In silico analysis of enantioselective binding of immunomodulatory imide drugs to cereblon

In silico docking simulations

The entire docking protocol was performed using MOE 2013.08 software package (Molecular Operating Environment; Chemical Computing Group, Montreal, Quebec, Canada) (MOE 2015). 2D molecular structures of IMiDs, including (S)-enantiomers [(S)-thalidomide, (S)-lenalidomide and (S)-pomalidomide] and (R)-enantiomers [(R)-thalidomide, (R)-lenalidomide, (R)-pomalidomide], pyrimidine nucleosides (uridine, cytidine, thymidine) were obtained from the Nikkaji web service (Nikkaji 2015), and the 3D structures were modeled using the conformational search program in MOE.

Finally, the ligands were docked with each receptor using ASEDock program (Goto et al. 2008). During this procedure, flexibility of the ligand atoms was allowed and the backbone atoms of the receptors were tethered. Following the docking scores (U_dock), the top 10 poses of each docking pose were retained and the top scores were evaluated in terms of docking scores and poses.

Docking results for IMiDs

The eleven crystal structures of cereblon were obtained from the PDB (Table 1), and these structures were docked with the (S)-enantiomers [(S)-thalidomide, (S)-lenalidomide, and (S)-pomalidomide] and the (R)-enantiomers [(R)-thalidomide, (R)-lenalidomide, (R)-pomalidomide] of IMiDs. Using eight out of the eleven structures (PDB ID: 4CI1, 4V2Y, 4CI2, 4TZ4, 4V30, 4CI3, 4V2Z, 4V31), we correctly replicated the enantiomeric selectivity, where the docking scores of (S)-enantiomers were lower than those of the respective (R)-enantiomers (Fig. 2a; Additional file 1: Table S1) and the docking poses of (S)-enantiomers were within 2 Å of RMSD value compared to the crystal structures (Additional file 1: Figure S1). However, we could not replicate the selectivity when using the other three structures (PDB ID: 4TZC, 4TZU, 3WX2) (Fig. 2b; Additional file 1: Table S1). In these cases, some docking results were in contrast to the earlier results (PDB ID: 4TZC and 3WX2), and the overall docking scores of (S)-enantiomers were higher compared to the correctly replicated structures. We classified these ‘appropriate’ 8 structures and ‘inappropriate’ 3 structures into type A and type B, respectively. A detailed structural analysis revealed that compared to type A, cereblon structures of type B have some differences at the hairpin connecting β10–β11 region (residue 351–355). This hairpin region in type A is next to the IMiDs binding site, whereas this region is disordered or absent in type B structures (Fig. 2c). This is in part due to the polymerization of cereblon in crystal structures, which distorts this hairpin. This phenomenon is unlikely to occur in a natural state because no evidence of cereblon monomer polymerization has been obtained so far (Chamberlain et al. 2014). Therefore, because of this unnatural hairpin region, the docking results of type B structures might be ‘inappropriate’. Except for type B structures, we confirmed the enantiomeric selectivity of IMiDs with high accuracy. These results might be due to the increased interactions of these molecules with cereblon by their amino-phthaloyl-substituted characteristics in type A structures.

Docking results for pyrimidine nucleosides

Next, using type A structures we also tried to replicate the selectivity of pyrimidine nucleosides i.e. the preference of uridine over cytidine and thymidine. We correctly replicated this selectivity and found that the docking scores of uridine were lower than those of the other two pyrimidine nucleosides (Additional file 1: Figure S2). As previous reports stated (Hartmann et al. 2014), these results are attributed to the structural similarity between the uracil moiety of uridine and the glutarimide moiety of IMiDs, and the structural discrimination of the pyrimidine moiety of cytidine and thymidine. The cytosine moiety of cytidine has a different functional group compared to the uracil moiety of uridine, resulting in inappropriate chemistry for the ligands to bind and in higher docking scores for cytidine. As for thymidine, the thymine moiety of the ligand has an additional methyl group compared to the uracil moiety of uridine, resulting in a steric clash with the tryptophan residues and higher docking scores for thymidine (Additional file 1: Figure S2c). Moreover, type B structures were also tested (Additional file 1: Figure S3). Because the hairpin connecting β10–β11 was incomplete or was oriented in an outer space which is further away from the TBD domain of cereblon compared with type B structures, an additional space was formed. When two of the type B structures (PDB ID: 4TZC and 4TZU) were used as receptor for docking, pyrimidine nucleosides were docked out of thalidomide binding pocket. While when type A structure (e.g. PDB ID: 4CI2) was used, nucleosides were docked into the thalidomide binding site (Additional file 1: Figure S3). One more point to mention is that according to the overlap of the structures (Additional file 1: Figure S3e), in the case of 4TZC, if a β10–β11 hairpin exists, the predicted poses should conflict with it. On the other hand, 3WX2 extended its hairpin aside compared to 4CI2, and this may due to the lack of NTD (Additional file 1: Figure S3d, e). In this case, uridine and thymidine were docked into the thalidomide binding pocket, and however cytidine was predicted in a reversed form (Additional file 1: Figure S3c). Since there was no evidence that pyrimidine nucleosides bind to a space other than thalidomide binding site, it is hard to tell the docking poses were correct and there is less meaningful to compare the docking scores of these results (Additional file 1: Table S2), but only compared the result of uridine/3WX2 with thymidine/3WX2, two poses into thalidomide binding pocket, uridine is prone to bind to cereblon, due to the score showing −40.6 versus −34.4.

The β10–β11 hairpin region contributes to the enantiomeric selectivity for (S)-enantiomers

As mentioned above, the hairpin connecting β10–β11 could be a key region in the enantiomeric selectivity of IMiDs. To better understand the role of this hairpin region, we superimposed a type B cereblon structure (PDB ID: 4TZC) binding docking poses of (S)-thalidomide and (R)-thalidomide, with a type A cereblon structure (PDB ID: 4CI1) (Fig. 3a, b). These structures revealed that the (R)-thalidomide partially clashed with the hairpin region of 4CI1. Thus, for 4TZC the ‘inappropriate’ docking results might arise from the widely expanded binding pocket due to the disordered hairpin structure. These findings also suggest that in case of type A structures, the binding of (R)-enantiomers is unfavorable due to the steric hindrance caused by this hairpin, next to the IMiDs binding site. Finally, this hairpin could contribute to the IMiDs binding affinities, because the docking scores of (S)-enantiomers are higher for type B structures than for type A (Fig. 2; Additional file 1: Table S1). To obtain a deeper insight, we superimposed a type A cereblon structure binding the docking pose of (S)-thalidomide with a type B cereblon structure (PDB ID: 4TZC) binding that of (S)-thalidomide. These structures revealed that in absence of the β10–β11 hairpin, the interaction between the (S)-enantiomer and the β10–β11 hairpin constructive residues disappear (Fig. 3c). Thus, the disorder or absence of β10–β11 hairpin could result in decreasing the IMiDs binding affinity. Taken together, in addition to previously reported the aromatic cage of three tryptophan residues (W380, W386, and W400), the hairpin connecting β10–β11 may play an additional role in the recognition of IMiDs.

The possible effects of the β10–β11 hairpin in substrate recognitions of cereblon

Cereblon, a component of DDB1/cullin4 E3 ubiquitin ligase complex, mediates the recruitment and degradation of its target proteins (Ito et al. 2011; Shortt et al. 2013). Recent experimental results imply that IMiDs act as both agonists and antagonists of cereblon (Ito et al. 2010; Krönke et al. 2014; Lu et al. 2014; Gandhi et al. 2014; Krönke et al. 2015). As an antagonist, the MEIS2 protein, which acts at various aspects of human growth (Capdevila et al. 1999), was identified as the substrate of cereblon that competitively binds to the same binding site as IMiDs (Fischer et al. 2014). As such, cereblon with bound IMiDs prevents the recruitment, ubiquitylation and degradation of MEIS2, resulting in teratogenicity (Fischer et al. 2014). As an agonist, IMiDs bound to cereblon complex promote the binding of some proteins including transcription factors implicated in multiple myeloma (such as Ikaros and Aiolos) (Krönke et al. 2014; Lu et al. 2014; Gandhi et al. 2014) or myelodysplastic syndrome (such as casein kinase 1α) (Krönke et al. 2015), which otherwise could not bind to cereblon. As a result, these events cause the degradation of these target proteins, resulting in multiple mechanisms of IMiDs action. Sequence alignment studies regarding these proteins binding to cereblon with or without IMiDs revealed that these proteins do not share high sequence homology, suggesting that the ways in which cereblon or the cereblon–IMiDs complex recognizes the substrates may have some differences. As mentioned above, the β10–β11 hairpin of cereblon has some difference between type A and type B structures due to the polymerization of cereblon in crystal structures, thus this hairpin region might play an important role with respect to recognition of substrate binding. A recent experimental study, which implies that this hairpin region only folds upon ligand binding, also supports this hypothesis (Lopez-Girona et al. 2012). We carried out a 25 ns MD computation of human cereblon structure to analyze the movement of this hairpin (Additional file 1: Figure S4). This hairpin did not move violently throughout our simulation duration as expected. It is suggested that although β10–β11 hairpin is useful in substrate recognitions, but high flexibility is not needed. Or the simulation duration was not long enough to observe the conformation change of this hairpin. Another point should be mentioned here is that all crystal structures other than the type A structures were derived from only the TBD domain of cereblon, and these structures do not cover the overall structures. More studies are required to understand the details, such as structural identification of the cereblon–MEIS2 complex, cereblon–IMiD-substrate complex, and so on.