Interaction Profiling Methods to Map Protein and Pathway Targets of Bioactive Ligands

Abstract

Recent advances in –omic profiling technologies have ushered in an era where we no longer want to merely measure the presence or absence of a biomolecule of interest, but instead hope to understand its function and interactions within larger signaling networks. Here we review several emerging proteomic technologies capable of detecting protein interaction networks in live cells, as well as their integration to draft holistic maps of proteins that respond to diverse stimuli, including bioactive small molecules. Moreover, we provide a conceptual framework to combine so-called “top-down” and “bottom-up” interaction profiling methods and ensuing proteomic profiles to directly identify binding targets of small molecule ligands, as well as unbiased discovery of proteins and pathways that may be directly bound or influenced by those first-responders. The integrated, interaction-based profiling methods discussed here have the potential to provide a unique and dynamic view into cellular signaling networks for both basic and translational biological studies.

Introduction

Bioactive small molecule ligands lie at the center of biological processes, serving as substrates and co-factors in metabolism, diffusible signals, structural components and pharmacologic agents. These roles all originate from direct interactions with other biomolecules within and between cells, tissues and organisms. Therefore, a central challenge in determining the mechanisms of action for a small molecule of interest is to identify the direct biomolecular target(s) with which it physically interacts at pharmacologically relevant concentrations. Beyond providing insights into mechanism of action, for example in the follow-up from high throughput screens, there are myriad examples where target identification has led to discoveries in basic biology and expanded opportunities for the development of new therapeutics. These include classic affinity purification methods, such as those used to identify the protein targets of natural products like FK506[1], Trapoxin[2], and Staurosporine[3]. A similar, contemporary example is the discovery of ubiquitin-ligase recruiting “molecular glues” [4] [5, 6], which has led to a potentially new class of small molecules that promote catalytic degradation of their protein targets [7–10]. Likewise, efforts to identify the targets of endogenous ligands like the “oncometabolite” (R)-2-hydroxyglutarate[11], anti-inflammatory metabolites like itaconate[12, 13], and reactive glycolytic metabolites like methylglyoxal[14, 15] have resulted in the identification of new signaling pathways that operate under normal and pathophysiologic contexts. Beyond those produced in mammalian cells, a recent surge in the discovery of novel microbial metabolites, including short chain fatty acids[16], modified bile acids[17], and diverse natural product classes [18, 19], suggests a rich and broad landscape of small molecule-mediated signaling between microbes and host tissues. While the relevant targets of bioactive small molecules span proteins, RNA, DNA and other biomolecules, the vast majority of annotated small molecules target proteins, and therefore the primary focus of this review will be on the emerging role of proteome-wide interaction profiling methods in network-level protein target identification for bioactive small molecules. We will discuss several new methods that yield “top-down,” global views of interactions in the proteome, as well as higher resolution “bottom-up” interaction networks surrounding proteins of interest. Specifically, we will discuss the potential to combine these interaction profiling methods to identify both the direct targets and surrounding pathway networks that are functionally relevant with regard to small molecule-mediated signaling in diverse biological contexts.

Top-Down Interaction Profiling Methods

Thermal Protein Profiling

The phenomenon of ligand-induced protein stabilization has been used to detect protein-ligand interactions for decades[20, 21]. An important aspect of these assays is the fact that no chemical modification of either the ligand or protein of interest is required, suggesting that it may be a general way to screen for protein binding. In 2013, Nordlund and colleagues introduced a variant of this classic assay, entitled Cellular Thermal Shift Assay (CETSA), which elegantly demonstrated that changes in the stability of a protein of interest could be monitored directly, for example by Western blot, in whole proteome samples after exposure to a ligand of interest[22]. A subsequent expansion on CETSA, Thermal Proteome Profiling (TPP; also referred to as MS-CETSA), integrated this concept with LC-MS/MS detection and quantification of protein unfolding and precipitation after exposure of either whole cell lysate or live cells to parallel, stepped pulses in temperature, typically between room temperature and 65–70° C (Figure 1A). Subsequent isotopic barcoding of whole proteome at each temperature provides a method to quantify the relative amount of each detected protein (measured at the peptide level) present in each temperature aliquot, ultimately yielding protein-specific melting temperatures (T_m’s) via curve fitting of isotope abundance across the temperature gradient [23, 24]. TPP datasets provided the first global view of protein stability across the proteome. A main limitation with the original TPP workflow is that it was largely limited to soluble proteins that display prototypical melting behavior. Modification of the sample processing workflow with specific surfactants has proven useful to query membrane-associated proteins[25, 26], ligand dose-responsive changes in protein stability (isothermal dose response-CETSA, ITDR-CETSA), and simultaneous changes in protein abundance and stability (2D-TPP). Also, to relax the requirement for a prototypical sigmoidal melting response, an method named proteome integral solubility alteration (PISA) combines proteome from across the temperature range in mock- and ligand-treated samples and directly compares the integrated ion intensity in one isotope channel, essentially capturing the area under the unfolding curves, regardless of their shape[27]. This modification of TPP should be very useful because it not only allows for comparison of proteins with atypical temperature response, but also increases the number of conditions that can be compared in a single mass spectrometry run by an order of magnitude.

Figure 1. Top-Down Protein Interaction Methods.

Mechanistic overview of “Top-Down”, global protein interaction profiling methods. A) General workflow of MS-based thermal profiling (e.g., TPP) to measure aggregate proteoform stability in live cells (i.e. the bulk population of a protein of interest). Bioactive ligands can cause altered protein stability, either due to direct ligand engagement or downstream effects on interactions with other biomolecules. B) Hotspot Thermal Profiling (HTP) combines posttranslational modification-specific enrichment with peptide-level thermal profiling to measure the stability of distinct modiforms, in contrast to TPP, which measures aggregate stability from many peptides in a protein of interest. HTP therefore permits interrogation of ligand-induced stability changes in native protein sub-populations directly in cells. C) Schematic depiction limited-proteolysis profiling (e.g., DARTS or LiP-MS). Ligand binding in lysates leads to stabilization or protection of specific protein surfaces against proteolysis. Quantitative LC-MS/MS detection and peptide-level mapping can identify proteins and specific sites that are protected, and thus candidate ligand binding sites.

The primary application of TPP assays has been the interrogation of ligand-induced, protein-specific stabilization or destabilization for thousands of proteins in the same experiment. This was first demonstrated with the pan-kinase inhibitor stauorsporine, which caused significant T_m shifts in dozens of known protein kinases when both drug treatment and temperature pulse exposure was performed in live Jurkat cells[28–30]. An important observation in this study was that both significant destabilization of target proteins (i.e. negative shift in T_m), as well as no shift of known staurosporine targets were present in the dataset. This trend has borne out in subsequent studies, and underscores the fact that each protein is stabilized by a unique set of biophysical interactions and will display equally unique overall changes in stability in response to altered intra- or intermolecular interactions. For this reason, there is significant potential for false negatives in TPP assays. Nevertheless, TPP has subsequently been deployed to detect the direct binding targets of pan-kinase inhibitors [29], PIP4K inhibitors a131 and a166[31], the CDK4/6 inhibitor Palbociclib[32], and the MTH1 inhibitor TH1579[33]. Beyond the mammalian proteome, TPP has been used to profile E. coli, S. cerevisiae, T. thermophilus[34], and P. falciparum, the latter of which led to the identification of nucleoside phosphorylase (PfPNP) as the elusive target of the antimalaria drugs quinine and mefloquine[35]. Because of the unbiased nature of the technique, TPP can capture off-target interactions in families not predicted to be targets of a molecule of interest, with notable examples including ferrochelatase (FECH) for the several kinase inhibitors in clinical use, as well as phenylalanine hydroxylase for the well-known HDAC inhibitor panobinostat[36].

Beyond pharmacologic small molecules, TPP has been successfully deployed to detect endogenous protein-metabolite and protein-protein interactions in cells and lysates [37, 38]. Several studies have used TPP and ITDR-CETSA to detect the protein targets of nucleotides [26, 39] and nicotinamide adenine dinucleotide coenzymes in cells and in lysates. Focused studies in lysates have drafted proteome-wide maps of ATP binding proteins, rediscovering many high affinity interactors alongside the discovery that the ATP solubilizes a large number of positively charged and intrinsically disordered proteins [38]. Each of these datasets also identified altered stability of proteins that were presumably not direct targets of the metabolites being studied, raising the possibility that secondary effects on protein interaction networks were being detected. Indeed, a recent approach named thermal proximity co-aggregation (TPCA) uses the similarity in protein aggregation behavior to infer protein-protein complex membership[40]. While the potential to detect the downstream effects of biological stimuli could be viewed as a confounding factor in interpreting TPP datasets (i.e. whether T_m shifts are due to direct or indirect interactions with a small molecule), this information has the potential to more completely map complexes and pathways that are affected by small molecule binding. For example, the impact of posttranslational modifications (PTMs) on proteins downstream of small molecule action or other stimuli results in the differentiation of a bulk population of protein into discrete proteoforms that carry out distinct functions in the cell. A recent method, termed Hotspot Thermal Profiling, combines PTM-based enrichment and peptide-level thermal profiling to measure the stability of specific modified proteoforms (modiforms) in live cells (Figure 1B)[30]. The presumption in this approach is that PTMs that significantly perturb a modiform’s stability relative to its parent “bulk” protein population may be “hotspot” modification sites that are uniquely functional under specific biological conditions. Several annotated and unannotated phosphorylation sites were explored in this first report, confirming that “hotspot” phosphorylation sites could be identified on the basis of modiform stability changes, and that these sites functionally affect intra- and intermolecular protein-protein interactions, as well as protein-small molecule interactions. In principle, this peptide-level mapping approach could be adapted to interrogate diverse modification types, protein mutants, and protein isoforms in native samples.

Differential Protease Sensitivity and Chemical Accessibility

Thermal profiling represents just one class of methods that can provide “top-down,” global proteome information, however there are significant caveats in translating protein-specific T_m shifts into direct or secondary effects on protein interaction networks. A subset of the proteome does not exhibit prototypical or significant melting behavior, and those proteins that do may not be highly stabilized or destabilized in response to bona fide binding events. In principle, any method that interrogates a generic molecular attribute of protein structure or chemical state could be used to detect altered protein interactions in response to small molecules or other perturbations. Several methods have been developed that take advantage of the increased protease resistance of protein surfaces that are engaged in intermolecular interactions with small molecules or other biomolecules. Drug affinity response sensitivity (DARTS) and limited proteolysis-mass spectrometry (LiP-MS) both compare the kinetics of proteolytic cleavage in whole proteome between conditions to identify proteins that are likely engaged in differential interactions (Fig. 1C)[41, 42]. In a recent study, LiP-MS using the promiscuous protease Proteinase K in E. coli lysate identified more than 1,000 metabolite-protein interactions. Moreover, because the assay results in altered sensitivity at specific sites within proteins (detected by altered daughter peptides from those sites; Figure 1C), this study was able to identify more than 7,000 putative binding sites involved in those interactions[43]. In addition to protease sensitivity, methods that directly probe the reactivity of protein functionalities have been developed to query protein-biomolecule interactions in native conditions. These include susceptibility to protein oxidation (SPROX)[44], protein “painting” [45], and more recent bioconjugation reactions such as labeling of accessible methionine residues in native proteome[46]. Activity-based probes focused on specific amino acid reactivity [47, 48] or photoaffinity chemical probes[49, 50], can also report on the occupancy of small molecules or other biomolecules within subsets of the proteome.

Bottom-Up Interaction Profiling: Intracellular Proximity Labeling

While thermal profiling, protease resistance and related “top-down” methods can provide a global view of the proteins that are impacted by a specific stimulus, there are significant caveats in categorizing direct or indirect targets within protein networks. Therefore, there is also great value in other methods that enable protein-specific interaction profiling in cells in response to small molecule perturbation. This has heralded the emergence of a new class of intracellular proximity profiling platforms. Published proximity profiling approaches invariably involve the expression of genetic constructs encoding an engineered enzyme or receptor fused to a target protein of interest (POI). The resulting fusion can accept endogenous or exogenous substrates and co-factors to generate a reactive molecule in the immediate proximity of the fusion protein (Figure 2A). Direct transfer or diffusion of this reactive tracer results in covalent labeling of proximal proteins with a specific chemical tag for enrichment prior to quantitative profiling by LC-MS/MS. An early example, BioID, employs an engineered biotin ligase-POI fusion, which generates a biotinoyl-5’-AMP product that can covalently label surface amines on proximal proteins[51–53]. An evolved variant termed TurboID, which exhibits significantly faster labeling rates, has been used to map proximal protein interactors in mammalian cells and whole organisms[53], and more recently in plants[54]. Other methods use similarly reactive thioesters within small peptides or proteins as the tagging element, and rely on direct enzymatic transfer to proximal proteins, perhaps providing more specific labeling profiles. These direct proximal transfer methods include the NEDDylator[55], PUP-IT[56], and EXCEL[57] systems (Figure 2A). These ligases efficiently label short consensus motifs or specific amino acids present on proximal proteins both on and inside of cells, and in the case of PUP-IT, for example, can capture even weak, transient protein-protein interactions[58].

Figure 2. Bottom-Up, Proximity Profiling Methods.

A) Schematic of proximity labeling-based LC-MS/MS approaches to study protein-protein interactions, complexes, and protein localization. Labeling strategies can be classified in two distinct categories based on whether the technique produces a diffusible probe with a reactive group or whether it involves direct ligation of the tag onto the interacting proteins. B) Overview of intracellular proximity methods that rely on production of a diffusible reactive probe, with key assay parameters listed.

To generate more reactive, diffused labeling elements, genetic fusions of horseradish peroxidase[59] and engineered ascorbic acid peroxidase enzymes (i.e., the APEX system), convert an exogenous biotin phenol probe into a reactive phenoxyl radical[60, 61] in the presence of heme and H₂O₂ to label proximal proteins (Figure 2B). The phenoxyl radical will react preferentially with surface tyrosines (and less so Trp, His, and Cys), and should provide a spatially-restricted proximity profile due to the short half-life of the radical in a biological environment. Due to these attributes, APEX has proven widely useful in labeling the proteome of sub-cellular compartments like mitochondria[61], membrane-associated proteins[62], and the endoplasmic reticulum[63]. More recently, a light-activated photoproximity protein interaction (PhotoPPI) profiling method has been developed to localize a masked and highly reactive carbene to a POI through a SNAP-Tag fusion and complementary O⁶-benzylguanine targeting element[64]. Subsequent activation by 365 nm light results in proximal protein labeling and interactome mapping with high spatiotemporal control (Figure 2C). This study focused on identifying steady state and dynamic binding partners of the redox sensor protein KEAP1, which would be challenging to interrogate with affinity purification and likely other proximity labeling approaches. The minimal requirements and modularity of the photoproximity chemical probe in this system, coupled with facile activation by light, may enable proximity profiling in cellular contexts that may not be suitable for other methods. Taken as a group, all of these methods require distinct co-factors to produce uniquely reactive products, and thus provide opportunities to query protein networks in different cellular compartments, under distinct kinetic regimes, and proximity radii.

Multi-Omic Interaction Profiling: Integrating Top-down and Bottom-Up Strategies

Given the unique information provided by top-down and bottom-up profiling methods, we posit that integration of these methods in matched experiments could be used to generate high-resolution maps of the proteins and pathways that are impacted by a small molecule or biological signal under investigation. Top-down approaches like TPP can be applied to identify putative target proteins and complexes that are significantly perturbed in response to a specific small molecule or other stimulus. For example, kinetic monitoring of protein T_m shifts could enable proteome-wide tracking of altered biophysical interactions that originate at sites of direct small molecule-target interactions (i.e. early responders), and subsequently flow outward within protein-protein and other protein-biomolecule interaction networks (i.e. neighboring protein complexes and interaction networks; Figure 3). This protein neighborhood-level information should afford greater likelihood of identifying relevant target sites, protein complexes and networks that are involved in small molecule mechanism of action directly in the cells of interest. Taking the next step of determining direct versus indirect target proteins and upstream versus downstream signal propagation events is currently a bottleneck in the interpretation of these global profiles. Generation of matched proximity profiles may represent a general and relatively rapid way to confirm whether the proximal binding partners of a protein of interest are being altered in response to a small molecule perturbation, as well as define the specific protein partners and pathways involved in that network (Figure 3). One avenue toward this would be to perform a kinetic series of top-down interaction profiles that mirror the phenotypic changes caused by a small molecule or other event, identify specific proteins and/or complexes that are differentially affected, and prioritize them for proximity profiling using one of the methods highlighted in Figure 2. Generating a matched kinetic profile around several prioritized “nodes” would generate a high-resolution profile of the protein interactions surrounding each protein, as well as how they change along the kinetic series (Figure 3). Bioinformatic integration of these multi-omic interaction profiles with interaction datasets generated by orthogonal methods, and other -omic datasets that measure the molecules that could be involved in signal propagation (e.g., metabolites or protein PTMs) should enable a systems-level view of the proteins and pathways that are impacted by bioactive small molecules or other biological signals.

Figure 3: Kinetic interaction profiling to identify small molecule target proteins and signaling networks.

Schematic depiction of cellular protein networks responding to a perturbation, in this case a bioactive small molecule. The direct (red) and secondary (blue) targets of the stimulus would theoretically be the most significantly altered proteins in top-down, and subsequent bottom-up profiling datasets in early timepoints, as depicted in hypothetical volcano plots. Subsequent timepoints can capture signal propagation as interactions ripple outwards through affected protein networks and pathways. Integrated tracking of proteome-wide interactions and phenotypic consequences could provide complementary and novel insights into small molecule mechanism(s) of action as well as basic topology of interconnected proteins and pathways.

Conclusion

The current challenges in defining the molecular makeup of biological systems have shifted away from defining the parts, and more toward understanding the dynamic interactions of those parts that mediate cellular signaling and function. There is no place where this paradigm shift is more relevant than in the study of the proteome, where the cellular locale, proximal protein complexes, posttranslational modification state and small molecule ligand interactions all converge to control function. Here we presented a review of global and local interaction profiling methods that can be combined to uncover the relevant mechanism of action for, among other perturbations, bioactive small molecules. As we have discussed above, we posit that label-free interaction profiling methods operating directly in live cells offer an alternative to traditional approaches because they do not require alteration of the chemical structure of target small molecules or performing assays in lysates. More importantly, they offer a global picture of both the direct and indirect interaction networks affected by a specific perturbation, which provides ample opportunity for basic biological discovery. Integration of protein interaction-based profiling datasets with existing –omic profiling methods can aid in drafting more complete signaling maps within and between pathways. While not explicitly explored previously in the literature, we posit that these approaches could also be applied to interrogate cell-cell and cross-species (i.e. microbe-host) signaling interactions. The integrated, interaction-based profiling methods discussed here should provide a unique and dynamic view into cellular signaling networks for both basic and translational biological studies.