The FKBP38 Catalytic Domain Binds to Bcl-2 via a Charge-sensitive Loop

Background: FKBP38 regulates the prosurvival activity of Bcl-2 in apoptosis signaling. Results: The catalytic domain of FKBP38 interacts with helix 4 in Bcl-2. Conclusion: An electrostatic interaction modulates transient complex formation between FKBP38 and Bcl-2. Significance: The protein complex provides a molecular basis for FKBP38-mediated Bcl-2 regulation. FKBP38 is a regulator of the prosurvival protein Bcl-2, but in the absence of detailed structural insights, the molecular mechanism of the underlying interaction has remained unknown. Here, we report the contact regions between Bcl-2 and the catalytic domain of FKBP38 derived by heteronuclear NMR spectroscopy. The data reveal that a previously identified charge-sensitive loop near the putative active site of FKBP38 is mainly responsible for Bcl-2 binding. The corresponding binding epitope of Bcl-2 could be identified via a peptide library-based membrane assay. Site-directed mutagenesis of the key residues verified the contact sites of this electrostatic protein/protein interaction. The derived structure model of the complex between Bcl-2 and the FKBP38 catalytic domain features both electrostatic and hydrophobic intermolecular contacts and provides a rationale for the regulation of the FKBP38/Bcl-2 interaction by Ca2+.

FKBP38 is a peptidylprolyl cis/trans-isomerase (PPIase 6 ; EC 5.2.1.8) that belongs to the family of FK506-binding proteins (FKBPs) and has been associated with a number of important cell regulatory processes (1). The multidomain protein consists of a Glu-rich segment at the N terminus, followed by the structurally well characterized PPIase domain (2, 3), a tetratricopep-tide repeat (TPR) domain, a calmodulin (CaM)-binding site, and a putative C-terminal membrane anchor, which is unique among the human members of the FKBP family (4). Moreover, an exceptional cation-binding site was recently discovered in the PPIase domain close to the putative active site (3).
The prosurvival Bcl-2 protein is a core regulator of apoptosis. Bcl-2 shares a common fold with other prosurvival Bcl-2 family members (Bcl-x L , Mcl-1, Bcl-w, and Bcl-B), containing four Bcl-2 homology (BH) domains and a C-terminal membrane anchor. Bcl-2 is localized in the membranes of mitochondria and the endoplasmic reticulum, just like FKBP38, and in the nucleus (1,20). Bcl-2 is associated with many diseases that are related to deviations in the cell cycle and can lead to pathological regulation of mitochondrial apoptosis, including breast and prostate cancers, schizophrenia, and autoimmune disorders such as type 1 diabetes (21,22).
Bcl-2 was found previously to interact with FKBP38 in a yeast two-hybrid system (8). In recent years, this interaction was further analyzed by several groups with partly contradictory results (1). Kang et al. (17) suggested that a long flexible loop between the BH4 and BH3 domains of Bcl-2 mediates the binding to FKBP38 and that this protein/protein interaction can prevent phosphorylation of the flexible loop, which would lead to apoptosis as a result of Bcl-2 degradation by caspase-3. Edlich et al. (9,13) identified a CaM-dependent binding interaction between the FKBP38 catalytic domain and Bcl-2, which gave rise to a pro-apoptotic effect in cell experiments. Also on the basis of cell experiments, Choi et al. (19) reported that the PPIase and TPR domains of FKBP38 bind to Bcl-2 by affecting primarily its flexible loop between the BH4 and BH3 domains.
Taken together, all of these results differ (i) in the localization of the contact regions, (ii) in the dependence on additional components, and (iii) in the pro-or anti-apoptotic nature of the FKBP38/Bcl-2 interaction.
In contrast with the above-described biological approaches, we present NMR-based data that aim to elucidate the structural features of the interaction between the FKBP38 catalytic domain (FKBP38(35-153)) and Bcl-2. On the basis of a peptide library derived from the Bcl-2 sequence, we identified a binding epitope that revealed specific binding to FKBP38(35-153). The corresponding Bcl-2(119 -131) peptide induced significant chemical shift changes in the charge-sensitive ␤5-␣1 loop of FKBP38. Further characterization of the respective binding interface in the transient FKBP38(35-153)⅐Bcl-2(1-211) complex revealed a predominantly electrostatic interaction. A structural model of the complex has been derived from docking calculations based on chemical shift perturbation (CSP) data.
The His-tagged Bcl-2(1-211) variants were mutated by overlay extension PCR (23), expressed in Escherichia coli BL21 by adding isopropyl 1-thio-␤-D-galactopyranoside at 30°C to a final concentration of 1 mM, and purified with a nickel-nitrilotriacetic acid affinity column (Qiagen, Hilden, Germany). The cell lysate was solubilized in 70 mM Tris buffer containing 0.5 M NaCl and 20 mM imidazole (pH 8.7). After loading the lysate on the column, it was washed using 70 mM Tris buffer containing 0.5 M NaCl and 100 mM imidazole (pH 8.7). The imidazole was eluted with 50 mM EDTA in 50 mM HEPES buffer containing 1 M NaCl (pH 8.2), followed by elution with 120 mM EDTA in 50 mM HEPES containing 1 M NaCl (pH 8.2).
Peptide Library Assay-Two peptide-coupled cellulose membranes representing a peptide library array of overlapping decamers that span the entire Bcl-2 sequence with a one-residue shift were synthesized as described previously (24). Each membrane was incubated overnight at 4°C with 50 ml of buffer (15 mM MES, 150 mM NaCl, 6.3 mM KCl, 5% (w/v) saccharose, and 0.05% Tween), 0.030 M human FKBP38(1-336), 0.072 M CaM, and 1.8 mM CaCl 2 (1 M stock solution). In addition, 0.018 M Bcl-2(1-211) was added to the second membrane for incubation. Both membranes were analyzed by Western blotting using anti-FKBP38 antibody.
For detailed analysis, intensities of the chemiluminescence signals were determined for individual spots using NIH ImageJ software. The intensities of spot signals in the presence of Bcl-2 were normalized using maximal signal intensity and background signal and subsequently subtracted from the corresponding normalized spot signals in the absence of Bcl-2. Analysis of the means Ϯ S.D. was performed with SigmaPlot 12.
NMR Experiments-All NMR data were collected at 10°C using a DRX500 spectrometer (500.13-MHz proton resonance frequency, 5-mm TXI probe with gradient capability) or an AVANCE 800 spectrometer (800.23-MHz proton resonance frequency, 5-mm TCI cryoprobe) (both from Bruker, Rheinstetten, Germany). Two-dimensional 1 H/ 15 N heteronuclear single-quantum correlation (HSQC) spectra were collected with the carrier placed on the water resonance, which was suppressed by a WATERGATE sequence. The NMR spectra were processed on a Silicon Graphics O2 workstation using the XWINNMR 3.5 software package with the DRX500 spectrometer or TopSpin 2.1 with the AVANCE 800 spectrometer. A 90°phase-shifted squared sine-bell function was used for apodization in all dimensions. After Fourier transform, a polynomial base-line correction was applied to the processed spectra in the directly detected 1 H dimension. All chemical shifts were referenced to external 2,2-dimethyl-2-silapentane-5-sulfonate to ensure consistency among the spectra (25). Peak picking of backbone amide resonances was performed with the program FELIX 2000 (Accelrys Inc., San Diego, CA). Chemical shift differences in the amide proton (⌬␦ 1HN ) and nitrogen (⌬␦ 15N ) resonances of the free and complexed protein forms were combined for each residue using the expression ((⌬␦ 1HN ) 2 ϩ (⌬␦ 15N /6.5) 2 ) 1/2 (26).
All experiments involving the Bcl-2(119 -131) peptides listed in Table 1 were performed in 10 mM HEPES buffer (pH 7.5). For the titration experiments, the Bcl-2(119 -131) peptides, FKBP38(35-153) (both non-labeled and 15 N-labeled), and CaM were all dialyzed as highly concentrated stock solutions in the same buffer solution. Furthermore, CaCl 2 was also solubilized in this buffer to prepare a Ca 2ϩ stock solution, and, to avoid false-positive effects due to Ca 2ϩ contamination, all protein samples had been pretreated with EDTA. Unless noted otherwise, all data points were derived from 1 H/ 15 N HSQC spectra obtained with the DRX500 spectrometer.
The resonance assignment of non-labeled Bcl-2(119 -131) was achieved using total correlation spectroscopy, rotatingframe Overhauser enhancement spectroscopy, and 1 H/ 13 C HSQC experiments performed with a 3 mM peptide sample. For the partly 15 N-labeled peptides, 1 H/ 15 N HSQC spectra were additionally recorded. The complete resonance assignment is listed in supplemental Table S1. For secondary structure prediction of the peptide, the chemical shift index consensus of the H␣, C␣, and C␤ resonances was determined using the CSI Program of the University of Alberta (27).
Docking Calculations-The three-dimensional model of the FKBP38(35-153)⅐Bcl-2 complex was obtained from docking calculations using the HADDOCK program (28). To drive the docking process, ambiguous interaction restraints (AIRs) were derived from the experimental NMR data. Based on CSP effects in combination with the results from the protein mutations, the residues mediating the electrostatic intermolecular interaction between FKBP38(35-153) and Bcl-2(1-211) were selected as "active AIR residues", i.e. Asp-92 and Asp-94 in FKBP38 as well as Arg-129 in Bcl-2. A surface accessibility cutoff value of 25% was generally applied. In addition, surface-accessible residues located next to the active AIR residues were selected as "passive AIR residues", i.e. Thr-89, Gly-91, and Cys-93 in FKBP38 as well as Gln-118, His-120, Thr-125, Gly-128, and Thr-132 in Bcl-2. Upper distance limit restraints were introduced between the residues responsible for the electrostatic interaction. The distance restraints from Arg-129 C (Bcl-2) to Asp-92 C␥ and Asp-94 C␥ (both FKBP38) were set to 4.5 Ϯ 0.2 and 7.0 Ϯ 0.2 Å, respectively, according to a previously determined cation-binding model of FKBP38(35-153) (3).
Surface accessibility calculations with the program NAC-CESS (29) and subsequent docking calculations were carried out using the atomic coordinates of FKBP38(35-153) (Protein Data Bank code 3EY6; x-ray structure) and Bcl-2(3-207) (code 1GJH; minimized average NMR structure). For the calculation of the complex, 400 conformers were initially generated using the rigid-body docking protocol of HADDOCK. Next, the 20 lowest energy structures were submitted to semiflexible simu-lated annealing in torsion angle space, with the side chains of residues that define the protein/protein interface kept flexible. The resulting top 20 structures were subsequently refined in an 8-Å shell of explicit TIP3P water molecules (30).
In a competitive membrane assay (Fig. 1), decapeptides derived from the Bcl-2 sequence were immobilized on two cellulose membranes. Each membrane was incubated with a mixture of FKBP38 and holo-CaM. The second membrane was incubated additionally with soluble Bcl-2(1-211) as FKBP38 competitor. Both membranes were analyzed by immunoblotting against FKBP38. Thereby, zones of consecutive peptide spots in the first but not the second membrane indicated specific FKBP38 binding to the membrane-bound Bcl-2 peptides, which was subsequently reduced due to competition when Bcl-2(1-211) was added to the solution.
Comparison of the resulting spot intensities in the absence and presence of Bcl-2 revealed one particular series of contiguous spots with pronounced intensity differences within the spot range 118 -123, i.e. a region of the Bcl-2 sequence that comprises residues 118 -132. In addition, minor binding interactions could be envisioned also in the spot ranges 21-25, 89 -91, 169 -172, and 187-191, even though none of these ranges featured more than two consecutive spots with above-average intensity changes (Fig. 1C), as would be expected for a potential binding epitope wherein several successive residues define the interaction site. Nevertheless, the corresponding peptides were all tested in a streptavidin matrix assay for FKBP38 binding and competition with Bcl-2(1-211). Although all peptides showed interaction with FKBP38, in analogy to the peptide array, Bcl-2 competed for FKBP38 binding only with the peptide that corresponded to spots 118 -123 (supplemental Fig. S1).
Interestingly, the long flexible Bcl-2 loop, which had been reported earlier to be predominantly responsible for the FKBP38 interaction (17), showed no notable effects in the membrane assay (Fig. 1). Furthermore, the peptide Bcl-2(51-80), representing the core region of this supposedly non-structured loop, exhibited no FKBP38 binding in additional binding assays (supplemental Figs. S2 and S3).
Hence, only the peptide Bcl-2(119 -131), derived from spots 118 -123, showed specific binding to FKBP38 (supplemental Figs. S2 and S3), with a K d value in the lower micromolar range as determined by isothermal titration calorimetry (supplemental Fig. S4). Moreover, this interaction between Bcl-2(119 -131) and FKBP38 was hindered by the presence of FK506 (supplemental Fig. S3), which indicates an involvement of the putative FKBP38 active site in the binding of the identified epitope, corroborating prior results on the interaction between FKBP38 and Bcl-2 (1,9). We therefore selected the Bcl-2 segment Leu-119 -Ala-131 as the primary FKBP38-binding epitope to be employed in further investigation of the FKBP38/Bcl-2 interaction. The CSP effects caused by the Bcl-2(119 -131) peptide in the ␤5-␣1 loop region of FKBP38 decreased significantly when apo-CaM was added (Fig. 2B) but were essentially restored upon subsequent addition of Ca 2ϩ to generate holo-CaM. These results suggest a reduction of the FKBP38(35-153)/Bcl-2(119 -131) interaction in the presence of apo-CaM, whose negatively charged Ca 2ϩ -binding sites (31) may compete for the positive charges in Bcl-2(119 -131) under these experimental conditions, which apparently is not the case in the presence of fully Ca 2ϩ -saturated holo-CaM.
To further explore the electrostatic interaction between FKBP38(35-153) and Bcl-2(119 -131), analogous peptides lacking one or both of the positively charged arginine side chains were synthesized, and their CSP effects in 15 N-labeled FKBP38(35-153) were determined (supplemental Fig. S6). The data revealed that binding of the peptide Bcl-2(119 -131) R127Q/R129Q to the catalytic domain of FKBP38 was abolished by the two arginine-to-glutamine substitutions (Fig. 2C). Minor residual shifts in the ␤5-␣1 loop region may be the result of hydrophobic contact(s), which are described below. However, this experiment identified the electrostatic attraction as the main driving force in the intermolecular interaction between Bcl-2(119 -131) and FKBP38(35-153).
To analyze the effects of the FKBP38 catalytic domain on the Bcl-2 epitope, the reverse experiment was performed with two different partially 15 N-labeled Bcl-2(119 -131) peptides, Npep1 and Npep2 (Table 1). Upon addition of non-labeled FKBP38(35-153), the 15 N-labeled residues of both Npep1 and Npep2 all showed significant chemical shift changes (supplemental Fig. S7), suggesting that the entire peptide sequence may be affected by the catalytic domain of FKBP38 (Fig. 3). The largest chemical shift changes occurred in the C-terminal half of the peptide, where the two arginines Arg-127 and Arg-129 are located. An additional indicator for the involvement of both arginines in the formation of the protein⅐peptide complex is the fact that the N⑀H side chain signals of Arg-127 and Arg-129 were observable in the 1 H/ 15 N HSQC spectra only in the presence of FKBP38(35-153) (supplemental Fig. S8), implying slowed solvent exchange of the N⑀ protons due to contact with the FKBP38 catalytic domain. Moreover, CSP experiments of 15 N-labeled FKBP38(35-153) with the peptide variants Bcl-2(119 -131) R127Q and Bcl-2(119 -131) R129Q, which both affect the residues in the ␤5-␣1 loop (supplemental Fig. S9), demonstrated that either of these positively charged Bcl-2(119 -131) side chains can participate in the interaction with FKBP38(35-153).
In contrast, upon addition of the FKBP38(35-153) D92N/ D94N variant, which renders the charge-sensitive ␤5-␣1 loop insensitive to cationic moieties (3), a reduction of the CSP effects in Npep1 and Npep2 was observed (Fig. 3), corroborating that aspartates Asp-92 and Asp-94 play a key role in this intermolecular interaction. Interestingly, however, most peptide residues still showed residual CSP effects, and Leu-121 even displayed the same shift as with wild-type FKBP38(35-153), suggesting a possible hydrophobic component to this intermolecular interaction.
According to the Bcl-2 structure (32), the positively charged guanidinium groups of Arg-127 and Arg-129, which are both integrated in ␣-helix 4, are ϳ15 Å apart. Therefore, as long as the helical structure of ␣-helix 4 persists, it seems highly unlikely that both arginine side chains act simultaneously as counterparts to aspartates Asp-92 and Asp-94 in the ␤5-␣1 loop of FKBP38. In case of the Bcl-2(119 -131) peptide, the fact that both arginines reveal contacts to the charge-sensitive ␤5-␣1 loop may thus be an artifact that is due to a lack of an organized structure in the peptide, as indicated also by the chemical shift index consensus of the H␣, C␣, and C␤ resonances (data not shown). Hence, the extensive data obtained from the above-described experiments with the Bcl-2(119 -131) peptides could not unambiguously resolve which arginine residue is mainly responsible for the electrostatic protein/protein interaction between FKBP38 and Bcl-2.
The single mutants of Bcl-2(1-211), with at least one of the two arginine residues intact, caused chemical shift changes in the ␤5-␣1 loop of FKBP38(35-153) that were smaller than in case of the wild-type protein but larger than in case of the Bcl-2(1-211) R127Q/R129Q double mutant (Fig. 5). Compared with the double mutant, which precludes an electrostatic interaction with the ␤5-␣1 loop but may feature hydrophobic interactions with FKBP38(35-153), the CSP effects were generally larger in Bcl-2(1-211) R127Q (with Arg-129 intact) than in Bcl-2(1-211) R129Q (with Arg-127 intact), suggesting that Arg-129 of Bcl-2 is the primary counterpart for the electrostatic interaction with the catalytic domain of FKBP38. However, Arg-127 nevertheless     Model of FKBP38(35-153)⅐Bcl-2 Complex-Based on the CSP data, docking calculations were performed to obtain a structural model of the FKBP38(35-153)⅐Bcl-2 complex. Distance restraints defining the electrostatic interaction between the ␤5-␣1 loop of FKBP38(35-153) and Arg-129 at the surface of Bcl-2 were employed as additional structural input. Based on structural convergence (Table 2), a cluster of five conformers out of the top seven calculated structures was chosen as representative for the FKBP38(35-153)⅐Bcl-2 complex (Fig. 6A). Average backbone root mean square deviations of 2.70 Ϯ 0.81 and 3.94 Ϯ 1.15 Å were obtained when the C␣ traces of Bcl-2 or FKBP38 were superposed, respectively. The values for the intermolecular energies reflect the predominantly electrostatic character of the protein/protein interaction, as the total energy of the complex consists mainly of the electrostatic energy contribution. The electrostatic character and the size of the buried surface area in the complex are typical for transient protein⅐ protein complexes (34).
All selected conformers featured a hydrogen bond between the Arg-129 side chain of Bcl-2 and the Asp-92 side chain of FKBP38(35-153), according to the predefined distance restraints that define this hot spot of the complex. Other characteristic intermolecular interactions in the cluster occur between Gly-128 (Bcl-2) and Asp-94 (FKBP38), Thr-125 (Bcl-2) and Asp-94 (FKBP38), and the aromatic rings of Phe-124 (Bcl-2) and Phe-88 (FKBP38). In fact, the hydrophobic side chain of Phe-124 (Bcl-2) is immersed in a hydrophobic pocket of FKBP38(35-153) that consists of Leu-70, Leu-86, Phe-88, Val-95, Ile-96, Leu-99, and Tyr-122 and corresponds to the active site in the prototypic FKBP12. Aside from a possible conformational change in the ␤5-␣1 loop upon complex formation, this structural binding arrangement could be yet another reason why the backbone amide signals of Val-95 and Ile-96 show significant CSP effects in the NMR titration experiments (Fig.  4A).
The interface of the protein⅐protein complex is thus characterized by a potential salt bridge between Asp-92 of FKBP38 and Arg-129 of Bcl-2 as well as by a non-polar contact of Bcl-2 residue Phe-124 with the putative active site of FKBP38 (Fig.  6B). The latter hydrophobic interaction might be too weak to produce significant CSP effects in the presence of the Bcl-2(1-211) R127Q/R129Q double mutant. However, a small CSP effect observed for the backbone amide signal of Tyr-122 (supplemental Fig. S10), whose side chain ring is also located in the putative active site, further supports such a hydrophobic binding contact.
Conclusions-Our results identify the molecular details of the interaction between Bcl-2 and the ␤5-␣1 loop in the catalytic domain of FKBP38. The binding interface consists of a salt bridge formed between Asp-92 of FKBP38 and Arg-129 of Bcl-2 as well as additional hydrophobic contacts at the putative active site of FKBP38. The predominant electrostatic interaction involves two negative charges in the ␤5-␣1 loop, which are unusual among the human FKBPs and mediate also the binding of divalent salt cations such as Ca 2ϩ (3).
Remarkably, the experiment with the Bcl-2(1-211) R127Q/ R129Q double mutant demonstrated that of 16 solvent-accessible arginine side chains in Bcl-2, only the identified epitope featuring Arg-127 and Arg-129 mediates the interaction with Asp-92 and Asp-94 in the ␤5-␣1 loop of the FKBP38 catalytic domain. This specificity of interaction between the catalytic domain of FKBP38 and the identified epitope of Bcl-2 is apparently defined by additional hydrophobic contacts between nearby Bcl-2 residues and the putative active site of FKBP38.
A second interaction interface between the two protein domains can be excluded based on the globular shape of both molecules and the lack of additional CSP effects in the catalytic domain of FKBP38. Previous studies have reported, however, that Bcl-2 is stabilized by interactions with both the catalytic domain and the TPR motif of FKBP38 (17,19). Moreover, the long flexible Bcl-2 loop, spanning 56 residues between the BH4 and BH3 domains from Asp-35 to Pro-90 (32), has been implicated to protect Bcl-2 from degradation by caspase-3. A close look at the Bcl-2 structure (supplemental Fig. S11) reveals that the here presented interaction site at Arg-129 and the long flexible loop are located at opposite ends of the Bcl-2 molecule, i.e. Ͼ25 Å apart. Hence, if this supposedly non-structured Bcl-2 loop should in fact interact with the TPR motif of FKBP38, thereby forming a "bidentate" protein⅐protein complex, this latter contact site may be extremely difficult to characterize on a structural level due to the very low solubility of Bcl-2 constructs featuring this flexible loop.
Electrostatic binding interactions, as observed here between Bcl-2 and the FKBP38 catalytic domain, are typical for transient complexes, which are often involved in signal transduction processes. Bcl-2 is a central regulator of mitochondrial apoptosis signaling, and the interaction of Bcl-2 with FKBP38 inhibits the prosurvival activity of Bcl-2 (9). Detailed structural insight into the binding process between both complex partners provides the molecular basis of FKBP38-mediated Bcl-2 regulation in human cells. The results presented here show that the formation of the FKBP38⅐Bcl-2 complex could be affected by the presence of other electrostatic factors such as Ca 2ϩ and thus provide a rationale as to why high Ca 2ϩ concentrations reduce the regulatory interaction between FKBP38 and Bcl-2 despite the requirement of CaM/Ca 2ϩ in cells (3,13).