IDPs (intrinsically disordered proteins) play crucial roles in many important cellular processes such as signalling or transcription and are attractive therapeutic targets for several diseases. The considerable structural flexibility of IDPs poses a challenge for rational drug discovery approaches. Consequently, structure-based drug design efforts to date have mostly focused on inhibiting interactions of IDPs with other proteins whose structure can be solved by conventional biophysical methods. Yet, in recent years, several examples of small molecules that bind to monomeric IDPs in their disordered states have been reported, suggesting that this approach may offer new opportunities for therapeutic interventions. Further developments of this strategy will greatly benefit from an improved understanding of molecular recognition mechanisms between small molecules and IDPs. The present article summarizes findings from experimental and computational studies of the mechanisms of interaction between small molecules and three IDPs in their disordered states: c-Myc, Aβ (amyloid β-peptide) and α-synuclein.
The cell machinery is controlled by a large number of interactions between proteins and nucleic acids. It is now well appreciated that a large number of proteins do not adopt a single well-defined structure under native conditions. Proteins that contain a segment of at least 30 consecutive disordered residues in their native state are typically classified as IDPs (intrinsically disordered proteins) . In comparison with globular proteins, IDPs tend to contain fewer hydrophobic residues, but are generally enriched in charged amino acids [2–4]. IDPs can adopt a broad range of conformations, ranging from collapsed to fully extended. The considerable flexibility of IDPs facilitates interactions with a broad range of proteins and explains why IDPs often play key roles in important cellular processes such as signalling or transcription [5,6]. Molecular recognition between an IDP and a partner protein can involve a disorder-to-order transition through a coupled folding upon binding mechanism, which produces high-specificity low-affinity complexes . There are, however, several examples of IDPs that remain disordered upon complex formation .
IDPs are attractive therapeutic targets as they are often implicated in a broad range of diseases, such as cancers, cardiovascular disease or neurodegenerative diseases. However, the considerable flexibility of IDPs presents a challenge for drug discovery approaches . Owing to their lack of a well-defined tertiary structure, it is generally not possible to determine the structure of isolated IDPs. So far, structure-based approaches to inhibit IDPs have targeted either partner proteins that are ordered or ordered complexes, in those cases where IDPs fold upon binding. For instance, the p53 tumour suppressor is an IDP that is involved in the progression of more than 50% of human cancers. The transcriptional activity of p53 is tightly regulated by its partner protein MDM2 (murine double minute 2) and cancer cells often overexpress MDM2 to inhibit p53 function . As the p53-binding domain of MDM2 is folded, crystal structures can be readily obtained and have been exploited to design several classes of small-molecule inhibitors of p53–MDM2 . Some of the most successful inhibitors have advanced into clinical trials .
However, several protein–protein interactions involve two IDPs whose structure cannot be solved in isolation. Even in those instances where two IDPs mutually fold upon binding, the structure of the complex may not reveal pockets to which small molecules could readily bind. Thus a more general route to inhibiting IDP function would be to directly target their disordered state with small molecules. Historically, this approach has not been considered feasible . However, this view has been challenged in recent years, with the realization that several small molecules inhibit IDP function by binding to their unfolded state [14–16]. The interactions of small molecules with IDPs challenge our understanding of molecular recognition and it is important to clarify the mechanisms of IDP–small molecule interaction before such proteins can be more routinely targeted. The present review article focuses on three well-studied systems: the oncoprotein c-Myc, Aβ (amyloid β-peptide) and α-synuclein.
The proto-oncogene protein c-Myc consists of 439 amino acids and contains an 88-amino-acid bHLHZip (basic helix–loop–helix leucine zipper) domain. In its monomeric form, c-Myc is intrinsically disordered . c-Myc has been shown to interact with a large number of other proteins. The specific interaction between c-Myc and the protein Max has been studied extensively because the c-Myc–Max heterodimer binds DNA and regulates gene expression . It has been shown that overexpression of c-Myc is frequent in many cancers, and disruption of the c-Myc–Max interaction is a possible anticancer strategy .
Structurally diverse small molecules inhibiting the formation of this complex were discovered through a yeast two-hybrid screen . Biophysical studies using fluorescence assays, NMR and CD measurements were performed to characterize protein–ligand interactions [17,19,20]. These studies suggest that the small molecules disrupt the c-Myc–Max interaction by stabilizing conformations in monomeric c-Myc that are incompatible with heterodimerization with Max. Three distinct binding sites, encompassing residues 366–375, 375–385 and 402–409, have been mapped on to the c-Myc bHLHZip domain . Remarkably, the three distinct c-Myc-binding sites can be occupied simultaneously by different ligands. These results suggest that the c-Myc–small molecule interactions are fairly localized and can be predicted from primary sequence analysis. Indeed, protein disorder prediction algorithms can locate approximately the small-molecule-binding sites of c-Myc, which tend to be enriched in hydrophobic amino acids in comparison with the rest of the domain . In addition, many of the small-molecule ligands can bind truncated c-Myc segments containing a single binding site with a binding affinity similar to that of the full c-Myc bHLHPZip domain. For instance, the small molecule 10058-F4 binds in a fluorescence polarization assay to c-Myc353–437 with a Kd of 5.3±0.7 μM and to c-Myc402–412 with a Kd of 13.3±1 μM . Furthermore, similar chemical shift perturbations were observed for c-Myc353–437 and c-Myc402–412 upon binding 10058-F4. NMR and CD studies suggest that c-Myc remains disordered upon binding 10058-F4. Ligand binding appears to lead to formation of a hydrophobic cluster between the ligand and the side chains of Tyr402, Ile403, Leu404 and Val406 (Figure 1). Molecular dynamics studies performed by our group reveal multiple distinct binding modes for 10058-F4, with frequent stacking interactions with Tyr402 as well as hydrogen-bonding interactions with the backbone of Tyr402, Val406 and Lys412 .
Summary of the main interactions observed in three IDP–ligand complexes: c-Myc–10058-F4, α-synuclein–dopamine and Aβ–Pep1b
AD (Alzheimer's disease) Aβ
AD is a neurodegenerative pathology characterized by the formation of senile plaques in the brain . The aggregation of Aβ is known to be one of the main components of those plaques and may be associated with the pathogenesis of AD [23,24]. Aβ (36–43 amino acids) is produced by the successive cleavage of the APP (amyloid precursor protein) by the enzymes β-secretase and γ-secretase. Although the role of APP is not completely characterized, it appears to be crucial for synapse formation and function . The aggregation of Aβ, as well as with other compounds such as apoliprotein E, induces the development of senile plaques. Aβ adopts a folded helical structure in membrane environments, but an aggregation-prone β-sheet conformation in aqueous solution .
Over the last few decades, many peptide and small-molecule inhibitors of Aβ aggregation have been discovered, primarily through in vitro assays . Current small-molecule inhibitors appear to inhibit Aβ aggregation through at least two distinct mechanisms. For instance, scyllo-inositol derivatives have been shown by electron microscopy experiments to bind and stabilize monomeric and trimeric forms, thus blocking aggregation [28,29]. On the other hand, compounds such as Thioflavin T or Congo Red appear to interact with Aβ aggregates, although decades of studies on these compounds have produced several conflicting models of binding mechanisms. Plausible hypotheses have been reviewed extensively by Groenning  recently.
Computational studies have attempted to clarify protein–ligand interactions. Molecular dynamics simulations were performed recently for ten small-molecule inhibitors in the presence of a truncated form of Aβ (Aβ12–28) . Although the small molecules did not exhibit a predominant binding mode and did not dramatically affect the secondary-structure preferences of Aβ12–28, a number of conserved interactions with Aβ12–28 could be observed. Most of the ligands interacted preferentially with the N-terminal portion of the peptide (residues 13–20). Energetic analysis revealed favourable electrostatic interactions with three amino acids (His13, His14 and Lys16). Additionally, favourable hydrophobic interactions are observed between the inhibitors and the entire N-terminal stretch, with the sites of highest interaction probability being near the side chains of Phe19 and Phe20. The binding affinities appear to be roughly correlated with the number of aromatic groups and charged groups present in the ligands. Molecular dynamics simulations have also been performed to examine the interactions of two small ligands, Pep1b and Dec-DETA, that were designed to stabilize the central helix in Aβ . Both ligands appear to stabilize the Aβ central helix (residues 15–24) in Aβ13–26 by interacting preferentially with two charged amino acids: Glu22 and Asp23. In addition, electrostatic interactions with His13 and Lys16 as well as hydrophobic interactions with Phe19 and Phe20 were also reported for Pep1b (Figure 1). It appears that the extended side-chain interactions between the ligands and Aβ disfavour intramolecular side-chain interactions that would destabilize the central α-helix. Recently, molecular dynamics simulations were used to study the interactions of inositol ligands with (Gly-Ala)4 modelled either as disordered or β-sheet aggregates of four peptides, or as an extended fibril-like oligomer . The ligands were observed to form predominantly one or two hydrogen bonds with the peptide backbone. The results suggested that inositol does not inhibit amyloid formation by dispersing preformed aggregates or by preventing aggregation, but is more likely to bind instead to the surface of prefibrillar aggregates . The computed dissociation constants of the ligands were two orders of magnitude higher than those measured experimentally, suggesting that additional side-chain interactions must contribute significantly to the binding affinity of the inositol ligands to Aβ aggregates .
The 140-amino-acid protein α-synuclein consists of three distinct domains. The central region of α-synuclein is known to be crucial for the aggregation of α-synuclein fibrils, one of the main components of Lewy bodies associated with many neurodegenerative diseases such as PD (Parkinson's disease) [34,35]. Under physiological conditions, α-synuclein normally adopts a helical conformation that is non-pathogenic and plays a role in neurotransmitter release. It is still not well understood how α-synuclein first forms soluble oligomers called protofibrils, followed by the development of β-sheet-rich α-synuclein fibrils. In the light of these observations, a deeper molecular-level understanding of interactions between monomeric, protofibril and fibril forms is important to facilitate the discovery of small-molecule inhibitors of α-synuclein fibrillization.
A few years ago, 15 fibrillization inhibitors were found by screening a small-molecule library using a fibrillization assay . Many of these inhibitors are members of the catecholamine family and include dopamine. There is controversy about the mechanisms of interactions between dopamine and α-synuclein. Conway et al.  have suggested that dopamine readily oxidizes into dopamine-derived orthoquinones that subsequently form a covalent adduct with α-synuclein by radical coupling to form dityrosine linkages or by nucleophilic attack of a lysine side chain. On the other hand, Norris et al.  failed to detect significant levels of dopamine–α-synuclein adducts and suggested instead that binding occurs through non-covalent interactions with the α-synuclein segment Tyr125-Glu-Met-Pro-Ser129. Herrera et al.  used docking calculations and molecular dynamics simulations to study the interactions of dopamine and several plausible oxidized derivatives with an NMR-derived structural ensemble of α-synuclein. In the majority of the simulated complexes, the ligands interacted through a broad range of hydrogen-bonding and hydrophobic interactions with the region Tyr125-Glu-Met-Pro-Ser129. Additionally, significant electrostatic interactions were computed between the ligands and Glu83 located in the non-β-amyloid region of α-synuclein. These predictions were tested by a series of biophysical experiments. Point mutations to alanine in the Tyr125-Glu-Met-Pro-Ser129 region did not prevent dopamine inhibition of α-synuclein aggregation in an in vitro fibrillization assay, suggesting that dopamine interacts non-specifically with this region. On the other hand, mutation of Glu83 to alanine strongly impaired the ability of dopamine to inhibit α-synuclein aggregation .
Non-catecholamine inhibitors of α-synuclein aggregation have also been identified. A broad range of biophysical methods were used by Lendel et al.  to characterize the interactions of Congo Red and lacmoid with α-synuclein. They concluded that these two small molecules interact broadly with the N-terminal and central region of α-synuclein as small oligomeric species .
Although small molecules have now been found to interact directly with several IDPs in their monomeric form, an important challenge is to clarify the specificity of the interactions. For instance, there are numerous proteins that contain a bHLHZip domain similar to that of c-Myc. Consequently, several small molecules that inhibit the c-Myc–Max complex also inhibit related bHLHZip pairs. To illustrate, the compound 10058-F4 has also been shown in a yeast two-hybrid assay to disrupt the complexes MyoD–E2-2, Mad1–Max and Mxi1–Max, although several other bHLHZip pairs were not inhibited . Several of the dopamine derivatives that inhibit α-synuclein aggregation have also been shown to also dissolve fibrils of Aβ in vitro . Congo Red and lacmoid bind readily to β-synuclein, a protein closely related to α-synuclein which does not aggregate under physiological conditions .
In several cases, relatively structurally diverse small molecules have been found to interact with similar regions in an IDP. Additionally, many studies suggest that the complexes between small molecules and IDPs remain disordered . This suggests that the binding of the small molecules is driven by a large number of weak interactions . Arguably, unlike proteins, small molecules are unlikely to induce IDP folding upon binding, as the relatively limited intermolecular contacts that they form are unlikely to overcome the large conformational entropy loss necessary to structure an IDP. Structure-based approaches to design ligands for IDPs will therefore have to explicitly consider multiple binding modes.
Although the mechanisms of IDP aggregation are still not well understood, a number of small-molecule inhibitors of IDP aggregation have reached clinical studies. For instance, methylthionium chloride, initially developed as an antimalarial agent, has been shown to inhibit in vitro the aggregation of the IDP tau . Results of a Phase II clinical trial reported that methylthionium chloride slows down cognitive impairment in patients suffering from AD, thus inhibiting the formation of tau aggregates is a promising strategy for the development of AD treatments .
Intrinsically Disordered Proteins: A Biochemical Society Focused Meeting held at University of York, U.K., 26–27 March 2012. Organized and Edited by Jennifer Potts (York, U.K.) and Mike Williamson (Sheffield, U.K.).
J.M. is supported by a Royal Society University Research Fellowship. This research was also part funded by a Marie Curie International re-integration grant [grant number PIRG-GA-2010-276785].