Biochemical Journal

Research article

Specific amino acids in the BAR domain allow homodimerization and prevent heterodimerization of sorting nexin 33

Bastian Dislich , Manuel E. Than , Stefan F. Lichtenthaler


SNX33 (sorting nexin 33) is a homologue of the endocytic prote-in SNX9 and has been implicated in actin polymerization and the endocytosis of the amyloid precursor protein. SNX33 belongs to the large family of BAR (Bin/amphiphysin/Rvs) domain-containing proteins, which alter cellular protein trafficking by modulating cellular membranes and the cytoskeleton. Some BAR domains engage in homodimerization, whereas other BAR domains also mediate heterodimerization between different BAR domain-containing proteins. The molecular basis for this difference is not yet understood. Using co-immunoprecipitations we report that SNX33 forms homodimers, but not heterodimers, with other BAR domain-containing proteins, such as SNX9. Domain deletion analysis revealed that the BAR domain, but not the SH3 (Src homology 3) domain, was required for homodimerization of SNX33. Additionally, the BAR domain prevented the heterodimerization between SNX9 and SNX33, as determined by domain swap experiments. Molecular modelling of the SNX33 BAR domain structure revealed that key amino acids located at the BAR domain dimer interface of the SNX9 homodimer are not conserved in SNX33. Replacing these amino acids in SNX9 with the corresponding amino acids of SNX33 allowed the mutant SNX9 to heterodimerize with SNX33. Taken together, the present study identifies critical amino acids within the BAR domains of SNX9 and SNX33 as determinants for the specificity of BAR domain-mediated interactions and suggests that SNX9 and SNX33 have distinct molecular functions.

  • Alzheimer's disease
  • BAR domain
  • endocytosis
  • molecular modelling
  • protein dimerization
  • sorting nexin


BAR (Bin/amphiphysin/Rvs) domain-containing proteins have been implicated in a variety of cellular functions, such as endocytosis, protein trafficking, cell polarity, regulation of the actin cytoskeleton, signal transduction, tumour suppression, learning and memory [13]. The BAR domain is 250–280 amino acids long and was named after the founding members of this family, Bin1, amphiphysin 1 and Rvs167. Most BAR domains consist of a three-helix bundle, which can dimerize and form a crescent-shaped structure. The positively charged concave surface of this dimeric structure senses and induces membrane curvature by binding to curved negatively charged membranes [3,4].

The BAR domain protein family includes several members of the SNX (sorting nexin) family, SNX1 and SNX2, SNX4–SNX9, SNX18, SNX30, SNX32 and SNX33 [5]. SNXs are a family of 33 cytosolic and membrane-associated proteins characterized by the presence of a SNX-type PX (Phox homology) domain, which is a subgroup of the phosphoinositide-binding PX domain superfamily [57]. In addition to the PX domain, SNXs may contain additional lipid or protein interaction domains. Few SNXs have been functionally studied, but are generally assumed to be involved in endosomal trafficking [7]. SNX9 (also known as SH3PX1), SNX18 and SNX33 form the SNX9-subfamily of SNXs and share the same domain structure. An N-terminal SH3 (Src homology 3) domain is followed by a variable linker region, the PX domain and the C-terminal BAR domain (Figure 1a). Among the three proteins, SNX9 has been best studied. It is involved in endocytosis and actin assembly and appears to couple actin dynamics to membrane remodelling during the endocytic process [8,9]. Through its N-terminal SH3 domain it binds to several different cellular proteins, such as the endocytic GTPase dynamin and WASP (Wiskott–Aldrich syndrome protein) [811].

Figure 1 Homodimerization of SNX33 requires its BAR domain

(a) Schematic domain structure of SNX33 and deletion mutants. (b) HEK-293 cells stably expressing FLAG–SNX33 (FLAG33) were transiently transfected with HA-tagged full-length SNX33 (33HA), or truncated SNX33 variants lacking the N-terminal SH3 domain (ΔSH3HA) or the C-terminal helix of the BAR domain (ΔBARH3HA). Aliquots of the cell lysates (lower two panels) show that ΔSH3HA is expressed at slightly lower levels than full-length SNX33 and ΔBARH3HA. HA-tagged proteins were immunoprecipitated (IP) and detected by anti-HA antibody (second top panel) or stained for co-immunoprecipitated FLAG–SNX33 (uppermost panel). 33HA and ΔSH3HA co-immunoprecipitate FLAG–SNX33. (c) HEK-293/FLAG–SNX33 cells (FLAG33) were transiently transfected with control vector or SNX33–HA. FLAG–SNX33 was immunoprecipitated (IP) and tested for co-immunoprecipitated SNX33–HA (uppermost panel). (d) HEK-293 cells stably expressing SNX33 lacking its SH3 domain (ΔSH3FLAG) were transiently transfected with constructs as in (c). Co-immunoprecipitation with ΔSH3FLAG was observed for full-length SNX33 (33HA) and ΔSH3HA, but not for ΔBARH3HA. (bd) Representative Western blots of three independent experiments.

The function of SNX18, and SNX33 is less well understood than that of SNX9. SNX18 appears to have a similar trafficking function as SNX9 and also binds dynamin [12]. SNX33 has been implicated in the endocytosis and processing of the amyloid precursor protein and the prion protein [13,14] as well as in actin polymerization [15]. In agreement with these findings, SNX33 binds through its SH3 domain to dynamin and WASP [12,14,15]. Additionally, SNX33 binds the metalloprotease ADAM15 (a disintegrin and metalloproteinase 15), but the physiological function of this interaction remains to be established [16,17]. SNX33 also appears to form homodimers [15], but it remains unknown whether this occurs through the SH3 domain, the BAR domain or both. Both domains can mediate protein homodimerization. Although the SH3 domain is well known to bind to proline-rich regions in target proteins and thereby link distinct proteins, it can also form homodimers, for example in the proteins IB1 and IB2 (also known as JIP1 and JIP2) and the tyrosine kinase Csk (C-terminal Src kinase) [18,19]. Additionally, BAR domains form dimers [4,20] and can mediate the homodimerization of the corresponding full-length proteins, such as amphiphysin 1 and SNX9 [21,22].

BAR domains are not only able to form homodimers, but in some cases also heterodimers between distinct BAR domain-containing proteins. This has been described for the close homologues SNX1 and SNX2 or amphiphysin1 and amphiphysin 2 [23,24]. Heterodimerization has also been observed between more distant BAR domain family members, such as SNX4 and amphyphysin 2 [25]. Heterodimerization may increase the functional versatility of the corresponding proteins, as the heterodimers may have different functions or subcellular localizations than the homodimers. However, the molecular mechanisms, which determine whether a BAR domain is able to form heterodimers, remain unknown. In order to address this question we studied SNX33. Previously, conflicting results have been reported for a potential heterodimerization of SNX33 with its homologue SNX9. One study did not find evidence for SNX9–SNX33 heterodimers in HeLa cells expressing the endogenous proteins [12]. This contrasts with another study reporting that transiently overexpressed SNX9 and SNX33 do form heterodimers in HEK-293 (human embryonic kidney 293) cells [15]. In the present study, using molecular modelling and mutational analysis we show that critical amino acids within the BAR domain of SNX33 determine the specificity of the interaction with other BAR domains. We find that the BAR domain allows SNX33 homodimer formation, but prevents heterodimerization with the BAR domains of the SNX33 homologues SNX9 and SNX18, and the more distant homologue SNX1.


Reagents and antibodies

The following antibodies were used: anti-HA (haemagglutinin) HA.11 (Covance), anti-FLAG FLAGM2 (Sigma), HRP (horseradish peroxidase)-coupled goat anti-mouse and anti-rabbit (Promega), and Alexa Fluor® 555-coupled anti-mouse (Molecular Probes).

Plasmid construction

SNX33 and SNX9 cDNAs have been described previously [14]. SNX1 and SNX18 cDNAs were obtained from A.T.C.C. cDNAs without UTRs (untranslated regions) and with C- or N-terminal fusions to HA- or FLAG-tag and/or lacking the SH3 and LC domains (PX-BAR) or part of the BAR domain (ΔBARH3) were generated by PCR and cloned into the vector pEAK12 (SH3 region corresponding to amino acids 1–60 in SNX33, PX-BAR region corresponding to amino acids 159–574 in SNX33 and 185–595 in SNX9, the last helix of the BAR domain corresponding to amino acids 510–574 in SNX33 and 531–595 in SNX9). In order to generate SNX33BAR9HA and SNX9BAR33HA, an EcoRV site was introduced before the start of the BAR domain in SNX33 (L371I) and SNX9 (L392I) by PCR. The obtained constructs (SNX33L371IHA and SNX9L392IHA) were subsequently used to swap the BAR domains by EcoRV digestion and cloning into the vector pEAK12. 9mod-HA was generated by PCR in multiple rounds and the resulting fragments were cloned into the vector pEAK12 via triple ligation (see Table 1 for individual mutations). The identity of all constructs obtained by PCR was confirmed by DNA sequencing.

View this table:
Table 1 Stabilization and conservation of the SNX9–SNX33 dimer interface

Cell culture, Western blot analysis and immunoprecipitation

HEK-293 [EBNA (Epstein–Barr virus nuclear antigen)] and HeLa cells were cultured as described previously [26]. HEK-293 cells stably expressing FLAG–SNX33 were generated using plasmids pEAK12-FLAGSNX33 using 0.5 μg/ml puromycin (Sigma). Transfections were performed using Lipofectamine™ 2000 (Invitrogen). At 1 day after transfection, cell lysates (in 50 mM Tris/HCl, pH 7.5, 150 mM NaCl, 2 mM EDTA and 1% Nonidet P40) were collected and analysed as described previously [2729]. Phosphatase inhibitors [50 mM NaF, 1 mM NaVO4 and phosphatase inhibitor (1:100; Sigma)] were added to cell lysates. The protein concentration in the cell lysate was measured and corresponding aliquots of lysate was separated by SDS/PAGE. For SNX33/SNX9 co-immunoprecipitation, lysates were incubated with 5 μg of antibody (HA.11, 1:100) for 2 h (4 °C) using Protein G dynabeads (Dynal). After washing with STEN-NaCl [STEN buffer (0.05 M Tris/HCl, pH 7.6, 0.15 M NaCl, 2 mM EDTA and 0.2% Nonidet P40) plus 0.35 M NaCl] and twice with STEN, bound proteins were resolved by SDS/PAGE. Western blots were quantified using the luminescent image analyser LAS-4000 (Fujifilm).


HeLa cells were plated on poly-L-lysine-coated glass coverslips and transfected 24 h later with SNX33 and SNX9 deletion constructs. Medium was changed 5 h after transfection. At 16 h after the medium change, cells were washed in PBS, fixed for 20 min in 4% paraformaldehyde/sucrose, quenched for 2 min with 50 mM NH4Cl and washed with PBS. Cells were permeabilized with 0.1% saponin in 8 mM Pipes, 0.5 mM EGTA, 0.1 mM MgCl2 and 2% BSA. Then cells were stained with HA.11 (1:1000), Alexa Fluor® 488 Phalloidin (6.6 μM, Invitrogen), washed with 0.05% saponin in PBS and incubated with Alexa Fluor® 555-conjugated secondary antibody (1:500). Cells were then washed in 0.05% saponin in PBS and water and fixed with Moviol. Fluorescence was imaged using a Zeiss LSM 510 Meta inverted confocal microscope, equipped with Zeiss LSM software and a Plan Apochromat 100× lens. Expression levels of individual cells were monitored using the imaging processing software Fiji (

Molecular modelling

The atomic structure of SNX9 dimers was analysed with the PROTORP server [30] and molecular graphics [31] using its published crystal structure (PDB code 2RAI) [20]. Amino acid differences to SNX33 and SNX18 and their effect on the formation of potential heterodimers were evaluated manually, thereby identifying key residues of the interface for mutational analysis. Figures were created using PyMOL (DeLano Scientific; The analysis of the potential SNX18–SNX33 heterodimer relies solely on the assumption that amino acid residues pairing in a sequence alignment between SNX18 or SNX33 with interface residues of SNX9 do also form the interface in SNX18 and SNX33.


Homodimerization and membrane tubulation requires the SNX33 BAR domain

First, we tested whether SNX33 is able to form homodimers. To this aim, HEK-293 cells stably expressing FLAG-tagged SNX33 were used. They were transiently transfected with SNX33–HA. SNX33 was present as a doublet band in the immunoblot at approx. 75 kDa (Figure 1b), which represents the phosphorylated (upper band) and non-phosphorylated (lower band) form of SNX33, as we demonstrated previously [14]. Immunoprecipitation of SNX33–HA from the cell lysate co-precipitated FLAG–SNX33 (Figure 1b). The immunoprecipitation was also possible in the opposite way; immuno-precipitation of FLAG–SNX33 co-precipitated SNX33–HA (Figure 1c). This demonstrates that SNX33 is able to homodimerize. Next, we analysed whether the homodimerization of SNX33 is mediated by the SH3 domain, by the BAR domain or by both. Some cytosolic proteins, such as IB1 and Csk dimerize through their SH3 domains [18,19]. Other proteins, such as several BAR domain-containing proteins, homodimerize through their BAR domains [21,22]. To test the involvement of the SH3 and the BAR domains, two SNX33 mutants were used and tagged with an HA epitope tag. One mutant lacked the N-terminal SH3 domain (ΔSH3). A second mutant lacked the third helix at the C-terminus of the BAR domain (ΔBARH3) (Figure 1a). This truncation may induce a misfolding and consequently a loss of function of the BAR domain, as shown previously for SNX9 [22]. A loss of function was indeed found in a membrane tubulation assay for the ΔBARH3 mutant (see below in Figure 2). Similar to full-length SNX33, the mutant lacking the SH3 domain (ΔSH3) co-immunoprecipitated FLAG–SNX33, whereas ΔBARH3 did not, although it was expressed at similar levels as SNX33 and ΔSH3 (Figure 1b). This experiment demonstrates that the homodimerization of SNX33 requires the intact BAR domain, but not the SH3 domain, of SNX33. This was confirmed in a second experimental setting. HEK-293 cells were used, which stably express SNX33 lacking the SH3 domain (ΔSH3–FLAG). As in the experiment with the full-length FLAG–SNX33 (Figure 1b), SNX33 or the two deletion mutants ΔSH3 and ΔBARH3 were transiently transfected and tested for co-immunoprecipitation with the stably expressed ΔSH3–FLAG. Again, co-immunoprecipitation was only observed for the full-length SNX33 and ΔSH3, but not for the mutant with the truncated BAR domain (Figure 1d). Taken together, these experiments demonstrate that the homodimerization of SNX33 requires the intact BAR domain, but not the SH3 domain.

Figure 2 Membrane-tubulating activity of SNX33 requires the intact BAR domain

(a) Constructs encoding the HA-tagged PX and BAR domains of SNX33 and SNX9 were expressed in HeLa cells (33PXBAR, 9PXBAR). As a control, an unrelated cytosolic protein (HA-tagged protein kinase Cα; PKCalphaHA) as well as the empty vector were expressed. Cells were co-stained with phalloidin in order to visualize the actin cytoskeleton. Immunofluorescence analysis revealed membrane tubule formation, which was abolished when the BAR domains lacked the C-terminal third helix of SNX33 and SNX9 (PXΔBARH3). The control protein PKCalphaHA showed diffuse cytosolic staining and no tubulation, similar to the PXΔBARH3 constructs. (b) Expression levels of individual cells were monitored using the imaging processing software Fiji. Only cells exceeding a predetermined signal intensity threshold (i.e. a mean grey value above background) were counted. The numbers above the lines indicate whether the construct derives from SNX9 or SNX33. PXB, PXBAR construct; PXΔBH3, PXΔBARH3 construct. (c) Cells were scored for the presence or absence of membrane tubule formation. (d) On the basis of the expression level of HA-tagged constructs the cells were divided into high-, medium- and low-expressing cells (based on the signal intensity measured in b). The percentage of cells with observed membrane-tubule formation within each category is shown. (bd) Means±S.D. of three independent experiments. (e) Western blot analysis of transfected HeLa cells showing similar expression levels of all four SNX9 and SNX33 constructs (with HA antibody) as well as of actin.

Dimerization of BAR domains generates the crescent-shaped structure, which is required for the membrane-binding and -tubulating activity of BAR domains [4]. Having found that the ΔBARH3 deletion mutant of SNX33 was not able to dimerize, we next tested whether it had also lost its membrane-tubulating activity. The intact BAR domains of SNX33 and SNX9 induced membrane tubulation in approx. 15% of HeLa cells when they were expressed together with their PX domains (SNX33 PX-BAR and SNX9 PX-BAR) (Figures 2a–2c), in agreement with previous reports [12,32]. The observed membrane-tubulating activity correlated with the expression level of the PX-BAR domain (Figure 2d). However, as expected, the deletion of the C-terminal helix (ΔBARH3) of SNX33 or SNX9 completely abolished the membrane-tubulating activity (Figures 2a–2c), demonstrating that the truncation of the BAR domain results in a loss of its membrane-tubulating activity in addition to the loss of its homodimerization capability (Figure 1b).

SNX33 does not form heterodimers

Next, we analysed whether SNX33 is able to form heterodimers with other BAR domain-containing proteins, in particular with its homologues SNX9 and SNX18, but also with the more distantly related SNX1. As a positive control, SNX33 was used. All four proteins were transiently expressed as HA-tagged proteins in HEK-293/FLAG–SNX33 cells and probed for co-immunoprecipitation with FLAG–SNX33. In contrast with SNX33, neither SNX1, nor SNX9 or SNX18 co-immunoprecipitated significant amounts of FLAG–SNX33, although all four proteins were expressed at similar levels (Figure 3a). This clearly demonstrates that SNX33 does not form heterodimers with its homologues SNX9 and SNX18 or with SNX1. The only condition where we observed heterodimer formation between SNX33 and SNX9 was upon strong transient overexpression of both proteins (Figure 3b). However, this interaction was much less than the co-immunoprecipitation observed for SNX33 (Figure 3b). We conclude that under conditions where SNX9 and SNX33 are only mildly overexpressed, there is no heterodimerization of both proteins, similar to what has been reported for both proteins expressed at endogenous levels [12]. What prevents the heterodimerization between SNX33 and the other proteins? In view of the finding that the BAR domain is required for the homodimerization of SNX33, we speculated that the BAR domain of SNX33 may also determine the specificity of interactions with other BAR domains. To test this possibility, we made domain swap experiments in which the BAR domains of SNX9 and SNX33 were exchanged (Figures 3c and 3d). Both proteins were expressed with an HA-epitope tag. In contrast with wild-type SNX33, an SNX33 mutant carrying the SNX9 BAR domain (33BAR9) had lost the ability to co-immunoprecipitate wild-type FLAG–SNX33, similar to the SNX33 mutant with a truncated BAR domain (ΔBARH3) (Figure 3d). Conversely, the SNX9 mutant carrying the SNX33 BAR domain (9BAR33) was able to interact with FLAG–SNX33. This demonstrates that the BAR domain determines the specificity of the interaction with other BAR domains. The mutant 9BAR33 co-immunoprecipitated less FLAG–SNX33 than the wild-type SNX33, but was also expressed at lower levels than wild-type SNX33 (Figure 3d, lower panel). This indicates that 9BAR33 is likely to be as efficient as wild-type SNX33–HA in forming dimers with FLAG–SNX33. As a control experiment, one more mutant of SNX9 and SNX33 was tested. For the exchange of the BAR domains, an EcoRV restriction site had been introduced into the cDNAs at the position where the BAR domain codons started. This resulted in a single amino acid change for SNX33 (L371I) and for SNX9 (L392I). Both mutants SNX33L371I and SNX9L392I were tested for co-immunoprecipitation with FLAG–SNX33 and showed the same result as the corresponding wild-type proteins (Figure 3d). This demonstrates that the single point mutations did not affect the binding behaviour to SNX33. Taken together, the BAR domain swap experiments for SNX9 and SNX33 demonstrate that the BAR domains control the specificity of the BAR domain dimerizations and prevent the heterodimerization between SNX9 and SNX33.

Figure 3 SNX33 forms homodimers, but not heterodimers

(a) HEK-293 cells stably expressing FLAG–SNX33 (FLAG33) were transiently transfected with HA-tagged constructs encoding SNX1, SNX9, SNX18 or SNX33. Co-immunoprecipitation was analysed as described in Figure 1(b). FLAG–SNX33 co-immunoprecipitated with SNX33, but not with SNX1, SNX9 or SNX18. (b) HEK-293 cells were transiently co-transfected with FLAG–SNX33 and either the empty vector, HA-tagged SNX33 or HA-tagged SNX9. Transient overexpression of both interaction partners leads to artificial co-immunoprecipitation of FLAG–SNX33 with SNX9–HA, which is not observed if only one of the interaction partners is stably expressed at lower levels (compare with a). (c) Domain structure of BAR domain swap constructs. SNX33 domains are shown in white, SNX9 domains are indicated in black. (d) FLAG–SNX33-expressing HEK-293 cells were transiently transfected with HA-tagged constructs encoding full-length SNX33 (33HA), SNX33 lacking the C-terminal helix of its BAR domain (ΔBARH3), SNX33 carrying the BAR domain of SNX9 (33BAR9), SNX9 carrying the BAR domain of SNX33 (9BAR33) or SNX33 and SNX9 carrying the indicated leucine-to-isoleucine residue point mutation (33L371I, 9L392I). SNX33 carrying the SNX9 BAR domain (33BAR9) did not co-immunoprecipitate with FLAG–SNX33, whereas 9BAR33 was able to co-immunoprecipitate. (a, b and d) Representative Western blots of three independent experiments. IP, immunoprecipitation.

Mechanism of prevention of heterodimer formation

In order to determine why the BAR domains of SNX9 and SNX33 are not able to heterodimerize, molecular graphics and modelling were used. The known crystal structure of the dimerized SNX9 BAR domain [20] was compared with SNX33, for which a crystal structure is not available. Within their BAR domains SNX9 and SNX33 are 36.3% identical (74 out of 204 amino acids) and share an even larger number of similar amino acid residues, strongly suggesting that the overall fold of the BAR domain of SNX9 is conserved in SNX33. A detailed analysis of the large dimer interface of SNX9 [~3000Å2 (1Å=0.1 nm) buried accessible surface area per protomer] using the PROTORP server [30] was carried out. Mapping of the amino acid conservation between SNX9 and SNX33 to the molecular interface (Figures 4a and 4b) revealed that the dimer interface region shows a degree of conservation (23 out of 64 residues, 35.9%) similar to that observed for the whole BAR domain. This lack of a higher degree of conservation at the dimer interface is typical for molecules that evolved independently from one another. Correspondingly, no evolutionary pressure forced the dimer interface to retain the ability for heterodimerization between SNX9 and SNX33. These findings are in excellent agreement with the lack of co-immunoprecipitation between SNX9 and SNX33 (Figure 3a) and indicate that both proteins may have different biological functions.

Figure 4 Molecular analysis and modelling of the SNX9–, SNX33– and SNX9mod–SNX33 BAR domain dimers

Conservation of amino acid residues within the BAR-domain dimer of SNX9 (a), SNX33 (b) and the mutant SNX9mod (c) carrying 19 SNX33-like mutations. The atomic co-ordinates were always taken from the PDB entry 2RAI [20]. Molecule B (MolB) is shown as a cartoon (red) in front of the solid surface representation of molecule A (MolA, yellow and blue). The surface patch of molecule A that is buried upon dimer formation is shown in blue. Panel (b) is rotated with respect to (a) by 180 ° around a horizontal axis. In (a) and (b), light colours indicate residues that are conserved between SNX9 and SNX33, whereas dark colours indicate exchanges across this comparison. (c) The comparison between the mutated SNX9-variant SNX9mod and SNX33, exhibiting a significantly larger surface area that is amenable to productive interactions (light blue/light red). (d) Close-up of the SNX9 dimer, showing the amino acid residues contributing to the strong, bifurcated Glu579–Arg586–Glu583 charged hydrogen-bonding network with all atoms in stick representation. Oxygens and nitrogens are shown in red and blue respectively, whereas the carbons and the cartoon representation of the helices are shown in orange and green for molecules A and B (MolA and MolB) respectively. Hydrogen bonding distances are given in Å. Glu583 does not directly participate in the dimer interface but is essential for this hydrogen-bond network. This Figure was rotated about a horizontal axis by ~90 °C with respect to (a), thus looking from the bottom at (a). (e) HEK-293 cells stably expressing FLAG–SNX33 (FLAG33) were transiently transfected with HA-tagged constructs encoding SNX33 (33HA), SNX9 (9HA) or SNX9 with a modified BAR domain (9modHA). Co-immunoprecipitation was analysed as described in Figure 1(c). FLAG–SNX33 co-immunoprecipitated with SNX33 and to a lower degree with 9modHA, but not with SNX9. IP, immunoprecipitation.

A deeper analysis of the interface showed that many strong intermolecular salt bridges, hydrophobic interactions and hydrogen-bonding networks are responsible for the formation of tight SNX9 homodimers (Table 1). Out of the 64 residues that line the symmetrical interface in each protomer, 24 each contribute to more than 2% of the total molecular interface area, making a total of more than 50% of the interface area. Across the interface, 24 hydrogen bonds are formed. A total of eight tight salt bridges link the two protomers further together, including the bifurcated Glu579–Arg586–Glu583 charged hydrogen-bonding network (Figure 4d). Many of the amino acids at the dimer interface of SNX9 are not conserved in the SNX33 sequence (Table 1). As a consequence, many of the interactions between the two BAR domains would be disrupted in the case of a heterodimerization between SNX9 and SNX33, providing a molecular explanation for the observed lack of heterodimerization between both proteins. In silico analysis was also carried out for the potential interaction between SNX9 and SNX18 as well as between SNX33 and SNX18 (Supplementary Tables S1 and S2 at In agreement with a previous study [12] and the results from the present study (Figure 3a), a heterodimerization of SNX9–SNX18 and SNX33–SNX18 does not appear possible due to the lack of conserved amino acids at the dimer interface. To test further the validity of the in silico analysis, the interaction between rat amphiphysin 1 and amphiphysin 2 was analysed (Supplementary Table S3 at In contrast with the SNXs, both amphiphysins have been shown to heterodimerize through their BAR domains [24]. In agreement with the experimental data, the in silico analysis revealed that the residues at the dimer interface are either conserved between both proteins or replaced by structurally tolerated amino acids.

The in silico analysis of the BAR domains of SNX9 and SNX33 suggests that it should be possible to induce heterodimerization between a modified SNX9 and SNX33, if relevant amino acids in the BAR domain of SNX9 are replaced by the corresponding amino acids of SNX33. To test this hypothesis, 19 amino acid residues, each making a major contribution to the dimer interface in SNX9, but not being conserved in SNX33, were mutated to their counterparts in SNX33 (see Table 1). Indeed, this SNX9 variant with the modified BAR domain (9mod) was able to co-immunoprecipitate SNX33 (Figure 4e), showing that the chosen amino acids were critical for dimer formation. The observed reduction in interaction strength between FLAG–SNX33 and SNX9mod as compared with either wild-type SNX33 or SNX9 homodimers shows that for an optimal BAR domain dimerization, the many smaller alterations across the BAR domain interface are also necessary. The introduced mutations significantly increase the number of compatible surface patches within the larger interface, but a significant number of ‘non-conserved’ residues still exist for the SNX33–SNX9mod dimer (Figure 4c).

To clarify further whether a heterodimer between SNX33 and SNX9mod can exist, we estimated the dissociation constant (Kd) for the SNX33–SNX9mod heterodimer in comparison with the SNX33 homodimer using a mild overexpression of the interaction partners in HEK-293 cells. First, the Kd of the SNX33 homodimer was estimated in relation to the SNX9 homodimer, which has a reported Kd of 7.9 μM [33]. The amount of precipitated SNX33–FLAG after immunoprecipitation of SNX33–HA was in the same range, but slightly higher, than the amount of SNX9–FLAG precipitated by SNX9–HA (Figures 5a and 5b). Densitometric quantification revealed ~2.5-fold higher levels of SNX33 dimer compared with SNX9 dimer, leading to an estimated Kd value of ~3 μM for the SNX33 homodimer. In contrast, the estimated Kd value of the SNX33–SNX9 heterodimer is ~30-fold higher (Figure 5c), based on the ~30-fold lower co-precipitation between SNX9 and SNX33 (Figure 3a). The mutations in SNX9mod increase the co-precipitation efficiency ~6-fold, leading to an estimated Kd of ~13 μM, which is ~1.5-fold higher than the Kd reported for the SNX9 homodimer [33]. This further supports our experimental finding of SNX33–SNX9mod dimers.

Figure 5 Estimation of SNX33–SNX33 and SNX33–SNX9 dissociation constants

(a) HEK-293 cells were cotransfected with FLAG- and HA-tagged SNX33 or SNX9 (33HA, 33FLAG, 9HA, 9FLAG respectively). As a control, empty vector was co-transfected with the HA-tagged SNX9/33. Co-immunoprecipitation was analysed as in Figure 1(b). IP, immunoprecipitation. (b) The relative amount of FLAG–SNX33 that co-immunoprecipitated with SNX33–HA was calculated (enrichment of FLAG-tagged constructs in HA immunoprecipitation divided by the immunoprecipitation efficiency=[(HA immunoprecipitation and Blot FLAG: loading control FLAG):(HA immunoprecipitation and Blot HA: loading control HA)]) and compared with the relative amount of SNX9–FLAG precipitated by SNX9–HA. (c) The relative amount of FLAG–SNX33 that co-immunoprecipitated with HA-tagged SNX9/SNX9mod was calculated (as in Figure 5b) and compared with the relative amount of SNX33–FLAG precipitated by SNX33–HA (compare with Figure 4e). Panels (b and c) show the means±S.D. of three independent experiments.


The present study shows that SNX33 forms homodimers but not heterodimers with its closest homologues SNX9 and SNX18. Using mutational analysis as well as molecular modelling we present a molecular mechanism by which the BAR domain allows homodimerization and prevents heterodimerization of SNX33. We expect that similar modelling approaches should predict the potential of other BAR domains for heterodimerization.

Dimerization of BAR domains generates the crescent-shaped structure, which is required for the membrane-binding and -tubulating activity of BAR domains [4]. BAR domain-containing proteins form homodimers [34], which we also found for SNX33, in agreement with a previous study [15]. Mutational analysis revealed that the BAR domain, but not the SH3 domain, is required for SNX33 homodimerization. This is in line with the BAR domain being the dimerization domain in other proteins [4,20,34]. A similar result was also reported for the SNX33 homologue SNX9, in which a truncation of the 13 C-terminal amino acids of the BAR domain resulted in loss of homodimer formation [22].

Although some BAR domains are able to form heterodimers with the BAR domain of other proteins, we find that SNX33 does not form heterodimers with its homologues SNX9 and SNX18 or with the more distant homologue SNX1. This lack of heterodimer formation is in agreement with a previous study, in which endogenous SNX9 or SNX33 were immunoprecipitated from HeLa cells, but failed to co-precipitate the other protein [12]. However, that study and our own data are in contrast with another study reporting heterodimer formation between SNX33 and SNX9 in HEK-293 cells [15]. In the present study, both SNX33 and SNX9 were transiently overexpressed. We also observed co-immunoprecipitation between SNX33 and SNX9 when both proteins were strongly overexpressed. However, the use of lower plasmid concentrations for the transfections as well as the generation of stable cell lines, expressing the target protein at lower levels, abolished the co-immunoprecipitation between both proteins. Thus we conclude that under physiological conditions there is no heterodimer formation between SNX33 and SNX9. This conclusion is further supported by mechanistic studies, including mutational analysis and molecular modelling. Swapping of the BAR domains of SNX9 and SNX33 revealed that the SNX33 BAR domain only forms dimers with the SNX33 BAR domain, regardless of whether the remaining part of the protein is of SNX9 or SNX33 origin. The molecular basis for this exclusive homodimer formation of the SNX33 BAR domain was revealed upon modelling of the SNX33 BAR domain structure in analogy to the SNX9 BAR domain structure, which was determined by X-ray crystallography [20]. In the SNX9 structure, we analysed the amino acids which form the dimer interaction interface. However, in the SNX33 sequence, several amino acids at the interface are not conserved and are exchanged to amino acids which are not compatible with an energetically favourable interaction with the BAR domain of SNX9. Indeed, mutation of several of these amino acids in the SNX9 sequence to the corresponding amino acids of SNX33 allowed the mutant SNX9 protein to heterodimerize with wild-type SNX33. This clearly shows that the wild-type BAR domains of SNX33 and SNX9 are incompatible for heterodimer formation, indicating that both proteins may have distinct cellular functions, for example by acting at different cellular membranes. This possibility is in agreement with the lack of cellular co-localization of both proteins [12]. However, SNX9 and SNX33 may act at their respective membranes by similar molecular mechanisms. SNX9 couples actin dynamics to membrane remodelling during the endocytic process [8,9]. Likewise, SNX33 has been implicated in endocytosis and actin remodelling and binds to proteins, such as dynamin, WASP and ADAM15, which are also binding partners of SNX9 [1012,1417,35].

At present, only a few BAR domain-containing proteins have been shown to form heterodimers through their BAR domains, such as the close homologues SNX1 and SNX2 and amphiphysin 1 and 2 [23,24]. But even for the two more distantly related BAR domains of SNX4 and amphiphysin 2, heterodimerization has been reported [25]. For many other BAR domain-containing proteins, it is not yet known whether they are able to form heterodimers, either with close homologues or with more distant homologues. On the basis of our modelling of the potential interaction between the BAR domains of SNX9 and SNX33, we expect that a similar modelling of the potential dimer interface should allow us to predict whether a heterodimerization between other BAR domains is possible or not. If the relevant amino acids at the dimer interface are not conserved between two distinct BAR domains, it is likely that they are not able to heterodimerize. In contrast, if the essential amino acids are conserved, a heterodimerization should be possible. This assumption was further validated by analysing the interaction among the other SNX9 family members. In silico analysis predicted that there is no heterodimerization between SNX18 and either SNX9 or SNX33. This is in agreement with our results in the present study and a previous study analysing the endogenous proteins [12]. In contrast, another recent study reported a co-immunoprecipitation between SNX9 and SNX18 [36]. However, that study used a transient transfection assay, such that the interaction may be due to the strong overexpression, as discussed above. In silico analysis also provided the correct prediction for the experimentally shown interaction between the two BAR domain-containing proteins amphiphysin 1 and 2 [24], which are unrelated to the SNX9 family. In this case, the amino acids at the dimer interface are either conserved or show conservative mutations, providing a molecular explanation for the heterodimerization between two distinct BAR domain-containing proteins.

The different examples show that the in silico analysis provides a valuable approach for the prediction of BAR domain dimerization. One limitation is that the structure of one of the two BAR domains should be known. Although distinct BAR domains may have a similar overall fold [34], the presence of non-conserved amino acids may subtly alter the structure, such that different amino acids constitute the dimer interface in different BAR domains. Amino acid conservation may not be the only factor determining successful interaction between two distinct BAR domains. Additional factors may be the length of the BAR domain monomers, the curvature of the crescent-shaped dimer structure, as well as kinks in the individual helices of the BAR domain monomers [34]. Although the extent and functional consequence of heterodimer formation between BAR domain-containing proteins remains to be determined, heterodimerization could be a means to increase the functional versatility of the corresponding proteins, as the heterodimers may have different functions or subcellular localizations than the homodimers. Additionally, heterodimerization may allow the formation of large molecular complexes consisting of the distinct binding partners of the two different BAR domain-containing proteins.


Stefan Lichtenthaler planned and guided the study, designed the experiments, analysed the data and wrote the paper; Bastian Dislich designed the experiments, performed the experiments and analysed the data; and Manual Than designed the experiments, performed the structural modelling, analysed the data and wrote the paper.


This work was supported by the Deutsche Forschungsgemeinschaft [SFB596 project B12 (to S.F.L.)], the Bundesministerium für Bildung und Forschung [project KNDD (to S.F.L.)] and the Molecular Medicine Program of the Medical School of the University of Munich (to B.D. and S.F.L).

Abbreviations: ADAM15, a disintegrin and metalloproteinase 15; BAR, Bin/amphiphysin/Rvs; Csk, C-terminal Src kinase; HA, haemagglutinin; HEK-293, human embryonic kidney 293; PX, Phox homology; SH3, Src homology 3; SNX, sorting nexin; WASP, Wiskott–Aldrich syndrome protein


View Abstract