Macrocyclic FKBP51 Ligands Define a Transient Binding Mode with Enhanced Selectivity

Abstract Subtype selectivity represents a challenge in many drug discovery campaigns. A typical example is the FK506 binding protein 51 (FKBP51), which has emerged as an attractive drug target. The most advanced FKBP51 ligands of the SAFit class are highly selective vs. FKBP52 but poorly discriminate against the homologs and off‐targets FKBP12 and FKBP12.6. During a macrocyclization pilot study, we observed that many of these macrocyclic analogs have unanticipated and unprecedented preference for FKBP51 over FKBP12 and FKBP12.6. Structural studies revealed that these macrocycles bind with a new binding mode featuring a transient conformation, which is disfavored for the small FKBPs. Using a conformation‐sensitive assay we show that this binding mode occurs in solution and is characteristic for this new class of compounds. The discovered macrocycles are non‐immunosuppressive, engage FKBP51 in cells, and block the cellular effect of FKBP51 on IKKα. Our findings provide a new chemical scaffold for improved FKBP51 ligands and the structural basis for enhanced selectivity.


Introduction
Proteins often cluster in families with similar structure. Thed iscovery of selective ligands that can discriminate between these close homologs remains af ormidable challenge in chemical biology as well as in drug development. Most proteins are flexible and differential dynamics have been suggested as aw ay to distinguish between otherwise very similar proteins. [1] Atypical example is the family of FK506-binding proteins (FKBPs) that possess ah ighly conserved binding pocket ( Figure 1A)b ut have diverged to perform diverse biological functions.T he larger homolog FKBP51 [2] is ar egulator of glucocorticoid receptor (GR) signaling [3] and has emerged as apotential target for depression, [4] obesity-induced diabetes [5] and chronic pain. [6] In contrast, the homologous proteins FKBP12, FKBP12.6 and FKBP52 are considered anti-targets due to their important roles in cardiology,s exual development and female infertility,emphasizing the need for selective inhibition. [7] Them ost advanced FKBP51 ligands are compounds of the SAFit class (Selective Antagonists of FKBP51 by induced fit), [8] which bind to atransient binding pocket unavailable to FKBP52 [1] and are up to 10 000-fold selective for FKBP51 over FKBP52. [8,9] However,S AFit-like ligands still bind FKBP12 and its isoform FKBP12.6 with substantial affinities. These FKBPs are cofactors of the ryanodine receptor [10] and play an important role in fine-tuning the excitability of smooth or heart muscle.F KBP12 knockout or knockdown lead to severe cardiac defects in mice, [11] underscoring the importance of selectivity for FKBP51 over FKBP12/12.6 in FKBP51-based therapies.
Macrocyclization is apopular approach to improve druglike properties for compounds outside the rule-of-five space [12] and is thought to be crucial for the unusually beneficial properties of the clinically used natural products FK506 ( Figure 1B), Rapamycin and Cyclosporin. [13] Macrocyclization was key to enhance the affinities or physicochemical properties of synthetic ligands for FKBP12 [14] and cyclophilins. [15] In ap ilot study on the macrocyclization of SAFit analogs, [16] we surprisingly observed ar earrangement of the FKBP51 binding pocket, which in turn allowed an unprecedented selectivity against the off-targets FKBP12 and FKBP12.6.

Results and Discussion
Based on the structure-affinity relationship findings of SAFit analogs [8,9] and the highly conserved binding mode ( Figure 1C), we chose to keep the pipecolate,t he chalconederived A/B rings,and cyclohexyl ring constant and to cyclize between the latter two.The synthesis started from compound 1, [8] where the ketone was reduced by an asymmetric Noyori catalyst to the chiral alcohol 2 and then coupled with allyl bromide or linker 3 to the alcohols 4a/b (Scheme 1). After coupling with Fmoc-S-pipecolate, [17] 5a/b were deprotected and coupled with 6. [9b] Thel inear precursors 7a/b were cyclized by RCM to yield 8a and b.F or 8a we were able to separate both E and Z isomer (ratio in crude mixture = 89:11) and for the larger macrocycle 8b we only observed and isolated the E isomer. Unfortunately,n one of these macrocycles showed detectable binding to FKBP51 in afluorescence polarization assay. [18] To introduce additional functionalities into the linker,w efurther derivatized the E isomers of 8a/b by Wacker oxidation, dihydroxylation or hydrogenation (9ag), which for 8a/b resulted in only one Wacker product (9b/f). After dihydroxylation of the smaller macrocycle,weobtained the diastereomers (9c/d)a nd an inseparable dihydroxylated diastereomeric mixture (9g,d r= 1:1b yN MR). Gratifyingly, the dihydroxylated derivative 9g of the larger macrocycle bound to FKBP51 with a K i of 1.2 mm,w hereas for 9a-f no binding to FKBP51 could be detected. To our great surprise, 9g did not bind to FKBP12 or FKBP12.6. Therefore,w es et out to investigate this finding in more detail, resorting to amino acids to rapidly explore the effect of the linker (Scheme 2). Using an Fmoc SPPS strategy,w e started from the immobilized SAFit1 precursor 10 [19] to introduce d-cyclohexyl glycine as the FKBP52-discriminating moiety,f ollowed by coupling with av ariety of amino acids yielding the immobilized intermediates.T he deprotection of 11 had to be optimized to suppress diketopiperazine formation. [20] Prior to derivatization of the intermediates 12 a an optional N-methylation sequence was included to probe the influence of the resulting amide groups (12 a-c). [21] Finally,the linear peptides 12 a or 12 c were cleaved from the resin and cyclized by macrolactamization (13 a-o).
All final compounds were screened for affinity towards the FKBPs 12, 12.6, 51 and 52 in ac ompetitive fluorescence polarization assay (Table 1). [18,22] Gratifyingly,m ost of the macrocycles with amino acidbased linkers bound to FKBP51 in the low to submicromolar range and as expected none to FKBP52 (not shown). The glycine derivative 13 a had an affinity of 2.3 mm,w hich gradually increased with increasing substitution ( In contrast, the l-Ala derivative 13 j bound more weakly, consistent with the substantially reduced affinity of the l-Pro derivative 13 k.Longer linkers such as b-Ala 13 l and GABA (no binding,n ot shown) displayed reduced affinity,w hich could be compensated by appropriate rigidification as in 13 m (1.8 mm ;o ther diastereomers were inactive,n ot shown). Notably,n one of the tested linear precursors bound to FKBP51, underlining the significance of the macrocyclization. Most importantly,however, none of the macrocycles did show any affinity towards FKBP12 or FKBP12.6.
Thea ffinity of 13 a and d for FKBP51 was confirmed by isothermal calorimetry (ITC), yielding an enthalpy-driven K d = 3.6 mm AE 0.9 mm for 13 a and K d = 0.6 mm AE 0.1 mm for 13 d,respectively ( Figure S1).
We also prepared the fluorescent analog 14 of the best binding compound 13 c (Figure 2A,synthesis see Scheme S1), which bound in af luorescence polarization assay with high affinity to FKBP51 (K d = 45 AE 7nm)b ut poorly to FKBP12, FKBP12.6 or FKBP52 ( Figure 2B). Theaffinity of the tracer was further confirmed in aF RET assay with af luoresceinlabeled FKBP51FK1 domain (K d = 80 AE 10 nm ;F igure S2).
To check if compounds of the new class of macrocycles were able to engage FKBP51 inside cells,w ep erformed aN anoBRET assay using at ransiently expressed FKPB51-NLuc construct. Theassay utilizes afluorescent tracer for the FKBP51-Nluc construct, which accepts the luminescent energy to generate aB RET signal. If compounds engage the FKBP-Nluc construct inside cells,the tracer is displaced, reducing the BRET signal and allowing direct quantification  SAFit1 No linker 0.004 AE 0.001 [b] 0.163 AE 0.009 [b] 0.019 AE 0.002 [b] FK506 Figure 1A 0 of FKBP51-NLuc occupation. Representative macrocyclic compounds 13 c, 13 e and 13 h as well as SAFit1 and -2 all dose-dependently competed with af luorescent NanoBRET tracer inside cells ( Figure 2C). Thel ower potencies of the macrocycles compared to SAFit1 and -2 are in line with the lower affinities of the macrocycles. FK506 works as an immunosuppressant by an FKBP12dependent gain-of-function mechanism. Ac ellular analysis showed that compound 13 d has neither immunostimulatory nor immunosuppressive properties on its own ( Figure 2D). Importantly,u nlike the pan-selective FKBP ligand [4.3.1]16h [21] (Figure S6A) 13 d also did not block the immunosuppressive activity of FK506, in line with its selectivity against FKBP12 ( Figure 2D).
We next explored if the macrocyclic ligands could interfere with the cellular functions of FKBP51. We therefore treated SIM-A9 cells,w hich were recently discovered as aSAFit-sensitive cellular model for stress-mediated secretory autophagy. [24] Compound 13 d ( Figure 2E)a sw ell as compounds 13 c, 13 e, 13 h and 13 i ( Figure S6B) all inhibited IKKa phosphorylation, similar to SAFit1 and SAFit2 [4f] (Figure 2F and S6C). Forcompounds 13 e, 13 i,SAFit1 and SAFit2 aclear dose-dependence was observed. For 13 c, 13 d and 13 h,t he apparent maximal inhibition was already reached at the lowest tested concentration of 1 mm,p ossibly reflecting the higher affinities and/or improved cell permeability of 13 c, 13 d and 13 h compared to 13 e and 13 i.T aken together,these results show that the here discovered macrocycles can penetrate human cells,i ntracellularly occupy FKBP51 and interfere with its function.
To clarify the structural basis for this unprecedented selectivity,wesolved the cocrystal structures of 13 a, 13 d and 13 h in complex with FKBP51 ( Figure 3A and S7A/C;P DB-ID:7AOU, 7AOT,7AW F). [25] As intended, the interactions of the pipecolate,the A-and B-rings,a sw ell as the cyclohexyl group with FKBP51 were completely conserved in comparison to previous FKBP51-SAFit co-crystal structures ( Figure 3A). This includes ad isplacement of F 67 ,w hich is responsible for the strong selectivity vs.F KBP52 of SAFit-like ligands. [8,9] However, the b3b strand, which contains F 67 and which we and others previously showed to display enhanced basal mobility, [1,26] was substantially rearranged ( Figure 3B and S7B/D). Strikingly, we observed that the carbonyl group of 13 a, 13 d and 13 h displaced D 68 and replaced it as ahydrogen bond acceptor for the e-hydroxy group of Y 57 .T he rearrangement of the b3b strand is stabilized by an inward flip of H 71 ,w hich partially replaces S 70 and substitutes the former as ah ydrogen bond donor for the backbone carbonyl of Y 57 .Asimilar inward flip of H 71 has previously been observed for FKBP51 in complex with Rapamycin and FRB (PDB-ID:4DRH). [27] Intriguingly, H 71 is replaced in FKBP12 and FKBP12.6 by an arginine (R 40 in FKBP12/12.6 numbering), which can be expected to be less efficient in stabilizing the 13 a-binding conformation, providing am olecular rationale for the discrimination vs.F KBP12/ 12.6 observed for the macrocycles.
To clarify if the structural rearrangement of the b3b strand was also stabilized in solution, we developed as et of conformation-sensitive assays [28] that are responsive to alterations of the b3b strand. Towards this end, we introduced environment-responsive dyes selectively at positions 58, 60 and 65 in the b2strand below the b3strand or in the b2-b3b loop ( Figure 3C). Remarkably,a ll three sensors clearly differentiated between ligands with canonical and aS AFitlike binding mode ( Figure 3D and S8A/B). When using the K 60 C-and K 65 C-based sensors,compound 13 d induced similar changes in fluorescence lifetime as SAFit1, but with lower potencyi na ccordance with its lower affinity.H owever,t he K 58 C-based sensor clearly differentiated macrocycle 13 d from in complex with 13 d (pink-colored sticks) superimposed to the cocrystal structure of the SAFit1-analog iFit4 in complex with FKBP51FK1 (pale green, PDB-ID:4 TW7, iFit4 has been omitted for clarity). The b3b strand is highlighted in cyan and green, respectively,and key residues Y 57 ,D 68 and H 71 are shown as sticks. The key carbonyl of 13 d displacing D 68 is highlighted in magenta and the new hydrogen bond between Y 57 and the 13 d carbonyl is indicated as orange broken line (distance annotated in ). C) Side view of FKBP51 from cocrystal structures with 13 d (pale cyan, 7AOU), with the SAFit analog iFit4 (pale green, 4TW7), and the conventional binding-mode ligand [4.3.1]-16h(pale magenta, 5OBK). The residues F 67 ,D 68 and H 71 are shown as pale sticks. K 58 ,K 60 and K 65 as the attachmentpoint for the environment-sensitive dye are highlighted as intense colored sticks. D) Fluorescence life-time analysis of DBD-labeled FKBP51FK1 K58C in the presence of the indicated ligands.
both SAFit1 as well as canonical FKBP ligands ( Figure 3D). This strongly suggests that the new peptide-based macrocycles stabilize an ew conformation in solution that is different from the known FK506-like or SAFit-like ligands.

Conclusion
Macrocycles have repeatedly been discussed to impart improved physicochemical properties.H owever,t hey have rarely been associated with selectivity.H ere we show that macrocycles can also provide the basis for subtype selectivity. In this particular case,t he macrocycles provide the scaffold for the proper positioning of the key carbonyl group that displaces Asp 68 ,which was not possible in the linear analogs. With regards to FKBP51, our results provide the first ligands that robustly discriminate between FKBP51 and FKBP12/ FKBP12.6 and provide as tructural basis for the rational design for further optimization regarding affinity,s tability, specificity and cellular activity.