Introduction

The proper functioning of eukaryotic cells requires the sorting and targeting of proteins synthesized in the endoplasmic reticulum (ER). From the ER, correctly folded/assembled proteins are transferred to the Golgi, which initiates the first step in protein sorting and trafficking. In plant cells, vacuolar sorting receptors (VSRs) are responsible for the sorting of proteins from the trans-Golgi network (TGN) to prevacuolar compartments (PVCs) and finally to their respective vacuoles1. VSRs are type-I transmembrane proteins with ~ 600 amino acids (80 kDa without the N-terminal signaling peptide)2. VSRs recognize specific signal sequences of the cargo proteins called Vacuolar Sorting Determinants (VSDs) through its N-terminal protease-associated (PA) domain3,4,5. This recognition by VSRs is known to be pH-dependent; the cargo protein binds to the VSR via the VSD in the TGN in a neutral pH, and the cargo protein is released upon exposure to an acidic pH6. An alternative model presented by Kunzl et al. suggests that the binding and release of cargo occur at the ER/Golgi and TGN/early endosome, respectively, through the influence of pH and Ca2+ changes7. In addition, PV72, a homolog of AtVSR1 in potatoes has a calcium-dependent cargo binding within the pH range of 5.5–78.

The VSDs of various vacuolar proteins are mainly classified into two groups, sequence-specific vacuolar sorting determinants (ssVSDs)9,10 and C-terminal vacuolar sorting determinants (ctVSDs)11. The proaleurain peptide (SSSFADSNPIRPVTDRAASTYC) present in a plant cysteine protease is the prime example of a ssVSD with the presence of a central ‘NPIR’ motif9,10. Any alteration from this specific sequence motif can negatively affect AtVSR1-binding. For instance, a mutated peptide with glycine in place of isoleucine in the NPIR motif renders the peptide unable to compete with the proaleurain peptide for binding to VSR19,10. On the other hand, the characteristics of ctVSDs appear to be contextual without any distinctive known sequence features, though it has been proposed that the VSR1-binding motif is approximately five residues long, with the last three being hydrophobic, and it is located at the C-terminus of a cargo protein5,12. A recent study by Park et al. proposed that soluble proteins carrying a ctVSD are transported by RMRs, not by AtVSR113.

The structural details of PA domain of AtVSR1 in complexes with both ssVSD and ctVSD were reported4,5,12. The structures show that the peptides interact with the PA domain of AtVSR1 in sequence non-specific manners, as the peptides bind the PA domain through their backbone only without any side chain involvement. Furthermore, the PA-ssVSD complex showed that the PA interacted with the preceding residues of the NPIR motif, while the motif itself was invisible, probably because of structural disorder4. Despite these structural studies, the involvement of the PA domain in recognizing VSD motifs remains unclear.

Herein, we report a crystal structure of full-length VSR1 luminal domain from Arabidopsis thaliana (AtVSR1). The structure displays a distinct organization of constituent domains with unique interactions among them. Although the structure does not contain a bound peptide, it still provides valuable insights into the ligand-binding mechanism of VSRs. Additionally, the crystal contacts and domain structures offer further information about the other characteristics and potential functions of this unique class of proteins.

Results

The overall structure

The luminal part of AtVSR1 from residue 1 to 560 was expressed from Drosophila S2 insect cells. The full sequence is disclosed in the Supplemental Information. The purified protein was crystallized in a space group P213, which diffracted up to 2.6 Å14. The crystal lattice packing indicated that one molecule in the asymmetric unit was closely associated with neighboring molecules through crystallographic three-fold symmetry. The resulting crystal structure was composed of an N-terminal protease-associated (PA) domain, a unique central region, and three C-terminal epidermal growth factor (EGF)-like repeats3 (Fig. 1A). The resolved structure had disordered regions at the C-terminus, where density could not be uniquely traced for proper model building after the first EGF-like domain. At the beginning of the refinement, a clear electron density attached to the sidechain of Asn289 in the central domain was observed and assigned as two N-acetylglucosamine (GlcNAc).

Figure 1
figure 1

The overall structure and the schematic of AtVSR1. (A) The overall structure of AtVSR1 is shown as a cartoon model. The color of each domain matches that of the schematic. Each domain is also labeled. (B) The schematic of AtVSR1 shows the domain organizations, the information of the crystallization construct, and the structure resolved portion. N-glycan indicates the location where glycosylation is observed. The black rectangle represents the signaling peptide.

Full size image

The structural model starts at Phe21 and ends at Ala465 and the overall structure is separated into three distinct domains: the PA domain (Phe21-Trp175), the thioredoxin (TRX) domain (Val186-Phe396), and the EGF-like domain (Glu409-Ala462) (Fig. 1B). The central domain is named as TRX domain, as a DALI search picked γ-interferon-inducible lysosomal thiol reductase (PDB ID 6NWX) as the closest 3D structure with a high Z-score of 14.6, followed by bacterial disulfide isomerase A (PDB ID 3BCI) with a Z-score of 14.015. The rest of the DALI results are all TRX-fold proteins, most of which are involved in protein disulfide bond reduction (Table 1).

Table 1 Structural similarity search by the Dali server.
Full size table

The PA and the TRX domains are connected by a 7-residue linker with no significant interaction between them. In contrast, the TRX and the EGF-like domains are connected through a 12-residue linker and display tight interdomain interactions between them through an extensive hydrogen bond (H-bond) network (Suppl. Table S1). At the core of the interdomain interface, the side chain of Asp303 in TRX forms H-bonds with the side chains of Arg440 and Tyr457 in EGF-like domains (Fig. 2A,B). At the periphery, an additional salt bridge between Lys 221 and Glu414 is observed (Fig. 2C), and the main chain carbonyl of Arg440 also forms an H-bond with the amine side chain of Lys213 (Fig. 2B). Notably, no significant hydrophobic interaction was found in this interdomain interaction. The Cys405 in the linker establishes a disulfide bond with Cys393 of the TRX domain potentially stabilizing the linker.

Figure 2
figure 2

Domain interaction between the TRX and the EGF-like domain. (A) The overall structure showing the interaction between the two domains. (B,C) Detailed hydrogen (H)-bond interactions are shown. Panel (A) and (B) are presented in the same view. Panel (C) provides a back side view of panel (A) and (B). The TRX domain and EGF-like domains are colored green and blue, respectively, and the hydrogen bonds are colored cyan. The key residues are depicted as sticks.

Full size image

The PA domain is composed of approximately 120 residues in length and organized as a central β-barrel with nine strands in two β-sheets, and two α-helices that are located at the periphery (Suppl. Fig. S1). Similarly shaped PA domains are known to exist in several protease classes such as subtilases, aminopeptidases, bacterial endopeptidases, as well as two families of sorting receptors, RMRs and VSRs16,17. However, the exact function of the PA domain has been examined in only a handful of proteins and speculated that the PA domain may serve as a protease-interacting or -regulating domain based on multiple studies17. Previous structural studies revealed that the PA domain of AtVSR1 harbors the binding sites for ctVSD and a portion of ssVSD4,5,12. The shallow pocket is established between β4-α3 loop and α4-α5 loop of PA domain where α4 unfurls and its residues are pushed away by the bound peptide4,5,12. The residues between β5 and α5 (Ser120-Asp137) appear highly flexible as they are disordered in our crystal structure (Suppl Figs. S1 and S2). A portion of the corresponding region is also reported to be disordered in the peptide-bound states4; however, the same portion is fully structured in the apo-form (PDB ID 4TJV) due to stabilization through interaction with the C-terminus and crystallographic contacts. Apart from this C-terminal linker, the PA domain of our AtVSR1 structure overlaps well with the previous apo-form PA domain structure (PDB ID 4TJV), with a root-mean-square-deviation (RMSD) of 1.22 Å over 148 Cα positions of the residues from 26 to 173.

The TRX domain of our AtVSR1 structure reveals four internal disulfide bonds plus an above-mentioned additional disulfide bond between Cys393 and Cys405, which connects it to the EGF-like domain. Although there is no direct interaction between the intramolecular PA and TRX domains, the two domains display extensive interactions with those domains of two neighboring molecules related by a crystallographic threefold symmetry, forming a cyclical domain-swapping (Fig. 3A). This homotrimer occurs without any conformational change in the PA domain, evidenced by the average RMSD of 0.64 Å between our structure and the structures of the PA domain alone (PDB ID 4TJV, 4TJX, and 8HYG).

Figure 3
figure 3

The crystallographic structure exhibits threefold symmetry through the swapping of the PA and TRX domains. (A) The asymmetric unit molecule is colored salmon and green, while the other two symmetry-related molecules are colored grey. (B) The detailed interaction between the two domains involves extensive hydrophobic interactions and tight van-der-Waals complementarity. The overall interface measures approximately 1,040 Å2. The PA and TRX domain residues responsible for the interactions are labeled and presented as sticks with transparent spheres (PA) and surface (TRX) (left and right panels). The PA and TRX domains are colored salmon and green, respectively.

Full size image

Alphafold2 model

The full length model of AtVSR1, generated by Alphafold2 (AF2), has been deposited in the AlphaFold protein structure database with an accession number AF-93026-F118. Most of the regions in the model have per-residue-confidence-scores (predicted Local Distance Difference Test, pLDDT) of 0.9 or higher, indicating that the model is highly reliable. Each domain of the model superimposes well with that of the crystal structure (Suppl Fig. S3A). However, the interdomain interactions of the AF2 model are completely different from those observed in our crystal structure (Suppl Fig. S3B,C). The inaccuracy of quaternary structure prediction is a well-known limitation of AF219,20. Despite the caveat, the AF2 model structure still offers valuable insight into how the regions not observed in the crystal structure may look like.

The oligomerization of AtVSR1

PDBePISA analysis of the crystallographic intermolecular interaction between the PA and the TRX domains reveals that the interface score is 1.0, indicating that AtVSR1 is highly likely to form a trimer in solution21. The total buried surface area between the PA and the TRX domains is approximately 1,040 Å2, and the interface is formed by 24.6% and 14% of the residues from the PA domain and the TRX domain, respectively. In comparison, the intramolecular interaction between the TRX and EGF-like domains has a PISA interface score of 0, indicating the observed interaction is not significant. The intermolecular interaction between the PA and the TRX domains displays notable hydrophobic interactions mediated by both ends of the loops connected to β3 of the PA domain (Fig. 3B). At one end, the residue Tyr52 of PA domain is nestled in a hydrophobic pocket in TRX domain established by Ile368, Pro370, Thr371, Tyr379, Gly381, Lys382 and Leu383. In addition, the side chain of Arg380 is packed against the nearby hydrophobic residue, Pro50, while also forming an H-bond interaction with the side chain of Thr145. The main chain carbonyl and amine of Thr55 exchanges H-bonds with the main chain of Leu369. The hydroxyl group of Tyr52 also forms a H-bond with the main chain carbonyl oxygen of Thr371 (Fig. 3B, left). On the other end, a loop formed by Pro84, Gly85, and Arg86 is nested within a valley fabricated by Trp234, Tyr235, and Gln354. Aggregation of hydrophobic residues, Leu56, Val57, Leu87, Pro88, and Ala113 connects the loops (Fig. 3B, right). Considering the resolved crystal structures of the TRX domain, the previously hypothesized intramolecular PA and TRX domain interaction through the same interface appears unlikely. The linker between the PA and the TRX domains is composed of 8 residues with approximately 26 Å in length. Even when we expanded the linker boundary in our modeling attempt, the proposed intramolecular interaction between the PA and the TRX domains was not possible without disrupting the adjacent residues and dihedral angle violations in linker residues (Suppl Fig. S4). This further supports the possibility of an intermolecular interaction that results in a domain-swapped trimer in vivo.

The EGF-like domain

It is noticeable that there is no significant intermolecular interaction by the EGF-like domains in the crystal lattice (Suppl Fig. S5). The two additional EGF-like domains that are tandemly located at the C-terminus are not visible in the crystal structure, indicating their high flexibility. It is likely that the string of EGF-like domains is oriented towards the ER membrane and provides flexibility for the PA and TRX domains, enabling them to promote intermolecular interactions (Suppl. Fig. S3A). The interaction between the TRX and EGF-like domains is held by multiple H-bond interactions. There are 4 residues each from the TRX domains and the EGF-like domains involved in this interaction. Together, they form 7 H-bonds, out of which 4 are salt bridges (Fig. 2, Suppl. Table S1).

Discussion

Trimeric nature of AtVSR1

Cao et al. conducted elaborate experiments digesting the luminal domain of AtVSR1 to define three protease-resistant domains and their involvement in ssVSD binding. Although the proposed multi-domain nature of AtVSR1 was accurate, interpreting the results based on the estimated molecular weights by SDS-PAGE has some drawbacks3. In the report, they suggested the luminal domain of AtVSR1 could be monomeric in solution based on its elution profile being similar to bovine serum albumin (BSA) on a gel filtration chromatography3. To have similar hydrodynamics to BSA, AtVSR1 must be tightly packed, suggesting that it could exist as a monomer with an intramolecular interaction between the PA and the TRX domains. In addition, all three EGF-like domains should be tightly packed with the other domains in the same molecule. However, our crystal structure shows that the AtVSR1 has an extended conformation with significant dynamic flexibility of the EGF-like domains. Significantly, the same chromatographic profile of Cao et al. also displayed a shoulder peak with a high molecular weight aggregate, which could be attributed to AtVSR1 multimers.

Evidence of AtVSR1 forming a trimer was reported by Kim et al. where hemagglutinin-tagged AtVSR1 (AtVSR1:HA) is expressed in transgenic plants, followed by fractionation of the protein extract using Superdex 200 HR 10/30, and identified the existence of 240 KDa and ~ 80–100 KDa species that responded to anti-HA antibody (Ab)22. Subsequent experiments confirmed the 240 KDa species to be homopolymer of AtVSR1. They further dissected the regions responsible for homotrimer formation and concluded that both the transmembrane and C-terminal cytosolic domains were necessary, while the luminal domain was not. This conclusion was based on a series of co-immunoprecipitation assays using deletion and substitution mutants of AtVSR1. It is possible that the detergents contained in the assay buffer were not suitable for the oligomerization of the luminal domain. Nevertheless, AtVSR1 with the C-terminal cytosolic domain mutants that cannot form oligomers suffer in vacuolar trafficking efficiency due to localization to the Golgi apparatus instead of the prevacuolar compartment.

Considering the structural evidence of homotrimer formation through the luminal PA and TRX domains presented in full-length luminal domain of AtVSR1, the formation of the oligomer through the transmembrane and C-terminal cytosolic domains might be enhanced by the homotrimer of the luminal domain.

Domain-swapping among homopolymers as observed in our AtVSR1 is not uncommon, in which proteins establish a dimer, a cyclic multimer, or an open-ended aggregate23,24. The swapping can be performed via exchanging a secondary structural element or an intact domain of the participating monomeric subunit. Diphtheria toxin is a classic example that forms a dimer through the latter method25, and the conformation in our AtVSR1 structure also swaps domains similarly. While diphtheria toxin can form a monomer through intramolecular domain interaction of the receptor domain, the only viable option for AtVSR1 is to form a homotrimer (or possibly a homomultimer) through intermolecular interactions, as opposed to the previously proposed monomer through intramolecular domain interaction. The presence of restrictive linker residues prevents the existence of a functional monomer with the compacted PA-TRX domains. Additionally, due to its cyclic domain swapping, AtVSR1 may also form additional oligomers.

Further analysis of VSR1 in solution using size exclusion chromatography-multi-angle light scattering (SEC-MALS) or sedimentation equilibrium (SE) analysis is necessary to clarify the oligomeric state in solution and the functional implications of oligomerization in the luminal domain.

VSD-recognition mechanism of AtVSR1

The previous structural study by Luo et al. showed that a part of the loop between β5 and β6 in the PA domain is displaced by an ssVSD peptide containing an NPIR motif4. Unexpectedly, this complex structure showed that the interaction between PA domain and a ssVSD peptide is not mediated by the NPIR motif, but rather by three preceding residues of the NPIR motif. Although sensitivity to mutation of the penultimate residue before the NPIR motif was shown by pull-down and other biological assays, a detailed structural analysis showed a sign of promiscuity in the ssVSD peptide binding, as little side chain involvement was observed.

The promiscuity of interaction is evidenced by another structural study by Tsao et al., where the authors found that a ctVSD peptide from CRU1 binds to the AtVSR1 PA domain in the same way as the ssVSD NPIR-peptide barley aleurain. ctVSD-peptides are composed of four C-terminal residues without a known consensus5,12. The structural study revealed a favored tendency; namely, a basic residue is preferred in position 1, and hydrophobic residues (except for proline) are favored in the three remaining positions. Intriguingly, the C-terminal carboxyl group of ctVSD is recognized by Arg95 of PA domain, which is conserved throughout the VSR isoforms. Therefore, the binding site of VSR appears to be attuned to a C-terminal region rather than an internal sequence motif.

Based on the crystal structure of PA domain, Luo et al. hypothesized that the internal NPIR peptide recognition by AtVSR1 is established by two separate domains: the three residues preceding the NPIR-binding motif in the PA domain and the NPIR-binding motif in the TRX domain4. As a mechanism to bring two motifs together of two separated domains from a same AtVSR1 molecule, the authors pointed out the 180° swing motion triggered by the three-residue peptide binding to the PA domain4.

Being estimated from our crystal structure of the luminal domain, the shortest distance between the ctVSD-binding site of the PA domain and the either intra- or inter-molecular TRX domain is approximately 30 Å, which is too far for an NPIR peptide to bind (Suppl. Fig. S6). If the domain-swapped trimer presented here exists in vivo and the NPIR-binding site is located in the TRX domain, it is still unlikely that the ctVSD-binding site at the PA domain is involved in ssVSD recognition.

The plausible function of the TRX domain

The TRX domain observed in our AtVSR1 structure shows a typical fold observed among protein disulfide isomerase, PDI-fold26, featuring a four-helix bundle attached between the classic TRX fold of β1α1β2 and β3β4α2 topologies (Suppl. Figs. S1 and S7). As noticed in many PDIs, β4-strand in the TRX domain of AtVSR1 is not prominent. The helical bundle observed in the TRX domain of AtVSR1 has two disulfide bonds; one between helices α1 and α4, and another between helices α2 and α3. There is a 40-residue-long insertion unique to the TRX domain between β1α1β2 and the helical bundle, which features an α-helix and a 310 helix with two disulfide bonds (Suppl. Fig. S7). The archetypal TRX-fold containing PDI enzymes have a dithiol CxxC active site at the beginning of α1 to catalyze the reduction of a disulfide bond. Superposition of the TRX domain with the PDBs retrieved by the DALI search shows that the dithiol CxxC motif is structurally conserved among those PDIs, even though AtVSR1 has an additional residue between the two conserved cysteines as 198CxxxC202 (Suppl. Fig. S7).

Some of the proteins that are sorted by VSR1 contain disulfide bonds. For example, cruciferin 1 and 3, which are 12S globulins in Arabidopsis, have two and one disulfide bond, respectively. Additionally, aleurain also has two disulfide bonds. Thus it is tempting to speculate that VSR1 engages in the quality control of the cargo proteins with disulfide before or during the sorting process. Furthermore, the formation of vacuoles in seed storage is a rapid process, which is typically completed within 24–48 h27. Rapid synthesis of the storage proteins may compromise protein folding, and the TRX domain of VSR1 may function as a secondary mechanism to ensure folding of proteins with proper disulfide bond configurations.

An intriguing aspect of TRX domains in general is their ability to recognize protein substrates. Whether the TRX domain of VSR1 is an active enzyme or nature is simply re-purposing the fold28 for target (ssVSD) binding needs further investigation.

Ca2+ and pH dependence of the EGF-like domain

Ca2+-coordination property of the EGF-like domains is achieved by 5 residues, 3 of which through conserved side-chain carboxyl/carboxamide and 2 through the main-chain carbonyls29,30,31. Comparison of the EGF-like domain of AtVSR1 to known Ca2+-binding EGF-like domains show those residue positions are conserved, suggesting the EGF-like domain of VSR1 is likely to bind a Ca2+ (Suppl. Fig. S8). Deleting the EGF-like domain in PV72, which is a homolog of AtVSR1 from pumpkin seed, displayed a tenfold decreased affinity for the NPIR peptide compared to the full length. Furthermore, the binding of the NPIR peptide of PV72 is dependent on Ca2+, supporting the significance of Ca2+-binding motif32.

Multiple studies have consistently shown pH-dependent cargo binding and unloading of VSRs. In a recent study, it was demonstrated that the NHX5 and NHX6 antiporters play a critical role in maintaining proper pH homeostasis in vacuoles, and VSR-cargo binding and trafficking rely on pH homeostasis maintained by the two vesicular antiporters6. The findings further highlight the importance of pH-dependent receptor-cargo interactions in protein trafficking and provide insights into the role of NHX antiporters in regulating VSR function.

Noticeably, the interaction between the TRX and EGF-like domains is held together by 6 H-bonds, 3 of which are salt bridges (Suppl. Table S1). The pKa values of Glu and Asp are 4.4 and 4.0, respectively. Therefore, at low pH, the strength of the salt bridges is likely to be reduced significantly. Likewise, binding of Ca2+ to the EGF-like domain is expected to cause a conformational change in the domain, which may influence the interaction with the TRX domain. Further study is needed to fully understand the involvement of the TRX and EGF-like domains, as well as the role of the salt bridges involved, in NPIR peptide binding.

Conclusion

Decades ago, when we embarked on determining the structure of AtVSR1 generated from the insect cell expression system, we anticipated that it would provide answers to questions regarding the protein’s function. The structure of the full-length luminal domain of AtVSR1 with glycosylation indeed provides fascinating insights into its function. It also raises more intriguing questions that require further investigation. Based on our current structural study, we argue whether the proposed model by Luo et al. accurately represents the physiological condition4, that is whether the ctVSD and ssVSD sites are in the PA and TRX domain in tandem at AtVSR1. We instead speculate that the ssVSD binding might be present near the TRX and EGF-like domain interface. This is due to the fact that ssVSD binding is affected by pH and Ca2+ ion, and the interaction between the TRX and EGF-like domain is dependent on multiple H-bonds. The Ca2+ binding site of the TRX domain may serve as a pH/Ca2+ sensor. Further research is needed to elucidate the specific region responsible for ssVSD binding in AtVSR1. Furthermore, whether the TRX domain of AtVSR1 functions as a disulfide bond isomerase needs to be investigated. The PFAM search did not find any other examples of a receptor protein having a chaperone domain. The VSRs may be a truly unique receptor system if the TRX domain functions as disulfide bond isomerase. Given the rapid nature of vacuolar sorting in cells, it is possible that quality control is not given high priority. In this regard, VSRs may serve a dual role of binding cargo and also playing a catch-up function of cargo refolding during the sorting process. Future research should investigate this unique plant system to enhance our understanding of the perspective it offers to all life forms.

Methods

Protein expression and purification

Information regarding cloning and protein production can be found in the literature2,3,14. In brief, the plasmid containing the luminal domain of AtVSR1 with a C-terminal 6XHis-tag was transfected into Drosophila S2 cells with the Drosophila Expression System kit. Stably transformed cell lines were selected with hygromycin and transferred to serum-free medium over a month period. Expression of the recombinant proteins was induced with 0.5 mM copper sulfate for 72 h. AtVSR1 was purified from S2 cell medium using His affinity column followed by a proaleurain peptide-affinity column.

Crystallography

The details of the crystallization method and data collection information were disclosed by Rogers et al.14. However, the 2.6 Å dataset featured in the paper was lost, and a dataset from a crystal diffracted to 3.5 Å Bragg spacings was used in this study and was successfully phased by molecular replacement using the PA domain of AtVSR1 (PDB ID 4TJV) as a search model33. The crystal contained one molecule in the asymmetric unit. The top solution of the MR search had a translation function z-score (TFZ) of 32.6, and the initial phase map showed boundaries for two additional domains as having positive density, all leading to the conclusion that the MR solution was correct. The additional model building was done with Coot, and the refinements were done in Phenix-refine34,35. Even at 3.5 Å resolution, most side chain electron densities were legible, and therefore the residues were placed in position. The final structure had a total of 427 residues with N-glycosylation at Asn289. Rwork and Rfree were 18.03%, and 21.60%, respectively. Regularization of the coordinate for structural analysis was done in the Maestro protein preparation workflow (Schrödinger, LLC). Structural figures were created with PyMol (Schrödinger, LLC). The data processing and refinement statistics are presented in Suppl. Table S2. The coordinate has been deposited (PDB ID 9B1R).