Introduction

Proteins play key roles in essential biological processes, ranging from the catalysis of biochemical reactions to the transport of nutrients, signal transduction and the maintenance of the cells mechanical structure [1]. Many proteins are modified post-translationally to regulate their structure, activity, subcellular localization and overall function. To date, over 80 000 unique sites bearing post-translational protein modifications (PTMs) have been reported, including acetylation, phosphorylation, glycosylation, methylation, and ubiquitination [2]. Many of these chemically different PTMs act mechanistically somewhat similar. Most PTMs are dynamic, whereby the modification is added to the protein by an enzyme, termed a writer, and removed again by another protein termed an eraser. The modifications can then be recognized by other proteins, referred to as readers [3, 4].

The most studied and, therefore, probably best characterized protein PTM is phosphorylation, a modification whereby a phosphoryl group is covalently linked predominantly to the hydroxyl groups of selected serine, threonine, and tyrosine residues [5]. Phosphorylation, however, has also been reported on histidine, lysine and arginine, or aspartate and glutamate [6, 7]. In the case of protein phosphorylation the writers are called kinases, the erasers phosphatases. At any given moment, approximately 30% of the human proteome is thought to be phosphorylated [8]. This phosphorylation can activate or deactivate a proteins function, affect its conformation or influence its localization [5, 8]. There are many readers of phosphorylation, such as those containing SH2 or WW-domains that can interact in a phosphorylation dependent manner [9, 10]. For example, several ubiquitin ligase complexes recognize specific, phosphorylated sequences on proteins known as phosphodegrons, that, once phosphorylated, target these proteins for degradation [11].

Another highly prevalent, yet lesser studied PTM, is O-GlcNAcylation. O-GlcNAcylation is a unique type of glycosylation whereby a single sugar moiety, O-linked N-acetylglucosamine (O-GlcNAc), is transferred typically to the hydroxyl groups of serine and threonine residues of proteins [12]. In sharp contrast to protein phosphorylation, for which hundreds of writers/kinases exist [13-15], there is only one writer for protein O-GlcNAcylation, namely the O-GlcNAc transferase (OGT). Moreover, in contrast to the plethora of coexisting protein phosphatases, there is also only one so-called eraser for O-GlcNAc, namely the O-GlcNAcase (OGA), and to date, although hypothesized to exist [16], no readers of O-GlcNAcylation have been identified. Analogous to phosphorylation, O-GlcNAcylation plays an important role in cell signaling and gene regulation [17-20] and more than 50% of all human proteins are expected to be glycosylated [21, 22].

Most protein PTMs occur on specific amino acids. Thus, evidently, a single protein can harbor multiple of the same PTMs (e.g., multiple phosphorylations), whereby the first modification may affect the addition or removal of the next. This is illustrated in the well-known example of substrates of the glycogen synthase kinase-3β (GSK3β), where prephosphorylation is needed, four amino acids upstream of the GSK3β phosphorylation site, before GSK3β will act to add an extra phosphate moiety to the substrate [23]. Similarly, a certain PTM can also affect a chemically different type of PTM, which has been coined as PTM crosstalk. This phenomenon is thought to be highly prevalent in cells, yet is still challenging to investigate. In general, one PTM can affect another in two ways, either promoting (i.e., a friend) or hampering (i.e., a foe) the presence of the other modification. This interplay between modifications can occur in many different ways, as summarized in Fig. 1. Crosstalk can be on the same writer, reader, or eraser, directly influencing the placement of a second PTM nearby or even one located much further away in sequence or in space. The placement of the second modification may then either enhance or hamper the activity of the targeted writer/eraser/reader. In recent years, several examples of crosstalk have been described for a wide array of proteins, ranging from kinases to transcription factors and from histone deacetylases to filament proteins [24-27]. Particularly well-studied examples of PTM crosstalk occur on histones/nucleosomes, highly conserved proteins that function in genome packaging. Histones are modified by at least eleven different PTMs, thus the number of types of PTM crosstalk on histones is immense, whereby the full extent of modifications and their crosstalk has been termed the histone code [28-31]. O-GlcNAcylation was also reported on histones, suggesting that O-GlcNAc is directly involved in the histone code [32-34]. However, detecting O-GlcNAcylation on histones is not straightforward and thus concerns have been raised as to the abundance of this PTM on histones in vivo [35].

In this review, we focus on the so far most documented form of PTM crosstalk, that between phosphorylation and O-GlcNAcylation. As both phosphorylation and O-GlcNAcylation occur mainly on serine and threonine residues, it is not very surprising that these two PTMs undergo crosstalk. Indeed, when over 800 phosphorylation sites were monitored upon the inhibition of OGA, phosphorylation on 280 sites decreased in abundance at the same time as phosphorylation on 148 sites increased. Also, changes in the O-GlcNAcylation pattern were observed upon inhibition of serine/threonine phosphatases [31]. The characterization of phosphorylated and O-GlcNAcylated proteins located at the Murine synapse revealed that 52% of the identified protein phosphatases were phosphorylated and 8% were O-GlcNAcylated. In addition, 66% of the identified kinases were phosphorylated, whereas 16% were O-GlcNAcylated, indicating that the modification of kinases and phosphatases is quite common [36]. These findings indicate that the relationship between phosphorylation and O-GlcNAcylation is complex and widespread. Here, we first describe in detail how O-GlcNAcylation is known to be regulated in cells. Secondly, we describe the effect that phosphorylation has on O-GlcNAcylation, detailing how these two modifications can affect each other. Indeed, many examples of crosstalk between phosphorylation and O-GlcNAcylation are emerging either occurring at the substrate level, or on the writers and erasers themselves. Most of these modifications directly modulate protein function. Finally, our review includes a description of the recent technological advances in the analysis and prediction of crosstalk between phosphorylation and O-GlcNAcylation.

Regulation of O-GlcNAcylation

Protein O-GlcNAcylation plays essential roles in the regulation of transcriptional activity [37-40], neuronal function [40], epigenetic regulation [41], modulation of protein–protein interactions [16], and the response to external stress factors [42]. Moreover, protein O-GlcNAcylation is very sensitive to the availability of uridine diphospho-N-acetylglucosamine, the GlcNAc donor for O-GlcNAcylation [43] and downstream metabolite of glucose, hence O-GlcNAcylation is often referred to as a nutrient sensor in cells [20, 43, 44]. Loss of O-GlcNAcylation was shown to result in loss of cellular function or even cell death [45] and the disturbance of normal O-GlcNAc function has been linked to many diseases, including Alzheimer's disease, diabetes, and other chronic illnesses [46, 47]. Therefore, the proper functioning of O-GlcNAcylation in cells is essential, however, how this regulation is fine-tuned is still not well understood. Only a single OGT enzyme is responsible for O-GlcNAcylation of over a thousand protein substrates [48]. OGT has three isoforms that differ in the number of tetratricopeptide repeats (TPRs) they harbor at their N terminus, a well-described protein–protein interaction-mediating domain [49]. Regulation may in part, therefore, be governed by (the number of) these TPR repeats. The OGT isoforms localize to different compartments of the cell. Short OGT (sOGT) and nucleocytoplasmic OGT (ncOGT) are found in the nucleus and cytoplasm, whereas mitochondrial OGT (mOGT) is found exclusively in mitochondria [50]. Localization of mOGT and ncOGT is directed by localization sequences. mOGT has an N-terminal mitochondrial localization sequence, whereas ncOGT contains a nuclear localization sequence between its TPRs and the catalytic domain located at the C terminus [51, 52]. Deletion of this sequence prevents transport of ncOGT into the nucleus, resulting in an increase of cytosolic O-GlcNAcylation [51]. The localization sequence in ncOGT is also present in the other isoforms, thus sOGT likely also has a localization sequence whose relevance is yet to be reported.

sOGT, mOGT, and ncOGT have 2.5, 9, and 11.5 TPRs, respectively. These TPRs are thought to be involved in substrate recognition [53], and removal of all TPRs abolished OGT activity towards protein substrates. Interestingly, the activity towards small peptide substrates in vitro remains unaffected upon TPR removal, indicating that the TPR repeats could play a specific role in mediating OGT substrate specificity [17, 54]. Indeed, interaction of OGT with myosin phosphatase target subunit 1 influenced OGT substrate specificity in vitro [55]. This principle is illustrated in Fig. 2, whereby OGT substrate specificity is defined by the initial complex formed with so-called adapter proteins. For example, OGT is recruited to specific promotors by the paired amphipatic helix protein Sin3A, resulting in transcriptional repression [56]. Furthermore, the interaction between OGT and p38, a mitogen-activated protein kinase, increases OGT activity by targeting OGT to Neurofilament H, as well as other targets [57]. Many other proteins have been shown to interact with OGT, however, the biological relevance of these interactions, and whether they play a regulatory role is often not yet fully understood [53, 55, 57-61].

Both writers and erasers act to regulate the presence of PTMs. The O-GlcNAc eraser, OGA, was identified by Hart et al. in 1994 and is mainly present in the cytosol [12, 62]. OGA catalyses the removal of O-GlcNAc from proteins via a substrate-assisted mechanism, utilizing the acetamido group of the sugar moiety [63, 64]. Thus, in addition to OGT, the regulation of OGA needs to be tightly controlled to prevent disease. Similar to OGT, the single OGA gene can encode two isoforms, namely a short isoform which appears inactive and the full length isoform [65]. The short isoform lacks a putative acetyl transferase domain located at the C terminus, however, OGA does not have acetyl transferase activity [12, 52, 66]. In addition, full-length OGA contains a caspase-3 cleavage site, although cleavage at this site does not reduce the enzymatic activity of OGA [67, 68]. Interestingly, OGA seems to form dimers and the intersubunit interactions differ between catalytically active and compromised OGA variants, indicating that the dimerization of OGA could also influence the binding of substrates and affect OGA activity [68-71].

The writers and readers of O-GlcNAcylation themselves, however, are not the only enzymes responsible for the regulation of O-GlcNAcylation in cells. Moreover, the interaction of OGT and OGA with various kinases and phosphatases hints that phosphorylation may additionally play a role in O-GlcNAc regulation. Indeed, as described above, it is known that writers and erasers can be modified themselves to increase or decrease their own enzymatic activity (Fig. 1A,E).

Phosphorylation/O-GlcNAcylation crosstalk at the enzyme level

There are ample examples reported of phosphorylation/O-GlcNAcylation crosstalk, whereby the modification of the writers and erasers of phosphorylation and O-GlcNAcylation themselves affect their own enzymatic capability (Table 1). For example, the calcium/calmodulin-dependent kinase IV (CaMKIV) was found to be phosphorylated at multiple phosphorylation sites, including Thr200 [72]. Phosphorylation at this site is required for CaMKIV activation [73, 74]. Dias et al. showed that CaMKIV is also O-GlcNAcylated on at least five specific residues, including Ser137 and Ser189. Mutation of Ser137 into an alanine resulted in a slight increase in CaMKIV phosphorylation at Thr200, however, mutation of Ser189 to alanine resulted in a significant increase of Thr200 phosphorylation. Moreover, a Ser189 CaMKIV mutant showed a five-fold increase in activity, indicating that O-GlcNAcylation of CaMKIV on Ser189 has an inhibitory effect on the kinase activity [72].

Another example was reported by Tarrant et al. [75], who demonstrated that phosphorylated Casein kinase 2α (CK2α) can readily phosphorylate Copine 1 isoform a, progesterone receptor membrane component 1 and the Golgin subfamily A member 4 protein, whereas O-GlcNAcylated CK2α could no longer phosphorylate these substrates. Inversely, O-GlcNAcylated CK2α could phosphorylate death-associated protein kinase 2, which was not phosphorylated by phosphorylated CK2α. Hence, the modification of CK2α by phosphorylation or O-GlcNAcylation affects CK2α substrate activity and thus specificity. Utilizing a more global approach, Hart and coworkers revealed that O-GlcNAcylation of kinases is a widespread phenomenon. They screened a protein array containing 152 kinases, confirming that 42 of them could be O-GlcNAcylated in vitro [76]. Recently, Shi et al. [77] showed that protein kinase B is O-GlcNAcylated at Thr308 and Ser473, which are known to be phosphorylated by phosphoinositide-dependent protein kinase 1 and mechanistic target of rampamycin complex 2, respectively [78]. Phosphorylation on these residues is required for protein kinase B activation [78, 79]. The overexpression of OGT revealed a decrease in phosphorylation at these sites, whereas overexpression of OGA led to an increase in phosphorylation, indicating a reciprocal relationship between O-GlcNAcylation and phosphorylation at these sites [77]. The actual relationship between OGT and protein kinase B is likely more complex than mere reciprocality, as OGT and OGA knockdowns have been reported to decrease and increase the level of phosphorylated protein kinase B, respectively [80, 81].

In addition to the extensive crosstalk on kinases induced by O-GlcNAcylation, the O-GlcNAc cycling enzymes OGT and OGA can also be post-translationally modified themselves, altering their activity. OGT can be phosphorylated on at least four different sites. Kaasik et al. [82] showed that GSK3β can phosphorylate OGT at Ser3 and/or Ser4, whereby these phosphorylations enhance OGT activity. Recently, Li et al. [83] showed that Checkpoint kinase 1 can phosphorylate OGT on Ser20, which seemingly improves OGT stability. In addition, AMPK was found to phosphorylate OGT on Thr444, which induced changes in OGT substrate specificity [34, 84]. Finally, tyrosine phosphorylation was reported for the rat homolog of OGT, possibly also occuring on Tyr989 on the human homolog [85]. Whelan et al. [86] showed that tyrosine phosphorylation of OGT is increased after insulin stimulation and that the isolated insulin receptor complex, as well as proto-oncogene tyrosine-protein kinase Src could phosphorylate OGT, resulting in an increase of OGT activity. Interestingly, OGT can also be O-GlcNAcylated on Ser3 and Ser4 [82], hinting at a potential reciprocal occupancy of these sites by O-GlcNAc and phospho modifications. OGA can be O-GlcNAcylated on Ser405 [87, 88], however, the functional effect of this modification is not yet known.

Phosphorylation/O-GlcNAcylation crosstalk at the substrate level

In addition to PTM crosstalk occurring on the ‘writers’ or ‘erasers’ themselves, crosstalk has also been reported whereby the same residue within a substrate protein sequence or residues in close proximity are modified by distinct PTMs (Fig. 1, Table 1) [89, 90]. Crosstalk whereby competition occurs for the same residue has been termed reciprocal crosstalk (Fig. 1D). Early evidence for the existence of reciprocal crosstalk came shortly after the discovery of O-GlcNAcylation by Hart et al. in 1984 when it was noted that O-GlcNAcylation occurs on known phosphorylation sites [91, 92]. The idea that such a reciprocal crosstalk mechanism could be more generic was validated by the fact that activation of protein kinase A and protein kinase C individually resulted in a global decrease of O-GlcNAcylation in cellular neurons and, inversely, inhibition of protein kinase A resulted in an overall increase of O-GlcNAcylation [93]. In line with these findings, the inhibition of GSK3β also resulted in an increase of O-GlcNAcylation in COS7 cells [94]. Now, many cases of reciprocal crosstalk between O-GlcNAcylation and phosphorylation have been identified, whereby several individual sites of modification have been mapped. Noteworthy examples include the modification of Thr58 by either phosphorylation or O-GlcNAcylation on the transactivation domain of c-Myc [95-97] and the modification of Ser16 on the N terminus of the murine estrogen receptor beta which functions to regulate its activity [98] (Table 1).

Phosphorylation/O-GlcNAcylation crosstalk also occurs on Ser/Thr residues that are in close proximity to each other along a protein sequence (Fig. 1B,F). In these cases, the O-GlcNAcylation or phosphorylation may have specific effects on the proteins function, whereby the PTM crosstalk can act to prevent or enhance one of these actions. For example, O-GlcNAcylation of p53 at Ser149 hampers phosphorylation at Thr155, a site targeted by the COP9 signalosome, finally resulting in p53 ubiquitination and degradation [99]. Indeed, this ‘proximity’ type of crosstalk features in many biological systems fine-tuning activity. For example, in the circadian clock, O-GlcNAcylation of the period circadian regulator 2 (PER2) at Ser662 in its regulatory region blocks casein kinase I-dependent PER2 phosphorylation [82]. Also in NF-κB signaling, p65 is O-GlcNAcylated at Thr305 and Ser319, where phosphorylation at Thr308 may prevent O-GlcNAcylation at Thr305 [100]. In addition, crosstalk was hypothesized to also occur between tyrosine phosphorylation and O-GlcNAcylation [101], which sites were later validated using a peptide microarray screening study by Pieters and coworkers [102]. One specific example occurs on STAT5, where O-GlcNAcylation of STAT5 controls tyrosine phosphorylation [103]. These disjointed examples show that phosphorylation/O-GlcNAcylation crosstalk at nearby sites in proteins is a generic feature in cellular signaling. The distance between the involved PTM sites in phosphorylation/O-GlcNAcylation crosstalk can be anything from a single amino acid apart [104] to six residues apart in the case of p53 [100].

Phosphorylation/O-GlcNAcylation crosstalk can even occur whereby the two PTMs are situated at quite distal sites with respect to one another (Fig. 1C,G). Although far away in terms of protein sequence, these modifications could still be spatially in close proximity, altering the ability of the ‘writers’ to interact with their substrates. A prime example of this comes from the insulin receptor substrate 1, whereby phosphorylation occurs at the N terminus and O-GlcNAcylation at its C terminus [86, 105, 106]. Alternatively, a PTM in one loop or domain may cause a conformational change in the protein effecting the whole protein. As in all types of crosstalk, distal crosstalk can both catalyze PTM addition, for example, in the case of the enzyme endothelial nitric oxide synthase, where phosphorylation at Ser615 enhances phosphorylation at Ser1177, and hamper PTM addition, for example the O-GlcNAcylation of Ser615 on endothelial nitric oxide synthase prevents phosphorylation at Ser1177 [107].

Prediction of phosphorylation/O-GlcNAcylation crosstalk

With serine and threonine being among the most frequently occurring amino acids in vertebrates, the number of residues that potentially could be modified by O-GlcNAcylation or phosphorylation, and thus be involved in PTM crosstalk, is in theory almost indefinite. Elucidating which residues can be modified, and when, is imperative for understanding cellular signaling and thus how cells function. Indeed, not all serine and threonine residues are modified. Most protein kinases target specific Ser/Thr residues to phosphorylate. Each kinase has evolved to recognize a specific amino acid sequence or structural motif. Developments in mass spectrometry (MS) and phosphopeptide enrichment methods [108, 109] have enabled thousands of phosphosites to be identified and site-localized within a single experiment [110]. These large datasets have been inferred to elucidate specific kinase motifs [111]. Thus, due to the wealth of information available on phosphosites, potential Ser/Thr phosphorylation on proteins can now be predicted with some degree of accuracy [112].

When it comes to O-GlcNAcylation, the substrate motif is more diffusely defined. The first prediction software for O-GlcNAcylation sites, termed YinOYang, was based on the observations that proline residues commonly occurred at the −4, −3, and −2 positions, with valines at the −1, +2, +4, and +5 positions and there was a high frequency of serine residues in the targeted stretch, typically at positions +1, +4, and +7 [113, 114]. It was also noted that leucine and glutamine residues were less frequently observed within the substrate motif. Although a reasonable start for O-GlcNAc prediction, this software was based on only 40 experimentally determined O-GlcNAcylation sites. A decade later, the number of O-GlcNAcylation sites identified on proteins had dramatically increased, primarily due to advances in MS-based detection of O-GlcNAcylation. From these data, a database of O-GlcNAcylated proteins (dbOGAP) was generated and a new prediction software emerged, termed O-GlcNAc scan, whose algorithm is based on ~ 800 experimentally determined O-GlcNAcylated proteins [115]. Although more accurate, both these methods are prone to false positive and negative predictions [116]. Most recently, by using machine learning to predict O-GlcNAc sites [117-119], O-GlcNAc sites identification has further improved with O-GlcNAc sites being predicted with supposedly up to 84% accuracy [119].

As the number of O-GlcNAcylation sites identified experimentally continues to rise, the motif (or motifs) for OGT will undoubtedly become clearer. To fast track the identification of OGT substrates, peptide microarrays have been developed, whereby thousands of proteins can be screened simultaneously for their ability to be OGT substrates [120, 121]. These arrays have helped to define many new important OGT substrates in vitro, which can subsequently be verified in vivo, although that is still much more laborious. For example, using such approaches OTX2, a transcription factor critical for brain development, whose misregulation is associated with the most malignant brain tumors in children, was found to contain multiple O-GlcNAcylation sites [121]. Observation of an O-GlcNAc site alone, however, does not provide information as to whether the protein of interest is a good or bad substrate of OGT. Indeed, good motifs for OGT should not only be based on sequence and detection alone but also on the efficiency at which these sequences can be O-GlcNAcylated on the native protein, ideally in vivo. Elegant work by van Aalten and coworkers addressed the O-GlcNAcylation efficiency, taking 720 peptides harboring a putative OGT site. They render a motif based on only the 32 most efficient O-GlcNAcylation substrates; giving [TS][PT][VT]S/T[RLV][ASY] [122]. Obviously, there are proteins O-GlcNAcylated that do confer to this rather strict O-GlcNAcylation motif. For example, Sox2, a transcription factor that functions in embryonic stem cell differentiation, has been found O-GlcNAcylated in mice at residue 248 within the sequence ₂₄₄SVVKSEASS₂₅₂ [39, 123], a sequence remarkably different to this ‘best substrate’ motif proposed. Such motif deviations hint at a more intricate involvement of the TPR domains of OGT in substrate specificity [53, 54, 124].

Although exceptions exist, sequence motifs can be defined wherein O-GlcNAcylation and phosphorylation can co-occur. Thus, crosstalk can be predicted by simply noting amino acid sequences within proteins that contain both the O-GlcNAcylation and phosphorylation motifs. Along these lines, Hart and coworkers developed software whereby sequences containing an O-GlcNAc motif were subjected to NetPhos, a computational software whereby phosphorylation sites are predicted based on the phosphorylation motifs of different kinases [125]. When a single residue had a positive hit for both O-GlcNAcylation and phosphorylation, that is, where reciprocal crosstalk could occur, it was termed a Yin–Yang site [113, 114]. This type of prediction works well, with Yin–Yang sites most commonly being predicted within PEST regions of proteins (peptide sequences rich in proline, glutamic acid, serine, and threonine) [114]. However, since the motif for O-GlcNAcylation is somewhat ‘fuzzy’, this yin-yang software likely also identifies false positive and false negative crosstalk sites. Thus, efforts more recently have turned to search for a reciprocal O-GlcNAcylation/phosphorylation crosstalk motif. By combining all experimentally identified PTMs, three motifs were extracted for O-GlcNAcylation/phosphorylation crosstalk; Pxx[S], Txxx[S], and [T]xxxxxxxxxP, whereby O-GlcNAcylation and phosphorylation occur on the same Ser/Thr site [126]. Interestingly, proteins containing these motifs were enriched in specific gene ontology terms such as nuclear transport, cytoskeleton and structure molecular activity, suggesting specific biological functions for PTM crosstalk [126]. Since many kinases are proline directed, a somewhat surprising observation came when Heck and coworkers observed that the presence of a proline residue at the P + 1 position compared to the O-GlcNAcylation site hampers O-GlcNAcylation, suggesting no reciprocal O-GlcNAcylation/phosphorylation crosstalk can occur on sites targeted by proline-directed kinases [127]. This work hints that reciprocal crosstalk likely only occurs with specific nonproline-directed kinases. However, there are notable exceptions. For example, c-Myc is O-GlcNAcylated at Thr58, a residue before a proline [95]. In addition, O-GlcNAcylation also occurs on the heptad repeats (YSPTSPS) in the C-terminal domain of RNA polymerase. This O-GlcNAcylation on RNA polymerase, however, only occurs efficiently when more than ~ 10 heptad repeats are present [128, 129], thus factors in addition to sequence alone likely govern which proteins can undergo reciprocal O-GlcNAcylation/phosphorylation crosstalk.

Predicting crosstalk on proximate sites is significantly more challenging than reciprocal crosstalk since the observation of O-GlcNAc and phosphate on a peptide in close proximity does not necessarily mean that crosstalk is occurring. Moreover, there are likely multiple factors governing whether crosstalk exists such as the location of the O-GlcNAc/phosphorylation site with respect to the other, and the order in which the specific sequence is post-translationally modified. Additionally, as mentioned previously, both positive and negative crosstalk can occur. Aside from this, patterns whereby crosstalk occurs on adjacent sites are starting to emerge. However, defining a motif for crosstalk on adjacent sites will take more effort since each site of modification must first be exactly site-localized and then the nature of the crosstalk (positive/negative) determined in a controlled manner. Kinetic-based MS assays have helped accelerate this process [127]. Leney et al. noted that the most common O-GlcNAcylation motif also contained a kinase motif whereby the serine/threonine at the −3 position with respect to the O-GlcNAcylation site could be phosphorylated. Thus, the rate of O-GlcNAcylation with and without a phosphate residue at the -3 position (and vice versa, that is, the rate of phosphorylation with and without an O-GlcNAc residue at the +3 position with respect to the phosphorylation site) was monitored using MS on several OGT substrate-mimicking peptides. In all cases, negative crosstalk was observed within the sequence [S/T]P[V/A/T][S/T]X_-p where X_-P represents any amino acid except proline. Indeed, this type of negative crosstalk is likely not uncommon with about one thousand sequences in the PhosphoSitePlus database containing precisely this putative crosstalk motif [127].

Future directions

The majority of experiments that investigate phosphorylation/O-GlcNAcylation PTM crosstalk to date either focus on how one protein's PTM affects another protein's structure/function, or look more globally at the cellular level on how phosphorylation affects O-GlcNAcylation or vice versa. In the case where PTM crosstalk is monitored extensively on a single protein or a few proteins, the proteins of interest are made recombinantly or overexpressed in large quantities enabling biochemical experiments such as enzymatic assays or structural studies using X-ray crystallography or nuclear magnetic resonance to be carried out. These data together provide a wealth of information on how each specific PTM alters the proteins structure alone and how this may be altered by PTM crosstalk. Moreover, this information is key to decipher how PTMs alter protein function. Indeed, there are no generic mechanisms on how a specific PTM alters protein function. For example, phosphorylation can both activate (e.g., in the case of some kinases whereby an activation loop needs to be phosphorylated for the kinase to be active [130, 131]) or inactivate proteins (e.g., in the case of Src tyrosine kinase whereby its phosphorylation results in a conformational change and its deactivation [132]). Although fundamentally important, expressing and characterizing each protein individually can be challenging and time consuming, and thus it will take decades to monitor each protein in the human genome. In addition, proteins act differently when removed from their cellular context where potential cofactors are absent. Thus, to accelerate the identification of PTMs and their roles in the context of the cell, we need to start to look at PTMs and PTM crosstalk at a system-wide level.

Mass spectrometry is an ideal method to detect multiple PTMs as each PTM can be uniquely detected based on a change in mass or fragmentation pattern [133, 134]. Using this technology, thousands of PTMs can now be detected within a single experiment [135, 136]. Unfortunately, the detection of some PTMs by MS is not straightforward. For example, on peptides the O-GlcNAc moiety is labile in the gas phase and thus O-GlcNAcylation went undetected for a long time before fragmentation techniques such as electron transfer dissociation and electron capture dissociation came about [137-139], as extensively reviewed [140-142]. This aside, PTMs are also dynamic modifications and are often present in low stoichiometric amounts hampering their analysis. To circumvent this, enrichment strategies are often employed. Thus, the development of methodology goes hand in hand with the identification of novel PTMs and PTM crosstalk sites. Also, enrichment methodologies employed in case of O-GlcNAcylation or phosphorylation have been extensively reviewed [108, 143].

To detect instances of where positive PTM crosstalk might occur on adjacent sites, peptides would ideally be detected harboring both modifications simultaneously. Enrichment techniques, however, are currently specialized to target a single PTM of interest. Thus, the chances of finding both modifications especially if they have different chemical properties are slim. In addition, proteomics methods commonly use enzymes such as trypsin for digestion that produce relatively short peptides. Thus, relatively few peptides that encompasses sequences that could contain multiple PTM sites are identified. In addition, through digestion, even if multiple PTMs are detected, information is lost as to whether these PTMs were colocalized on the same protein molecule. Thus, scientists need to think about developing methodologies to not only detect PTMs but that also aims to look specifically for PTM crosstalk.

To increase the chance of finding multiple PTMs on the same peptide and maintain information with regard to PTM co-occurance, one could employ MS techniques that analyze longer peptides, protein subunits, or even protein complexes [144-147]. In recent work by Sidoli et al., middle-down proteomics was used to analyze histone peptides of roughly 60 amino acids in length, about three times larger than regular tryptic peptides. These longer peptides enable the researchers to reveal many instances of co-occuring methylations and assess whether these PTMs are observed on proximal sites more or less frequently than expected based on their abundance and thus whether the PTMs undergo interplay [148, 149]. The analysis of intact proteins using top-down MS has also been successfully applied to help unravel histone code [150]. These techniques not only maintain valuable information on modifications co-occuring on the same protein but also require more purified samples than required in traditional, bottom-up approaches.

These examples show that MS is a very powerful tool to detect multiple, chemically distinct PTMs simultaneously and it can even reveal PTM crosstalk by the detection of modifications co-occuring on the same protein, by relying on less conventional MS-based techniques. The analysis of positive crosstalk, however, represents the simplest of cases. Negative PTM crosstalk adds yet another level of complexity as the lack of detection of one modification does not mean it does not exist. Hence, negative crosstalk can only be identified if the individual modifications are present at sufficient abundance in the sample and therefore it is currently the most challenging type of PTM crosstalk to study.

Summary

In conclusion, the crosstalk between PTMs is a general phenomenon, which occurs on many different proteins and protein complexes [29, 30, 82]. Crosstalk can occur on different levels, ranging from the modification of writers and readers (i.e., enzyme level) to proximal and distal sites on modified proteins, resulting in astounding combinatorial PTM complexity. Modifications can influence each other positively (friends) or negatively (foes) [29, 90]. Here, we have focused on phosphorylation and O-GlcNAcylation, briefly reviewed the regulation of these modifications, and discussed examples of the different types of crosstalk observed between these PTMs. Finally, we have addressed the advancements in the prediction of crosstalk and highlighted techniques that may aid in the future search for PTM crosstalk and its functions. Overall, we believe that researchers should not only look to control the presence of a single PTM in the prevention of disease but also be mindful that looking elsewhere for other PTMs involved in crosstalk either in close proximity or distally to the PTM of interest could offer an alternative therapeutic target. Indeed, if one particular PTM is proving challenging to control, why not target another whose mechanism is fully established?