Leveraging Fungal and Human Calcineurin-Inhibitor Structures, Biophysical Data, and Dynamics To Design Selective and Nonimmunosuppressive FK506 Analogs

ABSTRACT Calcineurin is a critical enzyme in fungal pathogenesis and antifungal drug tolerance and, therefore, an attractive antifungal target. Current clinically accessible calcineurin inhibitors, such as FK506, are immunosuppressive to humans, so exploiting calcineurin inhibition as an antifungal strategy necessitates fungal specificity in order to avoid inhibiting the human pathway. Harnessing fungal calcineurin-inhibitor crystal structures, we recently developed a less immunosuppressive FK506 analog, APX879, with broad-spectrum antifungal activity and demonstrable efficacy in a murine model of invasive fungal infection. Our overarching goal is to better understand, at a molecular level, the interaction determinants of the human and fungal FK506-binding proteins (FKBP12) required for calcineurin inhibition in order to guide the design of fungus-selective, nonimmunosuppressive FK506 analogs. To this end, we characterized high-resolution structures of the Mucor circinelloides FKBP12 bound to FK506 and of the Aspergillus fumigatus, M. circinelloides, and human FKBP12 proteins bound to the FK506 analog APX879, which exhibits enhanced selectivity for fungal pathogens. Combining structural, genetic, and biophysical methodologies with molecular dynamics simulations, we identify critical variations in these structurally similar FKBP12-ligand complexes. The work presented here, aimed at the rational design of more effective calcineurin inhibitors, indeed suggests that modifications to the APX879 scaffold centered around the C15, C16, C18, C36, and C37 positions provide the potential to significantly enhance fungal selectivity.

simulations, Aspergillus fumigatus, calcineurin, molecular dynamics, Mucor circinelloides, nuclear magnetic resonance, structural biology I nvasive fungal infections are a leading cause of death in immunocompromised patients. More than 1.6 million people die annually of infections caused by the major fungal pathogenic species of Aspergillus, Candida, Cryptococcus, and Mucorales (1). Due to rapidly emerging drug resistance to existing antifungals targeting the fungal cell wall and membrane, there is an urgent need to design more efficacious and highly selective antifungals targeting other critical fungal cellular pathways. However, this poses a fundamental challenge as both fungi and humans are eukaryotes and share many orthologous proteins and pathways (2). Recent structure-based inhibitor binding studies on the fungal heat shock protein 90 (Hsp90) have demonstrated the feasibility of increasing fungus-selective targeting of Hsp90 (3,4).
Calcineurin, the target of the immunosuppressive macrocyclic drug FK506 (tacrolimus) and the cyclic peptide cyclosporine (CsA), is a promising target for the development of effective antifungal drugs (5). Calcineurin plays central roles in fungal growth, pathogenesis, cellular stress responses, and drug tolerance/resistance (6). The calcineurin protein complex consists of a catalytic subunit, calcineurin A (CnA), and a regulatory subunit, calcineurin B (CnB) (7). The immunosuppressants first bind to their respective immunophilins, FKBP12 (12-kDa FK506 binding protein) and CypA (cyclophilin A), which subsequently bind to calcineurin in a groove between the CnA and CnB subunits. The immunophilin-immunosuppressant complexes inhibit calcineurin serinethreonine phosphatase activity, blocking the dephosphorylation of downstream targets, such as the human nuclear transcription factor NFAT (nuclear factor of activated T cells), involved in T-cell activation and interleukin-2 transcription, and the fungal transcription factor Crz1 (NFAT homolog), implicated in virulence, stress response, and thermotolerance (8)(9)(10)(11)(12). In humans, this leads to potent immunosuppression and is critical in preventing graft rejection but also precludes FK506 and CsA usage as antifungals in immunocompromised patients.
FKBP12 proteins are members of the FKBP PPIase (peptidyl-prolyl isomerase) superfamily and catalyze the cis-trans isomerization of proline imidic peptide bonds (13)(14)(15)(16)(17)(18)(19)(20). Our recent characterization of crystal structures of pathogenic fungal calcineurin-FKBP12 complexes bound to FK506 highlighted the overall conservation of the FKBP12-FK506 and calcineurin-FK506-FKBP12 structures (21,22). Molecular dynamic (MD) simulations utilizing these structures provided critical insight into differential binding of FK506 to human versus fungal FKBP12, especially in the 40's and 80's loops that define the FKBP12 binding cavity for FK506, enhancing our ability to specifically target fungal calcineurin to reduce mammalian immunosuppressive activity. Nuclear magnetic resonance (NMR) titrations and genetic mutations of the Aspergillus fumigatus calcineurin-FK506-FKBP12 complex identified a Phe88 residue in the 80's loop, not conserved in human FKBP12, as critical for binding and inhibition of fungal calcineurin. We leveraged our structural data and synthesized an FK506 analog, APX879, modified with an acetohydrazine moiety on the FK506-C 22 position (Fig. 1A). Based on our crystal structures, we proposed that APX879 would interact less favorably with the human FKBP12 (hFKBP12) His88 residue than with the corresponding A. fumigatus FKBP12 (AfFKBP12) Phe88 residue, thus potentially reducing immunosuppression (22). In fact, APX879 displayed a 71-fold-reduced immunosuppressive activity compared to FK506 (22). APX879 also maintained broad-spectrum in vitro antifungal activity against a wide range of human-pathogenic fungi, including A. fumigatus and Cryptococcus neoformans (MIC, 0.5 to 1 mg/ml), Mucor circinelloides (MIC, 2 to 4 mg/ml), and Candida albicans (MIC, 8 mg/ml), albeit reduced in comparison to FK506. In addition, in vivo testing in a murine model confirmed reduced immunosuppressive activity and confirmed efficacy in a cryptococcal model of invasive fungal infection (22). Taken together, a 4-fold improvement in the therapeutic window was noted when comparing the 71-fold reduction in immunosuppressive activity and a 17-fold reduction in the efficacy in C.
neoformans. These studies established the proof-of-concept of targeting fungal calcineurin for the design of more potent FK506 analogs to improve antifungal activity.
To guide the design of nonimmunosuppressive FK506 analogs selectively targeting fungal calcineurin, here we quantitatively compared FK506 and APX879 binding to the human and fungal FKBP12 proteins from A. fumigatus and M. circinelloides through a combination of genetic, structural, and biophysical approaches. As M. circinelloides is an emerging human pathogen recalcitrant to many current antifungals, efficient targeting of calcineurin aided by the molecular characterization of the FKBP12 protein (McFKBP12) is of utmost importance (23,24). Here, we report the first high-resolution crystal structures of McFKBP12 bound to FK506 in addition to the hFKBP12, AfFKBP12, and McFKBP12 proteins bound to APX879 (22). Through genetic studies, we demonstrate that McFKBP12 does not functionally complement AfFKBP12 and reveal key requirements of FKBP12 residue 88 for "productive" binding and inhibition of calcineurin. While FK506 binds to both human and fungal FKBP12 proteins with high affinity (2 to 5 nM), APX879 binds 40-to 100-fold less tightly (120 to 450 nM). Strikingly, the human and fungal FKBP12 proteins show different responses to APX879 binding as observed by NMR, isothermal calorimetric titration (ITC) experiments, and molecular dynamic (MD) simulations that are not readily apparent in the static X-ray crystal structures. Furthermore, MD simulations allowed quantitative comparison of the significance of the FKBP12-ligand interactions between the human and fungal proteins. This analysis reveals regions of the ligands that could be altered to enhance selectivity toward the fungal FKBP12 proteins. Our approach highlights the potential of combining structural, genetic, biophysical, and in silico methodologies to fully describe and identify variation in protein-ligand interactions involving structurally similar proteins. We take a rational approach to understand the balance between the immunosuppressive and antifungal activities of FK506 and a new, less immunosuppressive FK506 analog, APX879, in an attempt to broaden the therapeutic window for the development of efficacious antifungals.

RESULTS
Crystal structure of M. circinelloides FKBP12 bound to FK506. McFKBP12 shares 58% and 65% sequence identity with hFKBP12 and AfFKBP12, respectively. Sequence variations from hFKBP12 are located in (i) b2, back wall of the FK506-binding pocket leading into the 40's loop; (ii) the 50's loop; and (iii) b4, leading into the 80's loop ( Fig. 1B and Fig. 2A). Crystal structures of the apo and FK506-bound hFKBP12 and AfFKBP12 proteins have been reported previously (21,25,26). To correlate sequence variations with structure and identify potential differences between McFKBP12 and other FKBP12 proteins, we attempted to crystallize McFKBP12 in its apo form. All attempts failed to yield protein crystals, hinting at potentially high conformational flexibility of McFKBP12. However, crystals were obtained with the FK506-bound form (2.5 Å, P3 2 21), suggesting rigidification of McFKBP12 by FK506 binding (Table 1).
McFKBP12 shares the same fold as hFKBP12 and AfFKBP12 (Ca-RMSD [root mean square deviation], ;0.5 Å) with a 5-stranded b-sheet wrapping around an a-helix ( Fig. 2A) involving the backbone of residues Gln55 and Ile57 and 4 to the side chains of residues Tyr27, Asp38, and Tyr83 (Fig. 2B). Despite the sequence differences between McFKBP12 and AfFKBP12, they share high structural similarity, prompting investigation into their functional equivalence.
M. circinelloides FKBP12 does not functionally complement A. fumigatus FKBP12 in calcineurin inhibition. The deletion of the A. fumigatus AfFKBP12-encoding gene leads to FK506 resistance, establishing its central role for calcineurin inhibition (27). To verify if McFKBP12 can functionally complement AfFKBP12, an A. fumigatus strain expressing McFKBP12 was generated through genetic replacement of the Affkbp12 native locus with Mcfkbp12. Growth assays in the presence of increasing concentrations of FK506 revealed that McFKBP12 does not restore A. fumigatus FK506 sensitivity, indicating that McFKBP12 may not bind or may bind but not inhibit A. fumigatus calcineurin in vivo (Fig. 3A), consistent with previous analysis using hFKBP12 (22). Using structure and sequence alignments, a single mutation of hFKBP12-His88 to Phe (AfFKBP12 identity) was shown to restore FK506 sensitivity (22). Interestingly, in McFKBP12 residue 88 is a Tyr, potentially sterically hindering the ternary complex formation with A. fumigatus calcineurin. To test this hypothesis, we generated an McFKBP12-Y88F mutant and confirmed restoration of FK506 sensitivity in A. fumigatus (Fig. 3A). We also verified the calcineurin-FK506-FKBP12 complex formation in vivo using green fluorescent protein (GFP)-tagged McFKBP12 and McFKBP12-Y88F expression constructs by fluorescence microscopy (Fig. 3B). In the absence of FK506, as observed with AfFKBP12, the McFKBP12 and McFKBP12-Y88F proteins localize in the cytoplasm and nuclei. Upon FK506 addition, both proteins localize to the septum as observed with native AfFKBP12, demonstrating the formation of the ternary complex with calcineurin at the hyphal septum, independent of the Y88F mutation. This highlights the central role of residue 88 in the formation of a "productive" inhibitory complex with A. fumigatus calcineurin. Despite the binding of McFKBP12 (and hFKBP12) to  Crystal structures of the human and fungal FKBP12 proteins bound to APX879. The FK506 analog APX879, substituted on the C 22 -ketone with an acetohydrazine moiety (Fig. 1A), was hypothesized, based on our structural data, to introduce a steric clash with hFKBP12-His88, potentially benefiting fungal selectivity (22). To understand the differential binding of APX879 to the fungal and human FKBP12 proteins, we obtained X-ray crystal structures of each of these complexes ( Fig. 4 and Table 1). Crystals of hFKBP12, AfFKBP12, and McFKBP12 bound to APX879 diffracted at resolutions of 1.7, 1.6, and 1.9 Å, respectively. The FKBP12-APX879 structures showed minimal structural variations compared to their FK506-bound counterparts (Ca-RMSDs, ;0.5 Å), maintaining similar H-bond patterns (i.e., 4 to 5): two with the backbone of residues Glu h /Arg Af /Gln Mc 55 and Ile57 and two or three with the side chain of residues Tyr27, Asp38, and Tyr83.
Previous crystal structures of AfFKBP12 bound to FK506 required a Pro90Gly mutation in order to trap the ligand in the binding cavity. Without this mutation, an apo homodimer was captured that is hypothesized to result from a self-catalysis mechanism involving the Pro89-Pro90 motif (21). Here, we obtained crystals of AfFKBP12 bound to APX879 without the requirement for a P90G mutation by the addition of APX879 during purification. Interestingly, Pro90 adopted the cis conformation, as observed in the crystal structure of the calcineurin-FK506-FKBP12 complex (PDB 6TZ7) (22). In contrast, the human and McFKBP12 proteins, having a Pro89-Gly90 motif, do not show any evidence of an intermolecular self-catalysis-like binding event. The Pro90 cis conformer in AfFKBP12 allows the 80's loop to reduce the size of the FK506-binding  Table 1 presents data collection and refinement statistics.  (29)(30)(31). Our AfFKBP12-APX879 crystal structure further emphasizes that the Pro90 cis conformation might modulate the binding cavity volume.
We next assessed APX879 acetohydrazine moiety accommodation by the different FKBP12 proteins. Similar to the hFKBP12-FK506 structure, in the hFKBP12-APX879 crystal structure a C 21 -C 22 dihedral angle of 36°is measured, positioning the acetohydrazine in the same orientation as the FK506-C 22 ketone (Fig. 4B). In the AfFKBP12-APX879 crystal structure, two protein monomers bound to APX879 were resolved, one showing the acetohydrazine moiety in the same orientation as the FK506-C 22 ketone (C 21 -C 22 dihedral angle of 68°) and the other showing a 90°rotation of the acetohydrazine moiety with a C 21 -C 22 dihedral angle of 244° (Fig. 4C). This suggests conformational flexibility that might be restrained upon calcineurin binding. Interestingly, when bound to McFKBP12, the APX879 acetohydrazine moiety adopts a third conformation with a C 21 -C 22 dihedral angle of 292°leading to a ;130°flip compared to FK506-C 22 ketone (McFKBP12-FK506 C 21 -C 22 dihedral angle, ;32°) (Fig. 4D). The adoption of this conformation prevents steric clashes with McFKBP12-Tyr88, positioned ;3 Å away from APX879-C 60 when overlaid with McFKBP12-FK506. Altogether, the crystal structures suggest that the flexibility of the acetohydrazine moiety compensates, at least in part, for the bulkiness of the amino acids at position 88. The hFKBP12-His88 does not by itself prevent APX879 binding, but complex formation with calcineurin might add further conformational restraints on APX879. An overlay of the A. fumigatus and bovine calcineurin-FK506-FKBP12 crystal structures (PDB 6TZ7 [22] and 1TCO [32]) with hFKBP12, AfFKBP12, and McFKBP12 bound to APX879 shows that the conserved CnA residues Pro377 and Phe378 in the CnB binding helix (CnA-BBH) are less than 3 Å away from the APX879 acetohydrazine moiety, supporting the necessity for a rearrangement of either the ligand or the CnA-BBH in order to facilitate the formation of the calcineurin-APX879-FKBP12 complex.
The chemical exchange for both FK506 and APX879 binding occurs in the slowexchange regime (tight binding) for all three FKBP12 proteins leading, in most cases, to the observation of the chemical shifts of both the unbound and bound populations within the intermediate titration points (Fig. S3). Interestingly, peak doubling at the highest APX879/FKBP12 ratios (1:1 and 2:1) was observed for residues in the hFKBP12 40's loop (40 to 45 and Lys48), 80's loop (Thr86, Ile91, Ile92, and His95), and residues Glu62-Gly63, suggesting conformational dynamics on the NMR timescale for the bound protein not seen in the FK506-bound protein.
McFKBP12 also showed peak Selective and Nonimmunosuppressive FK506 Analogs ® doubling, not seen when bound to FK506, for an increased number of residues located in the 40's, 50's, and 80's loop (Ser39, Arg41, Arg43, Arg48 to -50, Arg52 to -53, Arg56 to -60, Arg83 to -84, Arg86, Tyr88, Leu91, and Glu103) when bound to APX879, while no peak doubling was observed for AfFKBP12 when bound to either ligand. Since neither of the crosspeaks in the doublets overlays the resonances observed in the unbound protein, these observations suggest that APX879 binds tightly to all three FKBP12 proteins with an intermediate to low exchange rate but that hFKBP12 and McFKBP12, when fully bound to APX879, experience conformational flexibility not observed in AfFKBP12 or when these proteins are bound to FK506.
MD simulations reveal structural differences relative to the X-ray crystal structures and suggest conformational flexibility. To relieve crystal contacts (Table S1) that might affect the protein and ligand conformation and to substantiate conformational flexibility suggested by NMR, six 500-ns MD simulations were performed for each protein-ligand complex (Fig. S4). Interactions as characterized by Ca-RMSD, the solvent-accessible surface area, center of mass measurements, and H-bond patterns were found to be generally similar between the different complexes ( Fig. S4 and Table S2). Contrastingly, the Ca-RMSF (atomic positions in the MD simulation fitted to the crystal structures) highlighted areas of the proteins and ligands exhibiting conformational differences ( Fig. S5 and S6). Both fungal FKBP12 proteins when bound to APX879 showed high RMSF deviations for residues 52 to 55 not observed for the hFKBP12-APX879 complex nor for any of the FK506-bound counterparts, suggestive of altered conformational states for these residues. The Ca-RMSF also revealed a common deviation of the 80's loop for all FKBP12-ligand complexes. The cores of the FK506 and APX879 macrocycles were found to be more similar in comparing the crystal structures and MD simulations (lower RMSF) than the solvent-exposed portions of the molecules (spheres in Fig. S5). For all FK506-bound complexes, the largest atomistic RMSF deviations occurred at atoms C 40 (FK506-C 21 allyl moiety) and C 45 (cyclohexylidene ring C 31 -O-methyl) and to a lesser extent for atoms C 33 , C 34 , O 11 , and O 12 of the cyclohexylidene ring that are directly involved in interactions with the 80's loop and are also at the interface for calcineurin binding. For complexes bound to APX879, the ligand showed significant deviations at the acetohydrazine moiety. In addition, significant deviations were observed for an extended region surrounding the cyclohexylidene ring (atoms C 28 to C 34 , C 42 , and C 45 ) and O 8 , and O 10 when bound to McFKBP12 or hFKBP12. These data suggest that the crystallization process might have artificially favored the adoption of one conformer in otherwise flexible regions of the proteins and ligands.
Significance of FKBP12 and FK506/APX879 contacts observed in the MD simulations. The significance of protein-ligand contacts observed throughout the MD simulations was quantified, allowing for identification of variations between the human and fungal complexes potentially informing the design of fungus-selective FK506 analogs. In all FKBP12-ligand complexes, a common core of conserved residues (Tyr27, Phe37, Asp38, Phe47, Val56, Ile57, and Trp60) and one nonconserved residue (E h /R Af /Q Mc 55) made a similar number of contacts to FK506 or APX879 (Fig. S7). In contrast, the 80's loop residues showed a varying degree of interaction to the ligands; the fungal FKBP12 proteins had a significantly larger number of contacts involving this loop.
Z-score metrics allowed for the quantification of the relative significance of the contacts for the human and the fungal proteins when binding either FK506 or APX879 ( Fig. 6 and Fig. S7). Residues Tyr83 and His h /Phe Af /Tyr Mc 88 shared an increased importance for FK506 and APX879 binding to both fungal FKBP12 proteins while residues Phe49 and Ile91 were more significant for hFKBP12. Interestingly, hFKBP12 also relied on Gly84 and Phe100 for FK506 binding while Arg43 was notably important for APX879 binding. Binding to FK506 also implicated residues Tyr27, Val56, and Trp60 as making differentiating contacts between hFKBP12 and AfFKBP12 and residue R h /T Mc 86 as making differentiating contacts between hFKBP12 and McFKBP12 that are not involved in differentiating contacts when binding APX879. Z-score metrics also pointed to a common core of atoms from the two ligands, approximating 65% of the molecules, demonstrating similar significance for binding to all three FKBP12 proteins ( Fig. 6 and Fig. S7). In the FK506-bound forms, atoms C 2 to C 4 of the pipecolate ring Selective and Nonimmunosuppressive FK506 Analogs ® and C 45 (cyclohexylidene ring C 31 -O-methylation) are more significant for interaction with hFKBP12, while C 35 (pyranose ring methylation) and C 37 (C 19 -methylation) are more significant for interaction with the fungal FKBP12 proteins. Interestingly, for the APX879-bound forms, atoms of the cyclohexylidene ring (C 33 , C 42 , and O 12 ), pipecolate ring (C 3 to C 4 ), and C 15 to C 19 region (C 15 to C 16 , C 18 , C 36 to C 37 , and O 8 ) are more significant for interaction with hFKBP12, while atoms C 11 to C 12 and C 35 of the pyranose ring; C 28 , C 30 , and C 45 of the opposite side of the cyclohexylidene ring; and C 61 of the acetohydrazine moiety show more significant interactions with the fungal FKBP12 proteins. The complexes bound to APX879 show nearly identical importance of the acetohydrazine moiety, and therefore, no significant differences are noted within this region with the exception of C 61 possibly making significant interactions with the fungal FKBP12 proteins (McFKBP12 Z-score right under 11, right above 11 for AfFKBP12).
Overall, these observations indicate that the fungal FKBP12 proteins rely generally on the same amino acids (most notably Tyr83 and Phe88) to interact with FK506/ APX879 while hFKBP12 relies significantly on the 40's loop and Ile91. The C 22 acetohydrazine moiety in APX879 increases the number of significant and differentiating contacts between the human and fungal FKBP12 proteins, suggesting that APX879 might be a better starting scaffold than FK506 for modifications to further increase selectivity for the fungal proteins and amplify the difference in the balance between the antifungal and immunosuppressive activities.

DISCUSSION
Targeting calcineurin is a promising approach toward the development of novel antifungals due to its central role in diverse cellular processes, including antifungal drug resistance and pathogenesis of the major human fungal pathogens (5,38). Although FK506 efficiently targets calcineurin, currently available formulations are immunosuppressive and, therefore, cannot be used as antifungal therapeutics. In-depth, atomistic-level understanding of calcineurin inhibition through FK506-FKBP12 binding has the potential to reveal unique features differentiating the human and fungal proteins that could be exploited to enhance fungal selectivity. The studies here, for the first time, identified distinct differences in selective binding determinants between the fungal and human calcineurin-inhibitor-FKBP12 complexes at atomic resolution that inform future inhibitor design.
Crystal structures of human and fungal FKBP12 proteins bound to APX879, a less immunosuppressive FK506 analog with an acetohydrazine moiety on FK506-C 22 , revealed similar interactions as when bound to FK506 ( Fig. 4 and Table 1) but, strikingly, also revealed a high degree of flexibility and rearrangement of the acetohydrazine moiety in order to prevent clashes with the FKBP12 residue 88. This mobility might well be restricted by the formation of the ternary complex with calcineurin. In solution, binding of APX879 revealed a 40-fold decrease in affinity for hFKBP12 and McFKBP12 compared to FK506 while AfFKBP12 showed a 100-fold reduction in affinity (Table 2 and see Fig. S2 in the supplemental material). These data lend support to our observed 71-fold reduction of interleukin-2 (IL-2) production in primary murine CD4 1 T cells exposed to APX879 and also the significant reduction in the ability of APX879 to stimulate T helper cell-dependent germinal center B cell response in comparison to the acute toxicity observed with FK506 treatment in animal models. Although we observed a 17-fold reduction in antifungal efficacy of APX879 in comparison to FK506, the C 22 modification in FK506 does demonstrate an improvement in the therapeutic index and, even more importantly, provides insights into potential modifications which can further improve the selectivity of calcineurin inhibitors.
Consistent with APX879 being an FK506 analog, the protein responses captured by the NMR chemical shift perturbations were similar when bound to either ligand but notably different when comparing the human and fungal FKBP12 proteins. Interestingly, when fully bound by APX879, both hFKBP12 and McFKBP12 showed several residues in the 40's, 50's, and 80's loops experiencing peak doubling not observed when bound to FK506 or in AfFKBP12 (Fig. 5 and Fig. S3). This suggests an increased conformational flexibility of regions of hFKBP12 and McFKBP12 central to the formation of the inhibitory ligand-protein interface, not captured solely by the crystal structures.
Areas of the FKBP12 proteins experiencing the greatest conformational difference between the crystal structures and the MD simulations contained a significant number of crystal contacts and are implicated in the formation of the interface for calcineurin binding (Fig. S4 to S6 and Table S2). The 80's loop in all complexes and the 50's loop in the fungal FKBP12 proteins showed high RMSF compared to the crystal structures. Though partially captured in the crystal structures, the MD simulations enabled the exploration of a broader and more detailed scope of conformational flexibility (Fig. S5). Altogether, our MD simulation results give credence to the rationale of using conformational ensembles rather than a single X-ray-characterized structure as a search model for rational structure-based ligand design (39)(40)(41)(42). A previous report utilizing 14 targets as a test system has benchmarked that using ,100 conformers to represent protein conformational flexibility versus a single conformer can increase the accuracy of the protein-ligand interaction predictions by more than 20% (43).
Comparison of the FKBP12-ligand interactions captured by the three structural biology methods used here (i.e., H-bonds in crystal structures, chemical shift perturbation by NMR, and H-bonds in MD simulations) revealed that the interaction between the FKBP12 Ile57 and the ligand atom O 2 is maintained in all complexes independent of the methodology (Table 3 and Table S2). These comparisons also allowed the identification of the 50's loop (residues 50, 52, 55 to 57, and 60) as being important in all complexes as captured at different levels by these methodologies. Interestingly, in contrast to crystal structures, both the MD simulation and the NMR titration studies determined that the interactions between the fungal FKBP12 proteins and the ligands are exceedingly similar to one another and notably different from those of the human FKBP12 protein.
Additionally, the MD simulations allowed for comparison of the significance of individual FKBP12-ligand interactions and identification of regions of the ligands where modifications may contribute to enhanced fungal selectivity ( Fig. 6 and Fig. S7). Similar free energy scores (DG°) for the binding of FK506/APX879 to the FKBP12 proteins were obtained from both the MD simulations and ITC experiments, further justifying their use in rational structure-based design (FK506, MD % 211.3 kcal/mol, ITC % 211.5 kcal/ mol; APX879, MD % 210.4 kcal/mol, ITC % 29 kcal/mol). The 80's loop residues Tyr83 and His h /Phe Af /Tyr Mc 88 are key for binding to the fungal FKBP12 proteins while residues Phe49 and Ile91 are more significant for FK506/APX879 binding to hFKBP12 (Fig. 6, Table 3, and Fig. S7). hFKBP12 relies more significantly on the pipecolate ring (C 2 to C 5 ) and C 31 -O-methylation through atom C 45 (cyclohexylidene ring) for FK506 binding while the C 35 and C 37 methylation sites in the vicinity of the pyranose ring are more significant for binding to the fungal FKBP12 proteins (Fig. 6A). These FK506 regions could therefore be targeted in an attempt to rationally enhance fungal FKBP12 specific binding; however, these regions are not structurally clustered and involve a small number of atoms.
The APX879 acetohydrazine moiety alters the interactions with the FKBP12 proteins in comparison to FK506 as captured by lower ITC binding affinities, NMR binding assays, and lower MD free energy values. Z-score analysis revealed that atoms of the cyclohexylidene (O 12 , C 33 , and C 42 ) and pipecolate (C 3 to C 4 ) rings are significantly implicated for hFKBP12 binding, as observed for FK506, but interestingly, several regions of APX879 become more significantly involved in binding to hFKBP12 (C 15 to C 18 , C 36 , C 37 , and O 8 ). In contrast, the pyranose (C 11 , C 12 , and C 35 ) and part of the cyclohexylidene (C 28 , C 30 , and C 45 ) rings show increased importance for interactions with the fungal FKBP12 proteins (Fig. 6B). APX879 makes a greater number of structurally clustered interactions with hFKBP12. Importantly, despite the fact that the C 22 modification to FK506 decreased its affinity toward both human and fungal FKBP12 proteins, we were able to identify notable differences in APX879 interactions with the human versus fungal counterparts which define it as a superior scaffold for future modifications aimed at enhancing fungal selectivity.
Previous studies have examined the effects of modifications to FK506 in the context of antifungal efficacy versus immunosuppressive activity. The reduction in immunosuppressive activity is primarily due to decreased affinity of the modified FK506 analog AfFKBP12 (PDB 5HWC and 6VCV)  I25  I25  H26 H26 McFKBP12 (PDB 6VRX and 6VCT)  I25  I25  H26 H26 a Residues in italic are observed by two methods while those in bold are observed by three methods for the same protein-ligand complex. For MD simulation H-bond Z-score calculations, see Table S2. toward calcineurin. For instance, the compound L-685,818, a C 18 -hydroxy, C 21 -ethyl derivative of FK506, is nonimmunosuppressive but still binds to FKBP12 and retains antifungal activity (44). Our findings corroborate medicinal chemistry studies performed using FK506 that identify the pipecolate ring as important for the immunosuppressive activity (45). Our results are also concordant with a recent report on FK506 analogs modified on the pipecolate and cyclohexylidene (C 31 ) rings showing reduced immunosuppressive activity while, in some cases, maintaining antifungal activity (46). In addition to studies under way exploring the effects of selective extensions to the FK506 scaffold centered around C 21 and C 22 (APX879) (22), the observed reduced immunosuppressive activity around C 31 (46), and the selective extensions produced biosynthetically around C 9 (26,27), the data presented here strongly suggest selective extensions or removal of branching atoms centered around C 15 , C 16 , C 18 , C 36 , and C 37 (Fig. 6). This region is chemically accessible through the creation of the C 18 -hydroxy analog to which various functional groups can be added (ketones, esters, carbamates, etc.). C 16 modifications can be achieved through the use of a C 18 -keto analog. While individual or combinatorial substitutions at these specific positions will impact the binding of modified FK506 analogs to fungal calcineurin-FKBP12 complexes, we expect that it is possible to strike a balance between antifungal activity and immune suppression. Further structural, biophysical, and NMR-based dynamics investigations of the calcineurin-APX879-FKBP12 complexes that are under way in our laboratories will greatly contribute to our understanding of the differential conformational flexibility in the human and fungal complexes noted here and guide the rational development of fungus-selective and nonimmunosuppressive FK506 analogs.

MATERIALS AND METHODS
DNA constructs, protein expression, and purification. Protein expression for crystallization, isothermal titration calorimetry (ITC), and NMR were performed as previously reported (31). Briefly, M. circinelloides, A. fumigatus, and human FKBP12 DNA constructs in the pET-15b plasmid (expression with an N-terminal hexahistidine tag [His 6 tag] cleavable by thrombin) and codon optimized for Escherichia coli expression were purchased from GenScript (Piscataway, NJ). Expression was performed using E. coli BL21(DE3) cells. Cells were propagated at 37°C with agitation to an optical density at 600 nm (OD 600 ) of 0.6 in modified M9 minimal medium containing 100 mg/ml ampicillin (Sigma-Aldrich, St. Louis, MO), 1 g/ liter NH 4 Cl (or 15 NH 4 Cl for NMR [Cambridge Isotopes, Tewksbury, MA]), and 55 g/liter of sorbitol (M. circinelloides only). Protein expression was initiated by the addition of 1 mM isopropyl b-D-1-thiogalactopyranoside (IPTG) and incubation for 16 h at 25°C with agitation. Cells were harvested by centrifugation at 4°C for 20 min at 6,000 Â g, and the pellet was stored at 280°C until purification. The cell pellets were resuspended in 30 ml of 50 mM sodium phosphate, 500 mM NaCl, pH 8.0, buffer. Lysis was performed by three cycles of 30 s of sonication at a power of 12 W with a 2-min rest interval on ice or by French press followed by the addition of 1 mM phenylmethylsulfonyl fluoride (PMSF). The lysate was clarified by centrifugation (4°C, 15 min at 20,000 Â g) and filtration using a 0.22-mm polyethersulfone (PES) syringe filter. The chromatography was undertaken at 4°C using an Äkta fast protein liquid chromatograph (FPLC) (GE Healthcare). The clarified supernatant was loaded onto a 5-ml prepacked Ni-nitrilotriacetic acid (NTA) column. Protein was eluted using a 0 to 1 M stepwise gradient of imidazole. Fractions containing FKBP12 proteins were identified by SDS-PAGE and Coomassie blue staining. Combined fractions were dialyzed into 50 mM sodium phosphate, 500 mM NaCl, pH 8.0, buffer to remove the imidazole (2 cycles in 1 liter for 2 h followed by 1 cycle in 2 liters for 16 to 18 h). The His 6 tag was cleaved for 16 h at 4°C using 1 U of thrombin (GE Healthcare) per 100 mg of total protein. The cleaved proteins were then run over the 5-ml prepacked Ni-NTA column to remove the cleaved His tag and any uncleaved protein.
For crystallography, in the presence of the ligand, APX879 or FK506 was added in a 1-to-1.5 molar ratio of FKBP12 to ligand using a solution stock of the ligand at 10 mg/ml in 100% dimethyl sulfoxide (DMSO). The total volume of DMSO added was limited to less than 5%. In all cases, protein solutions were concentrated to a volume of ;2 ml using a 3,000-molecular-weight-cutoff (MWCO) Amicon concentrator followed by size exclusion chromatography using a Sephacryl S100HR XK26/60 FPLC column. Fractions containing protein were identified by SDS-PAGE and Coomassie blue staining. Typical yields were 40 mg/liter of .98% pure protein.
Crystallization and determination of the structures of M. circinelloides bound to FK506 and M. circinelloides, A. fumigatus, and Homo sapiens FKBP12 bound to APX879. After size exclusion purification, the proteins were concentrated to 10 mg/ml using a 3,000-MWCO Amicon concentrator. For McFKBP12/FK506 and hFKBP12/APX879, crystals were grown at 22°C with a hanging drop vapor diffusion setting using a 1-to-1 ratio of the protein and reservoir solutions. McFKBP12/FK506 crystals were grown in 2,100 mM DL-malic acid, pH 7.0. Needle-shaped crystals of hFKBP12/APX879 were grown with 2.5 M ammonium sulfate, 0. For AfFKBP12/APX879, initial crystals were grown at 22°C using a hanging drop vapor diffusion setting and 10 mM morpholineethanesulfonic acid (MES) (pH 6.0), 200 mM zinc acetate, and 15% reagent alcohol as the reservoir solution. Protein drops were prepared using a 1-to-1 ratio of the protein and reservoir solution. A single crystal was then harvested to prepare a seed stock using the Hampton Research (Aliso Viejo, CA) seed bead kit and the classical method. Crystals were grown in 5 mM MES (pH 6.0), 200 mM zinc acetate, and 15% reagent alcohol using a 1-to-1 ratio of the protein and reservoir solution and streaking of the drop using the seed stock. All crystals were cryopreserved directly from the drop. Diffraction data were collected at the Advanced Photon Source using sector 22 BM and ID beamlines. The collected diffraction images were indexed, integrated, and scaled using HKL2000 (47). Initial phases were calculated by molecular replacement using Phenix.PHASER (48,49) and the PDB entries 5HUA (McFKBP12/FK506 or APX879), 2PPN (hFKBP12/APX879), and 5HWB (AfFKBP12/APX879) as search models (21,26). Iterative rounds of manual model building using Coot (50) and automatic refinement in PHENIX (48) were performed. Data collection and refinement statistics are summarized in Table 1.
Active-site volume estimation. The binding pocket cavity volume was estimated using 3V: Voss Volume Voxelator (28). The estimation was made with a small sphere with a 1.5-Å radius and a large sphere with a 7-Å radius.
ITC experiments. After the size exclusion purification, the proteins were concentrated to ;2 mg/ml using a 3,000-MWCO Amicon concentrator. Proteins were exhaustively dialyzed at 4°C in the isothermal titration calorimetry (ITC) buffer (50 mM sodium phosphate, 50 mM sodium chloride, pH 7.0). Experiments were performed using a VP-ITC instrument (MicroCal Inc., Northampton, MA) at 25°C. All solutions were degassed under vacuum at 25°C for 15 min immediately before use. For FK506, because of its high affinity, it was used in the sample cell at a concentration of 2 mM with constant stirring at 307 rpm. The protein was loaded into the syringe at a concentration of 25 mM and titrated in by a first injection of 2 ml followed by 22 injections of 6 ml. Following each injection, the cell was equilibrated for 3 min. For APX879, because of its low solubility restricting the usable concentration and its lower affinity, it was used in the syringe at a concentration of 150 mM. The proteins were loaded in the cell at a concentration of 10 mM with constant stirring at 307 rpm (25°C). APX879 was titrated in by a first injection of 2 ml followed by 29 injections of 8 ml and equilibrated for 3 min. The enthalpy of binding (DH, kilocalories/mole) was determined by integration of the injection peaks minus the controls for the heat of dilution (equivalent experiments without the ligand and without the protein). The MicroCal Origin software (OriginLab Corp., Northampton, MA) was used for a variety of binding models. to obtain a 1,290-bp PCR fragment and primers McFkbp12-Y88H-F (GAGCGCGGCTTCCCTGGACTT) and Hyg-R (GCCCATGAACTGGCTCTTAA) to obtain a 1,267-bp PCR fragment. Fusion PCR was then performed with pUCGH-2033-F and Hyg-R using a 1:1 mixture of the two PCR fragments (1,290 bp and 1,267 bp) to obtain the final 2,536-bp PCR fragment harboring the Y88F mutation. This 2,536-bp fragment was digested with NotI-KpnI and cloned into pUCGH at NotI-KpnI sites to obtain the pUCGH-McFkbp12promo-McFkbp12-Y88F vector. In the next step the Affkbp12 terminator was inserted at SbfI-HindIII as described for the pUCGH-McFkbp12promo-McFkbp12 vector followed by linearization and transformation into the A. fumigatus akuB KU80 strain. All the constructs were sequenced to confirm accuracy before being used for transformations. The transformants were verified for homologous integration by PCR, and accuracy of the Mcfkbp12 and Mcfkbp12-Y88F sequences was verified by sequencing and visualized by fluorescence microscopy. E. coli DH5a competent cells were used for subcloning experiments. A. fumigatus wild-type strain akuB KU80 was used for growth and recombinant strain generation experiments. A. fumigatus wild-type or recombinant strains were cultured on glucose minimal medium (GMM) or RPMI liquid medium at 37°C for 24-or 48-hperiods. In certain experiments, GMM agar or RPMI liquid medium was supplemented with FK506 (0.01 to 4 mg/ml) or APX879 (0.01 to 4 mg/ml). All growth experiments were repeated as technical triplicates, each also as biological triplicates.
Fluorescence microscopy. Conidia (10 4 ) from the recombinant strains of A. fumigatus were inoculated into 5 ml GMM and poured over a sterile coverslip (22 by 60 mm; no. 1) placed in a sterile dish (60 by 15 mm). Cultures grown for 18 to 20 h at 37°C were observed by fluorescence microscopy using an Axioskop 2 Plus microscope (Zeiss) equipped with AxioVision 4.6 imaging software. FK506 (100 ng/ ml) was added to the cultures in order to visualize the in vivo binding of FKBP12 to the calcineurin complex at the septum.
Molecular dynamic simulations. MD simulations were performed to provide a better representation of the protein's conformational flexibility and to more accurately characterize the proteins' solution structure bound to APX879 and FK506 (19,21). Crystal structures were used as the starting conformations: McFKBP12 bound to FK506 (PDB 6VRX) and APX879 (PDB 6VCT), hFKBP12 bound to FK506 (PDB 1FKF) and APX879 (PDB 6VCU), and AfFKBP12 bound to FK506 (PDB 5HWC, P90G mutant) and APX879 (PDB 6VCV). For the McFKBP12(Y88F) mutant bound to FK506, the wild-type crystal structure was used and mutated in silico accordingly. FK506 and APX879 small-molecule parameter and topology files were downloaded and created utilizing the Automated Topology Builder (ATB) and repository (52,53). All molecular dynamic (MD) simulations were performed with the GROMACS 5.0.1 software package utilizing 6 CPU cores and one Nvidia Tesla K80 GPU (54). The single starting conformations used for all of the MD simulations were resulting X-ray-characterized crystal structures described here or otherwise noted above. MD simulations were performed with the GROMOS54a7 force field and the flexible simple pointcharge water model. The initial structures were immersed in a periodic water box with a dodecahedron shape that extended 1 nm beyond the protein in any dimension and were neutralized with counterions. Energy minimization was accomplished through use of the steepest descent algorithm with a final maximum force below 100 kJ/mol/min (0.01-nm step size, cutoff of 1.2 nm for neighbor list, Coulomb interactions, and Van der Waals interactions). After energy minimization, the system was subjected to equilibration at 300 K and normal pressure for 1 ns. All bonds were constrained with the LINCS algorithm (cutoff of 1.2 nm for neighbor list, Coulomb interactions, and Van der Waals interactions). After temperature stabilization, pressure stabilization was obtained by utilizing the v-rescale thermostat to hold the temperature at 300 K and the Berendsen barostat was used to bring the system to 10 5 -Pa pressure. Production MD calculations (500 ns) were performed under the same conditions, except that the position restraints were removed, and the simulation was run for 500 ns (cutoff of 1.1, 0.9, and 0.9 nm for neighbor list, Coulomb interactions, and Van der Waals interactions). These MD simulations were repeated 6 times, with the exception of that for McFKBP12(Y88F)-FK506, which was performed once. Ca-RMSD, solvent accessible surface area (SASA), radius of gyration (Rg), and center of mass (COM) all confirmed stability and accuracy of the MD simulations by stabilizing after a 100-ns equilibration period (in most cases) allowing for the analysis of the last 400 ns of the simulations. Only three of the 36 calculations (6 complex repeated 6 times) were rejected from the analysis due to unstable Ca-RMSD, Rg, and/or COM (see Fig. S4 in the supplemental material).
GROMACS built-in and homemade scripts were used to analyze the MD simulation results and averaged over the 6 simulations. All images were produced using PyMOL (54). Atom indices for FK506 and APX879 are provided in Table S3.
Data availability. The refined structures have been deposited to the Protein Data Bank (https:// www.rcsb.org) under the accession codes 6VRX, 6VCT, 6VCU, and 6VCV.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.  beamline at the Advanced Photon Source, Argonne National Laboratory. We thank Duke's Research Computing staff for the use of Duke Computing Cluster and its support. We also thank Nathan Nicely and Priyamvada Acharya for the use of the equipment for protein crystallization and Michael Hoy, Zanetta Chang, Soo Chan Lee, and Anna Averette for support and discussions. Use of the Duke NMR Spectroscopy Center instrumentation is gratefully acknowledged. S.M.-C.G. designed and performed protein expression and X-ray crystallography and isothermal titration calorimetry (ITC) assays, analyzed and interpreted data, and wrote the manuscript. B.G.B. designed, performed, and analyzed the MD simulations and contributed to writing. P.R.J. supervised, designed, and performed genetic, biochemical, and microscopy experiments and contributed to writing. D.C.C. performed antifungal susceptibility experiments. R.A.V. and S.M.-C.G. designed, performed, and interpreted NMR experiments. J.H., W.J.S., and L.D.S. designed and supervised the overall study. All authors were involved in editing the manuscript.
We declare no competing interests.