Uncovering Post-Translational Modification-Associated Protein-Protein Interactions

Abstract

In living systems, the chemical space and functional repertoire of proteins are dramatically expanded through the post-translational modification (PTM) of various amino acid residues. These modifications frequently trigger unique protein-protein interactions (PPIs) – for example with reader proteins that directly bind the modified amino acid residue – which leads to downstream functional outcomes. The modification of a protein can also perturb its PPI network indirectly, for example, through altering its conformation or subcellular localization. Uncovering the network of unique PTM-triggered PPIs is essential to fully understand the roles of an ever-expanding list of PTMs in our biology. In this review, we discuss established strategies and current challenges associated with this endeavor.

Introduction

Although proteins are generally composed of only twenty canonical amino acids, the chemical space available to them is dramatically expanded in living systems by post-translational modifications (PTMs) of various amino acid side residues.[1] Such PTMs provide a powerful way to augment and regulate protein function.[1–4] The modified amino acid residue is often recognized by specific “reader” and “eraser” proteins (Figure 1), and these unique protein-protein interactions (PPIs) govern the complex regulation of the functional outcome associated with the PTM.[2,5] In addition, PTMs may also trigger unique PPIs indirectly; for example, through inducing dynamic structural changes or altered subcellular localization. Consequently, orchestrating new functionally important protein-protein interactions is a central mechanism underpinning the regulatory role played by many PTMs in our biology.

Figure 1.

PTMs can lead to novel PPIs either directly or indirectly

Over the last two decades, rapid advances in proteomics have dramatically expanded the catalog of known PTMs and the sites in the proteome that are subjected to them,[6,7] highlighting a significantly larger role for these modifications in our biology. To fully understand their function, it is essential to systematically identify the unique set of PPIs that are triggered in response to a particular PTM. Doing so would depend on two key abilities: 1) to model the presence/absence of the PTM on a target protein, and 2) to reliably characterize the PPIs associated with each state. However, these were historically challenging to achieve for multiple reasons.[8] The biochemical origins of many PTMs are poorly understood. Even when the mechanism is known, reconstituting them cleanly to generate homogeneously PTM-labeled proteins is frequently not possible. This challenge makes it difficult to investigate the consequences of the PTM, including the identification of PTM-associated interaction partners. Furthermore, the strengths of different PTM-dependent PPIs span a wide range,[9] and many are insufficiently strong to allow the isolation of non-covalently bound interaction partners from the complex cellular milieu via pull-down or immunoprecipitation.[10,11] Here we discuss contemporary strategies that are being developed to overcome these limitations. We first focus on approaches to reliably model the presence or absence of a PTM on proteins, followed by strategies to identify PTM-specific interaction partners.

Strategies to model the presence or absence of PTMs on proteins

Some proteins are post-translationally modified at significant levels in cells under known conditions, such as upon the activation of a signal transduction pathway. In such cases, the naturally modified protein can be used as a probe to identify both general and PTM-specific interaction partners. To uniquely identify the latter group, a control protein-probe without a PTM can be used, which can be generated either by mutating the site of modification, or by knocking out/down the ‘writer’ protein that installs the PTM. This strategy was used to reveal that BARD1 binds p50 when the latter is phosphorylated at Ser337,[12] and to show that cereblon binds and degrades glutamine synthetase when the latter is acetylated at lysine 11 and 14.[13] However, this strategy is limited to modifications that occur naturally at significant levels.

The use of short peptides harboring the PTM in its native context is one of the most popular approaches for investigating PTM-triggered PPIs. Such modified peptides with defined sequences can be readily accessed through chemical synthesis. Proteins that bind post-translationally modified residues often retain significant affinity and selectivity for such minimal recognition motifs. Such peptides probes have been used to systematically screen for binding by suspected protein domains in many studies, such as for identifying YEATS and double PHD finger domains as readers of crotonyllysine,[14,15] and GAS41 as a reader for succinyllysine.[16] Conversely, modified peptide probes have also been useful for identifying the target sequence for a particular PTM-binding domain.[17,18] Peptide probes have also been valuable for evaluating crosstalk between two different PTMs within the same motif.[19,20] Moreover, modified peptide-probes have also been employed in a high-throughput manner using, for example, peptide microarrays,[21–23] or in combination with quantitative mass-spectrometry following a pull-down.[24] There are countless other examples where peptide probes have been useful to characterize PTM-dependent PPIs.[25] Although peptide based probes are popular due to their ease of access, these do not capture the context of the full protein, which may partially or completely compromise their ability to bind native interaction partners.[26,27] Furthermore, peptide probes are not suitable to investigate PTM-triggered PPIs that do not directly bind the modified residue. These limitations have inspired efforts at generating full-length proteins homogeneously labeled with PTMs at chosen sites.[8]

Using a natural amino acid that ‘mimics’ the chemical properties of the modified residue is another popular strategy to model PTMs. For example, negatively charged aspartic acid and glutamic acid residues are often used to mimic the effect of phosphorylation, charge-neutral glutamine has been used to model lysine acetylation, and positively charged arginine is used to mimic an unacylated lysine residue.[12,28–32] However, such substitutions may not always accurately reproduce the consequences triggered by the actual PTM because of significant structural differences. Furthermore, this approach cannot be used for the vast majority of PTMs that are not structurally similar to one of the canonical amino acids.

A PTM can also be introduced onto proteins using the natural ‘writer’ enzyme(s) responsible for its installation, provided the pathway is established and can be reconstituted.[33,34] Engineered variants of such ‘writer’ proteins, many of which modify numerous substrate proteins, can be developed to target specific proteins. For example, it was possible to achieve protein-selective O-GlcNAcylation by fusing O-GlcNAc transferase (OGT) to a nanobody that selectively binds a target protein.[35] However, biochemical origin of many PTMs are either unknown or hard to reconstitute to modify only the desired site(s), compromising the general utility of this strategy.

Protein semisynthesis using expressed protein ligation (EPL) enables the assembly of full-length proteins from synthetic peptides and recombinantly expressed protein fragments.[36,37] Using this approach, it has been possible to access proteins site-specifically labeled with many different PTMs. For example, this strategy has been used to generate site-specifically succinylated,[38] glutarylated,[39] and ubiquitilated[40] histones to evaluate how these PTMs affect their interaction with other chromatin components. In another example, protein semisynthesis was used to generate Ubc9, an E2-ligase, that is site-specifically modified with SUMO, and further equipped with a photo-affinity probe and a biotin group. This trifunctional protein probe was used to covalently capture the E3 ligase RanBP2 that associates with this complex.[41] Although protein semisynthesis has been widely used to study PTM biology, there are several challenges limiting its scope. This strategy is often restricted to proteins that are smaller in size, and those that can survive the demanding workflow. For example, EPL is typically performed under denaturing conditions, and proteins that are not easily refolded may not be amenable to this technology.[37] Finally, application of this technology is restricted largely to in vitro experiments, precluding the investigation of delicate PPIs that require the context of a living cell.

Genetic code expansion (GCE) technology has emerged as a powerful strategy for generating full-length proteins homogeneously incorporating PTMs at predefined sites.[42,43] This technology uses engineered nonsense-suppressing aminoacyl-tRNA synthetase (aaRS)/tRNA pairs to enable co-translational site-specific incorporation of non-canonical amino acids (ncAAs) into proteins in living cells. Engineered aaRS/tRNA pairs have been developed to genetically encode numerous PTMs of interest, including several lysine PTMs,[44–46] as well as PTMs of tyrosine,[47–51] serine,[52,53] threonine,[54] and arginine.[55] It has been also possible to genetically encode ncAAs that structurally mimic the PTM, but are resistant to removal by endogenous machinery.[49,53,56] A significant advantage of the GCE technology is the ability produce homogeneously modified proteins directly in living cells, which may enable capturing delicate PPIs that require the cellular context. However, only a subset of known PTMs can be currently incorporated using this technology. PTMs with large and complex structures remains particularly challenging to encode, as it has been difficult to engineer existing aaRSs to recognize such ncAAs. The unpredictable ncAA incorporation efficiency is another challenge commonly encountered with the GCE approach.

GCE and other site-specific protein modification approaches have also been used to introduce unique functionalities into proteins that can be subsequently leveraged to chemically/enzymatically install a PTM mimic. For example, taking advantage of its unique reactivity and relatively low abundance, engineered cysteine residues have been chemoselectively functionalized to incorporate mimics of lysine acetylation and methylation.[57,58] Chemoselective modification of dehydroalanine, which can be generated from cysteine or selenium-containing ncAAs, has also been used to introduce mimics of various PTMs onto proteins.[59] Moreover, similar post-translational selective functionalization approaches have facilitated the introduction of more complex modifications such as glycosylation, [60,61], SUMOylation, [62] and uniquitination.[63]

Approaches to identify PTM-associated PPIs

Potential interactors for a particular PTM can sometimes be predicted based on existing biochemical, structural or genetic information, or through homology analysis with established PPIs targeting structurally similar PTMs. In such cases, the suspected proteins can be screened to confirm binding using peptide or protein probes (Figure 2A). For example, reader domains for novel lysine PTMs crotonylation and succinylation were confirmed by screening predicted protein domains for binding PTM-labeled peptide probes in vitro.[14–16] The limitation of this approach is the requirement of preexisting insight.

Figure 2.

A-C show different approaches to identify novel PTM-associated PPIs. D) Structures of ncAAs with photo-crosslinkers frequently used to capture PPIs. E) Novel lysine-derived ncAAs, harboring a diazirine group, which can be incorporated into proteins and used to capture interactors that bind this residue

Performing binding screens at a much higher throughput can enable identification of binding partners in the absence of a preexisting insight. Using synthetic peptide microarrays, higher throughput screening can be performed in vitro.[21,23] The ability to genetically encode various PTMs has opened the door to adapting established high-throughput cell-based PPI assays [64] for identifying PTM-dependent PPIs. For example, it has been possible to express a library of thousands of predicted phosphoserine-containing peptides from the human proteome in E. coli, using co-translational incorporation of phosphoserine, and screen these for binding potential interactors using a high-throughput fluorescence-based two-hybrid screen.[65] Other examples include the identification of Nop56p and Nop58p as arginine-methylation-dependent interaction partner of Nop1p through two-hybrid assay,[19] and of a PPI between RhPIP2 and RhPTM that is regulated by phosphorylation through a split-ubiquitin membrane yeast two-hybrid (MYTH) assay.[66]

Using the modified peptide or protein as a bait to pull-down or immunoprecipitate non-covalently bound interaction partners from the complex cellular milieu, followed by their identification using immunostaining or mass-spectrometry, is a popular strategy to identify unknown PPIs (Figure 2B). From such experiments, PTM-specific interactors can be identified by their differential enrichment upon using protein/peptide probes either with or without the PTM. For example, this strategy was used to identify readers of trimethyl-lysine,[24] to confirm that phosphorylation of PKM2 by ERK2 at Ser 37 recruits PIN1,[67] that phosphorylation of OPTN by TBK1 enhances its binding to ubiquitin chains,[32] that Wnt5a activation of ROR1 induces binding to 14-3-3ζ,[68] and that lysine-acetylated glutamine synthetase is a substrate for cereblon.[13] A key limitation of this approach is that many PTM-specific PPIs are not strong enough to survive the isolation step.

An attractive way to overcome this limitation is the use of photo-affinity probes, which can enable the capture of proximal non-covalent interactors through the formation of a stable covalent linkage (Figure 2C). Established photo-affinity probes such as aromatic azides, benzophenone, and diazirines (Figure 2D) can be readily incorporated into chemically synthesized peptides. Peptide probes harboring the photo-crosslinking functionality, in the presence or absence of the PTM, have been used in conjunction with quantitative MS-proteomics to identify PTM-specific interaction partners.[69–75] Fundamental limitations of peptide probes, as described earlier, restricts the scope of this approach.

The GCE technology has enabled site-specific incorporation of ncAAs harboring photo-affinity probes into full-length proteins (Figure 2D), which has been adapted for interrogating PTM-associated PPIs. For example, a lysine analog containing the small diazirine photo-crosslinker group on the sidechain was incorporated across the proteome (Figure 2E).[76] Two variants of the same lysine-analog were also incorporated site-specifically, carrying either a photocage or the crotonyl modification on N^ε (Figure 2E).[77] The small built-in diazirine group in both cases enabled capture of lysine-PTM-specific PPIs. Application of this approach is currently restricted to lysine PTMs. and it is conceivable that the built-in diazirine group may perturb the native binding interactions at the proximal N^ε in some cases. Incorporation of two distinct ncAAs – one encoding the PTM of interest and the other harboring the photo-crosslinking probe – has been recently reported.[78] This strategy allows flexible placement of the photo-crosslinker relative to the PTM in a full-length protein to systematically optimize capture efficiency, and it can be extended to other genetically encoded PTMs. Furthermore, this technology is uniquely suited to capture PPIs indirectly triggered by PTMs. However, application of this technology can be somewhat limited by the low efficiency of incorporating two distinct ncAAs. For protein-based PTMs such as SUMO, the photo-crosslinker ncAA has been directly incorporated into it to identify interacting proteins.[79] Conversely, genetically encoded photo-crosslinkers have also been incorporated into the binding site of known PTM-reader proteins to capture previously unknown targets.[80,81] It should be noted that photo-crosslinking groups typically have a short capture radius, making their crosslinking efficiency highly sensitive to the site of incorporation. Screening multiple sites are sometimes necessary to achieve efficient capture of unknown targets.[78] Finally, in addition to photoaffinity probes, genetically encoded chemical crosslinkers have also been used to capture interaction partners. For example, an electrophilic tyrosine derivative was used to covalently capture phosphatases that erase phosphorylation of distinct tyrosine residues of the HER2 receptor.[82]

Conclusion

Even though significant progress has been made in the last few decades to define the dynamic PPI networks that are regulated by PTMs, it is evident that our current understanding is limited. This is especially true for the rapidly expanding catalog of newly discovered PTMs. It is challenging to develop a single approach to address this issue that would be suitable for all PTMs, given the remarkable diversity of their structure and properties. As this review highlights, many different creative approaches have been developed to explore PPIs associated with different PTMs. In addition, computational approaches are also being explored to predict PTM-associated PPIs.[83–85] Expanding the current scope of these diverse approaches would be critical to broaden our understanding of the vast network of PTM-associated PPIs that underlie all aspects of our biology. In particular, it will be important to install previously inaccessible PTMs, or their structural analogs, into full-length substrate proteins, both in vitro and in living cells. A combination of engineered PTM-writers, the ability to enhance the endogenous PTM levels, an expanded toolbox of GCE platforms, and cellular delivery of modified proteins synthesized in vitro would be beneficial to this end. It will be also important to expand our capability to capture PPIs associated with the modified form of the protein through improved covalent capture methods, as well as advanced proteomic technologies for characterizing proteome-wide protein-protein interactions.

Table 1.

Strategies to model the presence and absence of PTMs on proteins

Strategy	Advantages	Limitations	Representative references
Proteins naturally modified in cells under known conditions	• Straightforward to access with few manipulations • The absence of PTM can be readily modeled by mutating the target site or knocking out/down the ‘writer’ protein	• Conditions leading to many PTMs are not known • Degree of endogenous modification is often heterogeneous • Site-specific modification may be difficult to achieve	[12,13]
Use of small synthetic peptides harboring the PTM in its native context	• Readily generated through chemical synthesis • Wide variety of modifications can be accessed • Possible to model multiple PTMs within the same peptide • Can be equipped with photo-crosslinkers • Can be used in higher throughput format such as peptide microarray	• Does not capture the context of the full-length folded protein, which may partially or completely abolish interaction with partners • Cannot be used to capture PTM-triggered PPIs that do not directly involve the modified residue	[14–25]
Use of natural amino acids that mimic a PTM	• Easy to implement through straightforward mutagenesis • Introduces modification at specific site(s)	• Significant structural differences often result in imperfect mimicry • Most PTMs cannot be modeled using a natural amino acid	[12,28–32]
Using the endogenous ‘writer’ protein(s) to install a PTM	• Can be relatively simple to generate using recombinant writers • Engineered writers can be generated to target specific substrate proteins	• Biochemical origin of many PTMs are unknown or hard to reconstitute • Can be difficult to homogeneously modify specific site(s) • Some PTMs are not enzymatically installed	[33–35]
Protein semi-synthesis through expressed protein ligation (EPL)	• Homogeneously modified full-length proteins can be generated • A wide variety of natural/synthetic modifications can be installed • Multiple, different modifications can be installed	• Technically demanding • EPL is performed under denaturing conditions; refolding the resulting protein may be challenging • Typically precludes experiments in living cells • Internal modifications on larger proteins are challenging to access	[36–41]
Genetic code expansion (GCE)	Homogeneously modified full-length proteins can be generated Close structural mimics of PTMs can be incorporated that are resistant to removal Modified proteins can be expressed in living cells The modified residue can be theoretically incorporated into any site of any protein that can be recombinantly expressed	A limited number of PTMs have been genetically encoded Efficiency of incorporation is site-dependent Incorporation of multiple modifications into one proteins can be challenging	[42,43,45–50,52–56]

Table 2.

Approaches to identify PTM-associated PPIs

Strategy	Advantages	Limitations	Representative references
Screening potential interaction partners for binding modified peptide/protein probe	• Straightforward workflow • Can be employed at a higher throughput using strategies like peptide microarray, or two-hybrid analyses	• Requires existing insight into possible potential interactors	[14–16,21–23,64]
Immunoprecipitation or pull-down of non-covalently bound interactors using modified peptide or protein probe	• Established workflow; widely used • Can be used in complex milieu, such as cell-free extract, to isolate native complexes • Can be coupled with quantitative MS experiments to characterize binding partners with high confidence • Enables the identification of unanticipated interactors	• Interactions that are weak or transient are challenging to detect, as these do not survive the isolation step	[13,24,32,67,68]
Small peptide probes harboring the PTM and a photo-crosslinker group	Almost any short sequence, with a wide variety of modifications, can be readily generated through synthesisEnables the identification of weaker binding interactionsCan be used in complex milieu such as cell-free extract to isolate interaction partnersCan be coupled with quantitative MS experiments to characterize binding partners with high confidence• Enables the identification of unknown interactors	• The photo-affinity probe may perturb the binding interaction • Photo-affinity probes have a short capture radius, making the cross-linking efficiency site-dependent. • Limitations intrinsic to small peptide probes still apply	[69–75]
Full-length proteins harboring genetically encoded photo-affinity probes: The same ncAA encodes both the PTM and photo-crosslinker	• Homogeneously modified full-length proteins can be generated in living cells • The proximity of the photo-crosslinker group to proteins directly binding the PTM is favorable for crosslink formation • Incorporation of a single ncAA introduces both the PTM and the photo-affinity probe	• Currently restricted to lysine PTMs • The photo-crosslinker group may perturb interaction with binding partners • Cannot be used to capture PTM-triggered PPIs that do not directly involve the modified residue	[76,77]
Full-length proteins harboring genetically encoded photo-affinity probes: Two different ncAAs encode the PTM and the photo-crosslinker group, respectively	• Homogeneously modified full-length proteins can be generated in living cells • Flexibility to optimally position the photo-crosslinker relative to the PTM for high capture efficiency • Enables the capture of PPIs indirectly triggered by the PTM	• Efficiency of incorporating two distinct ncAAs into one protein can be low • Can only be used for PTMs that have been already genetically encoded	[78]