Crystal structures of the phosphorylated BRI1 kinase domain and implications for brassinosteroid signal initiation.

Brassinosteroids, which control plant growth and development, are sensed by the membrane receptor kinase BRASSINOSTEROID INSENSITIVE 1 (BRI1). Brassinosteroid binding to the BRI1 leucine-rich repeat (LRR) domain induces heteromerisation with a SOMATIC EMBRYOGENESIS RECEPTOR KINASE (SERK)-family co-receptor. This process allows the cytoplasmic kinase domains of BRI1 and SERK to interact, trans-phosphorylate and activate each other. Here we report crystal structures of the BRI1 kinase domain in its activated form and in complex with nucleotides. BRI1 has structural features reminiscent of both serine/threonine and tyrosine kinases, providing insight into the evolution of dual-specificity kinases in plants. Phosphorylation of Thr1039, Ser1042 and Ser1044 causes formation of a catalytically competent activation loop. Mapping previously identified serine/threonine and tyrosine phosphorylation sites onto the structure, we analyse their contribution to brassinosteroid signaling. The location of known genetic missense alleles provide detailed insight into the BRI1 kinase mechanism, while our analyses are inconsistent with a previously reported guanylate cyclase activity. We identify a protein interaction surface on the C-terminal lobe of the kinase and demonstrate that the isolated BRI1, SERK2 and SERK3 cytoplasmic segments form homodimers in solution and have a weak tendency to heteromerise. We propose a model in which heterodimerisation of the BRI1 and SERK ectodomains brings their cytoplasmic kinase domains in a catalytically competent arrangement, an interaction that can be modulated by the BRI1 inhibitor protein BKI1.

INTRODUCTION

Plants have evolved a unique set of membrane receptor kinases, which contain an extracellular LRR ligand binding domain, a single membrane spanning helix and a cytoplasmic kinase domain (Shiu and Bleecker, 2001) (LRR-RKs). LRR-RKs sense diverse ligands, ranging from small molecules (Wang ) to peptides (Chinchilla ; Ogawa ) and entire proteins (Yang ; Zipfel ) to, for example, regulate growth (Li and Chory, 1997), development (Clark ; Shpak ) and interactions with the environment (Gómez-Gómez and Boller, 2000; Nishimura ). Many LRR-RKs in Arabidopsis remain orphan receptors, but selected members have been studied extensively. The LRR-RK BRI1 senses brassinosteroids, a class of polyhydroxylated steroid hormones (Fujioka and Yokota, 2003). A complete signaling pathway that connects brassinosteroid sensing by BRI1 to nuclear transcription factors has been uncovered (Zhu ).

BRI1 constantly cycles between the plasma membrane and endosomes (Geldner ). In the absence of steroid hormone, BRI1 is kept in a basal state by several mechanisms, which include auto-inhibition by its C-terminal tail (CT, residues 1161–1196, Figure1a) (Wang ), auto-phosphorylation on Thr872 in the kinase domain (Wang ) and interaction with the BRI1 KINASE INHIBITOR protein BKI1 (Wang and Chory, 2006; Jaillais ). BKI1 contains an N-terminal targeting motif for the plasma membrane and a C-terminal sequence that binds the BRI1 kinase domain, where it inhibits the association of BRI1 with SERKs, a protein family of smaller LRR-RKs essential for BR signal initiation (Li ; Nam and Li, 2002; Karlova ; Gou ).

Figure 1

Overall structure of the active BRI1 kinase domain and features of the nucleotide binding site.(a) Schematic overview of the BRI1 kinase domain constructs used in this study with construct borders included. The JM, KD and CT segments have been previously assigned using a BRI1 homology model (Jaillais ). (b) Ribbon diagram of the BRI1865–1160 kinase domain. The N-lobe (residues 865–956) is shown in light-blue, the hinge region (residues 957–959) in orange, the activation loop (residues 1027–1056) in yellow and the C-lobe (960–1160) in dark-blue, respectively. Four phosphorylation sites present in the structure are highlighted in bonds representation (with phosphorus coloured in cyan). (c) View of the adenine nucleotide binding pocket in BRI1 occupied by the non–hydrolysable ATP analogue AppNHp (gray, in bonds representation). The two Mn2+ ions are highlighted as magenta spheres, residues contacting the nucleotide are shown in yellow (in bonds representation). Hydrogen-bonding interactions of AppNHp with BRI1 are denoted as dotted lines (in black). The gatekeeper Tyr956 in the back pocket of the binding site makes a hydrogen-bond with Glu927, which in turn salt-bridges to Lys911 to keep the kinase domain in its active conformation. The genetic alleles bri1-1 (Ala909–Thr) and bri1-115 (Gly1048–Asp) in close proximity to the nucleotide binding site are highlighted as green spheres. (d) A complex structure with ATP (gray, in bonds representation) identifies the γ-phosphate of the nucleotide facing outwards, away from the catalytic Asp1009. An omit 2Fo–Fc electron density map contoured at 1.5 σ is shown alongside (blue mesh). A similar non-catalytic conformation of the nucleotide has been observed in a SERK3–AppNHp complex (in yellow, bonds representation, Protein Data Bank identifier (PDB-ID): 3uim).(e) The BRI1–ADP complex with ADP in grey (in bonds representation). Hydrogen-bonding interactions of the adenine base and ribose with the BRI1 hinge region (in orange) main chain atoms and with two water molecules (red spheres) are shown as dotted lines (in grey). An omit 2Fo–Fc electron density map contoured at 1.5 σ is shown alongside (blue mesh).

The initial events in brassinosteroid signaling have been described in molecular detail. Brassinolide (BL), a potent brassinosteroid in Arabidopsis, is sensed directly by BRI1 (Li and Chory, 1997; Wang ), specifically by its extracellular LRR domain (Kinoshita ; Hothorn ; She ). BL-binding to the BRI1 LRR domain causes ordering of a approximately 70-residue island domain (Hothorn ; She ), creating a docking platform for the smaller and shape-complementary LRR domain of a SERK co-receptor (Santiago ; Sun ). The hormone itself acts as a ‘molecular glue’, holding the BRI1 and SERK ectodomains together and bringing their C-termini in close proximity (Hothorn ; Sun ).

On the cytoplasmic side of signaling events, it has been established that the kinase domains of BRI1 and SERK3 (BAK1) can physically interact (Li ; Nam and Li, 2002; Oh ), and trans-phosphorylate each other in vitro (Li ) and in vivo, following BL stimulus (Wang ). Full-length BRI1 and SERK3 have been shown to interact in yeast cells (Nam and Li, 2002) and in planta, in a BL-dependent manner (Wang ). Also, BL application induces BRI1-catalysed phosphorylation of the BKI1 membrane targeting motif on Tyr211, thereby releasing BKI1 from BRI1 at the plasma membrane into the cytosol and allowing BRI1 and SERK3 to interact (Jaillais ). Sequential auto- and trans-phosphorylation of the BRI1 and SERK3 kinase domains enable BRI1 to phosphorylate immediate downstream signaling components, such as the BRI1 SUBSTRATE KINASES (BSKs) (Tang ; Sreeramulu ) and the receptor-like cytoplasmic kinase CONSTITUTIVE DIFFERENTIAL GROWTH 1 (CDG1) (Kim ).

It is presently not well understood how the BRI1 kinase domain switches from its basal to its activated state, making it capable of phosphorylating BKI1, SERKs and downstream components. BRI1 was first described as a canonical Ser/Thr kinase (Friedrichsen ; Oh ) and its auto-phosphorylation sites have been mapped (Wang ). Ser/Thr trans-phosphorylation has been demonstrated for brassinosteroid signaling components including SERK3, BSK1 and CDG1 (Tang ; Wang ; Kim ). BRI1 is a dual-specificity kinase as it also harbours significant tyrosine kinase activity and can both autophosphorylate on tyrosine residues within the kinase and juxtamembrane domains (JM; Figure1a) (Oh ) as well as trans-phosphorylate Tyr211 in BKI1 (Jaillais ) and tyrosines in other proteins (Wu ).

In this study we set out to obtain a three-dimensional map of the complex phosphorylation events taking place in BRI1. High resolution crystal structures of an active, phosphorylated BRI1 kinase domain now provide detailed insight into its kinase mechanism. Further, our structures rationalise the effects of previously characterised genetic missense alleles in the BRI1 kinase domain and define a surface area that could mediate interactions with early components in the brassinosteroid signaling pathway.

RESULTS

Overall structure of the BRI1 kinase domain

We systematically screened for crystallisable fragments of the BRI1 cytoplasmic domain (Figure1a) and obtained crystals for a catalytically active mutant (Thr872–Ala, (Wang ) which omits the JM (residues 814–865) (see The BRI1 nucleotide binding site). It is of note that the JM is crucial for full BRI1 kinase activity (Wang ; Oh , 2012a), and thus some aspects of BRI1's catalytic function cannot be fully rationalised using our crystallographic model (see below).

We improved the lattice packing of this crystal form by reductive protein methylation (Shaw ) and collected data extending to 2.48 Å resolution (Table1). We determined the crystal structure of BRI1865–1196 using the molecular replacement method. The solution comprises one BRI1 kinase domain per asymmetric unit. BRI1 folds into the canonical bilobal kinase architecture (Figure1b). Residues 865–1160 are well defined in the electron density map, including the activation loop (Figure1b, see below). The entire C-terminal tail however appears disordered and likely occupies large solvent channels in our hexagonal crystals (Figure S1). We speculate that the BRI1 CT is auto-phosphorylated (Wang ,b) and thus released from its auto-inhibitory conformation (Wang ). Consistently, we obtained well-diffracting crystals of a construct comprising only the BRI1 catalytic domain (BRI1865–1160 Thr872–Ala), in its apo-form and in different nucleotide-bound states (Table1).

Table 1

Crystallographic data collection and refinement statistics

	BRI1865–1196 ADP	BRI1865–1160 apo	BRI1865–1160 Mn2+/AppNHp	BRI1865–1160 ATP	BRI1865–1160 ADP
Data collection
Beam-line	SLS PXII	SLS PXIII	SLS PXIII	SLS PXIII	ESRF ID-29
Space group	P62	P62	P62	P62	P62
Cell dimensions
a = b, c (Å)	116.83, 50.56	116.14, 50.30	115.63, 49.64	115.82, 49.90	116.14, 49.74
Resolution (Å)	19.48–2.48 (2.62–2.48)	19.36–2.43 (2.58–2.43)	44.48–2.70 (2.86–2.70)	44.67–2.3 (2.44–2.3)	44.58–1.98 (2.10–1.98)
Rsym	0.097 (1.05)	0.062 (1.26)	0.08 (1.04)	0.043 (1.23)	0.036 (1.14)
I/σI	22.43 (2.63)	28.56 (1.99)	25.37 (2.27)	36.8 (2.5)	24.28 (1.71)
Completeness (%)	98.6 (92.5)	98.4 (91.5)	99.7 (98.3)	99.8 (98.9)	99.9 (99.6)
Redundancy	20.02 (18.66)	14.01 (13.55)	12.09 (11.98)	14.2 (14.3)	10.2 (10.1)
Refinement
Resolution (Å)	19.48–2.48	19.36–2.43	44.48–2.70	44.67–2.30	44.58–1.98
No. reflections	13 379	13 823	10 091	16 312	25 527
Rwork/Rfree	0.233/0.271	0.225/0.262	0.208/0.246	0.202/0.251	0.203/0.240
No. atoms
Protein	2156	2141	2128	2172	2176
Nucleotide	27	–	31	31	27
Mn2+	–	–	2	–	–
Water	13	8	15	16	23
B-factors
Protein	72.17	73.74	74.09	74.07	71.97
Nucleotide	96.34	–	72.63	99.07	120.92
Mn	–	–	77.58	–	–
Water	55.97	50.72	56.85	63.33	59.02
R.m.s deviations
Bond lengths (Å)	0.0125	0.0130	0.0136	0.0142	0.0191
Bond angles (°)	1.6429	1.6605	1.7956	1.7951	2.0264
PDB-ID	4OA2	4OA6	4OA9	4OAB	4OAC

Highest resolution shells are shown in parentheses.

The BRI1 nucleotide binding site

A 2.7 Å structure with the non-hydrolysable nucleotide analogue AppNHp and in the presence of Mn2+ ions reveals the substrate in the BRI1 nucleotide binding site, sandwiched between the N- and C-lobe (Figure1c). There are no significant conformational changes comparing the BRI1865–1160 apo and AppNHp bound states (root mean square deviation (RMSD) is approximately 0.4 Å comparing 283 corresponding Cα atoms). The ATP analogue adopts an active conformation as previously seen for example in Akt/PKB (Yang ), with two manganese ions bridging the α- and β-phosphate of the nucleotide with Asn1014 and with Asp1027 of the DFG motif (Figure1c). The γ-phosphate contacts the catalytic Asp1009 and the neighbouring Lys1011, the adenine base forms two hydrogen bonds with main chain atoms from the hinge region (Figure1c). In contrast, a complex structure with ATP presents the nucleotide in a catalytically incompetent configuration with the γ-phosphate rotated outwards away from the catalytic Asp1009, very similar as recently observed in crystals of the SERK3 kinase domain (Yan ) (Figure1d). Finally, a complex structure with ADP at 1.98 Å resolution reveals a well ordered adenine base and ribose but a largely flexible diphosphate moiety of the reaction product (Figure1e).

The kinase domain itself adopts an active conformation with a salt-bridge formed between Lys911 and Glu927 (Figure1c). Glu927 in turn establishes a hydrogen bond with the conserved gatekeeper tyrosine, a residue in the centre of the ATP binding site which determines the size of the ‘back pocket’ (Tyr956 in BRI1) (Figures1c and S2). It has been demonstrated previously that interfering with this interaction pattern (Lys911–Glu, Tyr956–Phe) renders the BRI1 kinase catalytically inactive (Oh , 2009, 2012a), as does mutation of Asp1027 from the DFG motif to asparagine (Jaillais ). The strong genetic BRI1 missense allele bri1-1 (Clouse ; Friedrichsen ) closely maps to the nucleotide binding pocket and the corresponding mutation Ala909–Thr likely interferes with adenine nucleotide binding (green sphere in Figure1c, compare with Figure S2). On the opposite site of the nucleotide binding pocket, the strong bri1-115 allele (Gly1048–Asp) may cause displacement of the catalytic Asp1009 (Figure1c), or disrupt substrate binding (Li and Chory, 1997).

BRI1 does not harbour guanylate cyclase activity

Our structure identifies that BRI1 is a canonical kinase with respect to its overall fold, structural motifs and catalytic residues. It has however been previously suggested that BRI1 and other plant receptor kinases contain a functional guanylate cyclase (GC) domain embedded in their kinase cores, enabling them to simultaneously exhibit kinase and guanylate cyclase activities (Kwezi , 2011; Qi ). We mapped the putative GC domain suggested for BRI1 (BRI1 residues 1012–1134, yellow in Figure2a) onto the structure of BRI1865–1160 (Kwezi ). The assignment includes the proposed GC catalytic motif (BRI1 residues 1071–1084, red in Figure2a as well as Asp1038 and 1087 suggested to be involved in magnesium ion coordination (Kwezi ). In the structure we find the envisioned GC domain to comprise large parts of BRI1's C-lobe, with the suggested GC catalytic core buried deep inside the hydrophobic core of the kinase domain and with Asp1038 and 1087 being positioned 35 Å apart from each other. Based on this analysis it is unlikely that a catalytically competent GC domain can form in BRI1. Consistently, we find that BRI1865–1160 efficiently hydrolyses ATP to ADP (Figure2b) and to a lesser extent also GTP to GDP and GMP (Figure2c), but cannot form either cGMP or cAMP in HPLC-based activity assays (Figure2b,c).

Figure 2

BRI1 does not contain a guanylate cyclase domain and has no detectable guanylate cyclase activity. (a) Ribbon diagram of the BRI1 kinase domain (in blue) and with AppNHp (gray, in bonds representation) and two Mn2+ ions (magenta spheres) bound in the active site. The GC domain fragment previously used in guanylate cyclase activity assays (Kwezi ) is shown in yellow, the putative catalytic GC motif in red (residues 1071–1084). Asp1038 and 1087 suggested to be involved in magnesium ion coordination are depicted in bonds representation. (b) High-performance liquid chromatography (HPLC) analysis of nucleotide products in BRI1 activity assays. (Top panel) Elution profile showing the retention times for different adenine nucleotides. (Bottom panel) HPLC analysis after incubating 10 μm BRI1865–1160 with 5 mm ATP for 0 (black line), 10 (blue) and 20 (red) min, respectively. (c) (Top panel) Elution profile showing the retention times for different guanine nucleotides. (Bottom panel) HPLC analysis after 10 μm BRI1865–1160 with 5 mm GTP for 10 min (black line) and over-night (red line), respectively.

Structural basis for BRI1's dual-specificity kinase activity

Next, we performed three-dimensional homology searches with the program DALI (Holm ). We identified BRI1865–1160 to be closely related to the human interleukin receptor-associated kinase 4 (IRAK-4) (Figure3a) (Kuglstatter ), a member of a group of kinases involved in animal host defence signaling (Li ). Similar to BRI1, IRAKs require two metal ions for catalysis (Hekmat-Nejad ) (Figure1c) and are the only members of the human kinome that use a tyrosine as gatekeeper residue (Figure3b) (Wang ). Indeed we find Tyr262 in IRAK-4 to engage in the same interactions as described above for the BRI1 gatekeeper Tyr956 (Figures1c and 3b). Structural comparison with the gatekeeper arrangement in SERK3 (Yan ) suggests, that hydrogen-bond interactions of the hydroxyl group of the invariant gatekeeper tyrosine with the conserved Lys/Glu salt-bridge (residue 911/927 and 317/334 in BRI1 and SERK3, respectively) are hall-marks of activated plant receptor-like kinases (Figures3b and S2).

Figure 3

Structural basis for BRI1 dual-specificity kinase activity. (a) BRI1 and human IRAK-4 are closely related kinases. Structural superposition of the apo BRI1865–1160 (blue Cα trace) and IRAK-4 (PDB-ID 2iob, in orange) kinase domains, sharing 35% sequence identity (RMSD is approximately 1.6 Å comparing 254 corresponding Cα atoms). (b) Arrangement of the gatekeeper tyrosine in human IRAK-4 and plant receptor-like kinases. Shown are Cα traces of the BRI1 (in blue), IRAK-4 (PDB-ID 2oib, orange) and SERK3 (PDB-ID 3uim, gray) kinase domains, with the gatekeeper tyrosine and the conserved Lys/Glu highlighted in bonds representation and interatomic interactions shown alongside (dotted lines). The kinase domains of BRI1 and SERK3 share 40% sequence identity and superimpose with a RMSD of approximately 1.5 Å comparing 264 corresponding Cα atoms. (c) Structural superposition of the BRI1 (blue) and IRAK-4 (orange) activation loops. The phosphorylation sites present in the BRI1 and IRAK-4 (PDB-ID 2oib) structures are shown alongside (in bonds representation). (d) Surface-exposed phosphorylation events in BRI1. Activation loop (in gold) Thr1039 and Ser1042 (in bonds representation) are in hydrogen-bond contact with His1040. The nearby Arg1062 and the phosphorylated Ser1060 are depicted in blue. (e) Structural superposition of the activation loop segments of BRI1 (in blue) and insulin receptor (orange, PDB-ID 1ir3, RMSD is approximately 2.5 Å comparing 253 corresponding Cα atoms). pSer1044 (blue, in bonds representation) in BRI1 binds to a positively charged surface pocket formed by BRI1 arginine residues 922, 1008 and 1032. Very similar interactions are made by insulin receptor Tyr1163 with arginine residues 1131 and 1155. (f) The P + 1 pocket in BRI1 (shown in blue, in bonds representation) perfectly superimposes with the corresponding segment in the classical Ser/Thr protein kinase A (in gold, PDB-ID 1jbp, RMSD is approximately 2.3 Å comparing 231 corresponding Cα atoms).

The BRI1 activation loop adopts a conformation rather similar to that seen in the IRAK-4 structure (Figure3a,c). It is of note that based on the conformation of its activation loop, IRAK-4 has been suggested to be a dual-specificity kinase (Wang ). Since BRI1 auto- and trans-phosphorylates on Ser/Thr and on Tyr residues (Friedrichsen ; Oh , 2009; Wang ; Jaillais ), we analysed the conformation of the BRI1 activation loop in detail: in planta, the activation loop is phosphorylated on Thr1039, Ser1042, Ser1044/Thr1045 and on Thr1049 (Wang ). A subset of these positions (Thr1039, Ser1042 and Ser1044) is found phosphorylated in our structures (Figures3c and S2). Thr1039 and Ser1042 are located at the surface of the kinase domain and possibly stabilise the conformation of the activation loop via interaction with His1040 (Figure3d). This network of interactions is extended by Ser1060 and Arg1062 adjacent to the activation loop. The conserved Ser1060 is phosphorylated in our structures (Figures3d and S2) and may also be phosphorylated in vivo (Wang ). Our analysis suggests that phosphorylation of Thr1039, Ser1042 and Ser1060 may affect the orientation of the activation loop and thus BRI1 kinase activity and its interaction with substrates. Consistently, Thr1039–Ala and Ser1042–Ala mutant plants exhibit intermediate brassinosteroid signaling phenotypes (Wang ).

In contrast, Ser1044 in the BRI1 activation loop displays a strong loss-of-function phenotype (Wang ). We find this residue folded back towards a positively charged pocket formed by Arg1008 from the HRD motif containing the catalytic aspartate, by Arg1032 in the activation loop and by N-lobe Arg922 (Figure3e). None of the arginines however is in direct contact with the phosphate group of Ser1044, while one would expect to find several direct protein–phosphate interactions in canonical Ser/Thr kinases (Wang ). Instead, the Ser1044 phosphate binding pocket in BRI1 is reminiscent of typical tyrosine kinases such as insulin receptor (Hubbard, 1997), as illustrated in Figure3(e).

The P + 1 pocket in BRI1, which critically determines the kinase substrate specificity, is very similar to the one described for protein kinase A, a classical Ser/Thr kinase (Madhusudan ) (Figure3f). Importantly, Thr1049, which forms the core of the P + 1 pocket in BRI1, can be differentially phosphorylated in vivo (Wang ). Thr1049 phosphorylation may thus control BRI1 specificity towards different substrates (Figure3f). The neighbouring Gly1048 is found mutated in bri1-115 (Figure1c). The strong phenotype associated with this mutant highlights the importance of the P + 1 pocket in brassinosteroid receptor function (Li and Chory, 1997).

Taken together, the BRI1 activation loop harbours structural features reminiscent of both Ser/Thr and tyrosine kinases, enabling it to act on both types of substrates (Friedrichsen ; Oh ; Wang ; Oh ; Jaillais ). Because the conformation of the activation loop and core phosphorylation patterns appear conserved among different plant receptor-like kinases (Oh , 2010; Klaus-Heisen ; Yan ), we speculate that most if not all of these enzymes could exhibit dual-specificity (Figures S2 and S3).

A three-dimensional map of BRI1 phosphorylation sites

A complex pattern of Ser/Thr and tyrosine phosphorylation has been identified for BRI1 (Wang ; Oh ). Many of the phosphorylation sites can be mapped onto the BRI1 kinase domain structure (Figure4a), where they occupy three different functional areas in the enzyme. Not surprisingly, most phosphorylation events take place within the activation loop (see above, yellow in Figure4a). Thr872, Thr880 and Ser887 form a second cluster in the BRI1 N-lobe, in a region N-terminal to the first β-strand (magenta in Figure4a). This N-terminal extension is not structurally conserved among different kinase families (Wang ), but is known to contribute to kinase function and in the case of BRI1 is essential for kinase activity (Oh ). In the human protein kinase Nek7, interactions between the corresponding segment and the conserved N-lobe region are essential for catalytic activity (Richards ). Similar interactions as previously seen with Nek7 are present in BRI1 (Figure4b). Phosphorylation of Thr872, Thr880 and S887 could disrupt these interactions, leading to reduced BRI1 activity, as previously demonstrated (Wang ). In line with this, we could only obtain crystals of BRI1 upon mutating Thr872 to alanine, thereby possibly stabilising the interaction of the N-terminal extension with the N-lobe β-sheet. It is of note that the N-terminal extension in BRI1 connects to the JM region (highlighted in blue in Figure4a which is again important for BRI1 activity (Wang ). Several Ser/Thr and tyrosine phosphorylation sites in the JM also negatively regulate brassinosteroid signaling (Wang ; Oh , 2011).

Figure 4

Phosphorylation sites in the BRI1 kinase domain. (a) Ribbon diagram of the BRI1 kinase domain, coloured according to Figure1(b). The known phosphorylation sites are highlighted by red spheres. They are grouped into three major clusters; the N-lobe of the kinase (in light-blue), the activation loop region (in yellow) and the kinase active site (in red). The region N-terminal of the first β-strand in BRI1 (residues 865–888) is highlighted in magenta. (b) Analogous N-terminal regions in BRI1 and the human kinase Nek7. Shown are Cα traces of BRI1 (in blue) and Nek7 (PDB-ID 2qwm, gold) with the BRI1 phosphorylation sites depicted as red spheres, and selected residues shown in bonds representation. Note that the area surrounding Thr872 in BRI1 is very similar in Nek7 and that the N-terminal region needs to unfold to allow for Thr880 and S887 to become phosphorylated. (c) Detailed view of the BRI1 kinase nucleotide binding site. pSer891 is located in the glycine-rich loop of BRI1 (residues 890–895, glycine residues shown as blue spheres). The gatekeeper Tyr956 is in hydrogen-bond contact with the critical Lys911/Glu927 pair and its phosphorylation is likely to inactivate BRI1.

A third cluster of phosphorylation sites is formed by Ser891 and Tyr956, which are located in the active site of the kinase (red in Figure4a). Phosphorylation of Ser891 is likely to affect the conformation of the glycine-rich loop in BRI1, which in turn could distort the triphosphate binding pocket for the nucleotide substrate (Figure4c). Consistently, a phospho-mimicking mutation at position 891 yields severely dwarfed plants, while the corresponding alanine mutant acts as a gain-of-function allele (Oh ,b). Phosphorylation of the critical gatekeeper Tyr956 is likely to completely inactivate BRI1 (Figure4c), based on its role in the formation of a catalytically competent enzyme (Figures1c and 3b).

A protein docking platform located in the BRI1 C-lobe

We next mapped the known BRI1 missense alleles onto the structure (Clouse ; Li and Chory, 1997; Friedrichsen ; Xu ). Most of the alleles target residues in the BRI1 kinase N-lobe (bri1-1, bri1-202), C-lobe (bri1-108, bri1-301) and activation loop (bri1-103, bri1-115) that are likely involved in the structural stabilisation of the enzyme (Figure5a). This finding is consistent with the strong loss-of-function phenotypes that have been reported for these mutants (Clouse ; Li and Chory, 1997; Friedrichsen ; Xu ). We have previously mapped the binding surface of the inhibitor protein BKI1, which binds to the BRI1 C-lobe, specifically to a region involving Ala1104 and Leu1106 (Jaillais ) (Figure5b,c). Binding of the BKI1 C-terminal tail to the C-lobe inhibits the interaction of BRI1 with SERKs, suggesting that the C-lobe could represent a docking platform for the co-receptor kinase domain (Jaillais ). Importantly, bri1-117 (Asp1139–Asn) closely maps to this envisioned interaction surface (Li and Chory, 1997; Friedrichsen ) (Figure5a–c). It is thus possible that in bri1-117 mutant plants, interaction between the BRI1 and SERK kinase domains is affected, which would be consistent with the strong loss-of-function phenotype (Li and Chory, 1997; Friedrichsen ).

Figure 5

Distribution of genetic alleles in the BRI1 kinase domain. (a) Ribbon diagram of BRI1, coloured according to Figure1(b). Genetic missense alleles are depicted as green spheres. (b, c) Surface representation of the BRI1 C-lobe in two different orientations. Residues affecting BKI1 binding (Ala1104, Leu1106) and the bri1-117 missense allele (Asp1139–Asn) are highlighted in orange.

The isolated cytoplasmic parts of BRI1, SERK3 and SERK2 homodimerise

We noted during the purification of BRI1 domain fragments (Figure1a) that a construct comprising the entire cytoplasmic portion (JM–KD–CT) eluted much earlier in size-exclusion chromatography experiments than expected for a monomer. We found the elution volume of BRI1814–1196 to be consistent with a homodimer of approximately 80 kDa (Figure6c) (Jaillais ). We systematically tested different BRI1 fragments (Figure1a) in analytical size-exclusion chromatography experiments and found that the JM and the CT appear to mediate the formation of BRI1 homodimers, as deletion of either segment renders the resulting kinase domain monomeric in our assays (Figures6c and S4). This situation holds also true for the cytoplasmic portions of SERK2 and SERK3, which again appear to be homodimers when expressed in isolation in E. coli (Figure6d and S4). Upon mixing recombinant BRI1814–1196 and SERK3250–615, there appears to be only a weak tendency to form larger heterooligomers (Figure6e,f).

Figure 6

Oligomeric state-analysis of individual domains in BRI1 and in SERKs. (a) Schematic overview of the SERK3 (in orange) and SERK2 (in yellow) kinase domain fragments with construct borders included. (b) SDS-PAGE analysis of the purified kinase domain fragments used in this study. (c) A 280 nm absorbance trace of an analytical size-exclusion chromatography on a Superdex 75 HR 10/30 column. Wild-type BRI1814–1196 elutes as a homodimer, while deletion of either the JM and/or the CT region renders the resulting kinase domain apparently monomeric. Void (V0) and total volume (V) are shown together with the elution volumes for molecular weight standards (A, aldolase, MW 158 000; B, conalbumin, MW 75 000; C, ovalbumin, MW 43 000; D, carbonic anhydrase, MW 29 000; E, ribonuclease A, MW 13 700, F, aprotinin, MW 6500). The calculated molecular weights are 42 800 (BRI1 JM–KD–CT), 39 000 (JM–KD), 37 000 (KD–CT) and 33 200 (KD). (d) Analytical size-exclusion chromatography of SERK2 and 3 kinase domain constructs performed as in (c). The calculated molecular weights are 41 700 (SERK3 JM–KD–CT), 33 200 (SERK KD), and 39 300 (SERK2 JM–KD–CT). (e) Analytical size-exclusion analysis of a mixture of BRI1 JM–KD–CT and SERK3 JM–KD–CT on a Superdex 200 HR 10/30 column. A, thyroglobulin, MW 669 000; B, ferritin MW 440 000; C, aldolase, MW 158 000; D, conalbumin, MW 75 000; E, ovalbumin, MW 43 000; F, carbonic anhydrase, MW 29 000). (f) Immunoblot of peak fractions from (e) using anti-BRI1 and anti-SERK3 antibodies. (g) Analytical ultracentrifugation of the isolated brassinolide-bound BRI1 ectodomain reveals its monomeric state. (h) Analytical ultracentrifugation of the BRI1–brassinolide–SERK1 ectodomain complex reveals a 1:1 stoichiometry. The purified BRI1 and SERK1 ectodomains are approximately 110 and 30 kDa, respectively.

We have previously reported that the spiral-shaped extracellular domain of BRI1 is exclusively monomeric in solution and has no tendency to homomerise upon BL-binding (Hothorn ). Instead we found that brassinosteroid binding creates a docking platform for the smaller and shape-complementary LRR-domain of SERKs (Santiago ). The steroid hormone acts as a ‘molecular glue’ and promotes the tight association of the receptor and co-receptor ectodomains (Santiago ). To assess the oligomeric state of the BRI1–SERK1 ectodomain complex, we performed analytical ultracentrifugation. Our experiments confirm that the BL-bound BRI1 ectodomain behaves as a monomer in solution, while addition of the SERK1 ectodomain results in the formation of a very stable heterodimer (Figure6g,h). We conclude that the extracellular domains in BRI1 and SERKs form heterodimers upon sensing brassinolide, while their cytoplasmic segments can form homodimers when expressed in isolation.

DISCUSSION

Evolution of dual-specificity plant receptor kinases

It has been previously suggested based on sequence comparisons and structural homology modeling that plant receptor kinases including BRI1 and SERKs are closely related to the animal Pelle family (Shiu and Bleecker, 2001; Klaus-Heisen ). The crystal structure of the BRI1 kinase domain further supports this hypothesis: First, there is a high degree of structural similarity between BRI1 and IRAK-4, the human orthologue of Drosophila Pelle (Figure3a). Second, BRI1 and SERK3 contain an unusual gatekeeper tyrosine residue, which in animal kinases is only found in the Pelle family (Figure3b) (Wang ). Third, the conformation of the activation loop is highly similar in BRI1 and in IRAK-4, and BRI1 and IRAK-4 are dual-specificity kinases (Figure3c). We thus speculate that the last common ancestor of Pelle/IRAKs and plant receptor kinases already had dual-specificity activity towards Ser/Thr and tyrosine substrates and that the plant kingdom preserved this feature. The BRI1 activation loop contains structural fingerprints from both Ser/Thr and tyrosine kinases (Figure3e,f), enabling BRI1 to phosphorylate rather diverse substrates (Oh , 2009; Wang ; Tang ; Jaillais b). It is interesting that Thr1049 in the P + 1 pocket of BRI1, which critically determines substrate specificity in other kinases, can be phosphorylated in planta (Wang ) (Figure3f). Thr1049 phosphorylation/de-phosphorylation could thus enable BRI1 to switch or modulate its substrate specificity. Alternatively, Thr1049 phosphorylation could be used to inactive the kinase (Wang ).

The BRI1 C-lobe surface is critical for brassinosteroid signaling

In our structure, residues involved in the binding of BKI1 cluster together with the known missense allele bri1-117 at the bottom of the BRI1 C-lobe (Li and Chory, 1997; Friedrichsen ; Wang and Chory, 2006; Jaillais ) (Figures5a–c and S2). We have previously demonstrated that a BKI1-derived peptide can bind to this surface and inhibit the interaction of BRI1 with SERK3 (Jaillais ). It is thus possible that the BRI1 C-lobe provides an interaction surface with SERKs. In the case of animal epidermal growth factor receptor (EGFR), formation of an asymmetric kinase homodimer activates the receptor (Jura ). In EGFR, this dimer is formed by the JM domain of one kinase partner interacting with the C-lobe of the second kinase partner (Jura ; Endres ). Interestingly, this interaction between two kinase domains can be modulated by the protein inhibitor MIG6 binding to the C-lobe of one of the partners (Zhang ). Based on this structural analogy, one may speculate that in the case of BRI1 and SERKs similar asymmetric dimers (in this case heterodimers) might be formed upon receptor activation and that BKI1 binding to the BRI1 C-lobe could prevent this interaction (Wang and Chory, 2006; Jaillais ).

Kinase homodimers could represent a ground state of the receptor

There is structural and biochemical evidence that the initial extracellular events in LRR receptor kinase signaling involve ligand-dependent heterodimersation of a receptor protein with a shape-complementary co-receptor (Figure6g,h (Hothorn ; She ; Santiago ; Sun ,b). Heterodimerisation of BRI1 and SERK1 or of FLAGELLIN SENSITIVE 2 (FLS2) and SERK3 brings the C-termini of receptor and co-receptor in close proximity in crystals (Santiago ; Sun ,b). We can thus speculate that interaction of BRI1/FLS2 with SERK ectodomains may also affect the arrangement of their cytoplasmic parts in planta. Experimental evidence for BL-dependent receptor co-receptor heteromerisation in the cytoplasm has been provided in vitro (Li ; Nam and Li, 2002; Oh ) and in vivo (Wang ), but also ligand-independent heterooligomers have been reported in intact cells (Bücherl ). Recently, it has been demonstrated that both the extracellular and intracellular segments of BRI1 and SERK3 are critical for the interaction and for brassinosteroid signaling (Jaillais ). It is even possible to swap the kinase domains of FLS2 and SERK3 and still obtain a fully functional receptor–co-receptor pair, strongly supporting the heteromerisation model (Albert ).

However, there is also substantial in vivo evidence for BRI1 (Wang ) and FLS2 (Sun ) homomerisation. In the case of FLS2 the kinase domain and the membrane helix appear to be critically involved in the formation of homooligomers (Sun ). Interestingly, BRI1 or FLS2 homomers appear to form independent of ligand stimulus, suggesting that this configuration could represent a resting state for these receptors (Hink ; Sun ). We speculate based on our biochemical assays, that BRI1 and SERK kinase domains are able to homodimerise using their JM and CT segments (Figure6a–f), although we cannot exclude that the full-length proteins may behave differently in intact membranes (Bücherl ). Association of the receptor and co-receptor extracellular domains could affect this resting state, by bringing the kinase domains of BRI1 and SERKs into close proximity. Further mechanistic studies in vitro and in planta will be required to fully dissect the relative contributions of BRI1 homomers and heteromers to brassinosteroid signaling.

EXPERIMENTAL PROCEDURES

Protein expression and purification

Either wild-type, kinase-dead (Asp1027–Asn) or hyperphosphorylated (Thr872–Ala) BRI1 receptor kinase domain fragments BRI1865–1160, BRI1814–1160; BRI1865–1196, BRI1814–1196, and SERK3250–615, SERK3272–566 or SERK2281–628 were recombinantly expressed in E. coli as previously described (Jaillais ). For protein purification, cells were thawed in lysis buffer (20 mm Tris pH 8.0, 500 mm NaCl, 4 mm MgCl2, 1 mm ATP and 2 mm β-mercaptoethanol [β-ME]) and lysed with an EmulsiFlex-C3 (Avestin, www.avestin.com). The lysate was centrifuged at 7000  for 60 min. The supernatant was loaded onto a 5 ml HisTrap HP Ni2+ affinity column (GE Healthcare, www.gelifesciences.com/), washed with 10 column volumes of washing buffer (20 mm Tris pH 8.0, 500 mm NaCl, 4 mm MgCl2, 1 mm ATP, 10 mm imidazole pH 8.0, 2 mm β-ME) and eluted in lysis buffer supplemented with 200 mm imidazole pH 8.0. The 6 × His tag was removed with recombinant tobacco etch virus protease (TEV) at 1:100 molar ratio for 16 h at 4°C during dialysis against lysis buffer. The kinase domain was then separated from the 6 × His tagged TEV protease by a second Ni2+ affinity step. We found that catalytically active BRI1 kinase domains were non-homogenously auto-phosphorylated when expressed in E. coli. These fragments were dialysed against 20 mm Tris pH 8.0, 100 mm NaCl, 4 mm MgCl2, 0.5 mm Tris(2-carboxyethyl)phosphine (TCEP) and loaded onto a 5 ml HiTrap Q HP anion exchange chromatography column (GE Healthcare). Elution was performed by a linear NaCl gradient to a final concentration of 1 m. The major peak eluted in a range of 150–300 mm NaCl and was purified further by preparative size-exclusion chromatography on a Superdex 75 HR26/60 column, equilibrated in 20 mm Tris pH 8.0, 300 mm NaCl, 4 mm MgCl2, 0.5 mm TCEP. Monomeric peak fractions were concentrated to about 10 mg ml−1 using a Amicon Ultra-15 concentrator (molecular weight cut-off 20 000) and stored on ice.

The isolated BRI1 and SERK1 ectodomains (residues 29–788 and 24–123, respectively) were expressed and purified as described (Santiago ). To form the BRI1–SERK1 complex, the purified BRI1 ectodomain was incubated at a concentration of 1.5 mg ml−1 with 0.05 mm brassinolide (Chemiclones Inc., http://www.chemiclones.com/) in 25 mm citric acid/NaOH pH 5.0, 100 mm NaCl and SERK1 was added in 1:1 molar ratio. The mixture was concentrated to 5 mg ml−1 and stored on ice.

Protein crystallisation and data collection

We tested all BRI1 kinase domain fragments in extensive crystallisation trials and obtained initial crystals for BRI1865–1196–Thr872–Ala (see above) in Morpheus (Molecular Dimensions, http://www.moleculardimensions.com/) screen condition E4. These hexagonal crystals initially diffracted to approximately 3.5 Å resolution. Reductive protein methylation of BRI1865–1196–Thr872–Ala (Shaw ) yielded improved crystals in 25% (w/v) PEG 4000, 8% (v/v) 2-propanol, 0.1 m sodium acetate diffracting up to 2.5 Å resolution. Reductively methylated samples of the shorter BRI1865–1160–Thr872–Ala crystallised in the same space group and with similar unit cell constants in Morpheus screen condition E5. Diffraction quality apo-form crystals were grown at room temperature by vapour diffusion from hanging drops composed of equal volumes (2 + 2 μl) of protein solution (10 mg ml−1) and crystallisation buffer (26% PEG 20 000/PEG 550 MME mix, 0.12 m ethylene glycols mix, 0.1 m HEPES pH 7.5) suspended over 0.5 ml of the latter as reservoir solution and using microseeding protocols. Crystals were cryoprotected by serial transfer into reservoir solution supplemented with 15% (v/v) ethylene glycol and snap-frozen in liquid nitrogen. Crystals of the BRI1–ADP complex were obtained by co-crystallisation, including ADP to a final concentration of 1 mm into the protein storage buffer. Crystals of the nucleotide analogue AppNHp complex were also formed by co-crystallisation (10 mm AppNHp and 2.5 mm MnCl2), while the ATP complex was obtained by soaking apo-form crystals in reservoir solution supplemented with 30 mm ATP for 5 min. Diffraction data were collected at the Swiss Light Source beam-line PXII and PXIII, Villigen, CH and the European Synchrotron Radiation Facility (ESRF) beam-line ID-29. Data processing and scaling was done in XDS (version: March 2013) (Kabsch, 1993) (Table1).

Structure solution and refinement

The structure of the BRI1865–1196–Thr872–Ala ADP was determined in space group P62 by the molecular replacement method as implemented in the program PHASER (McCoy ), and using the recently reported SERK3 kinase domain structure as search model (Yan ) (Protein Data Bank identifier 3 uim, residues 272–573). The solution comprises one molecule per asymmetric unit with a solvent content of approximately 53%. The resulting electron density map was readily interpretable and the model was completed in alternating cycles of manual model building in COOT (Emsley and Cowtan, 2004) and restrained TLS refinement as implemented in PHENIX.REFINE (Adams ). Analysis with MOLPROBITY (Davis ) suggested excellent stereochemistry, with no outliers in the ramachandran plot. Refinement statistics for BRI1865–1196–Thr872–Ala ADP, BRI1865–1160–Thr872–Ala apo, BRI1865–1160–Thr872–Ala AppNHp/Mn2+, BRI1865–1160–Thr872–Ala ATP and BRI1865–1160–Thr872–Ala ADP are summarised in Table1.

HPLC-based activity assays

Analysis of ATP and GTP conversion by BRI1 was carried out in buffer containing 20 mm Hepes pH 7.5, 100 mm NaCl, 4 mm MgCl2, 0.5 mm TCEP at 25°C using 10 μm BRI1865–1160 and 5 mm ATP or GTP. Twenty microlitre aliquots of the reaction sample were injected onto a Vydac. 218TP 5 μm C18 column (4.6 × 250 mm) and product elution was analysed at 1 ml min−1 by monitoring the absorbance at 259 nm. The isocratic elution buffer contained 100 mm potassium phosphate (pH 6.5), 6% (v/v) acetonitrile and 10 mm tetrabutylammonium bromide.

Analytical ultracentrifugation

The oligomeric state of the isolated BRI1 ectodomain and of a purified BRI1–brassinolide–SERK1 complex was investigated at 5 mg ml−1 in 25 mm citric acid/NaOH pH 5.0, 100 mm NaCl buffer at 4°C by monitoring its sedimentation properties at 280 nm using 100 000 in a Beckman Optima XL-A centrifuge fitted with a four-hole AN-60 rotor and double-sector Epon centerpieces. Molecular weight distributions were determined by the C(s)method (Schuck, 2000).

Generation of antibodies

For generation of BRI1 and SERK3 specific antibodies, purified BRI1865–1196 or SERK3250–615 was dialysed against phosphate-buffered saline (PBS) and injected into rabbits. The resulting sera were affinity-purified over BRI1- or SERK3-coupled Affigel 15 (Biorad, www.bio-rad.com) columns and eluted in 200 mm glycine pH 2.3, 150 mm NaCl.

Size-exclusion chromatography and immunoblotting

To assess homomerisation of the BRI1, SERK3 and SERK2 cytoplasmic portions, gel filtration was performed using a Superdex 75 HR 10/30 column (GE Healthcare) pre-equilibrated in 20 mm HEPES pH 7.5, 100 mm NaCl, 4 mm MgCl2, 0.5 mm TCEP. 100 μl of the isolated wild-type BRI1814–1196, BRI1814–1160, BRI1865–1196 and BRI1865–1160, SERK3250–615, SERK3272–566 and SERK2281–628 kinase domains were loaded sequentially onto the column and elution at 0.5 ml min−1 was monitored by ultraviolet absorbance at 280 nm. 0.5 ml peak fractions were collected alongside and analysed by SDS-PAGE.

Heteromerisation of the entire cytoplasmic portions of BRI1 and SERK3 was monitored by size-exclusion on a Superdex 200 HR10/30 column using the same running buffer and loading volumes. 0.5 ml peak fractions were analysed by immunoblot analysis using polyclonal antibodies raised against the isolated BRI1 and SERK3 kinase domains, respectively. Samples corresponding to each peak were mixed with 2× Laemmli buffer (v/v) (125 mm Tris–HCl, pH 6.8, 4% SDS, 20% glycerol, 2% mercaptoethanol, and 0.001% bromphenol blue), boiled at 95°C for 5 min, and run on a 10% SDS-PAGE gel. Proteins were then transferred onto nitrocellulose membranes (GE Healthcare) and blocking was performed using TBS–0.1%Tween buffer and 5% powder milk. Immunoblotting was carried out first incubating the membranes with αSERK3 or αBRI1 antibodies (1 h incubation at RT), and then with an anti-rabbit peroxidase conjugate (1 μg μl−1) antibody (Calbiochem, www.emdbiosciences.com). Detection was performed using the BM chemiluminescence blotting substrate (Roche, www.roche.com).