The sorting nexins SNX1 and SNX2 are members of the retromer complex involved in protein sorting within the endocytic pathway. While retromer-dependent functions of SNX1 and SNX2 have been well documented, potential retromer-independent roles remain unclear. Here, we show that SNX1 and SNX2 interact with the Rac1 and RhoG guanine nucleotide exchange factor Kalirin-7. Simultaneous overexpression of SNX1 or SNX2 and Kalirin-7 in epithelial cells causes partial redistribution of both SNX isoforms to the plasma membrane, and results in RhoG-dependent lamellipodia formation that requires functional Phox homology (PX) and Bin/Amphiphysin/Rvs (BAR) domains of SNX, but is Rac1- and retromer-independent. Conversely, depletion of endogenous SNX1 or SNX2 inhibits Kalirin-7-mediated lamellipodia formation. Finally, we demonstrate that SNX1 and SNX2 interact directly with inactive RhoG, suggesting a novel role for these SNX proteins in recruiting an inactive Rho GTPase to its exchange factor.
The sorting nexins (SNXs) are a family of approximately 30 proteins involved in sorting and trafficking events at multiple steps within the endocytic pathway. SNXs are characterized by the presence of a Phox homology (PX) domain that binds to phosphoinositides (PIs), and PI binding specificity is important for cellular function through recruitment to appropriate membrane domains (1). Many SNXs contain additional functional domains, with the largest subfamily (12 members) containing a C-terminal Bin/Amphiphysin/Rvs (BAR) domain, which forms a crescent-shaped dimer that acts as a sensor or inducer of membrane curvature (2,3). The combination of PX and BAR domains has been proposed to act as a coincidence detection module, with PI binding via the PX domain and curvature sensing through the BAR domain acting in concert to precisely target the SNXs to specific compartments (4,5).
A number of BAR domain-containing SNXs, including SNX1, SNX2, SNX5 and SNX6, have been identified as components of the mammalian retromer complex involved in retrieval of receptors from endosomes to the trans-Golgi network (6–8). Retromer was initially identified as a five-protein complex in yeast, with Vps26p, Vps29p and Vps35p forming a cargo-recognition subcomplex, and a Vps5p/Vps17p dimer acting as an additional structural subcomplex for retrieval of Vps10p (9). With the exception of Vps17p, all components of the yeast retromer complex have mammalian orthologues. SNX1 and its closest relative SNX2, which share greater than 60% identity, have been identified as putative orthologues of Vps5p (10). The mammalian retromer complex is involved in endosome-to-Golgi retrieval of the cation-independent mannose 6-phosphate receptor, the functional equivalent of Vps10p, suggesting evolutionary conservation of retromer function (11).
While retromer-dependent functions of SNX1 and SNX2 are well established, recent studies have reported a retromer-independent role for SNX1 in promoting degradation of the protease-activated receptor PAR1 or recycling of the P2Y(1) receptor (12,13). In addition, depletion of Vps35 in Drosophila S2 cells promotes aberrant signaling and actin polymerization events (14), suggesting that SNX1, SNX2 and Vps35 may have cellular functions that are independent of retromer or that retromer might be involved in processes other than endocytic sorting.
To better understand the function of SNX2, we previously performed yeast and bacterial two-hybrid screens to identify potential interactors, and found that SNX2 interacts with the RNA helicase Abstrakt to regulate its shuttling to the nucleus (15). In these screens, we also identified Duo, the human orthologue of rat Kalirin-7 (Kal7), a guanine nucleotide exchange factor (GEF) for activation of the actin-modulating small GTPases Rac1 and RhoG, as a potential binding partner for SNX2 (16,17). Kal7 belongs to a family of multifunctional Rho GEFs that arise from a single gene by alternative splicing, and consists of a Sec14p-like lipid binding domain, nine spectrin repeats, a Dbl homology/Pleckstrin homology (DH/PH) domain responsible for GEF activity, and a C-terminal PSD-95/discs large/ZO-1 (PDZ)-binding motif (17,18). Although mainly expressed in the brain, Kalirin expression has also been reported in tissues such as the spleen and kidney, indicating possible roles in non-neuronal cells (19). Through PDZ-binding motif-mediated recruitment to the postsynaptic density and GEF activity, Kal7 mediates formation and maintenance of actin-rich dendritic spines in response to ephrin-B/EphB signals (20–22). Human Duo was initially identified as a Huntingtin-associated protein-1 (HAP1)-interacting protein (23). Although HAP1 function is not fully understood, its role in endocytic trafficking has been well established (24–26). Potential interactions of Kal7 with HAP1 or SNX2 could thus implicate Kal7 in transport through the endocytic pathway. Alternatively, interaction of SNX2 with Kal7 might provide a role for SNX2 in modulation of the actin cytoskeleton.
In this study, we report that SNX1 and SNX2 play a role in modulating Kal7- and RhoG-dependent actin remodeling that does not require retromer function. Furthermore, the BAR domain of SNXs directly interacts with inactive, GDP-bound RhoG, suggesting a novel mechanism for recruitment of an inactive Rho GTPase for subsequent activation by its GEF.
Duo/Kalirin-7 interacts with SNX1 and SNX2
We initially identified the human protein Duo as a potential SNX2-interacting protein in a yeast two-hybrid screen of an adult human brain LexA cDNA library. We screened >106 transformants and obtained approximately 200 positive clones. Two clones with insert lengths of approximately 1.6 kb spanning the second through the seventh spectrin repeats of human Duo were identified. For all further studies, we used rat Kal7, the orthologue of Duo. Notably, the region of Duo that we isolated in our two-hybrid screen is expressed in most splice variants, indicating that SNX2 might interact with other spectrin repeat-containing splice variants of Kalirin as well. However, we focused solely on Kal7 in this study because it is the shortest naturally occurring isoform that contains the entire spectrin repeat region (27).
To verify the interaction of SNX2 and Kal7, we performed pulldown experiments using His6-myc-Kal7 and Flag-tagged SNX2. Given the high degree of similarity between SNX1 and SNX2, we also tested Flag-SNX1 for interaction with Kal7. When co-transfected CHO extracts were incubated with Ni-NTA beads, both SNX1 and SNX2 specifically co-purified with Kal7 (Figure 1A). Additionally, we found that the GEF-inactive Kal7 N1415A/D1416A (ND/AA) mutant could interact with SNX1 and SNX2, suggesting that Kal7 GEF activity is not required for interaction with either SNX.
To characterize the domains of SNX2 that are required for interaction with Kal7, we generated truncated SNX2 fragments consisting of the N-terminus and PX domain (N+PX) or the BAR domain (Figure 1B). Although both fragments were readily detected in CHO extracts, they failed to co-purify with Kal7, suggesting that neither SNX2 fragment is sufficient for stable interaction with Kal7 (Figure 1A). For the full-length and truncated SNX constructs, co-expression of Kal7 WT or ND/AA did not affect the level of Flag-SNX expression. We also attempted to pull down N-terminus, PX, or PX+BAR fragments of SNX2 with Kal7; however, these truncations did not express efficiently, and could not be used for interaction studies (data not shown). Overall, these data suggest that neither the N+PX nor the BAR domain is sufficient for interaction with Kal7, and that a combination of domains found in both SNX2 fragments likely comprises the interaction surface.
Simultaneous overexpression of Kal7 and SNX promotes lamellipodia formation
We next assessed the subcellular distribution of SNX1, SNX2 and Kal7 in CHO cells. The localization of endogenous SNX1 or SNX2 in CHO cells could not be assessed, because the SNX1 and SNX2 antibodies did not detect the hamster proteins (data not shown); thus, we made use of overexpressed Flag-tagged SNX1 or SNX2 in our initial studies. As reported by others (28,29), Flag-SNX1 and SNX2 partially colocalized with EEA1-positive structures when overexpressed in the absence of exogenous Kal7 (Figure 2A). Both wild-type and GEF-inactive myc-Kal7 also showed partial overlap with EEA1, indicating that a fraction of Kal7 is recruited to early endosomes. In all cases, we did not observe a significant change in EEA1 distribution compared to untransfected cells (data not shown). While a subset of SNX-or Kal7-positive structures did not colocalize with EEA1, the nature of these structures is not yet clear and merits future study.
While both Flag-SNX1 and SNX2 distribution partially overlapped with wild-type His6-myc-Kal7 in co-transfected cells (Figure 2B), we noticed a striking change in cell morphology. Co-transfected cells displayed a lamellipodia-like and membrane ruffling phenotype (hereafter referred to only as lamellipodia), with SNX1 and SNX2 partially redistributed to the leading edge of these structures (further characterized below and in Figure 3). Within these lamellipodia-like structures, SNX1 and SNX2 partially colocalized with Kal7 WT on discrete patches at the cell surface and on vesicular structures within the cell (Figure 2B, inset). Furthermore, Flag-SNX2 and myc-Kal7 WT showed colocalization on punctae in the perinuclear region, while Flag-SNX1 appeared to localize primarily to punctae at the cell periphery (Figure 2B, refer also to Figures 3B and S1). In contrast to other studies showing that GEF activity of Kalirin is not required for lamellipodia (30), cells overexpressing Kal7 ND/AA did not form these lamellipodia-like structures, either in the presence or absence of Flag-SNX1 or SNX2 (Figure 2B). Furthermore, neither SNX redistributed to the cell periphery when simultaneously overexpressed with Kal7 ND/AA, and little colocalization of either SNX with Kal7 ND/AA was observed (Figure 2B, inset), indicating that GEF activity is required for the redistribution of SNX1 and SNX2 and for colocalization with Kal7.
Because lamellipodia are F-actin-rich protrusions at the cell periphery, we used Alexa 568-conjugated phalloidin, which binds F-actin, to confirm that the observed structures are indeed lamellipodia. In cells singly transfected with SNX1, SNX2, Kal7 WT or Kal7 ND/AA, no obvious lamellipodia were detected (Figures 3A and S1A). However, F-actin-containing lamellipodia/membrane ruffles were readily detected in cells co-transfected with either SNX1 or SNX2 and Kal7 WT, but not with Kal7 ND/AA (Figures 3B and S1B). Furthermore, both SNX1 and SNX2 partially colocalized with phalloidin-labeled F-actin at the leading edge of lamellipodia in cells simultaneously overexpressing wild-type Kal7. Neither SNX1 nor SNX2 colocalized with F-actin in the absence of Kal7 WT or when simultaneously overexpressed with Kal7 ND/AA. Interestingly, we observed F-actin staining on SNX1- or SNX2-containing cytoplasmic punctae in cells simultaneously overexpressing Flag-SNX and myc-Kal7 WT (but not ND/AA), although F-actin was not found on cytoplasmic vesicles in cells overexpressing only Flag-SNX1 or SNX2, raising the possibility that actin and/or membrane remodeling may not be restricted solely to the plasma membrane.
To verify that lamellipodia formation is indeed because of simultaneous overexpression of Kal7 and SNX, we quantified the proportion of transfected cells that displayed this phenotype in a blinded protocol. When transfected with an empty vector, only a small fraction of cells (1.5 ± 1.1%) formed lamellipodia (Figure 4). Overexpression of Flag-SNX1 or SNX2 did not significantly stimulate lamellipodia, where 6.7 ± 0.9% and 4.4 ± 1.3% of cells displayed the phenotype, respectively. Although expression of wild-type Kal7 gave a small but significant increase in the incidence of lamellipodia to 10.9 ± 0.7%, co-transfection of Kal7 with SNX1 or SNX2 strongly enhanced the phenotype, with 42.3 ± 1.7% and 55.3 ± 4.4% of cells showing lamellipodia, respectively. This phenotype is dependent on GEF activity, as only 1.2 ± 0.4% of cells overexpressing Kal7 ND/AA developed lamellipodia, and simultaneous overexpression of SNX1 or SNX2 with Kal7 ND/AA only modestly promoted lamellipodia formation (3.8 ± 0.6% and 10.2 ± 1.1% of cells, respectively). Taken together, these results suggest that SNX1 and SNX2 can promote Kal7-dependent membrane remodeling, and that this phenotype is dependent on GEF activity of Kal7.
The combined action of SNX2 PX and BAR domains contributes to efficient Kal7-mediated lamellipodia formation
In numerous SNX proteins, the PX and BAR domains are thought to form a single functional module (4,5). Thus, we examined whether the proposed coincidence detection properties of the PX-BAR module might be important for the effects observed with SNX2. For this purpose, we generated the R182A/R183A/F184A (RRF) mutant that abolishes PI binding of the PX domain (29). In addition, we also created a K426A/R428A (KR/AA) mutant of SNX2, which targets analogous basic residues known to affect tubulation or curvature-sensing properties of the SNX1 BAR domain (4). We then performed Ni-NTA pulldown experiments with His6-tagged Kal7 as above, and found that both SNX2 mutants retained their ability to interact with Kal7 (Figure 5A). However, when tested for the ability to form lamellipodia, simultaneous overexpression of myc-Kal7 with Flag-SNX2 RRF and KR/AA mutants was significantly less effective (31.3 ± 1.3% and 32.9 ± 2.6% of cells, respectively) than with wild-type SNX2 (55.1 ± 1.7%, Figure 5B). As seen for wild-type SNX2, the RRF and KR/AA mutants were unable to induce lamellipodia in the absence of myc-Kal7 WT (data not shown). We also observed that mutation of either the PX or BAR domain did not significantly alter the membrane association of SNX2 at lower expression levels, such that mutant proteins still showed a punctate distribution that partially overlapped with EEA1 (Figure S2). At higher expression levels, SNX2 mutants mislocalized to the cytoplasm, as reported by others (4,28,29). We also tested a combined RRF, KR/AA mutant SNX2, and found that loss of both PX and BAR domain function in the same molecule could not further inhibit lamellipodia induction compared to mutation of either one of the two domains alone (data not shown). Taken together, these results highlight the importance of a combinatorial contribution of the two domains in lamellipodia formation.
An alternative model system demonstrates Kal7-driven, SNX-mediated lamellipodia formation at endogenous SNX abundance
To rule out a possible non-specific effect of SNX overexpression, we sought to determine whether depletion of endogenous SNX1 or SNX2 could attenuate Kal7-mediated lamellipodia formation. Because the sequences of hamster SNX1 and SNX2 are unavailable, we instead used HeLa cells for shRNA studies. We first verified that HeLa cells also formed lamellipodia when co-transfected with wild-type Kal7 and either SNX1 or SNX2, and that their responsiveness was similar to that seen for CHO cells (Figure 6A). Although the overall responsiveness of HeLa cells was slightly lower than that of CHO (Figure 4), we observed similar trends for lamellipodia formation in both cell lines, indicating that the effect is not cell line-specific.
Our experiments showing that SNX1 and SNX2 promoted lamellipodia formation in CHO and HeLa cells when simultaneously overexpressed with Kal7 were performed using wild-type Kal7 driven by the mammalian EF1α promoter. Under these conditions, expression of Kal7 alone did not efficiently induce lamellipodia. We thus examined whether increasing the level of Kal7 overexpression under the stronger cytomegalovirus (CMV) promoter could overcome the need for SNX overexpression and allow us to directly assess the effect of SNX depletion on Kal7-mediated lamellipodia. For this purpose, we obtained a plasmid encoding Kal7 ΔCT, which lacks the C-terminal PDZ-binding motif, under the control of a CMV promoter (Dr. Betty Eipper, University of Connecticut). Compared to EF1α-driven Kal7 WT or EF1α-driven Kal7 ΔCT, cells transfected with CMV-driven Kal7 ΔCT showed a five- to sixfold increase in Kal7 expression (Figure 6B). Importantly, Kal7 ΔCT still interacted with SNX1 and SNX2 when assessed by Ni-NTA pulldown (Figure 6C). Furthermore, Kal7 ΔCT was recruited to the plasma membrane and to internal punctae (Figure S3). While EF1α-driven Kal7 WT was not seen as prominently at the cell surface in cells simultaneously overexpressing SNX1 or SNX2 (Figure 2B), this might be due at least in part to the lower expression level with this promoter. Simultaneous overexpression of SNX2 with EF1α-driven Kal7 ΔCT resulted in a Kal7 distribution that was indistinguishable from that of Kal7 WT, indicating that enhanced plasma membrane localization was not a result of truncation of the C-terminus (data not shown).
We then compared the phenotype of HeLa cells expressing EF1α-driven Kal7 WT, EF1α-driven Kal7 ΔCT or CMV-driven Kal7 ΔCT. At endogenous SNX levels, overexpression of Kal7 ΔCT under the CMV promoter induced lamellipodia more efficiently than EF1α-driven Kal7 WT or EF1α-driven Kal7 ΔCT (45.4 ± 3.4% versus 10.9 ± 3.4% or 3.5 ± 0.5% of cells, respectively, n = 3, p < 0.01, Figure 6D). Notably, EF1α-driven Kal7 WT or EF1α-driven Kal7 ΔCT gave rise to similar levels of lamellipodia induction (p > 0.05), indicating that truncation of the C-terminus is not responsible for the increase in lamellipodia seen for cells expressing CMV-driven Kal7 ΔCT. In many cells expressing CMV-driven Kal7 ΔCT, we observed dorsal ruffles in addition to lamellipodia (Figure 7C), and both phenotypes were included in the analysis because they reflect activation of Rac1 and RhoG (31,32). Interestingly, endogenous SNX1 and SNX2 partially redistributed to the leading edge of lamellipodia and to dorsal ruffles in Kal7 ΔCT-expressing HeLa cells (Figure S3), as seen for epitope-tagged SNX1 and SNX2 in CHO cells. Taken together, these data suggest that the C-terminal PDZ-binding motif is not required for proper localization or function of Kal7 in epithelial cells, nor is it required for interaction with SNX1 or SNX2.
Depletion of endogenous SNX1 or SNX2 inhibits Kal7-mediated lamellipodia formation
We infected HeLa cells with lentiviral shRNA targets directed against SNX1 or SNX2, and found that shRNA infection caused a 50% decrease in SNX1 and SNX2 levels with little to no effect on expression of the cargo-selective retromer subunit Vps35 (Figure 7A,B). We also found that depletion of SNX1 resulted in a slight concomitant increase in expression of SNX2 and vice-versa, suggesting that loss of one protein may be compensated for by upregulation of the other. Co-infection with both lentiviral targets had a weaker effect than single infections, with a 27% and 38% reduction in SNX1 and SNX2 levels, respectively, and the weaker effect arising from co-infection might arise from the upregulation effect seen for one SNX isoform when the other is depleted. Infection with a control shRNA target against GFP had no effect on expression of SNX1, SNX2 or Vps35.
We next examined the phenotype of lentiviral shRNA-infected HeLa cells expressing CMV-driven Kal7 ΔCT. Control GFP shRNA had no effect on induction of lamellipodia or dorsal ruffles compared to mock-infected cells (40.3 ± 1.8% and 38.0 ± 1.6% of cells, respectively, Figure 7D). However, depletion of SNX1 or SNX2 caused an approximate twofold reduction in the proportion of cells displaying the lamellipodia phenotype (to 23.7 ± 0.7% and 17.0 ± 1.2% of cells, respectively). Knockdown of both SNX1 and SNX2 did not result in a further decrease in incidence of lamellipodia compared to depletion of SNX2 alone (p > 0.4), but showed a significant reduction compared to depletion of SNX1 alone (p < 0.05). It is thus possible that of the two SNXs examined, SNX2 might be more effective in mediating the lamellipodia phenotype.
Lamellipodia formation does not require retromer function
Because of the well-documented role for SNX1 and SNX2 in retromer function, we next asked if Kal7-mediated lamellipodia formation requires other components of the retromer complex. We first examined the colocalization of Kal7, SNX1 and SNX2 with Vps35, the cargo-selective subunit of retromer. We found that Vps35-containing punctae were labeled with Flag-tagged SNX1 and SNX2 in CHO cells (Figure 8A). Vps35 also showed limited colocalization with overexpressed wild-type Kal7. When Kal7 and either SNX1 or SNX2 were simultaneously overexpressed, both SNX1 and SNX2 still showed partial overlap with Vps35-containing structures within the cell (Figure 8B). However, localization of either SNX to the plasma membrane in cells co-expressing Kal7 did not result in a concomitant redistribution of Vps35 to the plasma membrane. Thus, the SNX1 or SNX2 that is recruited to the leading edge of lamellipodia is unlikely to be associated with other retromer subunits.
To further assess the potential role of retromer in Kal7 function, we used lentiviral shRNA infection to deplete HeLa cells of Vps35, and determined whether a loss of retromer function could impair lamellipodia formation by CMV-driven Kal7 ΔCT. Expression levels of the three members of the cargo-selective subcomplex of retromer are closely associated such that reducing levels of Vps26, Vps29 or Vps35 destabilizes other members of the subcomplex, without affecting expression of SNX proteins (11,33). Infection with a control GFP shRNA sequence had no effect on Vps35 expression, while Vps35 shRNA caused a decrease in its expression by 80–90% without appreciably affecting levels of SNX1 or SNX2 (Figure 9A,B). When HeLa cells expressing CMV-driven Kal7 ΔCT were depleted of Vps35, 45.1 ± 2.1% of cells displayed lamellipodia or dorsal ruffles, which was statistically similar to mock- or GFP shRNA-infected cells (43.2 ± 1.8% and 42.9 ± 2.3%, respectively, p > 0.5, Figure 9C). These results indicate that Kal7-mediated lamellipodia formation is likely retromer-independent.
Induction of lamellipodia by Kal7 and SNX is a RhoG-dependent phenotype
Previous studies have shown that the DH/PH domain of Kal7 acts through both Rac1 and RhoG (16,17). Because activation of either GTPase is known to promote formation of lamellipodia, we sought to determine which GTPase is involved in the phenotype observed upon simultaneous overexpression of Kal7 and either SNX1 or SNX2. We co-transfected wild-type myc-Kal7 with Flag-SNX1 or SNX2 and either empty vector or dominant-negative RhoAT19N, Rac1T17N, Cdc42T17N or RhoGT17N, and assessed the effect of the inactive GTPases on lamellipodia phenotype. Expression of all of the dominant-negative Rho GTPases was confirmed by western immunoblot (data not shown). In cells expressing Kal7 with either SNX1 or SNX2, co-expression of dominant-negative RhoAT19N, Rac1T17N or Cdc42T17N had only a minor inhibitory effect on lamellipodia formation compared to empty vector (Table 1). Strikingly, co-expression of dominant-negative RhoGT17N potently inhibited lamellipodia in CHO cells simultaneously overexpressing Kal7 and either SNX1 or SNX2 by seven- to eightfold. We also confirmed that RhoG is expressed in CHO cells at the mRNA level by reverse transcriptase-polymerase chain reaction (RT-PCR) as an indication that endogenous RhoG is present in these cells for activation by Kal7 (data not shown). Thus, the lamellipodia phenotype induced by simultaneous overexpression of Kal7 with SNX1 or SNX2 is likely to result from activation of RhoG, with only a minor, if any, contribution from Rac1 or other Rho GTPases.
Table 1. Effect of dominant-negative Rho GTPases on lamellipodia induction in CHO cells simultaneously overexpressing myc-Kal7 and either Flag-SNX1 or SNX2
His6-myc-Kal7 WT + Flag-SNX1
His6-myc-Kal7 WT + Flag-SNX2
aValues (presented as mean ± SE) correspond to the percentage of transfected cells displaying lamellipodia, and were scored under the same conditions described in Figure 3.
The BAR domain of SNX1 and SNX2 interacts specifically with inactive, GDP-bound RhoG
The BAR domains of Arfaptin-2 and Hob3p have previously been shown to interact with the small GTPases Rac1 and Cdc42, respectively (34–36). The ability of SNX1 and SNX2 to elicit the Kal7-dependent lamellipodia phenotype led us to examine whether SNX1 or SNX2 could interact with Rho GTPases. For this purpose, we used recombinant, His6-tagged SNX2 to pull down myc-tagged RhoA, Rac1, Cdc42 or RhoG from CHO cell extracts. Because the activation status of the Rho GTPases could influence any potential interaction with SNX, we initially loaded the extracts with equal amounts of the non-hydrolysable guanine nucleotide analogs GDPβS and GTPγS to obtain a mixed population of inactive and active GTPase states. Under these conditions, only RhoG specifically co-precipitated with SNX2 (Figure 10A), suggesting that this interaction is limited to RhoG.
We next sought to determine whether the activation status of RhoG could influence its interaction with SNX2. We prepared identical myc-RhoG cell extracts loaded with either GDPβS or GTPγS to maintain the pool of RhoG in its inactive and active states, respectively. Following co-precipitation with His6-tagged SNX2, interaction strongly favored inactive, GDP-bound RhoG (Figure 10B). The preference for inactive RhoG was eightfold greater than for activated RhoG, with 4.1 ± 1.0% of GDPβS-loaded RhoG versus 0.5 ± 0.1% of GTPγS-loaded RhoG associated with SNX2.
To identify the domains of SNX2 required for interaction with RhoG, we used recombinant, His6-tagged N+PX and BAR fragments of SNX2 to pull down GDPβS-loaded myc-RhoG from transfected CHO cell extracts. RhoG associated with full-length SNX2 and the BAR fragment but not with the N+PX fragment (Figure 10C), indicating that the BAR domain is critical for RhoG binding.
To rule out a contribution of other cellular factors in the interaction of SNX2 with RhoG, we tested whether recombinant SNX and Rho GTPases could interact directly using bacterially purified His6-tagged SNX1 or SNX2 and tandem-affinity purification (TAP)-tagged Rho GTPases isolated from yeast. When His6-SNX1 or -SNX2 were incubated with equimolar amounts of GDPβS-loaded RhoA, Rac1, Cdc42 or RhoG immobilized on immunoglobulin G (IgG)-Sepharose beads, both SNX1 and SNX2 interacted preferentially with RhoG compared to control IgG beads or other Rho GTPases (Figure 10D). In these experiments, SNX1 and SNX2 also appeared to associate with RhoA, although the RhoA interaction was weaker compared to RhoG. SNX1–RhoA association was 50.9 ± 3.3% that of SNX1–RhoG, while SNX2–RhoA association was 41.1 ± 8.3% that of SNX2–RhoG. Compared to RhoG, the relative association of Rac1 with Flag-SNX1 and Flag-SNX2 was 9.2 ± 3.1% and 7.7 ± 3.4%, respectively, while the relative association of Cdc42 with Flag-SNX1 and Flag-SNX2 was 4.2 ± 3.5% and 8.2 ± 5.5%, respectively. Thus, both SNX1 and SNX2 can interact directly with inactive RhoG, consistent with the results seen in pulldowns from cell extracts. A weaker interaction between recombinant SNX1 or SNX2 and RhoA might also occur; however, pulldowns from cell extracts indicate that this interaction may not occur in the context of a cellular environment.
In this study, we show that interaction of SNX1 and SNX2 with the actin-modulating RhoG GEF Kalirin-7 promotes membrane remodeling in two different cell lines. Although RhoG function is less clear than the related GTPases RhoA, Rac1 and Cdc42, emerging evidence supports a role for RhoG in specialized forms of endocytosis. In endothelial cells, activation of RhoG by the SH3-containing GEF (SGEF) regulates macropinocytosis and trans-endothelial migration (31,37). Furthermore, RhoG promotes phagocytic uptake of apoptotic cells in macrophages, as does its orthologue MIG-2 in Caenorhabditis elegans(38,39). Notably, both RhoG and SNX1 appear to regulate Salmonella entry and/or maturation of Salmonella-containing vesicles, where the bacterial PI phosphatase SigD recruits SGEF and SNX1 to entry sites (40,41). Although RhoG and SNX-BAR proteins have been implicated in trafficking events such as macropinocytosis (42,43), our attempts to monitor uptake of dextran, a fluid-phase marker commonly used to study macropinocytosis, have not revealed any difference between control and SNX1- or SNX2-overexpressing cells (our unpublished results). Notably, the described role for SNX-BAR proteins in macropinocytosis requires a functional retromer complex, while retromer does not appear to contribute to SNX1 and SNX2 function in the context of Kal7 and RhoG. Processes such as macropinocytosis and Salmonella entry require extensive membrane remodeling, and actin dynamics must be tightly regulated to effectively co-ordinate these events. Our finding that SNX1 and SNX2 can mediate membrane remodeling through the RhoG GEF Kal7 may thus represent a novel function for SNX1 and SNX2 that is distinct from their role in retromer.
BAR-containing proteins are involved in multiple endocytic events through curvature sensing and induction, and actin dynamics are closely coupled to BAR function (44–46). For example, actin depolymerization promoted membrane tubulation by the BAR proteins amphiphysin-1 and endophilin-1, among others (47). Furthermore, the BAR proteins Tuba and SNX9 can recruit dynamin and actin-nucleating proteins to sites of endocytosis (48–51). Thus, the BAR domain of SNX may provide spatial information for localized Rho-dependent actin reorganization. In the PX-BAR subfamily of SNXs, these two domains are likely to act as one functional unit. For both SNX1 and SNX9, mutation of either domain is sufficient to attenuate function (4,5,52). Our findings provide further support for PX and BAR domains of SNX proteins acting in concert. In particular, fragments containing either domain in the absence of the other do not interact with Kal7, and mutation of either domain reduces Kal7-dependent lamellipodia formation.
Although we observed only a partial loss of lamellipodia phenotype in cells co-expressing Kal7 with SNX2 loss-of-function mutants (RRF or KR/AA) compared to wild-type SNX2, the decrease in the number of mutant SNX-expressing cells forming lamellipodia was statistically significant. This residual phenotype may result from membrane recruitment (Figure S2) or partial activity of the SNX2 mutants. Numerous SNX-BAR proteins are known to form homo- and heterodimers (1). Thus, it is possible that mutant proteins can associate with endogenous SNXs and localize to the appropriate membrane compartment, consistent with our observation that PX- or BAR-mutant SNX2 can associate with membranes when expressed at low levels. Alternatively, because the BAR domain of SNX2 interacts directly with RhoG, overexpression of mutant SNX might recruit inactive RhoG to membrane compartments, where it could be activated by Kal7. To address a possible role for other SNXs in the formation of lamellipodia and membrane ruffles, we examined whether the BAR-containing SNXs 5 and 6 could promote Kal7-dependent membrane remodeling. We found that neither SNX5 nor SNX6 potentiated Kal7-dependent lamellipodia, suggesting that this property might not be shared with other BAR-containing SNX isoforms (our unpublished results). Importantly, endogenous SNX1 or SNX2 levels are sufficient to promote lamellipodia, because depletion of either SNX by RNA interference inhibited lamellipodia in cells expressing high levels of Kal7.
Expression of Kal7 is highest in the brain where it functions in formation and maintenance of dendritic spines (19,21). At the postsynaptic density, Kal7 acts downstream of the EphB2 receptor to promote dendritic spine formation in response to Ephrin-B1 (22). Co-culture of fibroblasts expressing Ephrin-B or EphB results in trans-endocytosis of the full-length transmembrane ligand or receptor in contacting cells to promote repulsion (53,54). Uptake of a transmembrane receptor is likely an engulfment response, because a portion of the apposing cell's membrane must also be internalized. Because RhoG and SNX1 are both implicated in engulfment, it will be interesting to determine if they regulate EphB function via Kal7.
In the present study, we made use of epithelial cell lines as a model for characterizing the interaction of SNX1 and SNX2 with Kal7 and RhoG as well as their effects on actin and membrane remodeling. It is worth noting that although expression of Kalirin is largely restricted to the brain, it is also expressed (at least at the mRNA level) at lower levels in other tissues such as the kidney and spleen (19). Although physiological studies of Kalirin function center on its role in neurons, non-neuronal cell types can also provide mechanistic insight. The RhoG-dependent effects we observed in epithelial cells co-expressing Kal7 and SNX1 or SNX2 could be more directly applied to non-neuronal cells that express Kalirin isoforms, as many cell types employ common mechanisms to mediate membrane remodeling and cytoskeletal rearrangement. For example, small GTPases such as Rac1 and RhoG promote formation of lamellipodial structures in epithelial cells as well as in neuronal growth cones, while Cdc42 promotes filopodia in both cell types. These structures contribute to cell spreading and migration, as well as neurite extension. In contrast, RhoA promotes stress fiber formation in epithelial cells and growth cone collapse/retraction in neurons. Additionally, the related RhoG GEF Trio, which shares a high level of sequence identity and similar domain organization to Kalirin, appears to be broadly expressed (55). Although an interaction between SNX1 or SNX2 and Trio has not yet been identified, it remains possible that such an association would be directly relevant to membrane and actin remodeling in epithelial cells. Of note, we have not observed significantly enhanced motility of Kal7- and/or SNX-overexpressing cells by live-cell microscopy or in scratch-wound assays (data not shown); however, they might contribute to cell migration in other contexts.
The lamellipodia phenotype we observed in epithelial cells could also be applicable to a neuronal Kal7-dependent process. Kalirin enhances neurite outgrowth by nerve growth factor (NGF) acting through the TrkA receptor in PC12 cells (56). Furthermore, RhoG is known to act downstream of TrkA in PC12 cells, which is also important for neurite outgrowth (57). We have begun to study the effects of SNX1 and SNX2, which are also expressed in PC12 cells (58), in Kal7-mediated neurite outgrowth. Our preliminary results indicate that SNX2 overexpression enhances Kal7-dependent neurite extension (our unpublished observations), and future studies will focus on fully characterizing this effect. While SNX1 and SNX2 are believed to share redundant functions, several studies suggest that they have some distinct functions as well. First, while transgenic mice lacking either SNX1 or SNX2 are viable, deletion of both genes is lethal (59). A single copy of SNX2 fully rescues the lethality and developmental defects while a single copy of SNX1 cannot. Secondly, the subcellular distribution of endogenous SNX1 and SNX2 is not identical, suggesting possible functions in different compartments (29). Finally, while SNX1 is required for lysosomal delivery of activated PAR1, SNX2 overexpression blocks receptor degradation by sequestering SNX1 (12). In light of the differential roles of SNX1 and SNX2 in trafficking of activated PAR1, it is possible that they may also differentially affect other activated receptors. Future experiments focusing on the relationship between the SNXs and Kalirin-dependent processes might provide insights to the regulation of neurite outgrowth and/or dendritic spine formation, both in PC12 models and in primary neurons.
There is currently little evidence for retromer-independent functions of SNX1 or SNX2. Studies of SNX1-dependent degradation of PAR1 provided the first evidence of retromer-independent activities, because Vps26 depletion had no effect on lysosomal delivery (12). In our study, depletion of Vps35, the cargo-recognition subunit of retromer, could not block Kal7-dependent lamellipodia, whereas SNX1 or SNX2 depletion inhibited this phenotype. Although we cannot rule out retromer function downstream of lamellipodia induction, SNX-dependent membrane remodeling appears to be independent of retromer.
Lamellipodia formation in our study is clearly a RhoG-mediated process. Direct binding to GDP-bound RhoG and its activator Kal7 suggests a role for SNX1 and SNX2 in maintaining specific membrane targeting of the activated Rho signaling complex. Other BAR-containing proteins are known to interact with Rho GTPases. For example, Arfaptin-2 and Hob3p interact with Rac1 and Cdc42, respectively (34–36). In both cases, GTPase interaction occurs with the BAR domain, which is consistent with our finding that the BAR domain of SNX2 interacts with RhoG. Furthermore, the Arfaptin-2 BAR domain shows structural similarity to the DH domain of the GEF Tiam, which also interacts with Rac1 (60). DH and BAR similarities are thought to have converged during evolution, and Rho GTPase interaction could be a common function of both domains. In Schizosaccharomyces pombe, Hob3p interacts with the Cdc42 exchange factor Gef1p to recruit Cdc42 to the site of fission (35). Thus, interaction of BAR proteins with small GTPases and GEFs could be a common theme, with interaction occurring in a cell- or function-specific context. At the very least, our findings extend the known interaction of BAR domains with small GTPases to the SNX subfamily of BAR-containing proteins. The interaction surface of BAR domains both with membranes and with small GTPases is located on the concave face, although it is currently unclear whether such interactions can be concurrent, sequential or mutually exclusive (3). Such a dual function certainly merits further investigation because it implies a mechanistic complexity that is beyond the currently accepted view of BAR domain function.
Overall, we propose a model where SNX1 and SNX2 directly recruit inactive RhoG to its GEF Kal7. Through coincidence detection of membrane composition and curvature, SNX could recruit Kal7 and RhoG to an appropriate compartment for activation. Alternatively, Kalirin and SNXs interact with receptors such as TrkA and EGFR (56,61), which could mediate recruitment to sites of receptor activation. Although the order of SNX and Kal7 recruitment remains to be determined, either scenario would result in formation of a membrane domain enriched in an inactive Rho GTPase and its activating GEF. A similar phenomenon has been described for the endosomal GTPase Rab5, where the effector Rabaptin-5 is required for recruitment of its GEF Rabex-5 (62,63). The complex formed between a Rab, its effector and its exchange factor is essential for signal amplification by generating domains that contain both Rab5 and its GEF. In our model, SNX1 or SNX2 could directly link inactive RhoG and Kal7 to create a localized enrichment and functional amplification of the Rho GTPase.
Materials and Methods
cDNA coding for full-length human SNX1 or SNX2 as well as SNX2 truncations coding for amino acid residues 1–269 (N+PX) or 270–519 (BAR) were amplified by PCR and cloned into the NotI and EcoRI sites of pFLAG-CMV2. SNX2 R182A/R183A/F184A (RRF) and K426A/R428A (KR/AA) mutants were generated by site-directed mutagenesis and cloned into pFLAG-CMV2 as above. Bacterial expression plasmids were generated by insertion of the respective full-length or truncated Flag-SNX fragment into the KpnI site of a modified pQE9 vector (Qiagen) containing an HA tag. Because of increased stability of the recombinant protein, residues 260–519 were used for the SNX2-BAR domain. His6-myc-tagged wild-type and GEF-inactive N1415A/D1416A (ND/AA) Kal7 in pEAK10, as well as psCEP/His6-myc-Kal7 ΔCT were generous gifts from Dr. Betty A. Eipper. pEAK10/His6-myc-Kal7 ΔCT was generated by site-directed mutagenesis using QuickChange (Stratagene) to introduce a stop codon immediately following nucleotide 4728, such that the resulting protein lacks the final 60 amino acid residues (21).
Wild-type RhoA, Rac1 and Cdc42 in pRK5-myc were obtained from Dr. Alan Hall. Wild-type and T17N mutant RhoG were inserted into the BamHI and EcoRI sites of pRK5-myc. Dominant-negative, HA-tagged RhoAT19N, Rac1T17N and Cdc42T17N in pcDNA3.1(+) were obtained from UMR cDNA Resource. For TAP of Rho GTPases, myc-tagged wild-type RhoA, Rac1, Cdc42 and RhoG were inserted into the HindIII site of pBS1761 (EUROSCARF). The resulting N-terminally TAP-tagged cDNA was then cloned into the KpnI and EcoRI sites of pYES2 (Invitrogen). All plasmid constructs were confirmed by sequencing.
Cell culture and transfection
Cells were maintained at 37°C in the presence of 5% CO2 in medium supplemented with 5% FBS, 100 units/mL penicillin and 100 µg/mL streptomycin. CHO cells were grown in minimum essential medium α (Invitrogen) and HeLa in Dulbecco's modified Eagle medium (DMEM, HyClone). Cells were transfected with LipofectAMINE (Invitrogen) according to manufacturer's instructions. 293TN cells (System Biosciences) were maintained in DMEM containing 10% FBS and 500 µg/mL G418, and were transfected with LipofectAMINE Plus (Invitrogen).
Immunocytochemistry and confocal microscopy
For visualization by confocal microscopy, cells were seeded onto 12-mm cover slips and transfected as above. Forty-eight hours after transfection, cells were fixed for 30 min with 4% paraformaldehyde in PBS. Cover slips were then incubated with blocking buffer (1% BSA, 2% normal goat serum, 0.6% saponin and 0.01% NaN3 in PBS) and stained with the following antibodies as required: mouse anti-EEA1 (Abcam), mouse anti-myc (Sigma), mouse anti-SNX1 or anti-SNX2 (Stressgen), mouse anti-Vps35 (Abnova), rabbit anti-Flag or anti-myc (Sigma) or rabbit anti-Kalirin (Upstate). Secondary goat anti-mouse and goat anti-rabbit antibodies conjugated with Alexa 488 (Invitrogen) or Texas Red (Jackson Laboratories) were used, and cover slips were mounted in SlowFade Gold reagent (Invitrogen). F-actin was visualized with Alexa 568-labeled phalloidin (Invitrogen).
Images were captured on a Zeiss LSM Confocor 3 NLO microscope equipped with a 63× Plan Apochromat, 1.4 numerical aperture (NA) oil-immersion lens, or with an Olympus IX70 microscope equipped with an MRC1024 confocal unit (BioRad) and a 60×, 1.4-NA oil-immersion objective. Images were acquired using lsm software v4.2 (Zeiss) or LaserSharp (BioRad) software and processed with ImageJ version 1.37v.
Purification of His6-SNX and TAP-Rho GTPases
Expression of recombinant His6-HA-Flag-SNX proteins in Escherichia coli was induced overnight with 1 mm Isopropyl β-D-thiogalactopyranoside (IPTG). Cells were resuspended in PBS containing 10 mm imidazole, 2 mmβ-mercaptoethanol (β-ME) and 2 mm phenylmethylsulphonyl fluoride (PMSF). Lysozyme was added to 20 mg/mL for 10 min, and cells were lysed by sonication in the presence of 1% NP-40. The lysate was then cleared by centrifugation at 30 000 ×g, and the supernatant was added to nickel-nitrilotriacetic acid (Ni-NTA)-agarose beads (Qiagen). Beads were then washed with 50 volumes of wash buffer 1 (0.1% NP-40, 10 mm imidazole and 2 mmβ-ME in PBS) and 50 volumes of wash buffer 2 (10 mm imidazole and 2 mmβ-ME in PBS), and eluted with 10 volumes of elution buffer (300 mm imidazole and 2 mmβ-ME in PBS).
For purification of TAP-tagged Rho GTPases, pYES2 plasmids were transformed into the EUROSCARF yeast strain FY1679-08A (MATa; ura3-52; leu2Δ1; trp1Δ63; his3Δ200; GAL2) and grown in SC-ura medium. Protein expression was induced overnight in SC-ura containing 2% galactose. Cells were disrupted by vortexing with 0.5-mm glass beads (Biospec) in 20 mm Tris–HCl pH 7.5, 150 mm NaCl, 5 mm ethylenediaminetetraacetic acid (EDTA) and 2 mm PMSF. The lysate was solubilized with 1% NP-40, cleared by centrifugation at 30 000 ×g, and TAP-tagged GTPases were immobilized on IgG-Sepharose beads (GE Healthcare), which were then washed with 100 volumes of wash buffer 1 (20 mm Tris–HCl pH 7.5, 0.1% NP-40, 300 mm NaCl, 2 mm EDTA) and 200 volumes of wash buffer 2 (20 mm Tris–HCl pH 7.5, 50 mm NaCl, 2 mm CaCl2).
Interaction of SNX1 and SNX2 with Kal7
CHO cells were co-transfected with Flag-tagged SNX and either His6-myc-Kal7 or empty vector. Transfected cells were scraped in PBS and solubilized in 1% 3-[(3-Cholamidopropyl)dimethylammonio]-1-propanesulfonate (CHAPS), 20 mm imidazole and 20 mmβ-ME in PBS containing protease inhibitors (Sigma). The lysate was cleared at 5000 ×g, and diluted to a final concentration of 0.1% CHAPS prior to incubation overnight at 4°C with Ni-NTA-agarose beads. Beads were then washed five times with 0.1% CHAPS, 20 mm imidazole and 20 mmβ-ME in PBS, and eluted with 1× SDS sample buffer. Retained proteins were analyzed by western immunoblot with mouse anti-Flag (Sigma) and mouse anti-myc antibodies.
shRNA depletion of SNX1, SNX2 and Vps35
Lentiviral transduction was performed to deplete HeLa cells of SNX1, SNX2 or Vps35. Briefly, 293TN (System Biosciences) cells were transfected with the pMD2.G and psPAX2 envelope and packaging plasmids (Addgene plasmids 12259 and 12260, Dr. D. Trono) as well as a mixture of five shRNA targets against SNX1, SNX2 or Vps35 in the pLKO.1 plasmid according to supplier instructions (System Biosciences). Details of the shRNA sequences are included in Table S1. A pLKO.1 GFP shRNA target (Addgene plasmid 12273, Dr. B. Weinberg) was used for control infection. Forty-eight hours after transfection, medium was collected and filtered through a 0.45-µm cellulose acetate filter. Based on the 293TN supplier's estimated titre of 2 × 106 particles/mL, a multiplicity of infection (MOI) of 20 gave optimal knockdown as determined by quantitative immunoblot.
For shRNA experiments, HeLa cells were seeded onto cover slips and transfected with His6-myc-Kal7 ΔCT. Twenty-four hours after transfection, cells were infected with GFP, SNX1, SNX2 or Vps35 shRNA in the presence of 5 µg/mL polybrene (Sigma), and cells were fixed and stained with anti-Kalirin antibodies 48 h after infection.
Interaction of SNX with RhoG
Interaction of SNX2 with Rho GTPases was determined by transfection of CHO cells with myc-tagged RhoA, Rac1, Cdc42 or RhoG. Forty-eight hours after transfection, cells were harvested in PBS and solubilized with 0.5% CHAPS, 1 mm MgCl2 and 10 mmβ-ME in the presence of PBS containing protease inhibitors. The lysate was cleared at 5000 ×g and diluted to a final concentration of 0.125% CHAPS, and was then loaded with 200 µm GDPβS, 200 µm GTPγS, or a combination of both nucleotides at 200 µm each, for 10 min at 30°C. One hundred picomoles of recombinant His6-HA-Flag-SNX2 was then added to the extracts, which were incubated for 16 h in the presence of Ni-NTA-agarose beads that were blocked with 0.5% gelatin. Beads were then washed five times with PBS containing 0.125% CHAPS, 150 mm NaCl, 1 mm MgCl2, 10 mmβ-ME and 20 mm imidazole before analysis by western immunoblot. For Figure 8C, beads were washed as above, but with 0.25% CHAPS.
For direct interaction, 45 pmol of recombinant TAP-myc-Rho GTPases immobilized on IgG-sepharose beads were blocked with 0.5% gelatin in equilibration buffer (0.125% CHAPS and 1 mm MgCl2 in PBS) prior to loading with 200 µm GDPβS in equilibration buffer for 10 min at 30°C. Beads were then incubated for 16 h with 45 pmol of recombinant His6-HA-Flag-SNX1 or SNX2, and were washed five times with 0.125% CHAPS, 150 mm NaCl and 1 mm MgCl2 in PBS prior to analysis by western immunoblot.
We would like to thank Betty Eipper and Alan Hall for providing plasmids and reagents used in this study, Mohammad Abdul-Ghani for technical assistance, and Yulia Artemenko for critical reading of the manuscript and helpful comments. We are also grateful to Michael McCaffery at the Integrated Imaging Center of the Johns Hopkins University. This work was supported by an operating grant to J. K. N. from the Canadian Institutes of Health Research. D. C. P. was supported by a Canada Graduate Scholarship from NSERC and by an Ontario Graduate Scholarship in Science and Technology.