De-Novo Design of Cereblon (CRBN) Effectors Guided by Natural Hydrolysis Products of Thalidomide Derivatives

Targeted protein degradation via cereblon (CRBN), a substrate receptor of an E3 ubiquitin ligase complex, is an increasingly important strategy in various clinical settings, in which the substrate specificity of CRBN is altered via the binding of small-molecule effectors. To date, such effectors are derived from thalidomide and confer a broad substrate spectrum that is far from being fully characterized. Here, we employed a rational and modular approach to design novel and minimalistic CRBN effectors. In this approach, we took advantage of the binding modes of hydrolyzed metabolites of several thalidomide-derived effectors, which we elucidated via crystallography. These yielded key insights for the optimization of the minimal core binding moiety and its linkage to a chemical moiety that imparts substrate specificity. Based on this scaffold, we present a first active de-novo CRBN effector that is able to degrade the neo-substrate IKZF3 in the cell culture.


■ INTRODUCTION
Classical immunomodulatory drugs (IMiDs) like thalidomide and its second-and third-generation analogues lenalidomide, pomalidomide, avadomide (CC-122), and iberdomide (CC-220) have constantly emerged to new therapeutic areas. Originally developed as a sedative 1 and banned in 1961 for its teratogenic effects when used during pregnancy, 2,3 thalidomide and a number of newly developed analogues are approved for the treatment of multiple myeloma (MM), 4 erythema nodosum 5 and myelodysplastic syndrome (MDS). 6,7 Because of their pleiotropic and especially anti-angiogenic properties, IMiDs have further been reported effective in many off-label indications as for Hodgkin's lymphoma, 8−10 light chainassociated (AL) amyloidosis, 11 and acute myeloid leukemia (AML). 12,13 Currently, IMiDs share a common glutarimide moiety (Figure 1), which is connected to a second moiety that is typically derived from phthaloyl. Via the glutarimide moiety, they are able to bind to a tri-tryptophan pocket within the thalidomide-binding domain 14−17 of cereblon (CRBN). CRBN is the substrate receptor of the Cullin RING E3 ubiquitin ligase CUL4-RBX1-DDB1-CRBN (CRL4 CRBN ) 17 and responsible for the recognition of endogenous substrates such as glutamine synthetase, 18 MEIS2, 17 and amyloid precursor protein (APP). 19 In the presence of IMiDs, however, its substrate specificity is changed. The solvent-exposed second moiety, the protruding moiety that is unique to each IMiD, recruits novel substrates to the CRBN surface for ubiquitination. The degradation of these neo-substrates accounts for most of the efficacy of IMiDs in MM (IKZF1 and IKZF3), 20 5q-deletion-associated MDS (CK1α), 21 and AML (GSPT1). 22 Recently, it has also been shown that the degradation of the neo-substrate SALL4 is linked to developmental malformations caused by thalidomide. 23,24 Crystal structures of the neo-substrate complexes CRBN· lenalidomide·CK1α, CRBN·pomalidomide·IKZF1, and CRBN· pomalidomide·ZNF692 provided first insights into the binding mode of neo-substrates. 25,26 Many of the identified neosubstrates possess no obvious sequence homology, but they all exhibit a structurally analogous β-hairpin, which binds to the surface around the IMiD-binding site, involving interactions with both surface residues and the IMiD itself. A potential prerequisite for the recruitment is the presence of a glycine at the tip of this β-hairpin; this glycine was found to be conserved for many neo-substrates, like CK1α 25 and GSPT1, 22 and appears in the common C-X(2)-C-G motif 24 in many transcription factors belonging to the C 2 H 2 zinc finger class, including IKZF3, 25 IKZF1, 25 ZFP91, 24 and SALL4. 24 While IMiDs seem to generally recruit several members of the C 2 H 2 zinc finger family, only lenalidomide was proven to recruit CK1α, indicating that the protruding moiety imparts substrate specificity.
The CRBN-binding ability of IMiDs has further been exploited for targeted protein degradation in an approach called proteolysis targeting chimera (PROTAC), coined in 2001 by Craig Crews and co-workers. 27,28 PROTACs are bifunctional small molecules with a binding moiety for a target protein linked to a binding moiety for an E3 ubiquitin ligase, thus inducing ubiquitination and proteasomal degradation of the target protein. Recent examples of successful PROTACs target the estrogen 29 and the androgen receptor 30 via a von-Hippel-Lindau (VHL) E3 ligase ligand but also BET bromodomains via linkage to thalidomide as a CRBN ligand. 31 Notably, also PROTACs with a VHL ligand on one and thalidomide as the CRBN ligand on the other end have been tested, leading to unidirectional degradation of CRBN. 32 In general, also the PROTAC approach has high potential to target the undruggable; a particularly illustrative example for this potential is exemplified by the recent discovery of PROTACs targeting the tau protein. 33−35 To date, essentially all CRBN effectorsIMiDs and CRBNbased PROTACsare derived from thalidomide and its derivatives. However, the chemical space of CRBN ligands ranges far beyond thalidomide: in first systematic characterization, we have previously revealed that a large variety of lactams and cyclic imides are potent binders, including several marketed drugs. 36 Specifically, we have shown succinimide to exhibit higher affinity than glutarimide, using the single-domain bacterial CRBN homologue from Magnetospirillum gryphiswaldense (MsCI4) in a FRET assay. 14 In this study, we set out to further explore and exploit the chemical space of CRBN binding by designing effectors based on succinimide and glutarimide, which we characterize with regard to their affinity, their structural binding mode, and their ability to induce proteasomal degradation of neo-substrates. Guided by hydrolyzed metabolites of thalidomide and of three of our designs, we present novel minimalistic motifs that are able to recruit and degrade neosubstrates and may serve as E3-recruiting ligands for future PROTACs.

■ RESULTS AND DISCUSSION
Biophysical and Structural Characterization of IMiD Analogues and Their Hydrolysis Products. As a starting point for new effectors, we chose the classical IMiD scaffold. Based on the finding that succinimide is able to bind to CRBN with a higher affinity than glutarimide (K i values of 4.3 μM vs 28 μM for MsCI4 36 ), we designed a first panel of derivatives based on glutarimide and succinimide, in which we probed the effect of different substitutions in the phthaloyl moiety ( Figure 2). The respective compounds 2a−5b were prepared by the synthetic route shown in Scheme 1. The imides 3a and 3b were synthesized from the commercially available N α -(tert-butoxycarbonyl)-L-asparagine (1a) and N α -(tert-butoxycarbonyl)-Lglutamine (1b), respectively, via an imide formation using N,Ncarbodiimidazole (CDI) and 4-dimethylaminopyridine  in tetrahydrofuran (THF) as well as a deprotection reaction in the presence of trifluoracetic acid (TFA). Compounds 4a−4d were obtained through coupling between the imides 3a or 3b and the commercially available 3nitrophthalic anhydride or 4-nitrophthalic anhydride. Catalytic hydrogenation of the molecules 4a and 4b using 10 wt % Pd/C in EtOAc produced the derivatives 5a and 5b.
The binding affinities of these compounds for MsCI4 were assessed in the FRET assay, starting with the glutarimide-based compounds 4c and 4d in comparison to the commercial thalidomide derivatives, lenalidomide and pomalidomide. Lenalidomide (K i = 3.1 μM) and pomalidomide (K i = 0.8 μM), both carrying an additional amino group in the R 2 position, show improved affinity to MsCI4 as compared to thalidomide (K i = 4.4 μM). When we exchanged this amino group for a nitro group (4c), we saw a significant drop in affinity into a range in which a precise value could not be obtained (K i > 40 μM); 36 moving this nitro group to the R 1 position (4d) had less impact on the affinity (K i = 8.9 μM, Figure 2A). Similar effects were observed in a recent study that was published during the preparation of this manuscript. 37 As both compounds, 4c and 4d, retained affinity to MsCI4, their overall binding mode is supposedly conserved and comparable to thalidomide, which is consistent with previous studies, showing that small modifications on the protruding moiety have little influence on the overall affinity to CRBN. 17,37 The binding modes of lenalidomide and pomalidomide have previously been reported to be virtually identical to thalidomide. 14,17 For the succinimide-based compounds 4a, 4b, 5a, and 5b, we obtained a similar picture as for the glutarimide-based compounds. Although they showed overall weaker binding in the FRET assay ( Figure 2B), substitutions in the R 2 position are less favorable for the affinity as substitutions in R 1 . For 4a and 5a, which have a nitro or an amino group in R 2 , respectively, we obtained K i values of >40 μM. For 4b and 5b, which have the respective groups in the R 1 position, the obtained K i values are 11 and 12 μM, respectively.
In a next step, we determined the molecular-binding determinants of the succinimide-based compounds via X-ray crystallography. To this end, crystals of the MsCI4·thalidomide complex were reproduced 36 and subsequently used for soaking experiments, in which the thalidomide molecules bound to MsCI4 may be displaced by the compound of interest. There are three chains, that is, three MsCI4·thalidomide complexes, in the asymmetric unit (ASU) of these crystals, in which the bound thalidomide molecules can potentially be displaced. However, as the three chains form different crystal contacts and differ slightly in their conformation, it is possible that thalidomide is only replaced in one or two chains of the ASU, depending on the particular compound. 36 These experiments yielded crystal structures for the four compounds tested, 4a, 4b, 5a, and 5b,

Journal of Medicinal Chemistry
Article all showing the classical IMiD binding mode with the basal main-chain interactions of the succinimide amino group with F77 and the carbonyl group with W79 of MsCI4. Also, the orientation of the protruding moieties of 4a, 5a, and 5b is very similar to that of thalidomide, despite the different ring size of glutarimide and succinimide. As a result, the hydrogen bond that is typically observed between the conserved N50 and a carbonyl group of the phthaloyl moieties for the classical glutarimidebased compounds is also found for 5b.
However, the crystal structures also held surprises. They did not only reveal the binding modes of the pure compounds but also that of hydrolysis products of 4a, 4b, and 5a with unambiguous electron density; only 5b was exclusively observed in the nonhydrolyzed form (Figures 3 and 4). For 4a and 5a, hydrolyzed metabolites were only observed in one chain of the ASU, with the second chain occupied with a nonhydrolyzed version and the third chain with thalidomide. For 4b, which showed the highest affinity with a K i of 11 μM, all three binding pockets in the ASU were occupied by a hydrolyzed metabolite, so the binding mode of nonhydrolyzed 4b could not be studied. In all cases, for the hydrolysis products of 4a, 4b, and 5a, ring opening of the phthaloyl group had led to the formation of a secondary amide and a carboxyl group, which are clearly resolved in F O −F C omit maps ( Figure 3). Both of these form important interactions with the binding pocket: first, the secondary amide rescues the hydrogen bond to the conserved N50, which is typically formed by a carbonyl of the phthaloyl moiety. Second, the additional carboxyl group replaces a conserved water molecule previously coordinated by W99 and engages in direct hydrogen bonding with the W99 side chain. In contrary, the primary amino group of hydrolyzed 5a and the solvent-exposed nitro groups of hydrolyzed 4a and 4b do not show additional interactions.
Binding Mode of CBG, a Major Hydrolysis Product of Thalidomide. Unintentionally, in addition to the hydrolysis products of 4a, 4b, and 5a, we were able to characterize the binding mode of a major hydrolyzed thalidomide metabolite. It is known that IMiDs can rapidly racemize in bodily fluids and water 38 and spontaneously hydrolyze under physiological conditions. 39−41 For thalidomide, being eliminated mainly by spontaneous hydrolysis in blood and tissues, a half-life of about 5.5−7.3 h at the physiological pH of 7.4 was reported. 42 Among several proposed hydrolysis products, the two main urinary metabolites are 2-phthalimidoglutaramic acid (∼50%) and α-(2-carboxybenzamido)glutarimide (CBG) (∼30%). 43 Of these, CBG has an unmodified glutarimide moiety and is also one of

Journal of Medicinal Chemistry
Article three major metabolites in human plasma. 44 As CBG was also reported to possess a higher TNF-α production-inhibitory activity (80%) than thalidomide (32%) at concentrations of 3 μM, 44,45 it is of great pharmacological interest, but its mode of interaction remained elusive so far. 46 During the course of this study, we obtained a crystal structure of CBG in the complex with a humanized mutant form of MsCI4 (Figures 3 and 4). This structure was the result of a cocrystallization trial of this mutant with thalidomide, which aimed at the characterization of the mutant protein. However, the mutant selectively bound CGB that was presumably present in traces in our crystallization setup with thalidomide. This mutant has a number of nonconserved residues in the direct vicinity of the thalidomide-binding pocket mutated to the residues in the human protein, including the substitution F56H. While this residue is not directly involved in classical IMiD binding in MsCI4 or the human protein, it plays . Binding mode of the initial compounds and their hydrolysis products inside the binding pocket. Thalidomide, 4a, and 5a are shown with their respective hydrolysis products. Ring opening of the phtaloyl moiety that leads to the observed hydrolysis product is indicated in red in the chemical drawings. 5b was exclusively found in the nonhydrolyzed form, whereas 4b was exclusively found as a hydrolysis product. PDB codes are 4V2Y (thalidomide), 6R0Q (CBG), 6R0S (4a), 6R0U (5a), 6R11 (5b), and 6R0V (4b).

Journal of Medicinal Chemistry
Article an important role in the binding of CBG: together with N50, it coordinates one oxygen of the carboxyl group in the CBGprotruding moiety; the other oxygen of this carboxyl coordinates the conserved water molecule bound to W99. However, in contrast to the hydrolysis products of the succinimide-based compounds 4a, 4b, and 5a, the amide moiety resulting from ring opening is not found to be involved in defined hydrogen bondingalthough superposing approximately with the phthaloyl moiety of thalidomide, it does not form the canonical hydrogen bond with N50, as the latter is engaged in the hydrogen bond to the carboxyl group.
Rational Design of Novel Succinimide Effectors Guided by Hydrolyzed Metabolites. In a next step, we aimed to exploit our knowledge on the binding mode of hydrolyzed metabolites for the design of novel effectors. The fact that the hydrolysis products were selected against their parent compounds in several cocrystallization or soaking experiments suggested that they pose binders of similar, if not superior affinity. Comparing the binding modes of CBG and hydrolyzed 4a, 4b, and 5a, we further hypothesized that the difference seen for the hydrogen bonding of their amide linkers should yield increased affinity for succinimide-based effectors. Consequently, we used this amide linker to connect different functional groups as protruding moieties to succinimide as the binding moiety. To this aim, we prepared derivatives 7a−7h, as shown in Scheme 2, by treating imide 3a with the corresponding acyl chlorides 6a− 6h and N,N-diisopropylethylamine (DIPEA) in THF.
All derivatives of this second panel were tested in the FRET assay. Only for compound 7e, with a 3,5-dinitrobenzol group, the affinity was decreased under the detectable levels of the FRET assay. For the compounds with an isobutylene (7a), benzyl (7b), or chlorobenzothiophene group (7g) as the protruding moiety, binding was well detectable but could not be quantified (K i > 40 μM). Better binding was observed for compounds with a styryl (7f) and dichloro-benzothiophene (7h) group, both showing K i values of 20 μM ( Figure 5A). Finally, the highest affinities were achieved with a 2,4,6trichlorobenzol moiety (7c, K i = 9 μM) and benzyloxy group (7d, K i = 4 μM), rendering 7d, the highest-affinity binder in this study, with a K i value comparable to an unmodified succinimide 36 ( Figure 5A).
We continued with characterizing the binding modes of compounds 7a−c and 7f via crystallography. For the best binder 7d, we performed cocrystallization screening, which yielded a new crystal form diffracting to 1.1 Å resolution; the other compounds were successfully evaluated in soaking experiments. All compounds revealed the expected binding mode as observed for the hydrolyzed metabolites of 4a, 4b, and 5a, with the succinimide moiety forming the canonical interaction within the binding pocket, and the amide linker forming the hydrogen bond with N50. Besides these hydrogen bonds of the binding and linking moiety, no defined interactions with the protein were found for the protruding moieties of any of the five compounds, including the best binder 7d. Figure 6 shows a superposition of the compounds of this panel and their conserved binding mode in the aromatic cage.
Based on these data, we hypothesized that the planar benzene connected via an oxygen as in 7d is favorable for the affinity. Consequently, we designed a third panel of derivatives carrying this feature based on succinimide and glutarimide ( Figure 5B). In this panel, in addition to affinity improvement, we also sought to increase the solubility in water via additional polar or charged groups. As described in Scheme 3, reaction of the intermediates 8, 13, and 17 with phosgene solution (15 wt % in toluene) in THF yielded to the corresponding chloroformates 9, 14, and 18. Derivatives 11a, 11b, 16a, 16b, 20a, and 20b were prepared by a coupling reaction between the imides 3a or 3b and the aforementioned chloroformates 9, 14, and 18 using DIPEA in THF followed by deprotection using TFA. Treatment of the derivatives 11a and 11b with succinic anhydride and triethylamine (Et 3 N) in DMF produced the molecules 12a and 12b, respectively.
Indeed, the final compounds with terminal amino groups were water-soluble (11a, 11b, 16a, 16b), while 12a and 12b, with a terminal succinyl group, were highly soluble in bicarbonate buffer (>200 mM). The compounds were subsequently tested in the FRET assay, apart from 20a, which still contained impurities (see Methods). The assay indicated that all derivatives retained high affinity for CRBN independent of the binding moiety and planarity of the substituent; a clear preference for either binding moiety was not recognizable.
In soaking experiments, we obtained crystal structures with the 5-ring members 11a, 12a, and 20a, and the 6-ring members 16b, and 20b, all forming the canonical interactions within the aromatic cage. As expected, the 5-ring effectors form the interaction of the amide linker with N50, which is also observed in one instance for the 6-ring effector 16b. Further interactions of the protruding moieties are not observed for any of the compounds. Consequently, the prolonged extensions are less resolved in the electron density map, which is especially evident for 12a ( Figure 7). As these prolonged compounds still retain high affinity, this confirms that the amidosuccinimide scaffold can serve as the universal CRBN-binding moiety, allowing great chemical variability on the protruding moiety.
Degradation of Neo-Substrates. After assessing the biophysical and structural parameters of our designs, we tested their potential for the degradation of neo-substrates in the MMderived human cell line OPM-2. To this end, we have selected the established neo-substrates IKZF3, which is targeted via a

Journal of Medicinal Chemistry
Article variety of IMiDs 20,47−49 and CK1α, which is so far only targeted via lenalidomide, 21 as two complementary targets. We treated OPM-2 cells for 24 h with the different compounds and assayed for the endogenous levels of both neo-substrates; for comparison, the classical IMiDs, thalidomide, lenalidomide, and pomalidomide were included in the test set. As anticipated, the results for the two neo-substrates were very different. None of the marketed IMiDs apart from lenalidomide was able to reduce the levels of CK1α significantly, 21 but also none of our compounds showed any effect on CK1α. This indicates that our compounds could neither supersede nor sufficiently mimic the interface for CK1α recruitment formed by lenalidomide, substantiating the notion of a very narrow specificity window for this substrate. 21 However, the situation was completely different for IKZF3. Here, in addition to all classical IMiDs, multiple of our designs were successful: significant effects were observed for 5a, 7d, and 7f, which are all compounds with a rather compact structure from our first and second panel ( Figure 8). The only successful compound from the first, phthaloyl-based panel, 5a, represents a

Journal of Medicinal Chemistry
Article direct succinimide analogue of pomalidomide. It reduced IKZF3 levels by almost 40%. Comparison of its structural binding mode to that of thalidomide or pomalidomide already suggested that it may be a possible functional substitute for pomalidomide. The fact that the other compounds of that panel did not show similar effects indicate that substitutions larger than an amino group in the R 2 position of the phthaloyl moiety, as for 4a and 4c, or any substitution in R 1 , as for 4b, 4d, and 5b, abolish IKZF3 recruitment.
Most interesting is the analysis of the successful compounds of the second panel, 7d and 7f, which are not derived from classical IMiDs but from the hydrolysis products of the first panel. 7d, which is also the highest affinity binder in this study, reduced IKZF3 levels by about 20%. Moreover, the analogous 7f, which is a weaker binder that only differs from 7d in one heavy atom in the linker, reduced IKZF3 levels by even 40%. Both compounds are more flexible and elongated than 5a, so it appears impossible that their protruding moieties adopt a conformation mimicking the classical IMiDs at the interface between CRBN and IKZF3. Their terminal benzyl groups inevitably project further away from the thalidomide-binding pocket, requiring another mode of interaction with the zinc finger motif than characterized for classical IMiDs. 26 This interaction mode is also very sensitive to chemical changes: shortening of the linker by only one atom, as for 7b, further modifications on the benzyl group, as for 11a or 12a, or any other variant we designed in the second or third panel abolished IKZF3 recruitment, which is indicative for a highly specific interaction. Therefore, 7d and 7f are the first representatives of a novel type of CRBN effectors with a recognition mode that is clearly distinct from that of the classical thalidomide-based IMiDs.

■ CONCLUSIONS
The development of CRBN effectors is a rapidly growing field, and novel IMiDs and PROTACS are reported frequently. To date, essentially all of these compounds are based on the classical thalidomide scaffold, which significantly restrains the chemical space available for the recruitment of neo-substrates. Here, following on from our previous characterization of CRBNbinding moieties, we have probed the chemical space for the linking and protruding moiety, taking advantage of the structural binding modes of hydrolyzed metabolites. Although not in the focus of this work, the apparent specificity of CRBN for these hydrolysis products, especially for the major thalidomide metabolite CBG, may inspire further research toward the understanding of their pharmacological relevance.
As a consensus from the binding modes of our initial designs and their hydrolysis products, we derived an amidosuccinimide scaffold as a minimal binding and linking moiety. 3-Amidosuccinimide can be used to mount almost arbitrary chemistry as the protruding moiety, while retaining affinities to CRBN in the range of classical IMiDs, rendering it an attractive CRBN-binding moiety for future PROTACS.
However, the most relevant aspect of our study concerns the versatility of the 3-amidosuccinimide scaffold for the design of IMiD-like CRBN effectors. As a minimal binding and linking moiety, it does not restrain the chemical space of protruding moieties, allowing for the design of substrate-recruiting motifs that cannot be realized on a classical IMiD scaffold. In our attempt to probe the effect of different protruding moieties mounted on the amidosuccinimide scaffold, two out of twelve compounds were able to recruit the neo-substrate IKZF3 without further optimization. Although we do not have structural insight into the IKZF3-recognition mode of these effectors, it is clearly incompatible with that of the classical IMiDs, which points at a large unexplored chemical space for the recruitment of new therapeutic targets. For the ongoing characterization of the proteome druggable via the IMiD approach, novel CRBN effectors are needed both for probing  (A) Immunoblot analysis of IKZF3 levels after treatment with compounds 7d, 5a, and 7f (100 μM) for 24 h, compared to DMSO (control) and lenalidomide (20 μM) (n = 3). Lenalidomide at 100 μM reduced IKFZ3 levels effectively by 100% (not shown). (B) Averaged IKZF3 levels from the three independent experiments, normalized against total protein loading control ( Figure S2). The significance of the data was tested comparing IKZF3 levels in the presence of the compounds and DMSO (p > 0.05).

Journal of Medicinal Chemistry
Article the space of neo-substrates, as well as for future pharmacological exploitation. With this work, we contribute first steps toward the rational design of a post-thalidomide generation of such effectors, toward unlocking the full potential of the IMiD approach.
■ EXPERIMENTAL SECTION Chemistry. General Procedures. Room temperature (RT) refers to 22°C. Reagents and anhydrous solvents were transferred via an ovendried syringe or a cannula. Flasks were flame-dried under vacuum and cooled under a constant stream of argon. Anhydrous solvents (toluene, dioxane, and DMF) were purchased from Sigma-Aldrich (anhydrous over molecular sieves). THF was dried over potassium. All other chemicals were purchased from ABCR, Acros, Alfa Aesar, Fluorochem, Merck, Sigma-Aldrich, and TCI Europe at highest commercially available purity and used without further purification. Compound 2b and 3b were prepared according to the procedures reported by Capitosti et al. 50 Compounds 8,13, and 17 were prepared according to the procedures previously reported. 51 Coupling constants J are reported in Hertz (Hz). Multiplicities are indicated by s = singlet, d = doublet, t = triplet, q = quartet, sep = septet, dd = doublet of doublet, dt = doublet of triplet, m = multiplet, and br = broad resonance. High-resolution mass spectra were obtained on Bruker Daltonics ESI-FT-ICR-MS APEX II. IR spectra were measured on ATI/Mattson Genesis FT-IR as thin film (in CCl 4 ) or KBr-disk. Absorbance frequencies are reported in reciprocal centimeters (cm −1 ). Melting points were measured on a Boetius-micro hot stage and are uncorrected. The purity of compounds was analyzed by detecting UV absorbance at 254 nm using a Poroshell 120 EC-C18 column on a 1260 Infinity II system (Agilent Technologies, Inc.) ( Figure S1). All compounds showed >95% purity with the exception of 7a (87.6%), 7b (91.5%), 7c (88.2%). 12a (89.4%), 12b (88.2%), and 20a (<50%).
General Procedure A. Preparation of the compounds 4a−d: to a solution of 3a or 3b (1.0 equiv) in anhydrous THF (4.60 mL/mmol), 4nitrophthalic anhydride (1.5 equiv) in anhydrous THF (0.5 mL/mmol) was added. CDI (1.7 equiv), Et 3 N (2.9 equiv), and catalytic amounts of 4-DMAP were added, and the reaction mixture was heated to reflux for 5 h. The solvent was evaporated, and the resulting crude oil was purified by column chromatography to give 4a−d.
General Procedure B. Reduction of the nitro group: to a solution of 4a or 4b (1.0 equiv) in EtOAc (100 mL/mmol), 10 wt % Pd/C (27 mol %) was added. The flask was degassed and let to stir under a hydrogen atmosphere for 20 h at RT. After completion of the reaction (TLC control), the mixture was filtered over Celite and washed with copious amounts of acetone. Afterward, the solvent was evaporated and 5a or 5b was obtained.
General Procedure C. Preparation of the compounds 7a−h (amide bond formation): to a solution of the TFA-salt 3a (1.0 equiv) in THF (5 mL/mmol), DIPEA (2.0 equiv) was added. The mixture was cooled down to 0°C, the acyl chloride 6a−h (1.0 equiv) was added, and the reaction mixture was left to stir under reflux for 2 h. Upon completion, the reaction was diluted with EtOAc and washed with 3 M HCl and brine. The organic layer was dried over Na 2 SO 4 , and the solvent was evaporated in vacuo. The resulting crude product was purified by flash column chromatography to give 7a−h.
General Procedure D. Phosgenation, carbamate formation, and Boc deprotection: to a 15% wt solution of COCl 2 in toluene (1.3 equiv) was added dry THF (0.58 mL/mmol) under argon. The solution was cooled to 0°C, and a solution of 8, 13, or 17 (1.0 equiv) in dry THF (0.6 mL/mmol) was added over 10 min. The reaction mixture was stirred at 0°C for 15 min and then warmed up to RT and stirred for two more hours. After completion of the reaction, the volatiles were completely evaporated under reduced pressure to give the chloroformates 9, 14, or 18, respectively, which were used for the next step without further purification. To a stirred suspension of the TFA-salt 3a or 3b (1.03 equiv) in dry THF (2 mL/mmol), DIPEA (0.35 mL/ mmol) was added at 0°C. Then, a solution of the chloroformate 9, 14, or 18 (1.0 equiv) in dry THF (0.35 mL/mmol) was added gradually to the suspension. The reaction was warmed up to RT, stirred for 30 min at RT, and then refluxed overnight. After completion of the reaction, the solvent was evaporated, and the crude product was purified by column chromatography giving the carbamates 10a−b, 15a−b, or 19 respectively. To a solution of 10a−b, 15a−b, or 19 (1.0 equiv) in DCM (4.5 mL/mmol), TFA (30.0 equiv) was added at 0°C. The reaction solution was warmed up to RT and stirred for 2 h. After completion of the reaction, the volatiles were completely evaporated to give 11a−b, 16a−b, or 20 respectively.
General Procedure E. Preparation of 12a−b (amide bond formation): derivatives 11a−b (1.0 equiv) were dissolved in DMF (2.8 mL/mmol), and Et 3 N (2.2 equiv) was added at 0°C. After 20 min of stirring, succinic anhydride (1.0 equiv) was added, and the reaction mixture was stirred at RT overnight. After the end of the reaction, the solvent was evaporated, and the corresponding acid 12a−b was afforded via flash column chromatography.
FRET Assay. FRET-based binding assay was performed using WT MsCI4 as described previously. 56 Compounds were dissolved in DMSO with the exception of 11a/b, 16a/b, and 20a/b, which were water-soluble, and compounds 12a/b that were dissolved in bicarbonate buffer. Because of remaining impurities, 20a was not tested in this assay and only used for structural studies. For compounds 4a, 4c, 5a, 7a, 7b, 7e, and 7g, binding was observed but could not be quantified, as saturation at higher concentrations was not appreciable in their binding curves. K i values are summarized in Table S1.
Crystallography. For crystallization, MsCI4 and humanized MsCI4 protein solution were concentrated to 17 mg/mL and mixed with either 3 mM thalidomide (MsCI4 and humanized MsCI4) or 7d (MsCI4). Crystallization trials for humanized MsCI4·thalidomide, which turned out to be humanized MsCI4·CBG, and MsCI4·7d were performed via the vapor diffusion technique at 294 K in sitting-drops. 400 nL of protein solution was mixed with 400 nL of reservoir solution with the Honeybee 963 robot (Genomic Solutions Ltd.). Diffraction-quality crystals of humanized MsCI4·CBG were obtained in a condition containing 20% PEG 3350 and 0.2 M ammonium acetate, and crystals of MsCI4·7d were obtained using 15% PEG 6 K and 5% glycerol. For all other compounds, crystals were prepared in hanging drop experiments by mixing 1 μL 0.4 M (NH 4 ) 2 HPO 4 ground solution with 1 μL MsCI4· thalidomide solution. Crystals grown in this setup were transferred to 3 μL ground solution spiked with individual compounds for soaking. After 36 h, crystals were cryoprotected by streaking through a 50% PEG 3350 solution. Crystals were flash-cooled in liquid nitrogen, and diffraction data were collected at 100 K and a wavelength of 1 Å on beamline X10SA at the Swiss Light Source using a PILATUS 6M-F hybrid pixel detector (Dectris Ltd.). Data were processed and scaled using XDS. 57 Structures of soaking experiments were solved based on the MsCI4·thalidomide coordinates (PDB 4v2y). The cocrystal structures of the complexes with CBG and 7d were of other crystal forms and were solved using molrep 58 with 4v2y as a search model, locating four and two chains in the ASU, respectively. All structures were rebuilt using Coot 59 and the integrated suite Lidia for chemical

Journal of Medicinal Chemistry
Article structures and generation of restraints. The models were finalized via cyclic modeling in Coot and refinement using REFMAC5. Molecular figures were generated using PyMOL. 60 Data collection and refinement statistics are summarized in Table 1.
Cell Culture and Western Blot. OPM-2 cells were routinely cultivated in 90% RPMI 1640 + 10% FBS and 1% penicillin/ streptomycin. Cells were split to 0.3 × 10 6 cells/mL 2 days prior to experiments. 4 mL of culture per well was pipetted in a sterile

Journal of Medicinal Chemistry
Article environment. Stock solutions of lenalidomide were prepared at concentrations of 80 and 400 mM and stock solutions of test compounds were prepared at a concentration of 400 mM. For the assay, 10 μL of stock solutions was added to the 4 mL cell culture. DMSO controls were prepared in the same manner, leading to a final DMSO concentration of 0.25%. After 24 h of incubation, cell solution was centrifugated at 500g. After one washing step with ice cold PBS, cell pellets were resuspended in 35 μL of lysis buffer (20 mM Hepes, 175 mM NaCl, 1% NP40, 2 mM MgCl 2 ) on ice and supplemented with 0.5 μL Benzonase. Samples were resolved by Mini-PROTEAN TGX gels (Bio-Rad) and activated under UV light, before transfer to lowfluorescence PVDF membranes, according to manufacturer's protocol. Membranes were imaged before blocking in 5% milk in PBS-T and incubating with primary antibodies anti-IKZF3 (15103S, Cell Signaling Technology, Inc.) and anti-CK1α (ab108296, Abcam plc). After overnight incubation, horseradish peroxidase conjugated secondary antibodies goat anti-rabbit (111-035-144, Jackson Immuno Research) and goat anti-mouse (115-035-003, Jackson Immuno Research) were used at 1:20 000 dilutions for detection of bands by chemiluminescence (ECL Vilbert). Protein bands were detected and integrated using the Bio-Rad analysis suite. Endogenous protein levels were normalized using Stain-Free Technology ( Figure S2) and analyzed by one-way ANOVA, according to the Holm-Sidak method integrated into SigmaPlot. The linearity of anti-IKZF3 mAb was tested via serial dilutions of cell extracts ( Figure S3).

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.9b00454.
HPLC traces of tested compounds, western blot quantification and normalization IKZF3, concentration response curve for IKZF3 mAb, and K i values of tested compounds (PDF) Molecular formula strings (CSV)