Topology, glycosylation and conformational changes in the membrane domain of the vacuolar H⁺-ATPase a subunit^†

Vacuolar-type H⁺-ATPases (V-ATPases) are highly conserved multi-subunit proton pumps. They are ubiquitous in eukaryotic cells where they perform essential housekeeping functions by acidifying endomembrane organelles [Beyenbach and Wieczorek, 2006; Saroussi and Nelson, 2009]. In specialized mammalian cells V-ATPases can be targeted to the plasma membrane to translocate protons out of the cell. This can be for the purpose of intracellular pH regulation [Nordström et al., 1995], or to maintain organismal acid–base balance [Wagner et al., 2009]. In some cases, proton secretion is a means to achieve low luminal pH in specialized extracellular compartments, such as the lumina of the epididymis [Brown et al., 1997], or osteoclast resorption lacunae [Manolson et al., 2003].

V-ATPases consist of two major subcomplexes, the V₀ sector, which is membrane-bound and involved in proton translocation, and the V₁ sector, which hydrolyzes ATP. V₀ consists of subunits a, c, c′, c″, d, and e, and V₁ is composed of subunits A through H [Zhang et al., 2008]. The energy of ATP hydrolysis is coupled to proton translocation by physical rotation of a central stalk structure that drives a hydrophobic c-barrel composed of (in yeast) four c subunits, one c′ subunit and one c″ subunit. The physical rotation of the c-barrel in contact with the membrane domain of the adjacent a subunit (Vph1p in yeast vacuoles) has been proposed to drive proton translocation by a mechanism that is analogous to F-ATPases “running in reverse.”

The a subunit is c. 100 kDa in size and consists of a cytoplasmic N-terminal half (NTa domain), and an integral membrane, hydrophobic C-terminal half (CTa domain). The NTa domain interacts with the V₁ sector and provides a stabilizing stator function [Landolt-Marticorena et al., 2000; Kartner et al., 2010; Muench et al., 2011]. The CTa domain plays a major role in proton translocation by forming the transmembrane proton channel [Toei et al., 2011]. Protons are thought to diffuse into a cytoplasmically accessible hemichannel, resulting in protonation of a passing glutamate residue situated in a transmembrane α-helix (TM) of one of the six c-type subunits within the rotating c-barrel. As a consequence of coupled ATP hydrolysis, the c-barrel rotates 360° in three 120° steps, each consuming one ATP molecule, to deliver the same proton to a putative second hemichannel with aqueous access to the lumen. Dissociation of the proton and diffusion into the lumen completes its translocation across the lipid bilayer [Toei et al., 2011]. Because there are six proton-carrying sites evenly spaced at the periphery of the c-barrel, two protons can be translocated for every molecule of ATP hydrolyzed. A critical arginine residue (R735 in yeast) is thought to interact with the c-barrel glutamic acid at the luminal hemichannel, possibly providing the necessary alignment and environment for protonation and deprotonation phases of the transport cycle. The general postulate is derived from one proposed originally to describe F-ATPase proton translocation [Vik et al., 2000].

The CTa domain of the a subunit has been reported as having six [Manolson et al., 1992], six or eight [Jackson and Stevens, 1997], seven [Leng et al., 1996; Leng et al., 1998; Peng et al., 1999; Ma et al., 2011], eight [Landolt-Marticorena et al., 1999; Wang et al., 2008; Toei et al., 2011], or nine [Leng et al., 1999] TMs. In addition to these discrepancies, topology models of the V-ATPase a subunit based on experimental accessibility of membrane-permeable and impermeable reagents, or protease accessibility, have not been unequivocal in their assignments of TM or inter-helical loop sequences [Jackson and Stevens, 1997; Leng et al., 1999; Wang et al., 2008], although a better defined model has emerged recently [Toei et al., 2011]. The present work seeks to clarify features of topology and function of the a subunit, using approaches complementary to those of previously published works, and in so doing add to its potential as a therapeutic target.

MATERIALS AND METHODS

Computed Topology Prediction

A topology model of Vph1p was derived by applying seven different, commonly used predictive algorithms to determine putative TM helices (Fig. 1). These were HMMTOP [Tusnády and Simon, 1998, 2001], TMHMM [Krogh et al., 2001], SOSUI [Hirokawa et al., 1998], TMpred [Hofmann and Stoffel, 1993], TopPred [von Heijne, 1992; Claros and von Heijne, 1994], DAS [Cserzo et al., 1997], and PHDhtm [Rost et al., 1995]. All were available on the Swiss Institute of Bioinformatics ExPASy Proteomics Server at the time that this work was done and were generally used with default settings. A description of the consensus model computation is given in Results Section.

RESULTS

Questions of Vph1p topology were addressed by first deriving computed models using seven commonly available TM prediction algorithms (see Materials and Methods Section). Their methodologies are similar; however, these algorithms employ somewhat different amino acid properties tables, are derived from various partitioning data, and comparisons are made with various surveys of both globular and membrane proteins, and some use hidden Markov models [von Heijne, 1992; Hofmann and Stoffel, 1993; Claros and von Heijne, 1994; Rost et al., 1995; Cserzo et al., 1997; Hirokawa et al., 1998; Tusnády and Simon, 1998, 2001; Krogh et al., 2001]. Consequently, there were considerable differences in the numbers, locations and spans of the predicted TM segments for Vph1p, depending on the algorithm used. Thus, it was felt that by building a consensus of results the most reliable features of this panel of TM-prediction tools could be incorporated into a model of greater accuracy than would be predicted by any single method.

To build a consensus, the seven different predictions were mapped to the primary sequence of Vph1p, and each amino acid in the sequence was given an unbiased score numerically equivalent to the number of predictive algorithms that flagged it as being within a TM. Stretches of flagged amino acids were identified as likely TMs if there were at least 18 contiguous amino acids with a score ≥4 (i.e., predicted by at least half of the algorithms used). Stretches of amino acids flagged by ≤3 of the seven algorithms over <18 contiguous residues were rejected as unreliable predictors of TM helices.

The resulting consensus computed topology prediction for Vph1p (see Fig. 1) was grossly defined by its well-established features: two nearly equally-sized domains, the largely hydrophilic, cytoplasmically oriented, amino-terminal domain (NTa) comprising residues M1 to N406, and the largely hydrophobic, integral membrane, carboxyl-terminal domain (CTa) comprising residues A407 to the C-terminus, S840. Furthermore, CTa was comprised of seven to eight predicted TM helices. There was ambiguity in the first hydrophobic stretch, which should have been a single TM by our stated criteria; however, further topological considerations, discussed below, require that it be treated as two TMs. The model also predicts putative cytoplasmic inter-helical loops, CL1–3; putative extracellular loops, EL1–4 (EL1 may not exist if TM1 and TM2 form a reentrant loop that does not penetrate to the lumen, as is suggested in Fig. 1); and the C-terminal tail domain, CTD. In Figure 1, the color-coded predictive scores (color legend, lower right) are mapped to the topology diagram, and TMs have been positioned within the phospholipid bilayer to maximize the predictive score within a 21 amino acid span.

There is abundant evidence that the NTa domain has a cytoplasmic orientation, in part because of its known interactions with a number of cytoplasmic V₁ subunits, such as A, B, and H [Jackson and Stevens, 1997; Landolt-Marticorena et al., 1999, 2000; Leng et al., 1999; Kartner et al., 2010], and there is evidence that the hydrophilic C-terminal “tail” domain (CTD) is oriented to the cytoplasm. The latter evidence comes from scanning cysteine mutagenesis and sulfhydryl reagent accessibility studies [Wang et al., 2008; Toei et al., 2011], but past investigations with similar methods have yielded contradictory results [Leng et al., 1998]. Additional support for a cytoplasmic orientation comes from the observation that the CTD interacts with the cytoplasmic glycolytic enzyme, phosphofructokinase-1 [Su et al., 2003, 2008]. With both the N-terminus and C-terminus in the cytoplasm, the number of TMs must be even. Thus, with a default of eight TMs, the first contiguous stretch of hydrophobic amino acids necessarily has to be divided into two TMs. Alternatively, since these two putative TMs are relatively short spans that do not fully meet our criteria for being transmembrane, it seems plausible that this segment forms a reentrant loop, as shown speculatively in Figure 1, rather than spanning the membrane twice; the implications for orientation of other TMs, however, remain the same. A third alternative is that the low-scoring hydrophobic region between TM5 and TM6 is a TM (while keeping the first hydrophobic stretch as one TM to satisfy the even-number requirement). This, however, would invert TMs 3–5 (as labeled in Fig. 1), drastically altering the topology. To determine which model is correct required further experimental evidence.

To test the model of Figure 1, it was necessary first to dispel any remaining doubts regarding the disposition of the C-terminus. Two constructs were prepared where the C-terminus of Vph1p was tagged, either with the 3HA epitope tag to yield Vph1p-3HA, or with GFP to yield Vph1p-GFP (see Materials and Methods Section and Table I). These constructs were integrated separately into the genome of a yeast strain with a VPH1 deletion, Δvph1. Since these tags are polypeptide additions to the Vph1p structure, and GFP has a substantial size (27 kDa), it was necessary to verify that the tagged Vph1p subunits were functionally expressed within the V-ATPase complex. This was accomplished with a growth assay in the presence of 4 mM ZnCl₂ (Fig. 2A). Zinc is toxic to yeast, but can be detoxified, possibly by sequestration within the vacuole, requiring a cation-exchange mechanism dependent upon the proton gradient generated by V-ATPase, or some other V-ATPase-dependent mechanism [Iwaki et al., 2004; Kane, 2006]. In yeast where Vph1p is not functional, V-ATPase activity is lacking in the vacuolar membrane and, consequently, the cells are sensitive to zinc toxicity and will not grow on media containing 4 mM ZnCl₂. This was observed for the Vph1p-deletion strain Δvph1 in a 4-day growth assay (Fig. 2A). In a dilution series, it was seen that Δvph1 grew poorly compared with the wild type VPH1 strain in standard medium (that it grows at all may be due to low expression of the paralogous subunit, Stv1p, which can partially complement the Δvph1 phenotype, though this remains speculative [Kane, 2006]), but in the presence of Zn²⁺, growth was completely eliminated. On the other hand, in the strain expressing the Vph1p-GFP fusion protein, growth was restored to a level indistinguishable from that of the wild type. Similar results were obtained with the Vph1p-3HA construct (not shown). Thus, neither the large C-terminal GFP fusion, nor the short 3HA epitope tag, significantly compromised Vph1p function.

To determine the orientation of the hydrophilic CTD of Vph1p, accessibility of either the 3HA epitope tag, or the GFP fusion tag, on the cytoplasmic surface of purified, intact vacuoles was tested. Upon purification of intact vacuoles from the Vph1p-GFP-expressing strain, GFP fluorescence was observed in densely packed vacuoles by fluorescence microscopy (Fig. 2B), revealing the characteristic ring-staining expected for a membrane-bound fluorescent marker. Treatment of vacuoles with 30 µg/ml proteinase K at 0°C for 20 min resulted in complete proteolytic quenching of GFP fluorescence. This suggested that the GFP-tag was accessible on the outer (cytoplasmic) surface of the vacuolar membrane and, therefore, that the CTD of Vph1p, to which it is fused, must reside in the cytoplasmic compartment; however, this conclusion depends entirely upon the integrity of the vacuoles to protease penetration.

To show that the vacuole preparation consisted of vacuoles that were impermeable to proteinase K, accessibility of the protease to pro-CPY, a luminal marker of the vacuolar compartment, was tested. Figure 2C (left panel) shows vacuolar proteins immunoblotted with anti-CPY antibody. It demonstrates that pro-CPY (66 kDa band) was not digested to mature CPY (61 kDa) in the vacuole preparation even under conditions of high protease concentration and extended digestion time (up to 40 µg/ml proteinase K, 20 min at 0°C); however, with the addition of 0.5% (w/v) of the membrane-disrupting, non-ionic detergent, Triton X-100, complete digestion of pro-CPY to CPY was observed. Furthermore, when probed with anti-GFP antibody, the blot of Figure 2C (right panel) showed that Vph1p-GFP was as completely digested in the absence of Triton-X100 as in its presence, even at low protease concentrations (right-most lane). These data support the notion that the GFP-tagged CTD of Vph1p is accessible at the cytoplasmic surface of the vacuole membrane.

Further evidence of the orientation of the C-terminus of Vph1p was sought using the HA epitope tag. To do this, a competitive ELISA was performed, where intact vacuoles were used to compete for anti-HA antibodies targeted to antigen-coated microtiter plate wells, as described in Materials and Methods section. Figure 3A shows results of competition with vacuoles from different sources. While intact and detergent solubilized vacuoles expressing Vph1p-3HA strongly competed for anti-HA antibodies at between 0.1 and 1 µg vacuole protein per assay, neither intact nor detergent-solubilized vacuoles expressing wild type Vph1p had any effect at up to 2 µg protein added.

Figure 3B shows vacuolar proteins resolved by SDS–PAGE and immunoblotted using anti-HA antibody. Vph1p-3HA was observed at its native size of c. 100 kDa in untreated vacuoles (left-most lane). Although vacuoles were “untreated” with respect to trypsinization in this lane, incubation was the same as in other lanes, but without addition of trypsin or detergent. Smaller bands were observed presumably due to autodigestion by vacuolar proteases during this incubation. While all lanes were of 100 µg initial vacuolar protein, this left-most lane appears overloaded relative to adjacent lanes where trypsin digestion resulted in significant loss of the HA epitope tag signal. When either intact vacuoles, or detergent-solubilized vacuoles, were treated with trypsin, similar digestion of Vph1p-3HA occurred, with the smallest HA-tagged band having a size of c. 10 kDa (arrowhead). There was no evidence of the 10 kDa band in the untreated control lane, indicating that this band was generated by trypsin digestion. No significant background bands were seen with wild type Vph1p (not HA-tagged). The 10-kDa band corresponds approximately to the size of the CTD, from its proposed junction with TM8 to, and including, the C-terminal 3HA-tag, which itself adds c. 5 kDa to the fragment size. That the fragment was released equally upon trypsinization in the presence or absence of detergent suggests strongly that it is accessible on the cytoplasmic face of the intact, trypsin-impermeable vacuoles. All of the above data, taken together, demonstrate unequivocally that the orientation of the hydrophilic CTD of Vph1p is cytoplasmic and is accurately depicted in the topology model shown in Figure 1.

In order to provide further evidence in support of the model presented in Figure 1, additional protease digestion experiments were done with endoproteinase Glu-C and intact vacuoles from yeast expressing Vph1p-3HA. Under the buffer conditions employed, endoproteinase Glu-C cleaves polypeptides on the carboxyl side of a glutamate residue [Sørensen et al., 1991] and, therefore, its protease activity is more selective than that of trypsin or proteinase K, making it a relatively infrequent cutter. Figure 4A shows an immunoblot probed with anti-HA antibodies after digestion of intact vacuoles with a range of Glu-C concentrations. A single major HA-positive band was released, with a size of c. 23 kDa (23.4 ± 0.3 kDa, n = 3; note that this includes the C-terminal 3HA epitope tag, with a predicted size of 5.0 kDa). This places the site of digestion near the N-terminal end of the large hydrophilic loop (green arrowhead in CL3 of Fig. 1). The indicated site would yield a predicted fragment of 23.7 kDa in size; the closest potential cleavage sites on the luminal side of the membrane would yield a fragment c. 10 kDa greater than the observed size. Figure 4B shows an immunoblot identical to that in Figure 4A, but probed instead with an antibody directed to the NTa domain of Vph1p. This yielded a single major band of c. 80 kDa (79.6 ± 1.7 kDa, n = 3), which complements the 23-kDa, HA-positive band of Figure 4A to give empirically a native protein size of 103.0 kDa ± 2.0 kDa, in good agreement with the apparent size observed in SDS–PAGE.

Further elucidation of Vph1p topology comes by inference from studies of V-ATPase a subunits in mammalian systems. Apps et al. [1989] showed over 20 years ago that a subunits of V-ATPases in bovine adrenal medullary chromaffin granules were sensitive to neuraminidase and PNGase F digestion. For the latter enzyme, half of the “120 kDa band” was reduced in size on SDS–PAGE gels when digested in the presence of ocytl-β-glucoside, but the other half could be digested only with the addition of 0.1% SDS, suggesting two distinct oligosaccharide chains (our interpretation). This was the first publication clearly to show N-linked glycosylation in any V-ATPase a subunit. Surprisingly, no further glycosylation studies appeared until recently, when Lee et al. [2010] published work on mouse neuronal autophagy that showed that the lysosomal V-ATPase a1 subunit, a highly conserved ortholog of Vph1p, was glycosylated.

The mammalian a3 subunit has often been referred to as the “116 kDa subunit” because of its relative mobility in SDS–PAGE, when in fact its predicted polypeptide size (in mouse) is 93.4 kDa. Many membrane proteins have anomalous mobility in SDS–PAGE, but our observations on native mouse a3 banding patterns strongly suggested that it had the characteristics of a glycoprotein with complex N-linked glycosylation, showing a diffuse band that was more than 20 kDa in excess of its predicted size. On examination of potential glycosylation signals (N-X-S/T, where X ≠ P) in the mouse a3 polypeptide sequence, we found that only two, at N484 and N504, within the putative second extracellular loop (EL2) were highly conserved in vertebrate species (Fig. 5A). Only one other site, at N41 in NTa could be informative regarding topology, but NTa was already clearly established to be in the cytoplasm, as discussed above. We felt that if we could precisely identify the glycosylation site, which must be luminal, it would help to clarify the orientations of TMs in the N-terminal half of CTa.

Figure 5B, shows immunoblots of a3-GFP constructs expressed in HeLa cells, with and without PNGase F treatment. The wild type a3 fusion protein was observed largely as an intense band at 152 kDa. The 152 kDa band was diffuse, which is typical of complex-glycosylated glycoproteins, and it was reduced to 128 kDa in size upon deglycosylation by PNGase F treatment. A faint band was observed additionally at 134 kDa, which we have recently characterized as the core-glycosylated protein [Bhargava et al., 2012], present at low steady-state levels in the ER prior to trafficking and processing in the Golgi (to the complex-glycosylated 152 kDa band). Similar observations were made with native a3 protein, endogenously expressed in BMM-derived mouse osteoclasts, but since this protein has no C-terminal GFP fusion, molecular weights of the identified bands were: complex glycosylated glycoprotein, 116 kDa; core-glycosylated glycoprotein, 102 kDa; and deglycosylated protein, 94 kDa. These latter observations, along with previously published work [Bhargava et al., 2012], support the notion that heterologous GFP-tagged a3 expression is identical, in terms of biosynthesis and trafficking, to that of the native protein. Site-directed mutagenesis (N → Q conversion) of the N41 site did not alter glycosylation of mouse a3-GFP relative to wild type, as was expected, and the a3^N41Q-GFP could be deglycosylated with PNGase-F in the same manner as the wild type protein. Individual mutagenesis of N484 and N504 also did not abrogate expression of the glycosylated band (data not shown); however, the double mutation N484Q/N504Q resulted in only a smaller-sized, unglycosylated band being observed at 128 kDa, similar to that seen after enzymatic deglycosylation of wild type a3-GFP (Fig. 5B). This demonstrated for the first time that the mouse a3 EL2 loop is glycosylated at two sites, N484 and N504. While there is no evidence that yeast Vph1p is N-glycosylated, nor are any glycosylation sites predicted in its EL2 sequence (Fig. 5A), the otherwise high degree of structural homology between mammalian and yeast a subunits suggests that the Vph1p EL2 loop must also be luminal. The further implications of this finding are that it supports the validity of dividing the first hydrophobic stretch of amino acids into TM1 and TM2 (or representing it as a reentrant loop, as shown in Fig. 1) and requiring an even number of TMs between EL2 and CL3, eliminating the possibility of an additional membrane-spanning TM between TM5 and TM6. As is speculatively indicated in Figure 1, however, this does not rule out the possibility that EL3 could also be a reentrant loop similar to what has been indicated for TM1 and TM2.

While performing experiments with Glu-C, it was observed that the cleavage of Vph1p with the protease appeared to be sensitive to Mg²⁺ concentration. Since the endoproteinase Glu-C itself is not Mg²⁺-dependent, this phenomenon was investigated further. Figure 6A shows results of a series of experiments where endoproteinase Glu-C cleavage of Vph1p (in vacuolar V-ATPase) was performed in the presence of 5 mM Mg²⁺, MgATP, or ATP. These data are further quantified and shown graphically in Figure 6B. At 5 µg/ml Glu-C concentration the yield of the 23-kDa, HA-tagged cleavage product was more than doubled in the presence of 5 mM Mg²⁺ and approximately doubled in the presence of 5 mM MgATP. ATP alone increased cleavage by c. 40%. This suggests that binding of substrate to the catalytic subunits (Vma1p and/or Vma2p; i.e., the A and B subunits) of the V-ATPase complex is able to effect an inter-subunit conformational change, transmitted from Vma1p/Vma2p to the Vph1p subunit, that affects the accessibility of the hydrophilic loop between TM6 and TM7 to endoproteinase Glu-C. Whether this is by direct interaction of the hydrophilic loop with the catalytic subunits, or by transmission through some intermediary subunit between the two, or allosterically by transmission through some other domain of Vph1p (likely NTa), is presently unknown.

The above results encouraged additional experiments with concanamycin A, a known inhibitor of V-ATPase function which is thought to interact directly with Vph1p at the junction between it and the c subunits of the c-barrel structure of the V-ATPase rotor [Huss et al., 2002; Bowman et al., 2004]. It was of interest to determine if this binding could also effect apparent conformational changes within Vph1p. Figure 7A shows results of a series of experiments, similar in execution to those of Figure 6A, but where endoproteinase Glu-C cleavage of Vph1p was performed in the presence of 1 or 10 µM concanamycin A. These data are further quantified and shown graphically in Figure 7B. Concanamycin A at 10 µM reduced the yield of the 23-kDa HA-tagged cleavage product of Vph1p to c. 70% of that seen in its absence, while at 1 µM the effect was somewhat less, at c. 80% of control. Concanamycin A at 10 µM concentration did not have any significant effect on Glu-C digestion of control protein (bovine serum albumin; data not shown). These data suggest that binding of concanamycin in the membrane domain of Vph1p has some allosteric effect on the accessibility of the hydrophilic CL3 loop between TM6 and TM7 to endoproteinase Glu-C. The apparent conformational changes observed here are consistent with the hypothesis that TM7 may undergo some physical displacement within the phospholipid bilayer during the catalytic transport cycle of the V-ATPase complex [Kawasaki-Nishi et al., 2003; Toei et al., 2011].

DISCUSSION

In the present work we attempted to refine the structural model for the Vph1p membrane domain by starting with a consensus of topology prediction algorithms. This approach yielded a model that was confirmed experimentally and is similar in organization to a recently published model that was also confirmed experimentally [Toei et al., 2011]. This supports the notion that using a consensus predictive approach is preferable to the use of any single algorithm. In contrast, a recent V-ATPase review features a Vph1p topology prediction using the TMHMM algorithm alone, which yields a model having seven TMs, with the CL3 and CTD domains (our designations) in the lumen [Ma et al., 2011]. Thus, the single-algorithm prediction has misplaced important features of V-ATPase a subunit topology that are now unequivocally established through many lines of experimental evidence.

Validation of the present topology model was performed using epitope-tagged Vph1p, and C-terminal fusion proteins. These experiments provided strong confirmation that the C-terminus of Vph1p is in the cytoplasm. Similarly, the orientation of the CL3 loop was verified by demonstrating its accessibility to endoproteinase Glu-C on the cytoplasmic surface of intact vacuoles. Additionally, homology modeling, using data from mouse a3 subunit glycosylation studies placed Vph1p EL2 in the lumen. Taking all these data into account supports both the number of predicted TM segments (totaling 8) and the dispositions of the hydrophilic loops and termini of the Vph1p topology model presented in Figure 1.

The current working hypothesis for V-ATPase proton translocation proposes that TMs of the a subunit form separate cytoplasmic and luminal hemichannels in the membrane bilayer. Evidence to support this comes from site-directed mutagenesis of polar and charged amino acids within potentially aqueous channel-forming segments that are part of, or are immediately peripheral to, TMs. These mutations block proton translocation and ATP hydrolysis without affecting V-ATPase assembly [Leng et al., 1996, 1998; Kawasaki-Nishi et al., 2001; Toei et al., 2011]. Physical evidence is also seen in 21 Å resolution single-particle electron microscopy of V-ATPase, where two openings are evident on the luminal side and one on the cytoplasmic side of the V₀ complex [Wilkens and Forgac, 2001]. A more detailed, dynamic model suggests that proton translocation requires not only rotation of the rotor c-barrel, but also intra-molecular, axial rotation of TM7, and/or its linear displacement, within the membrane domain of the a subunit [Kawasaki-Nishi et al., 2003; Toei et al., 2011].

In the present work, Glu-C protease accessibility experiments revealed conformational changes in the Vph1p cytoplasmic loop, CL3, in response to V-ATPase substrate or inhibitor binding. These data may lend support to the dynamic model for TM7 involvement in proton transport, as TM7 is contiguous with CL3. Thus, we speculate that conformational changes in the V-ATPase catalytic headpiece might be physically coupled to Vph1p CL3 through the NTa domain, which is in contact with the V-ATPase V₁ sector [Landolt-Marticorena et al., 1999, 2000; Kartner et al., 2010]. This could account for changes in the rate of CL3 proteolytic digestion in response to MgATP and suggests a physical transduction of conformational change from the V₁ sector to Vph1p CL3 that ultimately might be involved in displacement of TM7 in concert with the catalytic events of the proton transport cycle. Response to concanamycin A suggests that if the displacement of TM7 is inhibited (at the a subunit to c-barrel interface), then CL3 is constrained to a less protease-accessible conformational state. Consistent with this, Wang et al. [2005] found that bafilomycin and concanamycin resistance could be conferred by the mutations E721K, L724A, or N725F in Vph1p. These mutations are clustered at the CL3–TM7 interface, so they proposed that bafilomycin and concanamycin may be inhibitory by “locking” the V₀ domain into a conformation that prevents the dynamic association of TM7 and the c-barrel.

While the topology model presented here is in overall agreement with that of Toei et al. [2011] there are some notable differences. One surprising feature of the predicted topology shown in Figure 1 is that the critical R735 residue of TM7 is very near the cytoplasmic membrane interface, not near the middle of the TM, as has been proposed to accommodate its hypothetical interaction with the centrally located glutamate residues of the c-barrel, or even nearer the luminal membrane interface, as has been suggested in some models [Duarte et al., 2007; Wang et al., 2008; Toei et al., 2011]. TM7 has a surprisingly amphipathic sequence for a TM; nevertheless this is consistent with its proposed function as an aqueous channel-forming domain. Because of its amphipathic nature, in some orthologs of other species, TM7 is not detected as a TM by any of the TM-predicting algorithms used in the present work (data not shown). Thus, while we show the unrevised results of predictive modeling in Figure 1, it seems likely that the limits of TM7 cannot be accurately identified by hydropathy mapping alone. Cysteine accessibility scanning suggests that TM7 may include sequences as far upstream as C723, with a luminal border corresponding to A744, shifting the proposed TM7 α-helix upstream by a dozen amino acids with respect to the assignments shown in Figure 1 [Toei et al., 2011].

The possibility of axial rotation of TM7 to open alternately cytoplasmic and luminal hemichannels has been discussed by Toei et al. [2011]. We speculate that an alternative possibility might be that TM7 goes through a linear displacement, perpendicular to the membrane plane, in the direction of the lumen, allowing the hypothetical alignment of R735 with the rotating glutamic acid proton carriers of the c-barrel. The amphipathic nature of TM7 (including the upstream CL3 sequence) may allow such “bobbing” to occur (see Fig. 1), taking R735 from a position near the cytoplasmic interface to near mid-membrane and back again, in synchrony with the catalytic cycle, opening and closing aqueous channels, or promoting influx and efflux of protons in the lipid bilayer. This notion is in part supported by cross-linking studies suggesting that linear displacement of TM7 by more than one α-helical turn (at least four amino acids) may occur relative to the c-barrel [Kawasaki-Nishi et al., 2003]. A mechanism for this may be provided by the proposed conformational changes in CL3.

It is notable that many of the charged and polar amino acids suggested to form cytoplasmic or luminal hemichannels in the model of Toei et al. [2011] are actually in the cytoplasm (e.g., in the CTD domain), or in the lumen (in EL2), respectively, rather than in the TMs identified in the model of Figure 1. These charged and polar amino acids are thought to form hydrophilic segments of the proton hemichannels. To some extent the model of Figure 1 can accommodate these discrepancies, as the limits of TMs are arbitrarily constrained to a 21 amino acid window. Clearly some TMs will be somewhat longer, or shorter, but extending TM3, TM4, and especially TM8 to accommodate the working model proposed by Toei et al. would likely require significant tilting of α-helices to accommodate their length in the lipid bilayer. This is not an unusual feature of membrane proteins, but the three dimensional modeling of tilted helices would certainly complicate the predictions of dynamic helical rotation of TM7 in relation to other channel-forming helices.

Finally, we have suggested diagrammatically in Figure 1 that there is the possibility that EL3 could form a reentrant loop within the membrane. Reentrant loops are not common, but are seen in some transport proteins and may play regulatory roles. In the AE2 anion exchanger, for example, a reentrant loop is thought to act in acute regulation of anion transport by pH [Stewart et al., 2009]. The central hydrophobic segment shown in Figure 1 is flanked by charged residues, and Leng et al. [1999] have shown that the introduction of a factor X_a site (IEGR) into EL3 before P606 significantly reduced proton transport activity compared with wild type Vph1p. These observations are consistent with a possible role in proton transport for the EL3 domain.

CONCLUSION

In the present work we have demonstrated the utility of consensus predictive modeling, and have confirmed features of the predicted Vph1p topology model with experimental evidence. The amino and carboxyl termini of Vph1p and the CL3 loop domain have been unequivocally shown to be situated in the cytoplasm. We have shown for the first time that the mouse V-ATPase a3 subunit is N-glycosylated at two sites and by homology modeling deduced that EL2 of Vph1p is luminal. These observations and the constraints they place on topology resulted in the model of Figure 1, which is largely in agreement with the recently published experimentally determined topology of Toei et al. [2011]. Determination of CL3 orientation led to the observation that the conformation of CL3 is affected by V-ATPase substrate and inhibitor binding. This, and questions concerning the limits of TM7 suggest considering alternate models of TM7 dynamics in proton transport. Further experimental work will be required to resolve these possibilities.

With its important roles in housekeeping functions and disease processes, further elucidating the structure and function of the V-ATPase a subunit would be of benefit to both a greater understanding of V-ATPase cell biology and the potential development of novel V-ATPase-targeted therapeutics.