Crystal structures of Arabidopsis and Physcomitrella CR4 reveal the molecular architecture of CRINKLY4 receptor kinases

Plant-unique receptor kinases harbor conserved cytoplasmic kinase domains and sequence-diverse ectodomains. Here we report crystal structures of CRINKLY4-type ectodomains from Arabidopsis ACR4 and Physcomitrella patens PpCR4 at 1.95 Å and 2.70 Å resolution, respectively. Monomeric CRINKLY4 ectodomains harbor a N-terminal WD40 domain and a cysteine-rich domain (CRD) connected by a short linker. The WD40 domain forms a seven-bladed β-propeller with the N-terminal strand buried in its center. Each propeller blade is stabilized by a disulfide bond and contributes to the formation of a putative ligand binding groove. The CRD forms a β-sandwich structure stabilized by six disulfide bonds and shares low structural homology with tumor necrosis factor receptor domains. Quantitative binding assays reveal that ACR4 is not a direct receptor for the peptide hormone CLE40. An ACR4 variant lacking the entire CRD can rescue the known acr4-2 mutant phenotype, as can expression of PpCR4. Together, an evolutionary conserved signaling function for CRINKLY4 receptor kinases is encoded in its WD40 domain.


Introduction
The N-terminal β-propeller domain of ACR4 and PpCR4 folded into a seven-bladed WD40 domain 27 (Fig. 1d), as previously speculated 20 . Each blade is stabilized by a highly conserved disulfide bridge and connected by small loop regions, possibly an evolutionary adaptation to the extracellular environment (Fig. 1d, Supplementary Fig. 1). Cys180, which is found mutated to tyrosine in the acr4-7 mutant 20 , forms a disulfide bond in the 4 th blade (Fig. 1d). The N-and Cterminal blades are not connected by disulfide bonds (Fig. 1d). The most N-terminal β-strand is buried in the center of the propeller and is highly conserved among all known CRINKLY4 receptors 36 (Fig. 1d, Supplementary Fig. 1). Several small loops connecting the different blades of the WD40 domain appear partially disordered in our ACR4 and PpCR4 structures (Fig. 1d,e).
The C-terminal CRD comprises PpCR4 residues 313-401 and folds into a well defined βsandwich structure stabilized by six invariant disulfide bridges (Fig. 1e, Supplementary Fig. 1, see below). The WD40 and CRD domains are connected by a short linker region (Fig. 1e). Analysis of crystal lattice arrangements with the program PISA 37 and analytical size-exclusion chromatography experiments ( Supplementary Fig. 2) together indicate that the ACR4 and PpCR4 ectodomains behave as monomers in solution. All surface exposed cysteines in ACR4 and PpCR4 contribute to disulfide bond formation (Fig. 1d,e; Supplementary Fig. 1). The N-glycosylation pattern differs between ACR4 and PpCR4 (Fig. 1e, Supplementary Fig. 1). Taken together, a compact WD40 and a cysteine-rich domain represent structural fingerprints of monomeric CRINKLY4 ectodomains.
Structural homology searches against ACR4 WD40 using the program DALI 38 40 . ACR4 WD40 however lacks the UVR8 tryptophan cage involved in UV-B light sensing 40,41 and both BLIP-II and UVR8 are devoid of the buried N-terminal strand and the conserved disulfide bridge pattern present in ACR4 WD40 . Thus, the pore-filling N-terminus and the invariant blade disulfide bonds are unique structural features of extracellular CRINKLY4 WD40 domains.
We next studied the interaction of ACR4 WD40-CRD with its proposed ligand CLE40 [29][30][31][32] . As ACR4 has been previously reported to form hetero-oligomers with the LRR-RK CLV1, we sought to include the CLV1 ectodomain in these experiments, but we could not produce well-behaving protein samples of the AtCLV1 ectodomain by secreted expression in insect cells ( Supplementary   Fig. 4), and consequently could not use the CLV1 ectodomain for biochemical or crystallographic We found that CLE40 binds the AtBAM1 ectodomain with a dissociation constant (K d ) of ~1 μM ( Fig. 2b) but shows no detectable binding to the ACR4 ectodomain in quantitative grating-coupled interferometry (Fig. 2b) and isothermal titration calorimetry (Fig. 2c) assays. Thus, CLE40 does not represent a direct ligand for the ACR4 ectodomain.
Using the previously documented seed retardation phenotype of the acr4-2 mutant 18,20 we next carried out genetic complementation analyses using different constructs expressed from the ACR4 promoter. In agreement with an earlier report 20 , a construct in which the entire cytoplasmic domain of ACR4 had been deleted could not rescue the seed development phenotypes of acr4-2 plants (Fig. 3a). Full-length ACR4 lacking kinase activity partially restored seed development in acr4-2 plants (Fig. 3a). Strikingly, expression of full-length PpCR4, the ectodomain of which shares only 40% sequence identity at the amino-acid level with ACR4 WD40-CRD , from the ACR4 promoter could partially complement acr4-2 phenotypes as well. Together, these experiments reinforce an evolutionary conserved function for CRINKLY4 RKs, which are however not strictly dependent on the protein kinase activity of the receptor. The 2.7 Å crystal structure of the entire ectodomain from PpCR4 enabled us to further characterize the ~90 amino-acid CRINKLY4 CRD (Fig. 3b). A structural homology search with DALI 38 indeed identified several TNFR domains as top hits, but with very low DALI Z-scores (4.1-2.9). Structural superposition of PpCR4 CRD with the previously reported structure of a type I TNF receptor extracellular domain revealed that only a small portion of the CRINKLY4 aligns with canonical TNFR domains (r.m.s.d. is ~1 Å comparing 20 corresponding C ɑ atoms, Fig. 3b). The segment includes a small β-hairpin and two conserved disulfide bridges located at the center of the CRINKLY4 CRD (Fig. 3c). Structural superposition of the eight molecules in the asymmetric unit of our PpCR4 WD40-CRD crystal structure (Supplementary Table 1) revealed only subtle movements of the CRD versus the WD40 domain (r.m.s.d. is ~0.3-0.5 Å comparing 360 corresponding C ɑ atoms, Supplementary Fig. 5). In line with this, we located a small WD40 -CRD domain interface using PISA 37 (total buried surface area is ~900 Å 2 ). The interface is formed by the C-terminus of the CRD (PpCR4 residues 385-401) that makes mainly hydrophobic interactions with a small groove located between the N-and C-terminal blade of the WD40 domain ( Supplementary Fig. 6). Additional contacts originate from a small α-helix in the CRD and several loop regions in PpCR4 WD40 ( Supplementary Fig. 6).
Using the now experimentally determined domain boundaries of the ACR4 CRD (Supplementary Figs. 7 and 1), we re-performed complementation assays of the acr4-2 mutant with 6 a construct in which the entire CRD was omitted (ACR4 ΔCRD). A CRD). As previously reported 20,24 , we found that ACR4 ΔCRD). A CRD can rescue the seed development phenotype of acr4-2 plants (Fig. 3a).
Recently, mutation of the cysteine residues in ACR4 WD40 and ACR4 CRD involved in the formation of disulfide bonds in our structures ( Fig.1, 3b,c) to alanine resulted in a functional receptor for seed development 24 . We monitored migration of the purified ACR4 WD40-CRD ectodomain under oxidizing and strongly reducing conditions in analytical size exclusion chromatography experiments and found that reduction of ACR4 WD40-CRD did not induce aggregation of the receptor (Fig. 3d,e).
Together, the CRINKLY4 CRD only shares weak structural homology with animal TNFR domains, has a conserved domain interface with the WD40 domain and is dispensable for seed development.
The conserved disulfide bonds appear to be involved in structural stabilization. The domain interface between the WD40 domain and the CRD is conserved among CRINKLY4 receptors from different species (Supplementary Figs. 1,6).
While the CRD domain appears to be dispensable for at least some of ACR4's physiological functions, our and previous findings 20, 24 argue for an important role of the structurally unique WD40 domain in CRINLKY4 receptors. We located evolutionary conserved, surface exposed residues at the 'back side' of the ACR4 WD40 domain (Fig. 4a, Supplementary Fig. 1), which in our PpCR4 WD40-CRD structure is in contact with the CRD (Fig. 1e). We replaced individual residues by alanine or glutamine, respectively and assessed the ability of the resulting mutant proteins to complement the acr4-2 seed development phenotype (Fig. 4a-c). We analyzed three independent homozygous T3 lines per mutant receptor and found that most mutations behaved similar to wild type ( Fig. 4b) and that none of mutants tested displayed the strong loss-of-function phenotype of acr4-2 plants (Fig. 4b,c). Plants in which either Tyr157 or Asn158/Asn196 were mutated had seed numbers per silique that were significantly reduced compared to wild type (Fig. 4b). While there was no electron density for a N-glycan at position Asn158 in the ACR4 WD40 structure (see Methods), the corresponding Asn150 in PpCR4 was found glycosylated (Fig. 1e). ACR4 Asn196 is predicted to be N-glycosylated as well 42 , suggesting that the weak loss-of-function phenotypes observed in our Tyr157/Asn158 and Asn196 point mutants may be caused by an altered N-glycosylation pattern of the receptor (Fig. 4b,c).
We next analyzed the molecular surface of the 'front side' of ACR4 WD40 , which represents another canonical binding surface for peptide and protein ligands in many cytoplasmic or nuclear localized WD40 proteins 27 . We located a large binding groove in ACR4 WD40 formed by the WD40 domain core and by small surrounding loop regions, which appear similar in our ACR4 WD40 and ubiquitin ligase CONSTITUTIVELY PHOTOMORPHOGENIC 1 (COP1) (Fig. 4d) 43 . It may thus provide and interaction platform for high molecular weight ligand.

Discussion
Our crystal structures (Fig. 1d,e) and reverse genetic analyses (Fig. 3a) together reveal an evolutionary conserved domain architecture for plant-unique CRINKLY4 receptor kinases 36 . The CRINKLY4 WD40 domain differs from known cytoplasmic and extracellular WD40 domains 27,39,40,43 , with its seven blades being stabilized by disulfide bridges and the hydrophobic core of the domain being reinforced by insertion of the protein's N-terminus (Fig. 1d,e). We speculate that these unique structural features represent an adaptation to CRINKLY4 ectodomains being exposed to the plant cell wall environment. Previous 20 and our genetic data argue for an important function for the ACR4 WD40 domain in seed development (Fig. 3a). A large groove located on the 'front side' of ACR4 may be involved in the binding of a ligand (Fig. 4d). This ligand could be a small molecule, a protein or a peptide, and may be larger than the octameric peptide motifs recognized by COP1 (Fig. 4d). The low degree of sequence conservation of residues contributing to the formation of the binding groove in the WD40 domain (Fig. 4d, Supplementary Fig. 1) and the fact that PpCR4 can functionally replace ACR4's function in Arabidopsis seed development ( Fig.   3a) together indicate that CRINKLY4 receptors may sense a family of structurally conserved ligands.
Our quantitative binding assays reveal that the previously proposed peptide ligand CLE40 cannot directly interact with the ACR4 ectodomain ( Fig. 2), but we cannot rule out that CLE40 binds the CLV1 ectodomain in a signaling complex also containing ACR4 29,30,32 . The architecture and cellular functions of CLV1 -ACR4 signaling complexes remain to be elucidated, with recombinant expression and purification of the CLV1 ectodomain representing a significant challenge ( Supplementary Fig. 4). BAM1 cannot fully replace CLV1 in quantitative biochemical assays, as it binds CLE40 only with moderate affinity (Fig. 2). In contrast, the CLE family member CLE9 binds BAM1 with nanomolar affinity 44,45 . In solution and in the absence of ligand, CRINKLY4 ectodomains behave as monomers (Fig. 3e, Supplementary Fig. 2). The previously observed ACR4 homo-oliogomers 32 may thus be generated by ligand-induced oligomerisation of several CRINKLY4 ectodomains and/or be stabilized by interaction of the CRINKLY4 transmembrane helices, as previously suggested 46,47 .
Analysis of the CRINKLY4 cysteine-rich domain revealed only weak structural homology with animal TNFR domains (Fig. 3b) 14 . In line with this, we could not locate proteins with homology to tumor necrosis factors in the Arabidopsis or Physcomitrella patens genomes 48,49 . The CRINKLY4 CRD contains six conserved disulfide bridges (Fig. 1e, Supplementary Fig. 1), which in our PpCR4 WD40-CRD structure appear to be involved in structural stabilization (Fig. 3b,c). However, CRINKLY4 ectodomains can withstand reducing conditions (Fig. 3e), and thus the putative function of the CRD could indeed be regulated by changes in the cell wall redox environment 24 .
Our and previous 20 reverse genetic experiments suggest that the ACR4 cytoplasmic domain has to be present for normal seed development in Arabidopsis, yet its catalytic activity seems to be dispensable (Fig 1e). Similar observations have been made for CrRLK1L-family receptor kinases [50][51][52] . The mechanistic implications are poorly understood, but the involvement of protein phosphatases in both CR4 and CrRLK1L-mediated signal transduction 33,53,54 suggests that the cytoplasmic kinases domains of these receptors may act as scaffolding proteins that can become phosphorylated despite not requiring auto-and trans-phosphorylation activity themselves.
Genetic interactions between ACR4 and other receptor kinases such as ABNORMAL LEAF SHAPE 2 (ALE2) 55 the LRR-RKs CLV1 29,30,32 and GSO1/GSO2 45,56 have so far not yielded a mechanistic understanding of CRINKLY4's signaling functions. Also, no ligand candidate for ACR4 or for its homologs in Arabidopsis has emerged from forward genetic screens 21 . Our identification of a putative ligand binding pocket in ACR4 WD40 now reinforces the notion that bona fide ligands for CR4s may exist and that their identification may be achieved using a combination of genetic and biochemical approaches.

In vitro kinase phosphorylation assay
Coding sequence of ACR4 kinase domain (residues 497 -792) was amplified from A.  Table 1). Data processing and scaling were done with XDS and XSCALE 61 .

Structure solution and refinement
Nine consistent Pt sites were located in three wavelength MAD data using the program SHELXD 62 followed by site refinement and phasing in SHARP 63 . The resulting heavy atom sites and starting phases (FOM was 0.35 to 3.2 Å resolution) were input into phenix.autobuild 64 69 and restrained NCS refinement in phenix.refine 70 . Ile156 represents a Ramachandran plot outlier in each chain, but is well defined by electron density. Structural diagram were prepared in Pymol (https://sourceforge.net/projects/pymol/) and ChimeraX 71 .

Conflict of interest
The authors declare no conflict of interest.            a, Ribbon diagram of ACR4 WD40 (in blue) with surface exposed conserved residues shown in bonds representation (in orange) at the exposed surface. Blade numbers are indicated. b, Effect on surface point-mutations on ACR4-mediated seed production. Ten siliques per transgenic line from three independent homozygous T3 complementation lines were pooled and plotted as beeswarm plots with the bold line representing mean, whiskers indicating the standard deviation, and circles depicting the raw data. The plots for wild type, acr4-2 and ACR4 were generated from same data sets shown in Fig. 3a. Seed counts per silique significantly different from wild type were determined by simultaneous comparisons of several mutants against wild type using the Dunnett procedure (indicated by an asterisk). c, Molecular surface of the ACR4 WD40 β-propeller domain 'back side' (in light blue). The positions of the mutated residues are highlighted in orange. d, Comparison of the 'front sides' of the structurally related WD40 domains of COP1 (PDB-ID 6QTO 43 left panel) and ACR4 (right panel, r.m.s.d is ~3.5 comparing 205 corresponding Cɑ atoms). The COP1 VP-peptide ligand derived from the transcription factor HY5 is shown in yellow. Note the large and deep putative binding groove in the corresponding surface area in ACR4 WD40 .