Mcl-1–Bim complexes accommodate surprising point mutations via minor structural changes

Abstract

Mcl-1 is an antiapoptotic Bcl-2-family protein that protects cells against death. Structures of Mcl-1, and of other anti-apoptotic Bcl-2 proteins, reveal a surface groove into which the α-helical BH3 regions of certain proapoptotic proteins can bind. Despite high overall structural conservation, differences in this groove afford binding specificity that is important for the mechanism of Bcl-2 family function. We report the crystal structure of human Mcl-1 bound to a BH3 peptide derived from human Bim and the structures for three complexes that accommodate large physicochemical changes at conserved Bim sites. The mutations had surprisingly modest effects on complex stability, and the structures show that Mcl-1 can undergo small changes to accommodate the mutant ligands. For example, a shift in a leucine side chain fills a hole left by an isoleucine-to-alanine mutation at the first hydrophobic buried position of Bim BH3. Larger changes are also observed, with shifting of helix α3 accommodating an isoleucine-to-tyrosine mutation at this same position. We surveyed the variation in available Mcl-1 and Bcl-x_L structures and observed moderate flexibility that is likely critical for facilitating interactions of diverse BH3-only proteins with Mcl-1. With the antiapoptotic Bcl-2 family members attracting significant attention as therapeutic targets, these structures contribute to our growing understanding of how specificity is achieved and can help to guide the design of novel inhibitors that target Mcl-1.

Introduction

Antiapoptotic Bcl-2 proteins protect cells from programmed death and are implicated in deregulation of apoptosis that can lead to diseases including cancer. Death signals from a variety of cues, including DNA damage, growth factor withdrawal, and the presence of death ligands, are transmitted through proapoptotic BH3-only Bcl-2 proteins to induce apoptosis.1 BH3-only proteins inhibit the protective function of antiapoptotic Bcl-2 family members and probably also activate proapoptotic Bax and Bak proteins directly; details of the mechanism remain controversial.2 However, if BH3-only proteins, or BAX and BAK, are sequestered by antiapoptotic Bcl-2 proteins, death can be averted.3, 4 Thus, the interactions among antiapoptotic and proapoptotic proteins are crucial to the survival of the cell.

In humans, there are six protective Bcl-2 proteins: Bcl-2, Bcl-x_L, Bcl-w, Bcl-b, Mcl-1, and Bfl-1. Of these proteins, Bcl-x_L and Mcl-1 are the most extensively studied, but high-resolution structures now available for five of six reveal a structurally conserved hydrophobic groove, into which the BH3 α-helix from proapoptotic proteins can bind.5–9 Because of this interaction, we refer to these antiapoptotic proteins as “receptors.” It has been demonstrated that short peptides derived from the BH3 region of proapoptotic proteins are sufficient for binding to antiapoptotic Bcl-2 protein receptors and promoting mitochondrial permeabilization, a hallmark of apoptosis.3, 10, 11

The affinities of antiapoptotic receptor proteins for numerous BH3 peptides have been determined by several research groups.10, 12, 13 Although differences in reagents and methods have led to significant variability in reported results, certain trends are clear: (i) some BH3 peptides bind with high affinity to all antiapoptotic receptors; (ii) some bind Bcl-x_L, Bcl-2, and Bcl-w proteins with high affinity, but not the more distantly related Mcl-1 and Bfl-1; other BH3 peptides have the reverse specificity. Thus, Bcl-x_L and Mcl-1 have distinct BH3 interaction profiles. At a more detailed level, there seem to be unique binding profiles for each of the Bcl-2 family receptors, and these preferences have mechanistic implications for the regulation of apoptosis under different stimuli and the resistance of cancer cell lines to death-inducing signals.14

Although Mcl-1 has a very similar structure to other antiapoptotic Bcl-2 proteins (backbone root mean square deviation [RMSD] less than 2 Å), it only shares ∼25% sequence identity with other family members and has a significantly different BH3 binding profile. Most notably, Mcl-1 binds to the BH3 region of Noxa but not Bad, a reversal of specificity when compared with Bcl-2, Bcl-x_L, and Bcl-w.12, 13 Mcl-1 is a highly regulated protein at the level of expression, degradation, and posttranslational modification.15 It also has a short half-life, because of the role of its amino-terminal PEST domain in targeting proteins for proteasomal degradation.16 Mcl-1 was isolated from human myeloid leukemia cells17 and plays a key role in the regulation of apoptosis. For example, in healthy cells Mcl-1 (in addition to Bcl-x_L) sequesters Bak, preventing its proapoptotic function until competitive binding by a BH3-only protein releases it.4 Thus, Bcl-x_L and Mcl-1 share a similar function of protecting the cell from death, via similar structure and mechanism. However, these two proteins have very different binding specificities, expression patterns, and regulation.

Targeting the Bcl-2 family of proteins is a promising strategy for treating cancer. Evading apoptosis is one of the hallmarks of cancerous cells,18 and up-regulation of antiapoptotic Bcl-2 proteins is one way to avoid cell death. Small molecules capable of inhibiting the interactions of antiapoptotic proteins have been developed recently and are in clinical trials as anticancer therapeutics. One of the most promising Bcl-2 inhibitors was developed at Abbott Laboratories. ABT-737, and its orally available counterpart, ABT-263, have low nanomolar affinity for Bcl-2, Bcl-x_L, and Bcl-w and are in phase I/II clinical trials.19, 20 Reminiscent of BH3 peptide binding preferences, however, many small molecule inhibitors that bind the highly related proteins Bcl-2, Bcl-x_L, and Bcl-w show little or no binding to Mcl-1 and Bfl-1. This inability to inhibit Mcl-1 is significant, as upregulation of Mcl-1 seems to be a major source of resistance to the Abbott compounds.21, 22 An ability to target other Bcl-2 proteins would be beneficial to the advancement of cancer therapeutics, and the development of Mcl-1 inhibitors has become a high priority.

A variety of studies have begun to address how BH3 peptide binding specificities are determined at the level of receptor sequence and structure. The sequences of native BH3 peptides are highly variable. However, the helical nature of the BH3 region, when bound to the Bcl-2 receptor, places one face of an amphipathic helix in the receptor hydrophobic groove. The residues buried in the interface follow a heptad repeat, [abcdefg]_n, with “a” and “d” positions buried—a characteristic also found in coiled coils (Table I).16 Conserved features are largely restricted to a small residue at position 3e (i.e., the “e” position of the third heptad repeat), an aspartate residue at position 3f, and four hydrophobic residues, corresponding to the positions 2d, 3a, 3d, and 4a. The hydrophobic residues are mostly buried in the interface with Bcl-x_L, as shown in several structures of complexes with BH3 peptides. Sequence determinants of binding have been explored using alanine scanning and hydrophile scanning of Bim, as well as saturation mutagenesis to test 18 amino acids at Bim sites 3a and 4a.23–25 These studies revealed positions where differential effects are observed for binding to Bcl-x_L versus Mcl-1, particularly site 4a, which is much more tolerant of mutations in Mcl-1 complexes. Earlier studies speculated that crowding and electrostatics are important at several BH3 positions for determining the binding specificity for Bcl-x_L and Mcl-1.12, 16 For example, a large-to-medium tyrosine-to-isoleucine mutation in the buried 2d site of Bad significantly improved binding to Mcl-1.16 The 3d position was proposed to have the opposite effect on specificity, with large residues preventing binding to Bcl-x_L; in support of this model, mutation of the phenylalanine at 3d in Noxa to isoleucine improves binding to Bcl-x_L.12 The 3g position, conversely, was proposed to play a role in specificity determination against Bcl-x_L, based on electrostatics. This residue is typically negatively charged in BH3 sequences. However, in Noxa, this site is lysine and was hypothesized to prevent binding to Bcl-x_L due to an arginine (Arg 100) on the edge of the hydrophobic groove. Mutation of this lysine in Noxa to glutamate improved binding to Bcl-x_L.12 In addition, the highly conserved leucine at position 3a helped to confer specificity for Mcl-1 over Bcl-x_L when mutated to an alanine in Bim.23 However, when the same leucine in Noxa was mutated to an alanine, binding to Mcl-1 was lost,26 indicating that the effects of mutations can depend on their context.

Table I. Binding Affinity of Bim Peptides for Mcl-1 and Bcl-x_L

FP assay	BH3 peptide	Sequence				Mcl-1	Bcl-xL
		1	2	3	4
a The Bad peptide for direct binding was fluoresceinated at the bold, italicized lysine. Kd and IC50 values are reported as an average of triplicate binding assays with standard deviations. b Kd and IC50 values are calculated from duplicate runs.
						Kd (nM)	Kd (nM)
Direct	Bim					<2	<2
	Bim I2dA					<2	<2
	Bim I2dY					<2	<2
	Bim F4aE					<2	50.9 ± 1.7
	Bada					>5,000	15.2 ± 0.8
						IC50 (nM)	IC50 (nM)
Competition	Bim					14.0 ± 0.9	47.1 ± 6.6
	Bim I3dF					56.2 ± 9.3	28.7 ± 3.3b
	Bim E3gK					92.7 ± 8.9	81.8 ± 20.8
	Bad					>10,000	20.3 ± 1.8b
	Noxa					47.3 ± 0.2b	>10,000
	Noxa C2dY					58.7 ± 4.3	>10,000

Although many structures of Bcl-2 family proteins and complexes are now available, these do not yet provide a clear view of how specificity is achieved. In particular, although the overall structure of Bcl-2 family receptors is well conserved, modest changes have been observed that accommodate ligands in unanticipated ways. This is especially apparent in structures of Bcl-x_L in complex with small molecules.25, 27 Future efforts in drug discovery, and a mechanistic understanding of factors governing BH3 binding, would benefit from a better description of the structural states accessible to Bcl-2 family receptors. Herein, we examine the specificity of Bcl-2 family interactions with a focus on Mcl-1. We constructed several point mutations in a peptide corresponding to the BH3 region of human Bim that changed the physical/chemical nature of residues at the complex interface. These peptides maintained high affinity binding to Mcl-1. We report X-ray crystal structures of several peptides in complex with Mcl-1 that illustrate how certain substitutions can be accommodated. This is relevant to understanding how high affinity binding is maintained to peptides that differ from canonical BH3 sequences, and to elucidating structural determinants of binding affinity and specificity.

Results

The BH3 region of Bim has been used extensively for studies exploring the relationship between sequence, structure, and binding in the Bcl-2 family.23–25 This work focused on the analysis of two positions in Bim that have the potential to influence binding specificity. As defined in Table I, these are positions 2d (isoleucine) and 4a (phenylalanine). These sites, along with positions 3a and 3d, are largely conserved as hydrophobic in most BH3 sequences and have been the subject of many prior studies.

Human Mcl-1/Bim complex

To provide a structural context for interpreting changes in binding, we solved the structure of human Mcl-1, consisting of residues corresponding to the Bcl-2 fold without the PEST or transmembrane domains, in complex with a peptide consisting of the BH3 region from human Bim. Single-wavelength anomolous diffraction (SAD) phasing using a selenomethionine derivative provided the native structure at a resolution of 2.0 Å, with an R_free of 0.23. The model includes residues 172–197, 203–321 of Mcl-1, and 0–22 of Bim. There is one Mcl-1–Bim complex per asymmetric unit, and a disulfide bond connects two Mcl-1 molecules across a two-fold crystallographic interface via the sole cysteine residue. This is likely an artifact, as only a small amount of reductant (1 mM dithiothreitol [DTT]) was present in the protein stock, and none was added to the crystallization conditions. Several zinc ions are present in the asymmetric unit, coordinated by acidic residues, histidines, and water molecules. Zinc ions, positioned at crystal contacts, were essential for the crystallization of this complex.

The structure of Mcl-1 bound to wild-type Bim is similar to other Bcl-2-family complexes. The Mcl-1 protein has eight alpha helices (α1-α8), with α2-α5 and α8 forming a hydrophobic groove into which the BH3 peptide binds. The structure of Mcl-1, mostly human with some murine sequence, in complex with a human Bim-derived BH3 peptide has also been reported.6 Despite the murine sequence in this complex, the fully human and chimeric structures are nearly identical, with a backbone RMSD of 0.29 Å. Although these structures were solved independently, both used similar crystallization conditions, with zinc as a key factor. Interestingly, we observe no evidence of buried water near Bim BH3 positions 3a (Leu 10) and 3e (Gly 14) at the hydrophobic Mcl-1–Bim interface, as was observed in the chimeric structure.

Comparisons of the human/murine Mcl-1–Bim complex with unliganded Mcl-1 and with murine Bcl-x_L (mBcl-x_L) in complex with Bim have been reported previously6; only a few points are noted here. First, the α3 regions of Mcl-1 versus mBcl-x_L show interesting differences, with the helix longer and more helical in Mcl-1–Bim than Bcl-x_L–Bim (PDB IDs 2PQK vs. 1PQ1). In the recently solved human Bcl-x_L–Bim BH3 complex (3FDL), the third helical region is more helical than in murine Bcl-x_L–Bim (1PQ1), but this region still differs significantly from the conformation observed in human Mcl-1–Bim [Fig. 1(A)]. Most of the human Bcl-x_L complexes with BH3 peptides exhibit an α3 region that is less helical, more closely resembling the murine Bcl-x_L–Bim complex. With respect to BH3 binding specificity, a critical difference between Mcl-1 and Bcl-x_L is that the region of the Mcl-1 groove that binds to the C-terminal end of BH3 peptides, around the peptide 4a site, is much more open than the same region of Bcl-x_L. This is shown in Figure 2(A,B) and is discussed further later.

Figure 1. Structural changes in Mcl-1 versus Bcl-x_L on binding Bim BH3 peptides. (A) Superposition of Bim-bound complexes of Mcl-1 (red, peptide in pink, 2PQK) and Bcl-x_L (blue, peptide in light blue, 3FDL). (B) Structure of Mcl-1 uncomplexed (yellow, 1WSX) and bound to Bim BH3 (Mcl-1 in red and Bim in pink). (C) Structure of Bcl-x_L uncomplexed (green, 1MAZ) and bound to Bim BH3 (Bcl-x_L in blue and Bim in light blue). The gray arrows in (B) and (C) highlight the major conformational shifts upon binding.

Figure 2. Comparison of the Mcl-1 hydrophobic pocket around the 4a position of Bim BH3. (A) Wild-type Bim BH3 bound to Mcl-1 (2PQK). (B) Wild-type Bim bound to Bcl-x_L (3FDL). (C) Bim mutant F4aE bound to Mcl-1 (3KJ2). The surfaces of Mcl-1 and Bcl-x_L are represented using CPK coloring. The Bim BH3 helix is shown with side chains as sticks. The 4a position of Bim BH3 is shown in yellow. A zinc ion in the F4aE mutant complex is represented as a teal sphere.

Mcl-1 and Bcl-x_L undergo different conformational changes on binding Bim BH3. When Bim BH3 binds to Mcl-1, the hydrophobic groove widens, predominantly through reorientation of the carboxy-terminal end of α4 away from the Bim helix at the C-terminal-binding end of the groove [Fig. 1(B)]. Residues Ala 227, at the amino-terminus of α3, and Val 253, at the carboxy-terminus of α4, move away from one another on Bim binding. The distance between these residues increases from 11.2 Å (C_α-to-C_α atom) in the unbound state to 15.6 Å in the Bim-bound state. Conversely, the bottom of the N-terminal-binding end of the groove in Bcl-x_L restructures to accommodate the Bim BH3 helix, partially by means of an N-terminal shift in the residues that compose helix α3 [Fig. 1(C)]. The C_α of Leu 108 in α3 of Bcl-x_L shifts 7.2 Å from the unbound to the bound state.

Position 4a

Lee et al.25 have shown via exhaustive amino acid substitution that position 4a tolerates many different mutations when binding to Mcl-1. Before resolution of the Mcl-1–Bim structure, this was surprising, because this residue is conserved as hydrophobic in all native BH3s. As discussed earlier, however, it is now clear that the 4a pocket is much more open and solvent exposed relative to the same pocket in Bcl-x_L, explaining some of the permissiveness. Here, we report a solution K_d of less than 2 nM for Bim BH3 F4aE (Bim BH3 peptide with the phenylalanine at position 4a mutated to glutamate) binding to Mcl-1, which is comparable with wild-type Bim. However, this mutation was destabilizing for binding to Bcl-x_L (∼50-fold decrease in binding affinity, Table I). This is consistent with Lee et al. who measured interactions using a phage-ELISA assay, and with Boersma et al. who measured solution K_d values using an 18-mer peptide (K_d = 10 nM for Mcl-1 and no binding to Bcl-x_L up to 10 μM).24, 25 To explore in more detail how the charged glutamate residue is accommodated at the interface of the complex, we solved a 2.35 Å X-ray structure of Bim F4aE bound to Mcl-1 and compared it with our structure of human Mcl-1 bound to wild-type Bim BH3.

The F4aE mutant structure is very similar to the wild-type structure. The backbone RMSD over 154 of 166 common Mcl-1 and Bim residues in the two structures is just 0.45 Å. The glutamate at the 4a position (Glu 17) rotates out of the hydrophobic groove [Fig. 2(C)]. A tyrosine at position 4e shifts to partially fill the vacated space. Along with the glutamate at 4a, a glutamate at 3g in Bim (Glu 16) and His 224 in Mcl-1 coordinate a zinc ion present from the crystallization conditions. Interestingly, the glutamate at 4a is not required for zinc coordination; in a crystal structure of Mcl-1 bound to the double mutant Bim L3aA F4aA, (3D7V), a zinc ion is similarly coordinated without the glutamate at 4a.23 Although this coordinating conformation is likely nonphysiologic, it does illustrate how glutamate at this position can quite easily be accommodated in the shallow, open pocket in Mcl-1 at this site. It is likely that even in the absence of zinc, the glutamate at this position would have to rotate out of the protein core to avoid a severe desolvation penalty. Our structure illustrates how it can do so, with extremely minor perturbation to the rest of the complex structure.

Position 2d

Tyrosine at the 2d position is a distinguishing feature of the Bad BH3 sequence. Bad binds Bcl-x_L tightly, but does not bind with high affinity to Mcl-1. It was previously demonstrated that mutating tyrosine to isoleucine (Bad Y2dI) could increase affinity for Mcl-1.16 This might suggest that this single residue can act as a specificity determinant. Consistent with this, there is little room at the 2d position to accommodate a larger residue, such as tyrosine, in the crystal structure of Mcl-1 in complex with wild-type Bim. This leads to the expectation that Bim I2dY might not bind Mcl-1 with high affinity. However, direct binding experiments show that a Bim I2dY BH3 peptide retained low nM binding affinity (Table I). Tight binding was also retained to Bcl-x_L. We also made a similar mutation in Noxa, a BH3 sequence that binds to Mcl-1 and not Bcl-x_L. The Noxa C2dY mutant shows the same behavior, with a wild-type-like affinity for Mcl-1, as measured by a competition binding assay (Table I). This is consistent with the reported tight binding of Mcl-1 to murine NoxaA, which has phenylalanine at the 2d position, and of murine NoxaB with a glutamate-to-phenylalanine substitution at this site.12, 26

The native structure of Mcl-1/Bim shows that there is not sufficient room to accommodate a large tyrosine at 2d without a change in receptor and/or peptide conformation. To understand how high affinity binding to this mutant peptide is maintained, we solved the structure of the Mcl-1/Bim I2dY complex. The I2dY Bim mutant in complex with Mcl-1 crystallized in the P2₁2₁2₁ space group from 2-methyl-2,4-pentanediol (MPD) with Tris pH 7.5, in the absence of ions or disulfides, and diffracted to 1.7 Å. To further explore the range of Mcl-1 flexibility as a function of the size of the 2d position residue, we also solved the 1.95 Å structure of a Bim I2dA BH3 peptide in complex with Mcl-1. The I2dA mutant complex was crystallized under similar conditions as the wild-type and Bim F4aE mutant complexes. As expected, this structure had several zinc ions in the asymmetric unit and also included the same disulfide bond across the two-fold crystallographic interface that was seen in the other Mcl-1/Bim complexes.

The Mcl-1/Bim I2dA structure shows essentially no change in backbone from the wild-type Bim complex [all-backbone-atom RMSD = 0.28, Fig. 3(A)]. The most significant difference between the wild-type and I2dA mutant complexes is a shift of Leu 235 on Mcl-1 to fill the cavity created by the small alanine [Fig. 3(B)]. This is similar to the shift by Met 231 observed in the structure of Mcl-1 bound to double mutant Bim L3aA F4aA (3D7V).23

Figure 3. Structures of Mcl-1 in complex with Bim BH3 peptides mutated at position 2d. (A) Wild-type Mcl-1 bound to Bim BH3 (red, 2PQK), to Bim I2dY (teal, 3KJ0), and to Bim I2dA (purple, 3KJ1) are shown superimposed. The vertical helix in the front is the bound Bim BH3 peptide (pink for wild-type Bim, light cyan for Bim I2dY, and lavender for Bim I2dA). (B) The wild-type complex and the Bim I2dA mutant complex are shown superimposed, with the 2d mutation position and Leu 235 of Mcl-1 shown using sticks. (C) The wild-type complex and the Bim I2dY mutant complex, with the 2d positions shown using sticks, are superimposed. Colors in B and C are as in (A).

In contrast, the Mcl-1/Bim I2dY structure showed a larger conformational change. To analyze this change, we defined a scaffold of the most structurally invariant regions of Mcl-1 from several structures, excluding flexible loop regions and the α3-α4 region that reorganizes in different complexes. The conserved structural scaffold used for this comparison corresponds to residues 172–191, 204–221, 260–281, and 287–319. For Mcl-1–Bim I2dY compared with Mcl-1/Bim, these residues can be superimposed with an all-backbone-atom RMSD of just 0.58 Å, when compared with 1.18 Å for all backbone atoms. Structural differences are localized primarily to the N-terminus of the peptide and the α3 region of Mcl-1 (Table III). Based on superposition of the structural scaffold, the RMSD for residues 222-259, comprising α3, the α3-α4 loop, and α4, is 1.52 Å. The peptide moves away from the receptor by 1.2 Å (measured at the C_α of position 2d), and the α3 helix moves away from the peptide by 1.7 Å (measured at the C_α of Leu 235), to accommodate the larger tyrosine residue [Fig. 3(C)]. Thus, Mcl-1 accommodates a Bim mutant with a large tyrosine at the first buried position, with an affinity similar to that of wild-type Bim, by local conformational changes of the peptide and α3.

Table II. Mcl-1 Complexes Compared With Human Mcl-1–Bim BH3 Complex 2PQK Using a Core Structural Alignment

PDB ID	Peptide	Mcl-1		RMSD (Å)
		Species	Technique	Core	α3–α4	α3	All Mcl-1	Peptide
2NL9	Bim	Chimera	X-ray	0.268	0.347	0.230	0.298	0.233
3KJ0	Bim I2dY	Human	X-ray	0.580	1.518	1.203	1.184	0.938
3KJ1	Bim I2dA	Human	X-ray	0.175	0.435	0.526	0.285	0.174
3KJ2	Bim F4aE	Human	X-ray	0.355	0.634	0.934	0.456	0.407
3D7V	Bim 2A	Chimera	X-ray	0.384	0.597	0.962	0.469	0.302
3IO9	Bim L12Y	Chimera	X-ray	0.361	0.818	1.075	0.533	0.583
2NLA	NoxaB	Chimera	X-ray	0.673	1.690	1.066	1.285	2.306
2ROD	NoxaA	Mouse	NMR	0.748	1.977	1.392	1.462	0.624
2JM6	NoxaB	Mouse	NMR	0.809	1.778	1.199	1.407	0.719
2ROC	Puma	Mouse	NMR	0.831	2.185	1.316	1.595	0.976
1WSX	apo	Mouse	NMR	1.289	2.645	2.005	1.957	n/a

Table III. Bcl-x_L Complexes Compared With Human Bcl-x_L–Bim BH3 Complex 3FDL Using a Core Structural Alignment

PDB ID	Peptide/ligand	Bcl-xL		RMSD (Å)
		Species	Technique	Core	α3–α4	α3	All Bcl-xL	Peptide
1PQ1	Bim	Mouse	X-ray	0.346	3.005	4.787	1.589	0.829
2BZW	Bad	Mouse	X-ray	0.535	2.552	4.061	1.410	1.266
2P1L	Beclin	Human	X-ray	0.328	1.893	2.993	1.037	1.094
3IO8	Bim L12F	Human	X-ray	0.561	3.169	4.986	1.714	1.231
2YXJ	ABT737	Human	X-ray	0.315	1.403	2.000	0.802	n/a
3INQ:A	W1191542	Human	X-ray	0.487	2.989	3.551	4.277	n/a
3INQ:B	W1191542	Human	X-ray	0.392	1.909	1.240	4.050	n/a
1MAZ	apo	Human	X-ray	0.456	4.735	6.554	2.483	n/a
1G5J	Bad	Human	NMR	1.166	3.462	5.270	2.060	1.551
1BXL	Bak	Human	NMR	1.225	2.717	3.107	1.769	3.130
1LXL	apo	Human	NMR	1.204	4.734	6.817	2.682	n/a

Testing other candidate specificity determinants in Bim

Two other point mutations reported to affect Bcl-x_L versus Mcl-1 binding specificity are K3gE and F3dI. When made in the context of a Noxa BH3 peptide, each change restored some binding to Bcl-x_L. These two mutations could also be combined to give a greater effect.12 Lysine at 3g and phenylalanine at 3d are, therefore, candidates for disfavoring binding to Bcl-x_L. We introduced these changes individually into a Bim BH3 peptide. In a competition binding assay, Bim I3dF retained tight binding to Bcl-x_L, and binding of E3gK was only weakened by <2-fold in IC₅₀ (Table I). Using 18-mer peptides, Boersma et al.24 previously reported ∼10-fold weaker binding for the E3gK mutant to both Bcl-x_L and Mcl-1, compared with wild-type Bim.

Structural variation in Bcl-2 proteins

The past few years have seen a dramatic increase in the number of published Bcl-2 family protein structures. Taken together, these begin to paint a picture of the structural variability accessible to the antiapoptotic receptors, which is likely important for determining their binding specificity profiles. Lee et al.27 recently proposed that Bcl-x_L has greater structural plasticity than Mcl-1. To systematically evaluate the differences among published Mcl-1 and Bcl-x_L structures, we carried out structural comparisons using a conserved core scaffold. Defining a core where structural changes are minimal, and using this as the basis for structural superposition, made it easier to identify and describe the regions where changes do occur. We explored different core scaffold definitions, ranging from only helices 5 and 6 to all helices excluding helices 3 and 4. Similar results were obtained in all cases, supporting a low RMSD over core residues and higher RMSD in the localized α3-α4 region. In contrast, aligning over the full-length of the proteins, excluding only the highly flexible loop between α1 and α2, masked some of the flexibility of the α3-4 region (data not shown). Structural differences captured as all-backbone-atom RMSD values, based on our largest conserved core, are shown in Table II, where Mcl-1 structures are compared with human Mcl-1–Bim structure 2PQK as a reference. In Table III, Bcl-x_L complexes are compared with 3FDL, a human Bcl-x_L–Bim structure.28

Several things are apparent in these comparisons. First, the degree to which core structural scaffolds can be closely superimposed depends on the method used to solve the structure, but in all cases, the core scaffold structure is highly conserved. X-ray structures are more similar to one another than to NMR structures, with scaffold RMSD values of ∼0.4 Å versus ∼1.1 Å. These modest differences may originate from the similar crystal packing of many of the X-ray complexes or the use of existing X-ray models for phasing by molecular replacement. Differences also likely reflect the dynamic nature of proteins in solution. Despite this dependence on method, trends in structural variability can be discerned. The α3-α4 region is the most variable region in Bcl-2 family structures, outside of disordered loops. Table II shows that available Mcl-1 structures span RMSD values of 0.3–2.6 Å over 38 residues in this part of the structure (eight X-ray structures and four NMR). In contrast, Bcl-x_L structures vary more, with RMSD values in the 1.4–4.7 Å range (eight X-ray and three NMR structures). In both cases, most structural differences lie in α3, which has been the subject of discussion in the analysis of several individual complexes.27, 29, 30 This overall trend is apparent in manual inspection of structural superpositions, where the α3 region of Bcl-x_L varies significantly in its helical content in different complexes (Fig. 4). Structural differences in BH3 peptide structure and binding are also greater in Bcl-x_L complexes (see the “Peptide” column in Tables II and III, which reflects both the structure of the peptide and its orientation with respect to the receptor). In conclusion, existing structures show more structural variation for Bcl-x_L than for Mcl-1. This may be simply because a greater diversity of liganded Bcl-x_L complexes have been solved to date or it may reflect a real difference in the flexibility of these two proteins. Regardless, it means that for purposes of analysis and drug design, more snapshots of possible conformations are available for Bcl-x_L than for Mcl-1 at this time. The structures we have solved begin to address this imbalance, especially the Bim I2dY structure, which illustrates one way that Mcl-1 can adjust to accommodate changes in ligand size and shape.

Figure 4. Structural variation in Mcl-1 and Bcl-x_L complexes. Mcl-1 (A and B) and Bcl-x_L (C–E) structures are shown overlaid using core residues (in gray) as defined in the text. The α3-loop-α4 region of Mcl-1 or Bcl-x_L, and the BH3 binding peptide, if present, are shown in varying colors. All structures are in the same orientation, thus the α3, α4, and peptide regions identified in (A) can be applied to all panels. (A) and (C) show all structures compared in Tables III and IV. (B) and (D) show only the structures solved by X-ray crystallography. An alignment of peptide-bound crystal structures of Bcl-x_L is shown in (E). The Bim-bound reference structure is shown in gray: 2PQK for (A-B) and 3FDL28 for (C–E). In (A and B), the colors are as follows: 2NL9 in red,63KJ0 in teal, 3KJ1 in purple, 3KJ2 in orange, 3IO9 in beige,273D7V in pink,232NLA in blue,62ROD in green,262JM6 in dark blue,62ROC in light green,26 and 1WSX in yellow.16 In (C–E), the colors are as follows: 1PQ1 in dark blue,31 2BZW in red, 2P1L in orange,303IO8 in blue,272YXJ in dark green,253INQ:A in light green, 3INQ:B in cyan,271MAZ in purple,321G5J in pink,111BXL in yellow,5 and 1LXL in lavender.32

Table IV. Data Collection and Refinement Statistics for Mcl-1 Complexes

Data collection	WT native	WT Se-Met	I2dY mutant	I2dA mutant	F4aE mutant
X-ray source	24ID	24ID	24ID	CuKα	CuKα
Space group	I222	I222	P212121	I222	I222
Cell dimensions (Å)
a	53.09	52.88	40.38	53.07	51.46
b	71.85	71.20	52.94	72.01	71.49
c	118.14	118.12	69.37	117.85	119.44
Resolution (Å)	61.43–2.00	60.97–2.03	29.1–1.70	28.9–1.95	24.3–2.35
Highest bin	2.05–2.00	2.10–2.03	1.76–1.70	2.07–1.95	2.43–2.35
No. reflections	14761	14764	16317	16418	9515
Rsym	0.062	0.059	0.069	0.042	0.057
Highest bin	0.335	0.367	0.298	0.390	0.402
Completeness (%)	99.0	99.6	96.34	96.9	99.9
Highest bin	95.5	97.7	68.8	79.5	99.8
Redundancy	6.9	7.2	7.9	10.2	9.4
Highest bin	5.6	5.3	3.1	7.6	5.6
Refinement	2PQK		3KJ0	3KJ1	3KJ2
Program	Refmac		Refmac/Phenix	Phenix	Phenix
Rwork/Rfree	19.4/23.0		18.5/22.4	18.7/21.3	20.8/24.7
No. atoms	1480		1633	1550	1481
Protein	1358		1486	1424	1409
Ions	4		1	10	8
Water	118		139	116	64
B factors (Å^2)	36.1		22.6	42.3	44.3
Protein	35.1		21.6	42.4	44.3
Solvent	47.0		33.0	47.8	45.0
RMSDs
Bond lengths (Å)	0.012		0.003	0.009	0.002
Bond angles (°)	1.085		0.643	0.995	0.549
MolProbity40 statistics (%)
Ramachandran favored	98.8		98.9	98.3	99.4
Ramachandran outliers	0		0	0	0
Bad rotamers	1.4		0.65	0.66	1.33

Discussion

We have solved four structures of Mcl-1 in complex with high affinity Bim BH3 peptides. These structures exhibit conformational differences that explain the ability of Mcl-1 to bind a range of sequences and also inform speculation about possible determinants of Mcl-1 binding specificity.

BH3 peptides of diverse sequence bind the Bcl-2 receptors; however, Mcl-1 has very different interaction specificity from Bcl-2, Bcl-x_L, and Bcl-w. Determinants of these specificity differences are not yet fully understood, although prior studies provide some insights. For example, at the first buried position (2d), isoleucine and other medium-sized hydrophobic residues are found in most BH3 peptides capable of binding both Mcl-1 and Bcl-x_L. However, for Bad, which does not bind tightly to Mcl-1, this residue is tyrosine. This observation led Day et al.16 to hypothesize that crowding effects at this position might disfavor Mcl-1 binding by Bad, and indeed, mutation of tyrosine to isoleucine in the context of Bad improves binding to Mcl-1 while retaining binding to Bcl-x_L. The structure of Mcl-1–Bim also shows that tight packing of isoleucine at 2d in this region could potentially prevent a good fit for larger residues, although murine NoxaA, which has phenylalanine at this site, does bind tightly to murine Mcl-1.26 To test whether a steric exclusion model exerts a strong influence at the 2d site in human Bim, we made the I2dY mutant. This peptide bound Mcl-1 with high affinity. Therefore, steric clashes at position 2d do not provide a reliable single-site rule to govern specificity, and our structure demonstrates why this is the case. Mcl-1 accommodates tyrosine at the first buried position with a shifting of helix α3, whereas the mutant peptide bends modestly, to provide the necessary space for the aromatic residue. Furthermore, the I2dA mutant showed that small residues are also tolerated at this same position, where slight changes in residue conformation compensate for the alanine mutation, maintaining high affinity (Table I).

The impact of a given residue on Bcl-2 family binding specificity can vary with the context in which it is presented. In addition to the differential effects of isoleucine versus tyrosine in Bad and Bim presented above, we also found that isoleucine versus phenylalanine at the 3d position and, to a lesser extent lysine versus glutamate at 3g, gave different effects in the context of Noxa as compared with Bim. Similarly, it has been shown that mutation to any amino acid at the 4a position in Bim has little effect on Mcl-1 binding, yet Noxa with a leucine-to-alanine mutation at this site eliminates Mcl-1 binding.25, 26 It is not yet clear whether these differences indicate subtly different binding modes for the different peptides, synergy between different positions, or simple differences in the overall affinities of Bad, Noxa, and Bim for the various receptors. However, an emerging trend is that the high affinity Bim BH3 peptide can accommodate a wide variety of point mutations, and even surprising combinations of several alanine substitutions, while maintaining tight binding to anti-apoptotic receptors.23, 24 It would seem Bim has a large capacity to buffer such sequence changes.

The fourth buried hydrophobic position (4a) clearly exerts an important influence on Mcl-1 versus Bcl-x_L-binding specificity.24, 25 Saturating mutagenesis previously indicated that any amino acid at this site maintained Mcl-1 binding, but many charged or polar amino acid substitutions attenuated binding to Bcl-x_L.25 Our Bim F4aE mutant bound Mcl-1 with a K_d of less than 2 nM. For Bcl-x_L, the affinity was reduced more than 10-fold (Table I), consistent with a quantitative phage-ELISA assay that showed an approximately 13-fold reduction in affinity.25 This reduced affinity for Bcl-x_L was also observed in an 18-mer Bim peptide harboring the F4aE mutation that did not bind Bcl-x_L at concentrations up to 10 μM in a fluorescence polarization assay. The peptide length is likely responsible for the enhanced decrease in affinity, because 18-mer wild-type Bim showed a 6-fold reduction in affinity when compared with the 26-mer wild-type Bim peptide.24 Simple rotation of glutamate at this site out of the hydrophobic groove, observed in the Mcl-1–F4aE complex, demonstrates how Mcl-1 is able to maintain high affinity for peptides with a wide variety of residues at this site. However, Bcl-x_L is more constrained at this site [Fig. 2(B)], likely preventing similar side-chain repositioning.

The structures presented here reveal different types of conformational change induced by mutation of the Bim BH3 sequence. Some mutant structures are nearly identical to the wild-type structure of Mcl-1 with Bim BH3. The F4aE and I2dA mutant structures are reminiscent of other Bcl-2 family complex structures, where changes in sequence give rise to minimal structural adaptation. The structure of a double alanine mutant of Bim (L3aA F4aA) bound to Mcl-1 is nearly the same as the wild-type complex, with a backbone RMSD of 0.45 Å.23 Even more striking, the C_α RMSD of Bfl-1 bound to different native BH3 sequences in cocrystal structures is less than 0.7 Å.7 Each of these structures accommodates a mutation via modest side-chain rearrangements, without changes to the receptor backbone. However, for some mutants, such as I2dY, changes are more noticeable and involve backbone alterations. In the I2dY structure, the scaffold region of the receptor including BH regions 1–3 is relatively invariant, but the α3-α4 region shifts to accommodate the larger tyrosine residue. This change is similar to a shift in Mcl-1 that is observed on binding murine NoxaA, which also has a large aromatic phenylalanine at the 2d position.26 Structures of previously unobserved conformations for Mcl-1 are valuable for several reasons. They can be used as templates in docking studies to accelerate drug discovery and/or in computational design of novel BH3 ligands.27, 33 New structures can also lead to a better understanding of determinants of binding affinity and specificity, with a growing set of Bcl-2 complex structures revealing how a range of peptide ligands can be bound, sometimes in subtly different conformations.7, 29, 34 This type of information has value for developing a better, more predictive understanding of BH3 binding.

It is interesting to note that the variability of Mcl-1 structures observed so far is smaller than what has been seen for Bcl-x_L. Pair-wise comparison of different BH3-bound Mcl-1 complexes solved by X-ray crystallography with the wild-type Bim BH3 complex, when structurally aligned using a common core, show an average RMSD over the α3-α4 region of 0.86 Å, whereas similar comparison in Bcl-x_L shows an average RMSD of 2.7 Å (for peptide-bound Bcl-x_L crystal structures). Bcl-x_L structures have been solved in complex with a greater variety of ligands; however, it remains to be seen if Mcl-1 exhibits similar plasticity in as yet unseen complexes.

In summary, it is clear that modest flexibility of the structurally conserved Bcl-2 antiapoptotic receptor fold can accommodate binding to diverse BH3-like peptide sequences. The structural rearrangements that have been observed are difficult to anticipate and make it challenging to predict whether individual BH3 peptides will bind various Bcl-2 proteins. Primitive knowledge of how Bcl-2 receptor structures change on ligand binding also hampers structure-based drug design. It is likely that new modes of structural adaptation remain to be observed. As we move towards a more detailed and predictive understanding of Bcl-2 family binding, and eventually the design of new classes of therapeutic inhibitors, information about structural diversity will be valuable. The complexes we describe here contribute to the growing body of available structural data.

Materials and Methods

Bcl-2 proteins

The cDNA for Bcl-x_L and Mcl-1 was obtained from the Harvard Institute of Proteomics. Bcl-x_L (residues 1-209) and Mcl-1 (residues 172-327) were expressed as maltose binding protein (MBP) fusions from the pSV282 vector (provided by Dr. Laura Mizoue at Vanderbilt University). This Mcl-1 construct was designed to avoid problems with instability of longer Mcl-1 constructs under the purification conditions. Proteins were expressed in BL21 pLysS, and cultures were induced with 0.3 mM IPTG at an OD600 of 0.4–0.6. Proteins were purified by Ni-affinity chromatography, followed by cleavage with Tobacco Etch Virus (TEV) in protease 50 mM Tris, 50 mM NaCl, 0.5 mM EDTA, and 1 mM DTT at pH 8.0 for 3 hr at ∼25°C. The cleavage mixture was passed over a Ni-NTA column, and Mcl-1 in the flow-through was further purified by gel filtration chromatography on S75 resin. This construct is stable over the course of many days at ∼25°C in 20 mM NaPO₄, 50 mM NaCl, and 1 mM EDTA at pH 7.5.

Selenomethionine-derivatized Mcl-1, residues 172-327, was expressed using the same vector (pSV282) and cell line (BL21 pLysS) as the unmodified protein. Cultures were grown in autoclaved M9 media supplemented with sterile 100 mM MgSO₄, 66 μM CaCl₂, 4.2 μg/L FeSO₄, 0.4% glucose, 0.00005% thiamine, and 34 μg/mL kanamycin (for selection). Media prewarmed to 37°C was inoculated with soft pellets of 10 mL overnight LB (Luria Broth) cultures resuspended in warm prepared media. When the culture reached OD600 of 0.5, amino acid supplements were added: 100 mg of lysine, phenylalanine, threonine, and 50 mg of isoleucine, leucine, valine, and selenomethionine for 1 L. Thirty minutes after addition of the amino acids, protein expression was induced with 0.2 mM IPTG. Cultures were transferred to 30°C and incubated overnight, for ∼16 hr. Protein was purified as before, except for the addition of 10 mM β-mercaptoethanol in the first pass Ni-affinity purification buffers, and after the second Ni-affinity purification, DTT concentration was maintained at 5 mM. The selenomethionine expression protocol was modified from Van Duyne et al.35

TEV protease was expressed using plasmid pRK793 obtained from the Macromolecular Crystallography Laboratory at the National Cancer Institute in a BL21(DE3)pRP strain. Cultures were grown to OD600 of ∼0.5 at 37°C, and induced with 1 mM IPTG for 4 hr at 30°C. TEV was purified from cell lysate by Ni-affinity chromatography followed by size-exclusion chromatography on an S200 column.

BH3 peptides

Human Bim BH3 (NH₃-ggsgRPEIWIAQELRRIGDEFNAYYARRV-CONH₂) peptide was synthesized by SynPep of Dublin, CA. Lower case letters indicate linker sequence not part of the native BH3 sequence. Bim mutants I2dA and F4aE (CH₃CONH-RPEIWAAQELRRIGDEFNAYYAR-CONH₂ and CH₃ CONH-RPEIWAIQELRRIGDEENAYYR-CONH₂) were synthesized by the MIT Biopolymers Laboratory in the Koch Institute for Integrative Cancer Research and purified by reverse-phase high-performance liquid chromatography (HPLC) using a C18 column and a linear water/acetonitrile gradient in the presence of 0.1% trifluoroacetic acid (TFA).

Additional BH3 peptides were encoded in a pSV282 vector as an MBP fusion with a C-terminal FLAG tag. The peptides were expressed in BL21 and purified by Ni-affinity chromatography under native conditions. After cleavage by TEV protease, the peptides were passed over a Ni column, and the flow-through was further purified by reverse-phase HPLC. Wild-type Bim, Bim I3dF, Bim E3gK (gsggRPEIWIAQELRRIGDEFNAYYARRVFLNNYQg grdykddddk), wild-type Noxa, C2dY Noxa (gsgPAELEVECATQLRRFGDKLNFRQKLLNLISKLggrdykddddk), and wild-type Bad (gsggPNLWAAQRYGRELRRMSDEFVDSFKKGLPggrdykddddk) were all verified by mass spectrometry (mutation sites at underlined positions). Mutant Bim I2dY used for crystallography was shorter, due to degradation during native purification, as determined by mass spectrometry (gsggRPEIWYAQELRRIGDEFNAYYAR).

Fluoresceinated peptides were synthesized and purified by CHI Scientific (Maynard, MA). Bim (FITC-RPEIWIAQELRRIGDEFNAYYAR-CONH₂) and mutants I2dA, I2dY, and F4aE (mutations at underlined positions) were purchased at greater than 95% purity. Additionally, a purified fluoresceinated Bad peptide (CH₃CONH-NLWAARYGRELRRMSDKFVD-COOH), in which FITC was attached to the italicized K, was purchased from Calbiochem (San Diego, CA).

Fluorescence polarization binding assay

Direct binding assays were performed by serial dilution of Bcl-x_L or Mcl-1 in a 50 nM solution of fluoresceinated peptide. For the competition binding assay, a solution of 50 nM fluoresceinated Bim peptide and 50 nM Mcl-1 or Bcl-x_L was incubated with varying concentrations of unlabeled competitor peptide. The buffer used for both binding assays was 20 mM NaPO₄, 50 mM NaCl, 1mM EDTA, 5% DMSO, and 0.001% Triton X-100 at pH 7.5. The assays were performed in black, round-bottom, nonbinding, 96-well plates and incubated for at least 1 hr at ∼25°C before measuring on a SpectraMax M5 plate reader (Molecular Devices, Sunnyvale, CA). The K_ds or IC₅₀s of the interactions were fit in KaleidaGraph (Synergy Software, Reading, PA).

X-ray crystallography

Crystals of the wild-type Mcl-1–Bim BH3 complex were grown in hanging drops over a reservoir containing 0.2M zinc acetate, 0.1M imidazole pH 7.0, 17.5% polyethylene glycol (PEG) 3350 at ∼25°C. Crystals were flash frozen in liquid nitrogen directly from the mother liquor. Diffraction data for the wild-type complexes were collected at the Advanced Photon Source at the Argonne National Laboratory, NE-CAT beamline 24ID-C. Three selenium sites were identified using SHELX with a 2.03 Å dataset collected at the selenium absorption peak.36 SHARP was used to refine the selenium sites, and density modification was applied to produce an initial map.37 The model was built using COOT.38 The model from the selenomethionine peak diffraction data was used to phase the diffraction data collected from the native protein at 2.00 Å. The final structure, refined using REFMAC5 with TLS refinement,39 has an R_work of 19.4% and R_free of 23.0% and has been deposited in the PDB with the ID 2PQK (Table IV).

The I2dY mutant complex was crystallized by the hanging drop method at ∼25°C in 0.1M Tris pH 7.5 and 45% MPD. Crystals were flash frozen in liquid nitrogen directly from the mother liquor. Diffraction data for the P2₁2₁2₁ crystal were collected at the Advanced Photon Source, beamline 24ID-C, and scaled to 1.7 Å. Data were scaled with HKL200041 and phased by molecular replacement using PHASER,42 with chain A of structure 2PQK (Mcl-1 chain) as a model. This gave a single solution in which the density of the Bim mutant was clearly visible, though Bim was not present in the phasing model. Refinement of the I2dY mutant complex used REFMAC539 and PHENIX.43 COOT was used for model building. The structure has an R_work of 18.5% and R_free of 22.4% and has been deposited in the PDB with the ID 3KJ0 (Table IV). Both I2dA and F4aE cocrystals were grown in 0.2M zinc acetate and 0.1M imidazole, pH 7.0, at ∼25°C. Crystallization conditions for I2dA also contained 16% PEG 400, whereas F4aE had 2% PEG 3350. I2dA crystals were flash frozen with liquid nitrogen directly from the mother liquor, whereas F4aE crystals were first transferred into in a solution of the crystallizing liquor mixed with PEG 400 at a final concentration of 20% before freezing. Diffraction data were collected on a Rigaku MicroMax007-HF rotating anode source and an RAXIS-IV detector. The 1.95 and 2.35 Å data sets—I2dA and F4aE, respectively—were scaled with HKL2000.40 Data were phased using the Mcl-1 chain of the wild-type complex. Iterative rounds of refinement and model building were performed using PHENIX43 and COOT,38 respectively. The I2dA structure complex has an R_work of 18.7% and R_free of 21.3% and the F4aE structure has an R_work of 20.8% and R_free of 24.7% (Table IV). These have been submitted to the PDB with IDs 3KJ1 and 3KL2, respectively.

Structure comparisons

Mcl-1 structure cores were aligned to the core of 2PQK using the fitting.py script from Dr. Robert L. Campbell's PyMol script repository at Queen's University, Ontario, Canada (http://pldserver1.biochem. queensu.ca/∼rlc/work/pymol/). For data reported in Table II, the core was defined as Mcl-1 helices 1, 2, and 5–7 (chain A: 172–221, 260–281, 287–319); the α5-α6 core was defined as chain A: 261–281, 287–302. Residue numbering based on 2PQK. Pairwise RMSDs were calculated from those alignments for α3 (chain A: 222–234), α4 (chain A: 243–259), α3-loop-α4 (222–259), the Mcl-1 protein (chain A: 172–191, 204–319), and BH3 binding peptide (chain B: 4-21) using the rms_current.py script. The same protocol was applied to Bcl-x_L structures with 3FDL as the reference. This human Bcl-x_L–Bim complex has a nonphysiologic domain swap of the Bcl-x_L α1 helix.28 The model used for reference was built with the α1 in the standard Bcl-2-fold position from a crystallographically related Bcl-x_L. The Bcl-x_L core for Table III was defined as residues 4–19, 83–98, and 136–194. RMSDs for α3 (101–113), α3-loop-α4 (101–134), Bcl-x_L (4–19, 83–194), and BH3 peptide (88-105) were calculated. Residue numbers of reference structures shown in parentheses were matched to equivalent residues in the comparison structures. In cases when the peptide aligned to Bim was shorter than indicated earlier, RMSDs were calculated over the maximal region that could be aligned by sequence within the indicated section. Alternative conformations were removed for these comparisons.

Acknowledgements

We thank the staff at NE-CAT (beamline 24-ID) at APS, Argonne National Laboratories, for aid in collecting and processing X-ray diffraction data for the wild-type and I2dY mutant structures. We would also like to thank Sarah Mouradian and Jason Tan for help in generating reagents.