Profiling the C. elegans Phosphoproteome by MS in Advanced Approach
The phosphoproteome of C. elegans has not been surveyed rigorously; seeking to increase the coverage of the C. elegans phosphoproteome while aiming for high accuracy in both identification and quantification of phosphopeptides, we combined a number of technical elements and optimized the analytical workflow (Fig. 1A). These technical elements include extensive high-pH reverse phase fractionation coupled with interval pooling19, 20, polyMAC-Ti enrichment of phosphopeptides21, 22, high-speed and accurate-mass mass spectrometry, and stable isotope (15N) metabolic labeling, a highly accurate quantitative proteomics strategy (Fig. 1A).
From wild type (WT) C. elegans and the insulin signaling mutants (daf-2, daf-16, and the daf-16; daf-2 double mutant) — each analyzed in three or four biological replicates and two technical replicates — we identified a total of 15,443 phosphorylation sites with > 0.75 PhosphoRS site probability23. These phosphosites are represented by 22,536 phosphopeptides or 15,723 phosphoisoforms that belong to 4,418 proteins (Supplementary Fig. 1A-C). 9,949 phosphosites identified in this study are not covered by dbPAF, which is a comprehensive database dedicated to collecting phosphosites in humans, animals, and fungi17. Notably, the addition of these newly identified phosphosites close to doubles the current collection for C. elegans (Fig. 1B).
Although it is well established that phosphorylation is the primary means by which the IIS pathway transmits signals, very little is known about which sites are phosphorylated, even for the core components of C. elegans IIS. For example, dbPAF presently contains no phosphosites for the PI-3 kinase AGE-1. Here, our phosphoproteomics analysis uncovered 32 phosphosites in ten C. elegans IIS proteins, 17 of which have not been previously reported (Fig. 1C). These new, high confidence phosphosites (Fig. 1C, dark blue) are distributed throughout the pathway, from the upstream insulin-like ligands to the downstream FOXO transcription factor DAF-16, and for every kinase in between.
More than 15,000 phosphopeptides were quantified against their 15N-labeled cognate peptides, which were introduced as an internal reference standard by feeding C. elegans entirely on 15N-labeled bacteria (Supplementary Fig. 1B, see Methods). These peptides represent 10,705 quantifiable phosphoisoforms (Fig. 1D, Supplementary Fig. 1B-C, and Supplementary Table 1), about a quarter of which carry combinatorial information for two or more phosphosites. Clustering analysis of their abundance levels (relative to 15N-labeled peptides) across WT and IIS mutants indicated that the daf-2 mutant samples are clearly different from the WT, the daf-16, and daf-16; daf-2 samples. In other words, the quantitative phosphoproteomics data clustered according to the lifespan phenotype, not by batch (Supplementary Fig. 1E). The Spearman correlation coefficient between biological replicates of the same worm strain is in the range of 0.70–0.91. These results are indicative of high data quality for phosphopeptide identification and quantification.
Phosphorylation Changes Resulting from Genetic Disruption of IIS
Disrupting the activity of IIS induced abundance changes on 501 phosphoisoforms (> 1.5-fold in at least one of the IIS mutants relative to WT, Supplementary Fig. 1F). As expected, clustering and pathway enrichment analysis show that phosphorylation on proteins involved in FOXO signaling and longevity regulation were down-regulated in the daf-2 mutant. We also found that proteins related to protein synthesis or degradation had lower phosphorylation levels in the daf-2 mutant, but the changes were not preserved in the daf-16 or daf-16; daf-2 mutants. Notably, glycerolipid metabolism and glycerophospholipid metabolism were enriched for proteins with up-regulated phosphorylation in the daf-2 mutant. Up-regulation of lipid metabolism is a major phenotype of daf-2 mutants4, 24. Underlying this phenotype are gene expression changes25, 26 and protein abundance changes 8. Thus, our results should be highly informative in supporting characterization of biological processes regulated by IIS and are likely to extend the existing mechanistic understanding of this field to encompass PTM-level regulation.
Development of a Computational Strategy to Prioritize Putative Functional Phosphosites
A bottleneck in present-day biomedical research is a lack of efficient methods for extracting useful information from omics data27, 28. To facilitate the translation of phosphoproteomics data into biological insights, we developed a machine learning based method named iFPS (Inference of Functional Phosphorylation Sites) to predict whether a given phosphosite likely exerts a biological impact (Fig. 2A). Although the lack of suitable training data has to date prevented development of such a tool in C. elegans, note that a tool for similar predictive analysis of phosphorylation sites recently became available for Homo sapiens18. Our iFPS tool assesses six types of constraints for each phosphosite under examination (Supplementary Fig. 2A-H, see Methods), including 1) how many kinase families have consensus substrate sites that match its sequence context?; 2) how evolutionarily conserved a phosphosite is; 3) how many interacting domains are predicted to be influenced by phosphorylation at this site, 4) occurrence of a predicted acetylation site near the phosphosite (which could engage in PTM crosstalk); 5) relative surface accessibility; and 6) predicted secondary structure of the peptide containing the phosphosite.
For the initial iFPS training data, we searched in the literature for C. elegans phosphosites whose functions have been experimentally validated: the resulting 121 functional phosphosites served as the original positive training data set (Supplementary Table 2). The negative data set contained 605 (121 × 5) randomly selected phosphosites from dbPAF. Multinomial logistic regression (MLR), a widely used machine learning algorithm, was adopted for training the computational models, and 10-fold cross-validations were performed. The final model was determined automatically, and the highest area under the curve (AUC) value was 0.88 (Supplementary Fig. 2I). iFPS was applied to score all phosphosites identified in this study, which contain 31 known functional phosphosites from the positive data set (Supplementary Table 2). Distributions of iFPS scores show that functionally impactful phosphosites ranked higher than other phosphosites (Fig. 2B). Half of the functional phosphosites were among the top 5% iFPS scoring list.
Next, we focused on putative functional phosphosites regulated by daf-2. From the quantitation data reliability measured at least three times in both the daf-2 mutant and a control (WT or WT plus daf-16 mutants, see Methods), we found 222 down- and 226 up-regulated phosphoisoforms upon reduction of daf-2 activity (Fig. 2C). By overlapping the phosphosites regulated by daf-2 and the top 5% highest scoring iFPS phosphosites (Supplementary Table 2–3), we identified 27 high-priority phosphosites (i.e., with a high probability of being functionally impactful) (Fig. 2D). Notably, these sites do not represent a random set: the majority of the parent proteins harboring these sites function in either AMPK/insulin signaling, translation initiation/ribosome biogenesis, or cell cycle regulation. Further, 13 out of the 27 high-priority phosphosites are from eight proteins (AAK-2, AKT-1, CDK-1, DAF-16, EGL-45, MLT-3, MVK-1 and PDHA-1) known to regulate lifespan (phenotypic data from WormBase release WS275). Of note, the phosphorylation state of three conserved S/T residues of AAK-2, the catalytic subunit of C. elegans AMPK, was differentially regulated in the daf-2 mutant: phosphorylation of T597 and S601 increased whereas S553 decreased.
Any function(s) for most of the high-priority phosphorylation sites remain uncharacterized. In lifespan regulation, only S345 of DAF-16, a conserved AKT site, has been implicated: simultaneous mutation of S345 and other three predicted AKT sites induced nuclear accumulation of DAF-16, much like in the daf-2 mutant but without the extraordinary longevity phenotype13. To experimentally test the performance of iFPS and to flesh out the mechanism of lifespan extension by protein phosphorylation in response to reduced insulin signaling, we focused on several phosphosites for in-depth functional analysis (colored red in Fig. 2D). These are AKT-1 T492, EIF-2α S49, and CDK-1 T179, one in each of the three prominent protein function groups.
Constitutive Phosphorylation of AKT-1 T492 Promotes AKT-1 Activity
iFPS prioritized pT492 of worm AKT-1 (corresponding to pT450 of human AKT-1) (Fig. 3A). This site is positioned in a highly conserved turn motif near the AKT-1 C-terminus, and work in mammalian cells has shown that this site is co-translationally phosphorylated by mTORC2, supporting that this site may stabilize newly synthesized AKT29, 30. However, the functional impact of this site has not been confirmed.
Verifying the earlier suggestion, we found that phosphorylation of C. elegans AKT-1 on T492 is constitutive. The AKT-1 protein and T492 phosphorylation levels both doubled in the long-lived daf-2 mutant (FC = 2.2–2.4, daf-2/WT) as measured by shotgun proteomics (Fig. 2D and Supplementary Table 3) and by targeted quantitation assays using synthesized peptides bearing isotope labels (Fig. 3B). Whereas the T492-containing peptide of AKT-1 was undetectable in any of the four strains analyzed (Fig. 3B), the pT492-containing peptide of AKT-1 was readily detectable, and its abundance change followed that of the AKT-1 protein very closely. Thus, the T492 site is apparently constitutively phosphorylated following ATK-1 translation.
We used CRISPR/Cas9 to produce a T492A AKT-1 variant. Compared to the WT worms, those expressing the non-phosphorylatable T492A AKT-1 variant exhibited diverse phenotypes: akt-1-T492A mutant worms resembled weak IIS loss-of-function mutants such as akt-1(lf) or weak alleles of daf-2(lf). AKT-1-T492A caused nuclear accumulation of DAF-16::GFP in the intestinal cells of nearly 60% of worms, representing a 6-fold increase from the 9% detected in the WT animals (Fig. 3C) and indicating that phosphorylation of T492 promotes AKT-1’s ability to phosphorylate and thereby inhibit DAF-16. Consistently, the T492A mutation moderately but significantly extended the lifespan of WT worms by 8–17% (Fig. 3D), comparable to the 8–21% increase in lifespan conferred by the akt-1(lf) allele31. Notably, the pro-dauer formation effect of AKT-1-T492A is less obvious in akt-1(null): it failed to induce dauers at 27 °C. However, the T492A mutation did enhance the dauer formation phenotype in the sensitized background of daf-2(e1370) at 21 °C (Supplementary Fig. 3A).
To determine whether the loss-of-function phenotypes resulting from the T492A mutation are caused by destabilization of AKT-1, we used a knock-in approach to fuse a GFP reporter C-terminal to AKT-1. AKT-1::GFP and AKT-1-T492A::GFP were present in nearly all examined tissues, with no discernable difference in GFP intensity (Supplementary Fig. 3B-C), suggesting that T492A imparts no or little destabilizing effect on AKT-1. However, we did observe an effect related to the subcellular localization of AKT-1. Compared with AKT-1::GFP, there is more AKT-1-T492A::GFP in the nuclei of oocytes (Supplementary Fig. 3D). This T492A-induced localization change for AKT-1 was limited to the germline, and had high penetrance (84%). Moreover, this phenotype does not result from an overexpression artifact, because both AKT-1::GFP and AKT-1-T492A::GFP are expressed from the edited endogenous akt-1 gene locus. Since AKT-1 is normally recruited to the plasma membrane —where it transmits signals from receptor tyrosine kinases such as DAF-2—the nuclear translocation of AKT-1 may partially account for the observed loss-of-function effect of the T492A mutation. These results support that mutation of T492 to alanine impairs the activity of AKT-1, weakening AKT-1’s inhibition of DAF-16 and leading to both longer lifespan and a higher propensity for dauer formation. Thus, in WT animals, constitutive phosphorylation of T492 promotes the kinase activity of AKT-1.
AKT-1 is controlled by a negative feedback loop at the gene transcription level; that is, expression of the akt-1 gene is positively regulated by DAF-1632, while DAF-16 itself is negatively regulated by AKT-1. In the long-lived daf-2 mutant, activated DAF-16 induces transcription of akt-1, although the akt-1 mRNA level is elevated by only 10%33. However, this elevation is strikingly higher when examined at the protein level: the AKT protein level is elevated by around 140% as measured by quantitative proteomics6, a finding validated by our data for AKT-1::GFP in the present study (Supplementary Fig. 3E). Our phosphoproteomics analysis thus reveals T492 phosphorylation as a previously unknown layer of regulation in a complex regulatory network. Recalling that AKT-1 is phosphorylated at T492 immediately following its translation and that this PTM promotes AKT-1’s activity, our work at the phosphoproteomics level underscores how a negative IIS feedback loop is intricately controlled at multiple regulatory layers, including gene transcription, protein synthesis, and post-translational modification (Fig. 3E).
EIF-2α pS49 Potently Regulates Protein Synthesis and Lifespan in the daf-2 Mutant
Down-regulation of the processes that support protein synthesis (e.g., translation initiation and ribosome biogenesis) has been associated with longevity in previous studies34, 35, 36. The same down-regulation trend was evident in our phosphoproteomics data: phosphorylation of multiple eukaryotic initiation factors (EIF) was generally reduced in the long-lived daf-2 mutant (Supplementary Fig. 4A). The only exception to this trend was EIF-2α. Phosphorylation of EIF-2α at S49, which is an iFPS prioritized site, nearly doubled in the daf-2 mutant relative to WT worms (Fig. 2D and 4A), and this was verified by western blotting (Fig. 4B).
C. elegans EIF-2α S49 is a highly conserved site and is equivalent to human eIF2α S51, whose phosphorylation is known to block global mRNA translation37, 38. We thus asked whether mRNA translation is suppressed in the daf-2 mutant through hyper-phosphorylation of EIF-2α S49. We engineered an EIF-2α S49A mutation in the C. elegans genome using a CRISPR/Cas9 mediated gene editing method. Indeed, the S49A mutation, which locks EIF-2α in the dephosphorylation state, markedly increased the poly-ribosome fraction in the daf-2 mutant, albeit short of restoring it to the WT level (Fig. 4C). Further, the EIF-2α S49A mutation, which had no effect on WT lifespan, suppressed daf-2 longevity by 30% (Fig. 4D). These results suggest that enhanced phosphorylation of EIF-2α S49 in the daf-2 mutant may promote longevity by suppressing protein synthesis.
Next, we asked which kinase is responsible for hyper-phosphorylation of EIF-2α S49 in the daf-2 mutant. Mammalian eIF2α S51 may be phosphorylated by PERK, GCN2, HRI, or PKR38, among which only PERK and GCN2 have orthologs in C. elegans. We found that the gcn-2(lf) mutation significantly reduced EIF-2α S49 phosphorylation in the daf-2 mutant, while deletion of pek-1 had a weaker effect (Fig. 4E). Consistently, gcn-2(lf) suppressed daf-2 longevity (Fig. 4D) whereas pek-1(null) did not (Supplementary Fig. 4B). Therefore, we conclude that the GCN-2 kinase is responsible for the increased phosphorylation of EIF-2α S49 we observed in the daf-2 mutant and that GCN-2-mediated hyper-phosphorylation of EIF-2α S49 slows down protein synthesis in the daf-2 mutant to delay ageing.
Of note, two lines of evidence suggest that phospho-EIF-2α has a potent effect. First, a tiny amount of EIF-2α pS49, so low that it was undetectable by MS unless the phosphopeptides were enriched beforehand, is sufficient to generate the protein synthesis and lifespan phenotype. The S49 containing peptide generated by trypsin digestion from endogenous EIF-2α was only detectable and quantifiable by MS in the non-phosphorylated form in whole worm lysate samples (Supplementary Fig. 4C-E). Second, overexpression or knock-in mutation of the phospho-mimic EIF-2α S49D/E is lethal, suggesting a strong dominant effect of EIF-2α S49 phosphorylation. These target quantitation and genetics results both support that EIF-2α S49 phosphorylation has a potent inhibitory effect on protein synthesis and contributes substantially to daf-2 longevity (Fig. 4F).
Notably, our quantitative phosphoproteomics data also suggest that the observed EIF-2α pS49 increase of the daf-2 mutant (daf-2/WT = 1.83) may occur independently of daf-16: the EIF-2α pS49 increase was still observed upon deletion of daf-16 (daf-2; daf-16/ WT = 1.91) (Supplementary Fig. 4F). Pursuing this, it was surprising when we found that among the 448 phosphoisoforms which were differentially regulated in the daf-2 mutant (Fig. 2B), 124 apparently require daf-16, while 123 do not (Supplementary Table 3). This apparently very-well-balanced distribution of daf-16 dependent vs. daf-16 independent phosphorylation changes seems quite unique; to our knowledge, most of the documented changes in daf-2(lf) worms are dependent on daf-16. For example, two thirds or more of the protein abundance changes seen in the daf-2 mutant were suppressed by daf-16(lf)9.
Beyond EIF-2α, we characterized another EIF protein C37C3.2 (C. elegans eIF5). iFPS did not rank EIF-5 pT376 and pS380 among the top 5% (Supplementary Fig. 4G). The phosphorylation level of pS380 or pT376 pS380 either decreased or had no change, respectively, in the daf-2 mutant (Supplementary Fig. 4G). Simultaneous mutation of EIF-5 T376 and S380 to T375A S380A (2A) or T375E S380E (2E) by CRISPR/Cas9 had no or little effect on WT lifespan, and did not alter the lifespan of daf-2 (e1370 or RNAi) worms (Supplementary Fig. 4H-I). These findings indicate that, at least in the context of insulin-signaling-mediated lifespan extension, the two phosphosites of EIF-5 are not functionally impactful. At minimum, this result helps validate the utility of iFPS ranking as a hypothesis-generating tool to efficiently inform prioritization of candidates for functional studies.
CDK-1 and Other Germline Phosphoproteins Contribute to Lifespan Determination
CDK-1 is a master regulator of the cell cycle. For C. elegans germ cell division, CDK-1 is specifically required for entry into the M phase39. iFPS prioritized worm CDK-1 pT32, pY33, and pT179 (Fig. 2D), which respectively correspond to human CDK1 pT14, pY15, and pT161, (Fig. 5A). CDK-1 activity is inhibited by phosphorylation of T14 and Y15 by WEE1/MYT1, but is activated by phosphorylation of T179 by CAK40. In the daf-2 mutant, both inhibitory phosphorylation (pT32, pY33) and activating phosphorylation (pT161) of C. elegans CDK-1 decreased by 34–49%, while the CDK-1 protein level was about the same as that in WT worms (Fig. 5B).
Since the inactive form of CDK-1 (pT32 pY33) is not dominant negative, a reduced level of pT179 can be interpreted as a reduction of CDK-1 activity in the daf-2 mutant. Note that interpretation is supported by elaborate study of the daf-2 germline which reported a cell cycle delay in G2 in the proliferative zone; that is, proliferating daf-2 germ cells are slow to enter the M phase41. Importantly, all of our phosphoproteomics samples were synchronized to adult day one—a stage at which germ cells are the only dividing cells— so we can confidently assume that any detected CDK-1 activity must come from the germline.
We then asked whether the reduction of CDK-1 pT179 or CDK-1 activity in the daf-2 germline contributes to longevity. Mutating CDK-1 T179 to either A or E by gene editing was predictably unsuccessful: experimentally locking CDK-1 into either a completely inactive or a constitutively active state prevents cell cycle progression, causing lethality. We then took advantage of a temperature sensitive allele of cdk-1(ne2257ts) harboring an I173F mutation five amino acids away from T179 in the activation loop. We found that shifting cdk-1(ne2257ts) worms from the permissive temperature of 15 °C to the restrictive temperature 22.5 °C on adult day one significantly extended WT lifespan (by 11–30%), and noted that this extension was daf-16 dependent (Fig. 5C). We also found that temperature-shift-induced inactivation of CDK-1(I173F) at earlier time points extended WT lifespan (Supplementary Fig. 5A). Likewise, we observed an extended lifespan of 20–30% upon knockdown of cdk-1 starting from adult day one in the rrf-1(pk1417) mutant (in which RNAi is restricted in the germline, intestine, and some hypodermal cells42), whereas no extended lifespan phenotype resulted from intestine- or hypodermis-restricted cdk-1 RNAi in these animals (Supplementary Table 4). These results support that reduced CDK-1 activity in the adult germline is sufficient to promote a moderate lifespan extension.
Next, we investigated whether reduced CDK-1 pT179 in the adult germline is necessary for lifespan extension upon DAF-2 depletion. Using both gene editing and auxin-induced protein degradation (AID) technologies43, we were able to selectively degrade DAF-2 or WEE-1.3, or both, in the adult germline with high spatiotemporal precision. Degradation of WEE-1.3, the C. elegans ortholog of human WEE1/MYT140, should eliminate inhibitory phosphorylation of CDK1 on T32 and Y33 to drive an elevation of CDK-1 activity. Indeed, degrading WEE-1.3 specifically in the adult germline significantly shortened the lifespan of worms lacking germline DAF-2 (Fig. 5D). Moreover, both adult-specific and germline-specific degradation of DAF-2 slightly increased the mean lifespan and the maximal lifespan in two independent experiments, but not in a statistically significant manner. We thus conclude that reduced CDK-1 pT179 in the adult germline may confer a small contribution to daf-2 longevity.
It was highly striking that germline expression was predicted for the parent proteins of more than 70% of the iFPS-prioritized phosphosites (Supplementary Fig. 5B). Further, it was conspicuous that proteins of the reproductive system were highly enriched among the hypo-phosphorylated proteins detected in the long-lived daf-2 mutant (Fig. 5E). These findings motivated us to conduct a small-scale RNAi screen in the rrf-1(pk1417) mutant background to explore how germline phosphoproteins may affect ageing of the soma (Supplementary Fig. 5C). Interestingly, we found that adult onset RNAi of genes that promote mitosis or meiosis generally extended lifespan, whereas RNAi of genes that limit the genesis of germ cells or gametes shortened lifespan (Fig. 5F and Supplementary Fig. 5C). These results are in line with reports of lifespan extension through germline ablation44, and echo with the antagonistic pleiotropy theory of ageing. They also suggest that, although reduced CDK-1 pT179 alone contributes marginally to daf-2 longevity, the phosphorylation changes among all germline proteins may collectively confer a sizable contribution to lifespan extension (Fig. 5G).
Reduction of Casein Kinase 2 (CK2) Activity Prolongs Lifespan
Based on the hyper- and hypo-phosphorylated sites we detected in the daf-2 mutant, and in light of kinase-substrate relationships predicted with the iGPS algorithm45, we explored which kinases are likely to be more or less active upon reduced insulin signaling. Specifically, we used hypergeometric tests followed by Benjamini-Hochberg adjustment to assess whether the predicted or potential substrate sites of a given kinase were enriched among the differentially regulated phosphoisoforms of the daf-2 mutant. The hypo-phosphorylated sites displayed significant enrichment for putative substrate motifs (of 22 kinases), whereas no enrichment for kinase binding motifs was evident among the hyper-phosphorylated sites (Fig. 6A). There are studies for 5 of the 22 kinases reporting that RNAi or loss-of-function mutation extend lifespan (Fig. 6A), including investigations of C. elegans mTOR kinase LET-363 and the MAPK activated kinase MAK-2 and MNK-1.
Casein kinase 2 (aka CK2), was among the 22 kinases with a predicted activity decrease in the daf-2 mutant. In fact, there were two kinase binding motifs in CK2 which were significantly overrepresented among the hypo-phosphorylated sites found in the daf-2 mutant (Fig. 6B). Further implicating the likely impact of CK2 phospho-status in daf-2 longevity, an motif-x analysis46 detected CK2 but none of the 21 other kinases from our initial iGPS analysis. The C. elegans CK2 holoenzyme is composed of KIN-3, the catalytic subunit, and KIN-10, the regulatory subunit. CK2 has been shown to slow down ageing in C. elegans47, but studies in yeast revealed an opposite effect, reporting that the Saccharomyces cerevisiae CK2 accelerates both chronological and replicative ageing48, 49. We examined the lifespans of worms treated variously with kin-3 RNAi, kin-10 RNAi, or the CK2 inhibitor TBB. kin-3 or kin-10 knockdown during adulthood moderately but significantly extended WT lifespan in four independent trials (Fig. 6C and Supplementary Table 4). More strikingly, 24 and 48-hour TBB treatment (from adult day one) extended WT lifespan by 21–27% (Fig. 6D and Supplementary Table 4). These results demonstrate that inhibition of CK2 in young adults promotes longevity in C. elegans.