Structural insights into BCL2 pro-survival protein interactions with the key autophagy regulator BECN1 following phosphorylation by STK4/MST1

ABSTRACT

BECN1/Beclin 1 is a critical protein in the initiation of autophagosome formation. Recent studies have shown that phosphorylation of BECN1 by STK4/MST1 at threonine 108 (T108) within its BH3 domain blocks macroautophagy/autophagy by increasing BECN1 affinity for its negative regulators, the anti-apoptotic proteins BCL2/Bcl-2 and BCL2L1/Bcl-x_L. It was proposed that this increased binding is due to formation of an electrostatic interaction with a conserved histidine residue on the anti-apoptotic molecules. Here, we performed biophysical studies which demonstrated that a peptide corresponding to the BECN1 BH3 domain in which T108 is phosphorylated (p-T108) does show increased affinity for anti-apoptotic proteins that is significant, though only minor (<2-fold). We also determined X-ray crystal structures of BCL2 and BCL2L1 with T108-modified BECN1 BH3 peptides, but only showed evidence of an interaction between the BH3 peptide and the conserved histidine residue when the histidine flexibility was restrained due to crystal contacts. These data, together with molecular dynamics studies, indicate that the histidine is highly flexible, even when complexed with BECN1 BH3. Binding studies also showed that detergent can increase the affinity of the interaction. Although this increase was similar for both the phosphorylated and non-phosphorylated peptides, it suggests factors such as membranes could impact on the interaction between BECN1 and BCL2 proteins, and therefore, on the regulation of autophagy. Hence, we propose that phosphorylation of BECN1 by STK4/MST1 can increase the affinity of the interaction between BECN1 and anti-apoptotic proteins and this interaction can be stabilized by local environmental factors.

Abbreviations: asu: asymmetric unit; BH3: BCL2/Bcl-2 homology 3; DAPK: death associated protein kinase; MD: molecular dynamics; MST: microscale thermophoresis; NMR: nuclear magnetic resonance; PDB: protein data bank; p-T: phosphothreonine; SPR: surface plasmon resonance; STK4/MST1: serine/threonine kinase 4

Introduction

Macroautophagy (hereafter referred to as autophagy) is an evolutionarily conserved catabolic process in which intracellular vesicles engulf cytoplasmic proteins and organelles and target them to the lysosome to be removed and recycled. The protein encoded by the BECN1 gene initiates autophagosome formation in response to cellular stresses, such as starvation, by interacting with the class III phosphatidylinositol 3-kinase PIK3C3/Vps34 in a complex containing other proteins such as PIK3R4/Vps15 and ATG14/Barkor. BECN1 possesses a BCL2 homology 3 (BH3) domain [1–7] that mediates interactions with anti-apoptotic members of the BCL2 family of proteins. Accordingly, it has been shown that in addition to inhibiting apoptosis, anti-apoptotic proteins can also negatively regulate autophagy by engaging the BECN1 BH3 domain [1,8], preventing its interaction with PIK3C3/Vps34 and blocking autophagosome formation. Recently, the importance of BCL2 proteins in regulating autophagy was brought into question [9,10], though a subsequent study has disputed that conclusion [11]. Hence, this area of BECN1 regulation remains controversial.

Cardiomyocytes are critically dependent on autophagy under both basal conditions and following myocardial stress [12–16]. BECN1 plays a central role in cardiomyocyte survival where its activity is regulated by the serine threonine kinase STK4/MST1 (serine/threonine kinase 4) [17]. A recent report showed that STK4/MST1 can phosphorylate BECN1 within its BH3 domain at threonine 108 (T108), which apparently enhances its interaction with anti-apoptotic proteins BCL2 and BCL2L1 [17]. This conclusion was based on immunoprecipitation studies, though no affinity measurements were reported. As a result, BECN1-PIK3C3/Vps34 activity is reduced in cardiomyocytes where STK4/MST1 is highly expressed, and autophagy suppressed. At a structural level, it was postulated that the interaction of BECN1 with BCL2 proteins formed by the phosphorylated T108 (p-T108) is mediated by an electrostatic interaction with a conserved histidine residue found on both BCL2 and BCL2L1, thereby increasing the stability of the complex [17]. Here, we report the biophysical and structural consequences of BECN1 T108 phosphorylation with respect to interaction with BCL2 anti-apoptotic proteins. Our results suggest that T108 phosphorylation does slightly, but significantly, impact on the affinity of the interaction with BCL2 proteins, though the analysis of multiple BECN1:BCL2 pro-survival protein structures also suggests that other extraneous factors may contribute to how this post-translational modification influences the binding between these proteins, and hence, autophagy regulation.

Results

Phosphorylation of T108 of BECN1 BH3 results in a small but significant increase in binding affinity for BCL2L1 and BCL2

The previous report showing that phosphorylation of T108 enhances the interaction of BECN1 with BCL2 and BCL2L1 was based on co-immunoprecipitation studies [17]. To gain insight into the extent of the affinity increase afforded by this modification, we initially performed surface plasmon resonance-based solution competition binding studies using long (34mer) synthetic peptides corresponding to the BECN1 BH3 domain with T108 replaced with a phosphothreonine (p-T) residue. The wild-type BECN1 BH3 yielded IC₅₀ values consistent with what we and others have reported previously for binding to BCL2L1 [3,4,18–20], while the p-T108 peptide bound slightly, though significantly, tighter (2.0 μM versus 1.2 μM, p < 0.0001) (Table 1, Figure S1(a,b)). The results for BCL2 were very similar with the phosphorylation of T108 resulting in an almost identical relative increase in binding affinity (3.2 μM versus 1.9 μM, p = 0.0033) (Table 1, Supplementary Figure 1(c,d)). As a control, we also tested a BECN1 BH3 domain peptide in which T119 was substituted for a phosphothreonine to mimic a phosphorylation event mediated by DAPK (death associated protein kinase) that has been reported to negatively affect BCL2 protein binding [21,22]. This peptide showed no measurable binding to BCL2L1 at the highest concentration tested (20 μM) (Table 1). As an additional control to confirm that our pro-survival proteins were fully active, we showed that they bound a BCL2L11/Bim BH3 domain peptide with very high (low nM) affinity (Table 1), similar to that previously reported [23]. Finally, we also performed a direct binding analysis of the interaction between the BECN1 BH3 peptides and BCL2L1 using microscale thermophoresis (MST) to confirm that the competition assay accurately reflected the strength of the binding interactions. Here, the K_D values were essentially identical to those obtained in the competition assay, with the difference between them very close to significant (2.1 μM versus 1.4 μM, p = 0.055) (Table 1, Supplementary Figure 1(e,f)). Hence, we can conclude that the phosphorylation of T108 on BECN1 has a significant, though relatively minor, effect on the interaction with BCL2L1 and BCL2, resulting in an ~1.7-fold increase in affinity for both proteins.

Table 1.

Binding affinities of BCL2L1 and BCL2 for BH3 peptides as determined by SPR (using a Biacore instrument) and MST. ND: not determined.

	Biacore (IC50 µM)		MST (KD µM)
	BCL2	BCL2L1	BCL2L1
BECN1 BH3	3.2 ± 0.5	2.0 ± 0.4	2.1 ± 0.4
BECN1 BH3 p-T108	1.9 ± 0.1	1.2 ± 0.08	1.4 ± 0.3
BECN1 BH3 p-T119	ND	> 20	ND
BCL2L11 BH3	0.0010 ± 0.0001	0.0030 ± 0.0003	ND

Structural analysis of the wild-type BCL2L1:BECN1 BH3 interaction

While the binding studies indicated that BECN1 phosphorylation at T108 increases the affinity of BECN1 BH3 for anti-apoptotic proteins, consistent with the findings reported in Maejima et al. [17], this effect was relatively small. In that paper, the authors used the previously reported crystal structure of BCL2L1 bound to a BECN1 BH3 synthetic peptide (Protein Data Bank [PDB] code 2P1L) [4] to explain how T108 phosphorylation influences binding to anti-apoptotic proteins. They conjectured that due to the apparent proximity of T108 of BECN1 to histidine 113 (H113) on BCL2L1, phosphorylation of T108 would result in formation of a new electrostatic interaction with H113 that increases the affinity of BECN1 for BCL2L1. To obtain the BCL2L1:BECN1 BH3 crystal structure, a construct in which the unstructured loop between helices α1-α2 on BCL2L1 is deleted was used [4]. This modification results in BCL2L1 dimerization through an α1 helix ‘domain swap’. In the crystals used for structural determination, there were two copies of this dimer in the asymmetric unit (asu), thereby providing four ‘views’ of how BCL2L1 engages the BECN1 BH3 domain. However, our close analysis of that structural model and associated electron density of 2P1L revealed that there was density associated with H113 in only two of the four BCL2L1 molecules (Chains A and C, 2F_o-F_c maps contoured at 1 sigma). In both of these chains where the conformation of side-chains was obvious, the BCL2L1 H113 side-chain (Nϵ2) is 6.4–6.6 Å from the T108 side-chain (Oγ1) (Figure 1(a)). Although, this distance is significantly larger than a typical electrostatic interaction, it does not take into account the size and charge contributed by the phosphate group on T108, hence, it is feasible such an interaction could occur.

Figure 1.

Histidine 113 on BCL2L1 adopts multiple conformations. (a) In the published crystal structure of BCL2L1 and BECN1 BH3 domain (PDB 2P1L), where electron density is observed for H113 on BCL2L1 (white), it is >6 Å from T108 on BECN1 BH3 (blue). Bottom panel shows an enlarged view of this interaction. (b) In the 20 lowest energy NMR structures of BCL2L1 and BECN1 BH3 domain (PDB 2PON), H113 adopts multiple conformations ranging from 3.1–8.9 Å from T108 (average 6.0 Å). Bottom panel shows an enlarged view of this interaction.

As the structure of BCL2L1 bound to BECN1 BH3 has also been determined by nuclear magnetic resonance (NMR) (PDB code 2PON) [5], we were able to gain further insights into the structural dynamics within the complex, particularly in the region close to T108 on BECN1. No inter-residue NOE interactions involving H113 were assigned, suggesting this residue makes few interactions with the remainder of the protein and is likely solvent exposed. Alternatively, all protons on the histidine side-chain could not be characterized due to spectral overlap. From the coordinates of the 20 lowest energy models, we can infer the flexibility of this region by measuring the minimal distance between the H113 Nδ1 and Nϵ2 atoms on BCL2L1 and the BECN1 BH3 T108 Oγ1 atom (Figure 1(b)). This distance varied significantly from 3.1 Å to 8.9 Å (mean 6.04 ± 1.67 Å) with most of the variation due to the substantial movement of H113. As such, these data demonstrate that H113 can approach T108 close enough to engage in an electrostatic interaction.

Structural analysis of the BECN1 BH3 p-T108 and T108D:BCL2L1 interactions

To more conclusively establish the effect of BECN1 BH3 T108 phosphorylation on its interaction with pro-survival BCL2 proteins, we next solved crystal structures of BCL2L1 with BECN1 BH3 peptides in which T108 was either phosphorylated (p-T108) or replaced by the phosphomimetic residue, aspartic acid (T108D). Here, we used the domain-swapped BCL2L1 described previously for co-crystallization with wild-type BECN1 BH3 [4] and other BH3 domain peptides [4,24–26]. The crystals for both modified BECN1 BH3 peptide complexes grew in different conditions to the wild-type peptide complex and crystallized in a different space-group (Table 2). Overall, the structures for all complexes were almost identical, with the mutations to the BECN1 BH3 domain having no major effect on the overall binding mode of the peptide (Figure 2(a)).

Table 2.

Crystallographic statistics. Highest-resolution shell data are shown in parentheses.

	BCL2L1: BECN1 p-T108 PDB: 6DCN	BCL2L1: BECN1 T108D PDB: 6DCO	BCL2: BECN1 PDB: 5VAU	BCL2: BECN1 p-T108 PDB: 5VAX	BCL2: BECN1 T108D PDB: 5VAY
DATA COLLECTION
Wavelength	0.9537	0.9537	0.9537	0.9537	0.9537
Resolution range	48.93–2.444 (2.532–2.444)	48.92–2.198 (2.277–2.198)	34.32–1.754 (1.817–1.754)	48.78–2.0 (2.072–2.0)	44.44–1.804 (1.869–1.804)
Space group	P 4 21 2	P 4 21 2	P 1 21 1	P 21 21 21	P 1 21 1
Unit cell	109.42 109.42 97.13 90 90 90	109.39 109.39 96.96 90 90 90	85.50 53.34 91.66 90 107.838 90	66.47 87.17 126.63 90 90 90	85.21 53.19 91.41 90 108.393 90
Total reflections	241436 (22616)	430807 (33941)	296602 (28787)	365599 (34845)	262126 (25645)
Unique reflections	22390 (2143)	30476 (2914)	78943 (7775)	50378 (4930)	71187 (6963)
Multiplicity	10.8 (10.5)	14.1 (11.6)	3.8 (3.7)	7.3 (7.1)	3.7 (3.7)
Completeness (%)	99.73 (98.12)	99.56 (96.07)	99.75 (99.36)	99.72 (98.66)	98.95 (97.44)
Mean I/sigma(I)	11.13 (1.14)	20.63 (1.09)	17.17 (1.92)	12.87 (1.0)	14.41 (1.97)
R-merge	0.1632 (1.679)	0.09929 (2.122)	0.06028 (0.746)	0.1169 (2.112)	0.06172 (0.6618)
CC1/2	0.998 (0.533)	0.999 (0.483)	0.999 (0.654)	0.999 (0.522)	0.998 (0.705)
REFINEMENT
Reflections used	23378 (2142)	30439 (2883)	78932 (7774)	50317 (4916)	71180 (6962)
R-work	0.1886 (0.2849)	0.2028 (0.3131)	0.1873 (0.2648)	0.2091 (0.4176)	0.1850 (0.2714)
R-free	0.2286 (0.3587)	0.2311 (0.3466)	0.2146 (0.2975)	0.2403 (0.4656)	0.2268 (0.3084)
Number of atoms	2664	2648	5915	5408	5762
- macromolecules	2620	2602	5596	5332	5506
- solvent	44	43	319	76	256
Protein residues	332	333	645	651	654
RMS (bonds)	0.008	0.007	0.007	0.007	0.007
RMS (angles)	0.80	0.80	0.98	0.84	0.98
Average B-factor	58.04	59.30	29.55	45.71	33.43
- macromolecules	58.15	59.37	29.30	45.72	33.21
- solvent	51.26	55.21	33.95	45.42	38.09

Figure 2.

Crystal structures of BCL2L1 in complex with BECN1 BH3 T108 mutant peptides. (a) Overlay of the crystal structures of BCL2L1:wild-type BECN1 BH3 domain (white, blue), BCL2L1:BECN1 BH3 p-T108 (light orange, dark orange) and BCL2L1:BECN1 BH3 T108D (cyan, dark green). (b) Crystal structure of BCL2L1 (light orange, ribbon) bound to BECN1 BH3 p-T108 peptide (dark orange, tube). No electron density was visible for imidazole group of H113 though electron density was apparent for adjacent amino acids and for p-T108 on the BECN1 BH3 peptide. (c) Similarly, in the crystal structure of BCL2L1 (cyan, ribbon) bound to BECN1 BH3 T108D peptide (dark green, tube), no electron density was visible for the imidazole group of H113 though electron density was apparent for adjacent amino acids. Only weak partial density was observed for the D108 phosphomimetic substitution on the BECN1 BH3 domain peptide. Electron density in B and C is represented as blue mesh. (d) Summary of molecular dynamics simulation plotting distance between H113 and p-T108 over time and representative views of the conformations p-T108 and H113 at different time points. The residues only rarely approach each other close enough to form an electrostatic interaction.

In the BECN1 BH3 p-T108 complex (PDB code 6DCN, Table 2) there was just a single copy of the domain-swapped dimer. Interestingly, while there was obvious electron density associated with the p-T108 residue on the peptide ligand, no density was observed for the BCL2L1 H113 side-chain beyond the Cβ (Figure 2(b), Supplementary Figure 2(a)) although density could be discerned for the side-chains of adjacent residues (i.e. Q111, L112, L114, T115) (Figure 2(b)). The lack of electron density about the imizadole of H113 indicates that it is mobile in the complex and not engaged in any significant interaction with p-T108.

As Maejima et al. used a BECN1 T108D phosphomimetic substitution in a number of their experiments that provided the same outcome as when T108 was phosphorylated (i.e. enhanced the interaction with anti-apoptotic BCL2 proteins) [17], we also determined the crystal structure of BCL2L1 bound to a BECN1 BH3 T108D peptide (PDB code 6DCO, Table 2). This complex crystallized in the same space-group as the p-T108 peptide complex, with one copy of the domain swapped dimer in the asu. As with the p-T108 structure, no electron density was visible for the H113 side-chain beyond the Cβ atom in either molecule in the asu (Figure 2(c), Figure S2(b)). In addition, only very weak partial electron density was visible for the D108 side-chain on the BECN1 BH3 peptide (Figure 2(c)). Hence, as with the BCL2L1:BECN1 p-T108 BH3 peptide complex, there was no readily detectable interaction between H113 of BCL2L1 and the BECN1 D108 phosphomimetic side-chain.

In summary, these crystal structures suggest that phosphorylation of T108 in BECN1 (or substitution with the phosphomimetic, aspartate) has no discernible impact on the ability of the BECN1 BH3 to form an electrostatic interaction with H113 on BCL2L1. However, it is possible that an interaction that only occurs with low frequency, consistent with the small increase in binding affinity we observed, might not be detected in a crystal structure. Indeed, molecular dynamics simulations on the BCL2L1:p-T108 structure performed over 200 ns indicate that such an interaction can occur, though rarely (Figure 2(d)).

The wild-type BECN1 BH3:BCL2 structure

To further investigate the structural consequences of the phosphorylation of BECN1 T108 and pro-survival proteins, we next examined the complex between BCL2 and BECN1 BH3. To facilitate crystallization, we used a previously reported BCL2 construct in which the large unstructured loop between α-helices α1 and α2 was replaced with the (shorter) corresponding region in BCL2L1 [27,28]. This loop does not participate in any contacts with bound BH3 ligands, and should not influence any interpretation of how BCL2 engages BECN1.

We first solved a structure of BCL2 bound to wild-type BECN1 BH3 (PDB code 5VAU, Table 2). This structure has four molecules in the asu, affording four views of the interaction (Figure 3(a)). By comparison with an NMR structure of apo-BCL2 [27], it is apparent that binding of BECN1 BH3 causes a conformational change in the protein involving widening of the canonical BH3 peptide binding groove through movement of the α3 helix, and some reorganization of the α3-α4 loop (Figure S3(a)). This is in contrast to the BCL2L1:BECN1 BH3 complex where significant movement of both α3 and α4 helices were observed to accommodate the ligand [4] (Figure S3(b)). Using LIGPLOT [29], we were able to create an ‘interaction map’ of all hydrophobic and polar interactions between BECN1 BH3 residues within 4Å of BCL2 (Figure S4(a)). Given the very similar affinities of the interactions between BECN1 BH3 with both BCL2 and BCL2L1, it is probably not surprising that the number of contacts and types of interactions (mostly hydrophobic) were conserved (Figure S4(a,b)), and that the buried surface area for both complexes is also very similar (1962 Ų versus 1820 Ų). Contacts include the characteristic bidentate electrostatic interaction between BECN1 D121, an aspartate residue conserved on all BH3 domains, and a similarly conserved arginine found in the BH1 domain of all pro-survival proteins (R146 in BCL2 and R139 in BCL2L1). The conserved hydrophobic residues located at positions generally designated h1, h2 and h4 on the BECN1 BH3 are also buried within mostly hydrophobic pockets along the BCL2 binding groove. For example, the absolutely conserved leucine at residue at h2 (L116) contacts M115, V133, F153, though the h4 residue F123 also makes hydrophobic contacts with the aliphatic portion of the BCL2 R137 side chain.

Figure 3.

His 120 on BCL2 only interacts with BECN1 p-T108 or T108 phosphomimetic when its mobility is constrained. (a) Overlay of the four BCL2 (white):wild-type BECN1 BH3 (blue) complexes in the crystallographic asu. His120 adopts multiple conformations while electron density or partial density for T108 is only apparent in two of the four complexes. (b) Close-up view of the structure in (a) highlighting different conformations of H120 in the four BCL2:wild-type BECN1 BH3 complexes in the crystallographic asu. (c) In some BCL2 (light orange, ribbon) with BECN1 BH3 p-T108 (orange tube) complexes within the crystallographic asu, H120 can interact with p-T108. p-T108 also makes an electrostatic interaction with R129 on BCL2. (d) In the crystal structure of BCL2 (light orange, ribbon) in complex with BECN1 BH3 p-T108 peptide (dark orange, tube), strong electron density (blue mesh) is apparent for both H120 and p-T108. However, there are also close crystal contacts near H120 with a BCL2 molecule (green) in an adjacent molecule in the crystal. (e) Overlay of the four BCL2 (white): BECN1 BH3 D108 (blue) complexes in the crystallographic asu. Only in one complex does H120 interact with D108. In the others, the side-chains are too far apart or no electron density was apparent for one or both side chains.

Only one other crystal structure of BCL2 in complex with a BH3 peptide, that of BAX, has been determined [30]. This interaction is approximately 100-fold higher in affinity than with BECN1 BH3 (in assays using similar-length BH3 peptides) [30], which is consistent with the 12% larger buried surface area (2244 Ų) of that complex. It is also noteworthy that BECN1 BH3 contains a threonine residue (T119) at the h3 position which is normally occupied by a larger, hydrophobic residue in all pro-apoptotic BH3 domains. In BAX BH3, the isoleucine (I66) h3 residue makes contacts with multiple bulky hydrophobic residues (F104, Y108, F112) compared to T119 in BECN1 BH3 that makes only a hydrophobic contact with F104 via its Cβ atom) (Figure S4(c)). Together with the somewhat larger number of contacts for the BAX BH3 complex, this likely underlies the higher affinity of that interaction. Consistent with this hypothesis, smaller hydrophobic residues (e.g. alanine) at the h3 position of pro-survival BCL2 protein BH3 domains contributes to their weaker affinity for pro-survival proteins [25] and, moreover, a T119I mutation in BECN1 BH3 enhanced binding to BCL2L1 in immunoprecipitation experiments [5].

Focusing on the specific interaction required for STK4/MST1 inhibition of autophagy, electron density was visible for H120 of BCL2 (analogous to H113 in BCL2L1) in each BCL2:BECN1 BH3 complex in the asu, however, it appears BECN1 BH3 is highly unstructured up to residue T108. Only in one of the four complexes (Chain F) is the entire T108 side-chain visible, while in two chains (Chains G and H) there is electron density only up to the Cβ atom, and in chain E electron density is only apparent beyond the next residue (M109). Nevertheless, we used this information to estimate that the minimum distance between the side-chains of BECN1 T108 to BCL2 H120 ranges from 4.0 Å to over 14 Å in one complex (chains B and F) in which the α3 helix of BCL2 adopts a different conformation to the other three molecules in the asu. (Figure 3(b)).

Structural analysis of the BECN1 BH3 p-T108 and T108D:BCL2 interactions

To determine whether phosphorylation of T108 on BECN1 could influence its interaction with BCL2 (especially with H120), we next solved the crystal structure of BCL2 bound to the BECN1 BH3 p-T108 peptide (PDB code 5VAX, Table 2). This structure was informative as it was the first of the BECN1 BH3 complexes we examined in which the histidine residue (H120) approaches close enough to BECN1 BH3 p-T108 to engage in an electrostatic interaction (Figure 3(c)), as proposed by Maejima et al. [17]. Interestingly, of the four molecules present in the asu., this occurred in only two of the chains (chains C and D) where there are also direct crystal contacts with the side-chain of H120 that restrict its mobility (Figure 3(d)). Accordingly, the electron density associated with H120 in those chains is also much stronger than in the other complex structures we have examined. In chains A and B where there are no crystal contacts, the electron density for H120 is weaker (chain A) or not visible (chain B) suggesting these side-chains are mobile. Of note, BECN1 BH3 p-T108 appears to engage in an electrostatic interaction with R129 (3.4 Å separation) (Figure 3(c)) on BCL2 in the four BCL2 molecules in the asu.

Finally, we also solved a structure of BCL2 bound to the BECN1 BH3 T108D phosphomimetic mutant (PDB code 5VAY, Table 2). Again there were four copies of the complex in the asu, though with a different crystal lattice than the BCL2:BECN1 BH3 p-T108 structure. In three of the BCL2 molecules within the crystal, the density for H120 is partial or absent, and in those molecules where electron density for D108 is apparent, it is too distant to engage in any interaction with H120. Only in chain C does H120 hydrogen bond with D108 (Figure 3(e)), though again there are crystal contacts directly stabilizing the histidine side chain.

In summary, our structures of BCL2:BECN1 BH3 complexes suggest that residue T108 only interacts with H120 (either in the phosphorylated state or when mutated to the phosphomimetic aspartate) when crystal contacts are present to restrict H120 side-chain movement.

Nuclear magnetic resonance studies on BECN1 N-terminal domain

To gain further structural insights into the effect of T108 modification, we performed NMR studies on the entire N-terminal domain of BECN1 (i.e. residues 1–150). Phosphate groups associated with threonine residues have sometimes been observed in N-terminal helix ‘cap’ motifs resulting in increased helix stability [31] that could in turn, potentially lead to increased binding affinity. To determine whether T108 phosphorylation influences propensity for BH3 helix formation within BECN1, we compared ¹H/¹⁵N HSQC NMR spectra of the wild-type BECN1 BH3 N terminus (residues 1–150) [19,32] with the T108D phosphomimetic mutant (Figure 4). We have previously reported the entire N-terminal region of BECN1 is intrinsically disordered [19,32], and while we observed minor changes in resonances associated with residues close to D108, there was no evidence of major structural rearrangement due to the mutation of T108 to the phosphomimetic aspartate suggesting that the entire N-terminal domain of BECN1 likely remains unstructured.

Figure 4.

NMR studies on the BECN1 N-terminal domain. Overlay of ¹H-¹⁵N HSQC spectra of the BECN1 N-terminal domain (residues 1–150; black) with the T108D mutant form (red). Only minor shifts were observed for residues proximal to T108 following substitution with aspartate, but no major changes that would suggest a change from an intrinsically disordered state. Numbered residues are those within the BH3 domain.

Effect of detergent on interactions between BECN1 BH3 and pro-survival proteins

In the paper by Maejima et al. [17], the increased interaction between BECN1 and pro-survival proteins following BECN1 T108 phosphorylation was detected using co-immunoprecipitation experiments. These involve extracting the proteins from the cells using detergent-containing buffers. As different detergents have been shown to affect interactions between BCL2 family members [33–35], we next examined whether they also influence binding interactions with BECN1 BH3. For these studies, we performed MST direct binding experiments in the presence and absence of 0.5% (v:v) Triton X-100, as used in the previously reported co-immunoprecipitation experiments. To achieve this, it was also necessary to fluorescently label the pro-survival protein (BCL2L1) as the inherent fluorescence of the detergent interfered with the assay. Using this approach, the K_D values for the interaction between BCL2L1 and the wild-type BECN1 BH3 domain or the p-T108 variant in the absence of Triton X-100 (Table 3) were within 2-fold of those observed in the competition SPR assay or the MST assay using the label-free approach (Table 1, Figures S1(a,b) and S5(a,b)) suggesting the fluorescent label did not have a major impact on the interaction. More importantly, the phosphorylated peptide also bound with the same increased affinity (1.5-fold, p = 0.04) compared to the wild-type peptide, as seen in the other assays. Interestingly, the presence of Triton X-100 resulted in a 5–6-fold higher affinity of both peptides for BCL2L1, though the difference (0.74 μM versus 0.43 μM in its absence) between the non-phosphorylated and phosphorylated peptides only increased marginally (1.8-fold versus 1.5-fold) (Figure S5(c,d)). As detergents act as membrane surrogates, these results suggest that the interaction between BECN1 and BCL2 pro-survival proteins could be influenced by factors such as local membrane environment.

Table 3.

Binding affinities (K_D in μM) of BECN1 BH3 peptides for BCL2L1 in the presence and absence of Triton X-100 as determined by MST.

	- Triton X-100	+ Triton X-100
BECN1 BH3	3.7 ± 0.4	0.7 ± 0.09
BECN1 BH3 p-T108	2.5 ± 0.6	0.43 ± 0.02

Discussion

Phosphorylation of threonine residues within the BH3 domain of BECN1 has pro- and anti-autophagic effects through modulation of interactions with anti-apoptotic BCL2 family proteins. While previously published structural and binding affinity data (together with the results reported here) support the conclusions that T119 phosphorylation by DAPK promotes autophagy by reducing the affinity of BECN1 for BCL2 [21,22], biophysical and structural evidence for the mechanism by which the kinase STK4/MST1 attenuates autophagy have been lacking. Based on previously published immunoprecipitation data [17], it was concluded that T108 phosphorylation increases BECN1 affinity for BCL2, which was postulated to occur by promoting an electrostatic interaction with a proximal conserved histidine residue on the α3-α4 connecting loop in BCL2 and BCL2L1. Although the authors provided evidence that the T108 phosphorylation also affected assembly of the PIK3C3/Vps34 complex (and hence PIK3C3/Vps34 kinase activity), their associated data supported the hypothesis this was directly due to increased association of BECN1with BCL2 pro-survival proteins. This is, to some degree, supported by recent structural data showing that the BECN1 N terminus does not likely contact other components of this complex [36]. In this report we have provided a detailed structural analysis of this proposed mechanism and the interaction between BECN1 and BCL2 pro-survival proteins following T108 phosphorylation.

Unexpectedly, we observed that T108 phosphorylation only had a weak, but statistically significant, effect on the measured BECN1 BH3 peptide binding affinity for either BCL2 or BCL2L1. Structural analysis of this or a T108D phosphomimetic peptide in complex with BCL2 and BCL2L1 provided very little evidence that BECN1 BH3 p-T108 or D108 interacts to a significant degree with the conserved histidine residue proposed by Maejima et al. [17], although there was data demonstrating that such an interaction was possible in certain contexts. Indeed, in most of the new structures with either of the T108-modified peptides there were multiple ‘views’ of the complex in the asu, and in many of these, either the conserved histidine residue, the modified threonine, or both, had only weak electron density associated with them; this is consistent with a lack of significant stable interaction between the residues. However, our crystal structures of BCL2 bound to BECN1 BH3 p-T108/T108D did show evidence of an interaction between these residues when crystal contacts restrict the flexibility associated with the H120 side-chain.

The regulation of autophagy by BCL2 proteins has recently become a topic for debate with new evidence indicating that this might occur indirectly rather than through direct interactions between BECN1 and BCL2/BCL2L1 [9,10,18]. However, the observation that phosphorylation of BECN1, such as mediated by STK4/MST1 (or DAPK [22]), can affect autophagic responses by influencing interactions with pro-survival proteins provides a reasonable counterargument to this. This is an important issue as the survival of cardiomyocytes is dependent on appropriate regulation of autophagy through the action of STK4/MST1. While the increased affinity we observed when BECN1 BH3 T108 is phosphorylated is statistically significant and consistent with the mechanism proposed by Maejima et al. [17], it is relatively minor (< 2-fold). However, it is very difficult to predict how such a change in the strength of a single protein:protein interaction influences a biological response as many factors are involved including the intracellular concentrations of the molecules, their turnover and their localization. Nevertheless, while it is certainly feasible that this relatively small change could be sufficient to affect the autophagic response of the cell, it might be expected that a larger increase in affinity would be required to account for the significant phenotype associated with aberrant STK4/MST1 expression in vivo.

One important point to consider is that our binding studies were performed in aqueous solution, however, there is evidence for BECN1 engaging BCL2L1 and BCL2 on membranes such as those associated with the endoplasmic reticulum [1] or mitochondria [37]. As membranes can influence BCL2 protein structures [38,39], it is possible that in such an environment the flexibility of the critical histidine residue is restricted and adopts a conformation where it is more readily engaged by p-T108 once the BECN1 BH3 enters the ligand-binding groove. Because our structural data demonstrate that direct restriction of the movement of the conserved histidine side-chain (as occurs due to crystal contacts in our BCL2:BECN1 p-T108 structure) promotes the interaction with p-T108, this would likely lead to increased binding affinity between the molecules. This type of structural restraint is required as the histidine is highly mobile and adopts multiple conformations in every structure of BCL2L1 or BCL2 with BECN1 BH3 domain solved to date, including the five new structures reported here and in our molecular dynamics analyses. Extra interactions between p-T108 and R129 on BCL2 were also observed in that complex which could also contribute to tighter binding affinity.

Interestingly, we observed that the presence of Triton X-100 detergent micelles significantly impacts on the affinity of the interaction between BECN1 BH3 and BCL2L1. While there was not a dramatic change in the differential between the affinities of the phosphorylated and non-phosphorylated BH3 peptides in these conditions, it is possible that certain lipids associated with cellular membranes could lead to a more pronounced difference in binding affinity. (Notably, we also found that another detergent, CHAPS, resulted in a ~ 20-fold loss in affinity between BCL2L1 and BECN1 BH3 peptides; Supplementary Figure 5(e,f)). Detergent micelles, including those formed by Triton X-100 are often used as membrane surrogates and have previously been shown to significantly affect the structure and behaviour of BCL2 family proteins [33–35]. Hence, the increased affinity we observed between BECN1 BH3 and BCL2L1 in the presence of Triton X-100 could more accurately reflect the true affinity of this interaction. While this affinity is still somewhat lower that that seen between BH3-only proteins and pro-survival proteins, it is somewhat greater that the low micromolar affinity that has been used as an argument against the physiological relevance of this interaction.

In summary, we have now provided biophysical data supporting the proposal that STK4/MST1-mediated phosphorylation of T108 in BECN1 increases its affinity for BCL2 and BCL2L1, though only by a factor of ~2-fold. While such a change in affinity could be physiologically relevant, crystal structures of these complexes mostly failed to demonstrate significant interaction between p-T108 and the conserved histidine residue on the pro-survival proteins to which it was proposed to bind, except when the mobility of the histidine side-chain was restricted by crystal contacts. The small increase in affinity is consistent with it occurring at low frequency, and therefore would not necessarily be captured in a crystal structure. It may also be the case that binding in solution does not entirely capture intracellular conditions where local extraneous local factors such as membranes could similarly act to restrict the mobility of the histidine, and thereby promote the interaction. Our data showing membrane surrogates such as detergent micelles can influence the affinity of this interaction supports this hypothesis. Hence, it will be of particular interest in future studies to focus on how intracellular components, such as membranes, influence the structure of BCL2 and BCL2L1 in a way that could impact on their interactions with associated ligands such as BECN1.

Materials and methods

Recombinant protein expression and purification

Recombinant BCL2 pro-survival proteins with C-terminal truncations (i.e. BCL2L1ΔC24 or BCL2ΔC22) used for binding studies were expressed and purified as described previously [40]. The human BCL2 construct (BCL2ΔC32ΔLoop) used for crystallization was identical to that reported previously for structural studies [27,30]. This construct has the large unstructured loop between residues 35−91 replaced with residues 29−44 of BCL2L1. The C-terminal 32 residues corresponding to the hydrophobic transmembrane domain were also deleted to improve protein solubility. The human BCL2L1 construct (BCL2L1ΔC24ΔLoop) used for crystallization studies was identical to that used by in previous X-ray structural studies with BECN1 BH3 [4]. This construct has the large unstructured loop between residues 27–82 deleted as well as a 24 residue C-terminal truncation to improve protein solubility.

Synthetic peptides

Peptides were synthesized and purified by reverse-phase HPLC to >90% purity (Mimotopes). The sequences used were:

BECN1 BH3: GEASDGGTMENLSRRLKVTGDLFDIMSGQTDVDH

BECN1 BH3 T108D: GEASDGGDMENLSRRLKVTGDLFDIMSGQTDVDH

BECN1 BH3 p-T108: GEASDGGpTMENLSRRLKVTGDLFDIMSGQTDVDH (where pT represents phosphothreonine).

BECN1 BH3 p-T119: GEASDGGDMENLSRRLKVpTGDLFDIMSGQTDVDH (where pT represents phosphothreonine).

BCL2L11/Bim: DMRPEIWIAQELRRIGDEFNAYYARR

BCL2L11/Bim4E: DMRPEIWEAQEERREGDEENAYYARR

Solution competition assays

Solution competition assays were performed by surface plasmon resonance approaches using a Biacore 3000 instrument as described previously [24]. Briefly, peptides were incubated with recombinant proteins in running buffer (10 mM HEPES, pH 7.2, 150 mM NaCl, 3.4 mM EDTA, 0.005% [v:v] Tween-20 [Sigma, P2287-500ML]) for at least 2 h prior to injection onto a CM5 sensor chip on which either wild-type BCL2L11/Bim peptide or an inert BCL2L11/Bim mutant peptide (BCL2L11/Bim4E) was immobilized. Specific binding of BCL2L1 or BCL2 to the chip surface in the presence and absence of peptide competitors was quantified by subtracting the signal obtained on the BCL2L11 mutant channel from that obtained on the wild-type BCL2L11 channel. IC₅₀ values were calculated by nonlinear curve fitting of the data using GraphPad Prism (GraphPad Software).

Microscale thermophoresis

Initial affinity measurements using MST were carried out employing a Monolith NT.LabelFree instrument (NanoTemper Technologies) using a similar approach as described previously [41,42] where purified recombinant BCL2L1ΔC24 (1 µM) in 20 mM Tris pH 8.0, 150 mM NaCl, was incubated with varying concentrations of BH3 peptide. Affinity measurements with Triton X-100 (Sigma, X100-500ML) were carried out with a Monolith NT.115 instrument (NanoTemper Technologies) as described previously [41,42] with purified recombinant BCL2L1ΔC24 labelled using the NHS RED NanoTemper labeling kit (MO-L001) according to the manufacturer’s instructions. For the assay, labelled protein was mixed with unlabelled peptide in the presence or absence of 0.5% (v:v) Triton X-100. All experiments were incubated for 30 min before applying samples to Monolith NT standard treated capillaries (NanoTemper Technologies). Thermophoresis was measured at 25°C with laser off/on/off times of 5s/30s/5s. Experiments were conducted at 40% LED power and 40% MST infrared laser power. Data from three independently performed experiments were fitted to the single binding model via the NT.Analysis software version 1.5.41 (NanoTemper Technologies) using the signal from Thermophoresis + T-Jump.

Protein complex crystallization

In all cases crystals were obtained by mixing the pro-survival protein with the BH3 peptide at a molar ratio of 1:1.3 and then concentrating the sample to 10 mg/ml. Crystallization trials were performed at the Bio21 Collaborative Crystallization Centre. Crystals were grown by the sitting drop method at room temperature in the following conditions: BCL2:wild-type BECN1 BH3: 0.1 M Bis-tris chloride, pH 5.5, 0.2 M ammonium acetate, 25% (w:v) PEG 3350 (Sigma, 202444-250G); BCL2:BECN1 BH3 p-T108: Bis-tris chloride, pH 4.5, 0.2 M ammonium acetate, 25% (w:v) PEG 3350; BCL2:BECN1 BH3 T108D: 0.1 M Tris, pH 6.5, 0.2 M ammonium acetate, 25% (w:v) PEG 3350; BCL2L1:BECN1 BH3 p-T108: 2M NaCl, 10% (v:v) glycerol; BCL2L1:BECN1 BH3 T108D: 0.1 M Tris, pH 7.0, 0.2 M MgCl₂, 2.5 M NaCl. Prior to cryo-cooling in liquid N₂, crystals were equilibrated into cryoprotectant consisting of reservoir solution containing 15% (v:v) ethylene glycol. Crystals were mounted directly from the drop and plunge-cooled in liquid N₂.

Crystal diffraction data collection and structure determination

Diffraction data were collected at ~100 K at the Australian Synchrotron MX1 and MX2 beamlines (Victoria, Australia) (wavelength for all structures was 0.954 Å). The diffraction data were integrated with XDS [43]. The BCL2L1 complex structures were obtained by molecular replacement with PHASER [44–46] using the previous crystal structure of BCL2L1 from the BECN1 BH3:BCL2L1 complex [4] (PDB entry 2P1L) with the BECN1 peptide removed from the structure, as the search model. For the BCL2 complex structures we used the structure of BCL2 from the BCL2:BAX BH3 complex (PDB entry 2XA0) [30] with the BAX peptide removed as the search model. Multiple rounds of building in COOT [47] and refinement in PHENIX [43] led to the final models.

NMR studies

The ¹⁵N-labelled BECN1 N-terminal domain (residues 1–150) with all cysteines mutated to serine and a tyrosine included at the C-terminal of the construct for quantitation by UV spectroscopy, was identical to the material previously used for backbone chemical shift assignments and conformation studies [19,32]. This protein, and the T108D mutant (both ~180 μM in 50 mM sodium phosphate, pH 6.8 containing 50 mM NaCl and 0.02% [w:v] sodium azide), were expressed and purified as described previously [19]. NMR spectra were acquired on either a Bruker Avance 500MHz, or Avance II 800MHz spectrometer, both equipped with a TXI cryoprobe and a single-axis field gradient (Gz). NMR spectra were processed using TOPSPIN (Version 3.2, Bruker) and analyzed using XEASY (Version 1.3) [48].

Molecular dynamics studies

Molecular dynamics simulations were performed using the GROMACS (v4.5.5) [49] suite of programs and the OPLS-aa force field [50]. Atomic charges for the phosphate group of pT were obtained from CHELPG electrostatic charges calculated for CH₃PO₄^2- calculated at the B3LYP/6-31G(d) level using the GAUSSIAN-09 program (Revision A.02, Gaussian, Inc., Wallingford CT). Briefly, the simulation consisted of an initial steepest decent minimization, a short 50ps positionally-restrained MD holding the protein fixed, and finally unrestrained MD for 20 ns. The system was solvated using the TIP4P water model in a cubic box extending 10 Å beyond all atoms and utilized periodic boundary conditions. The system was made electrically neutral to a final ionic strength of 0.1 M with the addition of sodium and chloride ions; ionizable residues were assumed to be in their charged state at a pH of 7.0. The temperature of the system was maintained by coupling the protein and solvent independently to a velocity rescaling [51] thermostat at 300 K with a time constant of 0.1 ps. Pressure was controlled to 1 bar using a Berendsen barostat [52] with a coupling of 0.5 ps. A cut-off of 10 Å was used to account for non-bonded interactions, applying a neighbour-list update frequency of 10 steps (20 fs). The particle-mesh Ewald method [53] was used to account for long-range electrostatics applying a grid width of 1.2 Å and a fourth-order spline interpolation. All bond lengths were constrained with the LINCS algorithm [54]. Coordinates of the model were archived every 50 ps.

Funding Statement

This work was supported by an NHMRC of Australia Project Grant 1049949, Career Development Fellowship 1024620 and Australian Research Council Future Fellowship FT150100212 all to E.F.L. T.P.S.C. was supported by NHMRC Fellowship 1091976 and N.A.S was supported by an Australian Government Research Training Program Scholarship. Infrastructure support from NHMRC IRIISS grant #361646 and the Victorian State Government OIS grant is gratefully acknowledged. We also acknowledge the La Trobe University-Comprehensive Proteomics Platform for providing infrastructure.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplementary data for this article can be accessed here.