Schneider et al (2012)	https://imagej.nih.gov/ij/
R (version 3.6.0)	R Core Team	https://www.r-project.org/
Circos-0.69-9	Krzywinski et al (2009)	http://circos.ca
MotifeR	Wang et al (2019)	https://www.omicsolution.org/wukong/motifeR/
GRAVY Calculator	Stephan Fuchs	http://www.gravy-calculator.de/index.php
Predict_Property standalone package (v1.01)	Wang et al (2016)	N/A
Coral	Metz et al (2018)	http://phanstiel-lab.med.unc.edu/CORAL/
David Functional Annotation Tool v6.8	Huang et al (2009)	https://david.ncifcrf.gov/summary.jsp
ssGSEA2.0/PTM-SEA	Krug et al (2019)	N/A
IceLogo	Colaert et al (2009)	https://iomics.ugent.be/icelogoserver/
ggplot2 package in R	Thomas Lin Pedersen	https://www.rdocumentation.org/packages/ggplot2
pheatmap package in R	Raivo Kolde	https://www.rdocumentation.org/packages/pheatmap/versions/1.0.12
LSD package in R	Bjoern Schwalb	https://www.rdocumentation.org/packages/LSD

Highlights.

A proteomic method reveals site-specific phosphorylation impacts protein degradation

Many phosphorylation sites delay protein turnover in growing cells

Phosphosites involved in cell cycle and cell fitness destabilize protein expression

Glutamic acids surrounding phosphosites significantly delay protein turnover