Introduction

Fast neurotransmission is mediated by ionotropic receptors, which contain several subunits arranged pseudo-symmetrically around a central ion pore [17, 30, 36, 39, 44, 47, 55, 59, 65]. Neurotransmitter binding to the receptor causes a conformational change in the receptor, which leads to the opening of the pore and a subsequent change in the firing rate of the target neuron. There are three major families of ionotropic receptors: the ‘Cys-loop’ receptor family, which includes the nicotinic receptor for acetylcholine, the 5-HT3 receptor, the GABAA receptor and the glycine receptor [17, 36, 44, 59, 65], the glutamate receptor family, comprising the α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid, kainate and N-methyl-d-aspartic acid receptors [30, 47], and the P2X receptor family [39, 55]. Receptors from all three families are often constructed from multiple types of subunits, raising the question of the subunit stoichiometry and arrangement within the receptor.

By far the best characterised of the Cys-loop receptors is the nicotinic acetylcholine receptor, largely thanks to the extensive electron microscopy (EM) studies of the Torpedo electroplaque form of the receptor by Unwin et al. [49, 70] and also the recent discovery of the acetylcholine binding protein in the snail [13, 18]. This latter protein is homologous to the extracellular domain of the nicotinic receptor and, unlike the intact receptor, is amenable to study by X-ray crystallography. The nicotinic receptor assembles as a pentamer, composed of four types of subunits, α, β, γ and δ, in the order α, γ, α, δ, β, anti-clockwise around the pore, when viewed from the synaptic cleft [37]. The other members of the Cys-loop family are much less well characterised. Isolated GABAA and 5-HT3 receptors have been imaged by EM and appear as cylinders of length 11 nm and diameter 7–8 nm, with a vestibule of diameter 2–3 nm [11, 12, 52]. The pentameric structure of the channels is clearly visible in the EM images. By analogy, the glycine receptor is also assumed to be pentameric [17]. The glutamate receptors are all tetramers [30, 47]. Fortuitously, they have a large extracellular domain and a modular structure, which has permitted the application of both X-ray crystallography and homology modelling, resulting in the production of extensive structural information [30, 47]. The P2X receptors are relatively poorly characterised. Cross-linking studies have indicated that they are trimeric [2, 54], although there have also been suggestions that they are tetramers [40], hexamers or even nonamers [2, 54]. Isolated receptors have been imaged both by EM [48] and by atomic force microscopy (AFM) under fluid [50]. Their large extracellular domains appear to be arranged in the form of crown-shaped structures, perhaps containing an inner vestibule.

Despite the availability of quite detailed structural information, the arrangement of subunits within the ionotropic receptors is often unclear. Even in the case of the nicotinic acetylcholine receptor, agreement about the subunit arrangement was not reached until fairly recently [37, 49, 70], and the composition of the other members of the Cys-loop family is still largely unclear. In the case of the GABAA receptors in particular, the subunit arrangement is of therapeutic significance, as drugs such as the benzodiazepines are known to bind at subunit–subunit interfaces [65]. The glutamate receptors are often heteromeric, although the subunit arrangements are again unknown [30, 47]. P2X receptors are known to exist as both homomers and heteromers [2, 27, 34, 41, 54, 55], and the heteromers have different characteristics from the parent homomers [27, 34, 41], indicating that the properties of endogenous receptors may vary in subtle ways depending on the subunit composition.

One method to determine the arrangement of subunits within ionotropic receptors involves generating constructs in which the subunits are joined together by linker sequences, to form concatemers. This procedure forces the subunits into predetermined arrangements. The functionality of various arrangements can then be tested in heterologous expression systems such as Xenopus oocytes or transfected mammalian cells. This method has been successfully applied to the α1β2γ2 form of the GABAA receptor. Different combinations of subunits were concatenated as trimers and dimers, and various trimer/dimer pairs were expressed in Xenopus oocytes [8]. Many combinations did not give functional receptors. From the minority of combinations that did lead to functional expression, it was possible to deduce that the likely subunit arrangement was αβαγβ, anti-clockwise when viewed from the outside the cell. This arrangement was later confirmed by the concatenation of all five subunits in this order [9]. A similar result has recently been reported for the neuronal nicotinic acetylcholine receptor containing α3 and β4 subunits. It was shown that a ββαβα concatemer had characteristics very similar to those of receptors built from monomeric α- and β-subunits [31].

Subunit concatenation undoubtedly provides valuable information about the assembly of ionotropic receptors. However, the construction of the concatemers is rather laborious, and progress using it has been slow. In addition, there is concern that receptors might be forcibly produced that do not exist in vivo. We have sought an alternative method that more directly examines the architecture of receptors constructed form individual subunits. The AFM-based method described here is rapid, robust and generally applicable to a wide variety of multi-subunit proteins.

General features of the AFM-based method

In our method, epitope tags (one of which must be His6) are engineered onto individual receptor subunits, and the receptors are expressed exogenously by transfection of a suitable cell line (tsA 201). The cells are lysed, and a membrane fraction is prepared. The membrane fraction is solubilised in detergent, and the receptors are isolated from the detergent extract through the binding of the His6 tag to Ni2+-agarose beads. The receptors are eluted from the beads by incubation with imidazole. The receptors are then bound to a mica substrate and imaged by AFM. The molecular dimensions of a large number of the bound receptor particles are measured and used to calculate particle molecular volumes. A frequency distribution of molecular volumes is produced, and the peak volumes are then compared with the volume expected for an individual subunit, calculated on the basis of its molecular mass. This process produces an estimate of the likely subunit stoichiometry of the receptor. The receptors are then incubated with antibodies against the epitope tags on the subunits. The receptor–antibody complexes are bound to mica and again imaged by AFM. Receptors decorated by two antibodies are identified in the AFM images, and the angles between the bound antibodies are measured. A frequency distribution of angles is produced. The peak angles define the geometry of the receptor, which in turn reveals its subunit stoichiometry. When the receptor contains more than one type of subunit, the antibody tagging is repeated for each subunit. In this way, the subunit arrangement within the receptor can be determined.

Receptor isolation

In a typical experiment, we transfect tsA 201 cells growing on five 162-cm2 culture flasks with the appropriate complementary deoxyribonucleic acid(s) (cDNA). We then wait for 48 h to allow receptor expression. Cells are lysed, and nuclei and cell debris are pelleted by low-speed centrifugation (700 × g for 5 min in a cooled Eppendorf bench centrifuge). The supernatant from this step is spun at 21,000 × g for 15 min in the same centrifuge. The pellet from this spin is dissolved in a detergent solution (usually 1% CHAPS) by incubation at 4°C for 1 h on a rotating wheel. Unsolubilised material is precipitated by centrifugation at 100,000 × g for 1 h, and the supernatant is then incubated with pre-washed Ni2+-agarose beads at 4°C for 30 min. The beads are washed three times in buffer, and the bound receptor is eluted with two 500-μl batches of 200 mM imidazole followed by one 500-μl batch of 400 mM imidazole. The receptor is usually found mainly in the second 200-mM imidazole batch.

We find that CHAPS efficiently solubilises the receptors that we have studied. However, there is no particular reason why this detergent should be used. Our advice would be to optimise the procedure by trialling a number of detergents, checking for yield and stability of the oligomeric receptor during isolation. The protein yields in our receptor preparations are low. We typically concentrate a fraction of the eluted receptor, using centrifugation in Centricon™ columns, and then run immunoblots of the concentrated material to check for the presence of the receptor subunits. It is also possible to use silver-stained gels to check for the purity of the sample. An alternative method for checking purity is to run a batch of non-transfected cells in parallel with the receptor isolation and to compare the AFM images of samples obtained from transfected and non-transfected cells. In our experience, the images given by samples from non-transfected cells are usually almost featureless, in contrast to the receptor isolates, which appear as spreads of protein particles (Fig. 1).

Fig. 1
figure 1

Typical AFM images of samples prepared from non-transfected (a) and transfected (b) tsA 201 cells. The image in b, adapted from Barrera et al. [6], shows a homogenous spread of particles in a sample isolated from cells transfected with cDNA for the P2X2 receptor subunit. The mean molecular volume of these particles is 409 nm3, close to the volume expected for a homotrimer (390 nm3)

Full size image

Receptor imaging

Isolated proteins are diluted to a final concentration of about 40 pM, and a 45-μl solution of the protein is added to a poly-l-lysine-coated 1-cm mica disc attached to a steel puck. The proteins are allowed to adsorb to the mica for 10 min. The surface of the mica is then washed with Milli-Q water and dried under nitrogen. Imaging is performed with a Multimode atomic force microscope (Digital Instruments). Samples are imaged in air, using the tapping mode. The silicon cantilevers used have a drive frequency ~300 kHz and a specified spring constant of 40 N/m (Mikromasch). The applied imaging force is kept as low as possible (target amplitude ~1.6–1.8 V and amplitude setpoint ~1.3–1.5 V).

The molecular volumes of the protein particles are determined from particle dimensions based on AFM images. After adsorption of the receptors onto the mica support, the particles adopt the shape of a spherical cap. The heights and half-height radii are measured, and the molecular volume is calculated using the following equation:

$$V_{\text{m}} = \left( {{{\pi h} \mathord{\left/ {\vphantom {{\pi h} 6}} \right. \kern-\nulldelimiterspace} 6}} \right)\left( {3r^2 + h^2 } \right)$$
(1)

where h is the particle height and r is the radius [62].

Molecular volume based on molecular mass can be calculated using the equation

$$V_{\text{c}} = \left( {{{M_0 } \mathord{\left/ {\vphantom {{M_0 } {N_0 }}} \right. \kern-\nulldelimiterspace} {N_0 }}} \right)\left( {V_1 + dV_2 } \right)$$
(2)

where M 0 is the molecular mass, N 0 is Avogadro’s number, V 1 and V 2 are the partial specific volumes of particle and water, respectively, and d is the extent of protein hydration [62]. Because ionotropic receptors are glycoproteins, the volume contributions of core protein and attached oligosaccharides should ideally be calculated separately. Partial specific volumes for protein (0.74 cm3/g) and carbohydrate (0.61 cm3/g) have been reported previously [26]. The extent of glycosylation of a receptor subunit can be determined by comparing the mobilities on gels of fully glycosylated protein and protein that has been deglycosylated with N-glycanase F (e.g. [6]).

When the receptors are dried down, they flatten extensively on the mica substrate, likely because of electrostatic attractions between the protein and the poly-l-lysine used to coat the mica. Consequently, the shape of the protein will become distorted into the spherical cap normally observed. It has been shown previously, however, that there are no significant differences in molecular volumes determined under fluid and in air [62]. Thus, for the extent of protein hydration, we routinely use the value of 0.4 g of water per gram of protein reported for a typical globular protein (human serum albumin) in solution [29]. Particle radii are measured at half the maximal height of the particle in an attempt to compensate for the tendency of AFM to overestimate this parameter when the radii of both particle and tip are similar (i.e. in the nanometre range). By using this method, a very good correlation has been obtained between predicted and calculated molecular volumes for proteins of widely varying molecular masses [62]. Nevertheless, given all the caveats associated with this procedure (i.e., distortion of the protein during attachment, the likely presence of detergent, convolution of the image because of the geometry of the scanning tip), we feel that it is unsatisfactory to rely on volume measurements alone to determine the stoichiometry of the receptor. In this respect, antibody tagging is a powerful corroborative technique.

Imaging of receptor–antibody complexes

We have imaged receptors containing a number of different epitope tags—His6, haemagglutinin (HA), V5, Myc and FLAG. Suitable monoclonal antibodies for receptor decoration are commercially available for all of these tags. Tags can be attached either to the N terminus or the C terminus of the receptor subunit, and the choice of location will depend on where a tag can be added without affecting the assembly of the receptor. Whether or not the tag has any effect on the properties of the receptor should ideally be tested functionally. To some extent, the incubation conditions between receptor and antibody need to be determined by trial and error. The ideal concentrations of receptor and antibody will depend on the yield of receptor from the isolation procedure, which can vary. Typically, we incubate the receptor at a concentration of around 0.2 nM with antibody at around 0.4 nM. The incubation is overnight (about 14 h) at 4°C. After the incubation, the proteins are diluted to a final concentration of about 40 pM, before addition of a 45-μl aliquot to the mica. This procedure will give a spread of receptors decorated with antibodies. Typically, around 20–25% of the receptors have one bound antibody, and about 7–10% of them have two bound antibodies. Few receptors are seen with more than two antibodies bound, as expected from the binomial distribution. Of course, it is theoretically possible to increase the efficiency of decoration by using a higher concentration of antibody. However, because most receptors have more than one subunit that is recognised by the antibody, the receptors and antibodies will tend to form a network, resulting in the production of protein aggregates at higher concentrations of antibody. The method that we use results in the production of a satisfactory number of doubly decorated receptors, which is the key to the success of the method. One possible way to circumvent the problem of aggregate formation is to use Fab fragments of antibodies, which are monovalent. We have found recently that Fab fragments do bind to epitope tags on receptors. However, we have not yet explored their use in producing more efficient receptor decoration.

The concentrations of receptor and antibody detailed above result in the production of a spread of reasonably well-separated particles. The AFM images are analysed, and the numbers of receptors that are undecorated or decorated by either one or two antibodies are counted. Of course, it is necessary to be sure that an antibody is really bound to a receptor and has not simply attached next to it on the mica. To check this, we run two control experiments—imaging receptors alone and after incubation with a control antibody against an epitope tag not present on the receptor. We then compare the tagging profiles for the various conditions and confirm that the positive conditions give a significantly greater number of antibody-tagging events. We also measure the height of the ‘saddle’ between the receptor and the antibody and consider only those events where the saddle is at least 0.2 nm high to be genuine tagging events.

Once a reasonable number (i.e. at least 40) of receptors doubly decorated by antibodies have been identified, we then measure the angles between the two bound antibodies by joining the height peaks of the antibody to the height peak of the receptor. We then produce a frequency distribution of the angles and determine where the peaks of the distribution lie. The angle peaks provide the key information about the subunit stoichiometry and arrangement within the receptor. The angle peaks are typically broad, likely reflecting both flexibility in the attachment of the antibody to the receptor and the fact that the receptors attach to the mica in a variety of orientations. This latter complication is discussed further below.

Automated recognition of receptors and receptor-antibody complexes

Our AFM analysis involves the measurement of the dimensions of receptor particles using the AFM software. In addition, we measure the angles between antibodies attached to the receptor subunits, performing a geometric analysis of complexes. Initially, we carried out this analysis manually, which is an extremely laborious process. For a large data set, an operator will have to interpret several hundred images, while a typical image may contain more than a hundred receptors and/or receptor–antibody complexes.

We have now developed a framework for the automation of our AFM analysis [3] and validated it on P2X6 receptors, which we had already characterised extensively. The receptors were isolated and subjected to AFM analysis. We compared the results obtained by the conventional manual analysis with those given by the automated scheme. We found that the two methods generated very similar results, suggesting that this automated method can be used for determining the architecture of other multi-subunit protein complexes. We now apply the method routinely in our analyses.

Homo-oligomeric receptors

We first applied our method to a relatively simple ionotropic receptor system—P2X receptor homomers [6]. We began by addressing the stoichiometry of the P2X2 receptor homomer. A His6 tag was engineered onto the N terminus of the P2X2 receptor subunit, and the receptors were expressed in tsA 201 cells and isolated as described above (Fig. 1). When the receptors were imaged, the molecular volume distribution had a single peak at 409 nm3. The predicted size of a single P2X2 receptor subunit is 130 nm3. Hence, the measured size was close to the value expected for a trimer (390 nm3). The small discrepancy between measured and expected values is likely a consequence of the presence of detergent attached to the isolated receptors. To confirm the trimeric arrangement of subunits, we incubated the receptor with an antibody against its epitope tag and imaged the receptor–antibody complexes. We found that the frequency distribution for angles between pairs of antibodies bound to the receptor had a single peak at 123° (Fig. 2). This result (close to 120°) again pointed to a trimeric structure for the P2X2 receptor.

Fig. 2
figure 2

Analysis of complexes between P2X2 receptors and antibodies against an N-terminal His6 tag. a Representative AFM image of a receptor liganded by two antibodies. b Distribution of angles between pairs of bound antibodies. The mean angle is 123°, close to the value expected for a homotrimer (120°). The figure is adapted from Barrera et al. [6]

Full size image

Unlike the P2X2 receptor, the P2X6 receptor, when expressed exogenously in cells such as Xenopus oocytes or cultured olfactory bulb neurons, is not efficiently delivered to the plasma membrane [2, 10]. Instead, transport of the receptor is arrested in the endoplasmic reticulum, suggesting that it might be retained by the ‘quality control’ machinery of the cell. A possible reason for this retention is that it is not being correctly assembled, perhaps because the P2X6 receptor subunits are unable to oligomerise. To test this idea, we produced P2X6 subunits tagged at their C termini with a His6 epitope tag. When we imaged the receptor by AFM, the receptor particles were clearly smaller than those seen with P2X2, and the molecular volume frequency distribution had a single peak at 145 nm3, close to the predicted value of 97 nm3 for a single P2X6 receptor subunit. This AFM analysis, therefore, confirmed that the P2X6 receptor is unable to oligomerise [6].

To account for the failure of the P2X6 receptor subunit to oligomerise, we looked for anomalous features in the P2X6 receptor sequence. Of the seven P2X receptor isoforms, the P2X6 subunit has a particularly hydrophobic N terminus. We speculated that this region might interact with an (unknown) ER chaperone protein in such a way as to prevent its assembly into receptor oligomers. Assuming that this interaction was likely to be hydrophobic, we added positive charge to this N-terminal region by mutating two serine residues (residues 3 and 11) to lysines. We found that the introduction of these mutations caused a fourfold increase in cell surface expression of the receptor. When we expressed the mutated sequence in tsA 201 cells, isolated the receptors and imaged them by AFM, we found two types of particles—large and small [57]. The frequency distribution of molecular volumes now had two peaks, at 120 and 340 nm3, consistent with the presence of both monomers and trimers. Analysis of the numbers of particles within the two peaks indicated that 42% of the particles were receptor trimers. We also imaged the isolated mutant P2X6 receptors after incubation with an anti-His6 antibody that recognises the C-terminal His6 tag. We found that some of the larger particles were decorated with one, two or occasionally three antibodies. The angles between pairs of bound antibodies were measured, and a frequency distribution of angles was constructed. The distribution had a single peak at 124°, confirming that the mutant receptor was able to form homotrimers.

Our results suggest that the N terminus of the P2X6 receptor is normally preventing its assembly into trimers. It is known that the P2X6 subunit can associate with both P2X2 and P2X4 to form heteromeric receptors, which have properties distinct from the corresponding homomers [27, 41]. Furthermore, P2X6 is co-localized with both P2X2 and P2X4 in many parts of the central and peripheral nervous systems [22, 42, 43, 61, 64, 66, 69, 72]. It is possible, therefore, that the P2X6 subunit operates as a modulatory subunit rather than a receptor in its own right. The retention of P2X6 in the endoplasmic reticulum would then prevent non-functional subunits from reaching the cell surface and provide an intracellular pool of subunits ready to be incorporated into heteromeric receptors. Furthermore, the role of P2X6 as a modulatory subunit might be regulated in response to changes in circumstances. For example, expression of P2X6 is known to change under pathological conditions such as cancer and zinc deficiency [20, 51, 58, 71]. We have begun to examine the control of P2X heteromer assembly, and these results are discussed below.

Another receptor homomer that we have studied is the 5-HT3A receptor. This receptor belongs to the Cys-loop superfamily, and EM imaging of isolated native 5-HT3 receptors has revealed the pentameric structure characteristic of this receptor superfamily [11, 12]. Two 5-HT3 receptor subunits have been identified—A [46] and B [23, 25]. The A-subunit forms functional receptor homomers, whereas the B-subunit forms 5-HT3A/B receptor heteromers but not homomers. We expressed 5-HT3A subunits bearing a Myc-His6 tag at its C terminus in tsA 201 cells [5]. When the isolated receptor was imaged, the frequency distribution of molecular volumes was broad, with a peak at 757 nm3. The value predicted on the basis of the molecular mass of a 5-HT3A receptor pentamer was 511 nm3. Hence, the discrepancy between observed and predicted molecular volumes was much greater that was seen for the P2X2 receptor (above). One likely reason for this is the presence of twenty transmembrane domains in a 5-HT3 receptor pentamer (compared with only six in a P2X2 receptor trimer), which would attract a greater amount of bound detergent. Incidentally, this result does emphasize the difficulty in trying to estimate the subunit stoichiometry of a transmembrane receptor on the basis of molecular volume calculation alone. We next imaged the receptor in complex with either anti-Myc or anti-His6 antibodies. In both cases, singly and doubly tagged receptors were seen, and the occurrence of these complexes was far more common than was seen when a control antibody (anti-V5), or no antibody was used, indicating that both anti-Myc and anti-His6 antibodies specifically decorate the receptor. For both antibodies, the frequency distribution for angles between pairs of bound antibodies had two peaks, one at around 72° and one at around 144°, consistent with the decoration of either adjacent or non-adjacent subunits in a receptor pentamer. The areas under the two angle peaks were approximately equal, indicating that there was no preference for the binding of the pairs of antibodies at either an acute or an obtuse angle. Hence, there was no evidence for steric hindrance in the binding of the antibodies to the receptor.

Our results for homomeric receptors indicate that our method gives accurate information about subunit stoichiometry. We now go on to consider the application of this technology to more interesting examples where the receptor contains two or more different subunits.

Receptors containing two types of subunit

We alluded above to the fact that P2X6 subunits are able to form hetero-oligomers with both P2X2 and P2X4. We decided to examine the stoichiometry of P2X2/6 heteromers and specifically to test the idea that the stoichiometry of a P2X heteromer might be plastic and dependent on relative levels of subunit expression. To do this, we engineered different tags on the two subunits—initially His6 on P2X2 and HA on P2X6. When we compared the expression of the two subunits in cells that had been transfected with a mixture of equal amounts of the two cDNAs by immunoblotting, we found that the ratio of expression of the two subunits was about 4 P2X2:1 P2X6 [4]. When His6-containing proteins were isolated and imaged, the frequency distribution for molecular volumes had three peaks, at 115, 217 and 360 nm3. Because the predicted volumes of P2X2 and P2X6 subunits are around 130 and 97 nm3, respectively, the peaks likely represent receptor monomers, dimers and trimers. Each particle in the AFM images could be assigned on the basis of its size to one of these peaks. When the receptors were incubated with either anti-His6 or anti-HA antibodies, large particles, corresponding to receptor trimers, decorated by the antibodies could be identified. We found that there were about five times as many receptors decorated by two anti-His6 antibodies as by two anti-HA antibodies. Hence, the numbers of receptors doubly decorated by one antibody or the other were in the approximate ratio of subunit expression. Furthermore, when the epitope tags on the subunits were switched, to produce HA-tagged P2X2 and His6-tagged P2X6, the antibody decoration profile was reversed. Next, we changed the ratio of cDNAs used in the transfection so that 2.5 times as many P2X6 subunits as P2X2 subunits were expressed. When the receptors were isolated, we found that about 2.5 times as many receptors were doubly decorated with anti-HA antibodies (against the tag on the P2X6 subunit) as by anti-His6 antibodies (against the tag on the P2X2 subunit). Our conclusion from these results was that the subunit stoichiometry of the P2X2/6 hetero-oligomer depends on the relative levels of subunit expression.

The antibody decoration profile of the 5-HT3A receptor homomer has been described above. Although 5-HT3A homomers probably exist in vivo, most endogenous 5-HT3 receptors have properties consistent with the presence of both A- and B-subunits [24, 32, 35, 73]. For instance, the 5-HT3A homomer has a much lower single-channel conductance than the 5-HT3A/B heteromer, and the Hill slope for agonist activation of the homomer is higher than that for the heteromer [23, 25, 38]. We were interested, therefore, in determining the subunit stoichiometry and arrangement of the 5-HT3A/B receptor heteromer. Because the receptor is a pentamer, the possible permutations containing two subunits are more complex than for the trimeric P2X receptor. Specifically, there are six ways in which a pentamer containing A- and B-subunits can be built (Fig. 3). To solve the receptor architecture, we engineered different epitope tags on the two subunits—Myc-His6 on the A-subunit and V5-His6 on the B-subunit [5]. We transfected tsA 201 cells with equal amounts of the two cDNAs, and isolated receptors from the transfected cells. After AFM imaging of the receptor alone, the frequency distribution of the molecular volumes was broad (as for the 5-HT3A receptor) and had a single peak at 704 nm3, smaller than the value for the 5-HT3A homomer, which was expected because the B-subunit has a lower molecular mass than the A-subunit. When the receptors were incubated with either anti-His6, anti-Myc or anti-V5 antibodies, doubly decorated receptors were observed (Fig. 4a). The frequency distributions of angles between pairs of bound antibodies had two peaks, at about 72 and about 144°, for both anti-His6 and anti-V5 antibodies, but only one peak, at 144°, for the anti-Myc antibody (Fig. 4b). These results indicate that pairs of B-subunits (with the V5 tag) can either be adjacent or non-adjacent but that pairs of A-subunits (with the Myc tag) could only be non-adjacent. The only subunit stoichiometry and arrangement that is consistent with this antibody decoration profile is B-B-A-B-A (arrangement ‘e’ in Fig. 3).

Fig. 3
figure 3

Six possible subunit arrangements in a 5-HT3A/B receptor heteromer

Full size image
Fig. 4
figure 4

Analysis of complexes between 5-HT3A/B receptors and subunit-specific antibodies. a Gallery of images, adapted from Barrera et al. [5], of receptors that are either unliganded (top row) or liganded by either one (middle row) or two anti-V5 antibodies (bottom row), directed against the B-subunit. b Schematic illustration of the distribution of angles between pairs of anti-Myc antibodies, directed against the A-subunit (top) and anti-V5 antibodies (bottom). In the study reported in Barrera et al. [5], 40 receptor–antibody complexes were analysed in each case. The angle distributions indicate that the subunit arrangement around the central ion pore is B-B-A-B-A

Full size image

Receptors containing three types of subunit—determination of the orientation of the receptor

Many types of GABAA receptor contain three different subunits [45]. The most common form of the receptor in the brain is the α1β2γ2 form [45], which has the subunit stoichiometry 2α:2β:1γ [28]. The subunit arrangement within the receptor has been shown by the use of subunit concatenation to be α,β,α,γ,β, anti-clockwise when viewed from the outside of the cell [8, 9]. The AFM-based method described above (e.g. for the 5-HT3A/B heteromer) should be able to demonstrate that neither the two α-subunits nor the two β-subunits are adjacent. Indeed, we showed several years ago, by placing a His6 epitope tag on the α-subunit and imaging complexes between the isolated receptor and anti-His6 antibodies, that this receptor contains two non-adjacent α-subunits [53]. However, determination of the absolute subunit arrangement by AFM analysis requires finding the position of the γ-subunit (for example) in relation to the α- and β-subunits, to distinguish between the arrangements αβαγβ and αβαβγ. Two further advances are needed to accomplish this task: (1) decoration of the receptor simultaneously with distinguishable ligands for two types of subunit and (2) determination of the orientation of the receptor on the mica support.

We have addressed these challenges using the α4β3δ form of the receptor as a model system. The α4β3δ receptor, although a minor component of the total GABAA receptor population, has some interesting properties, such as an extrasynaptic location [14, 56, 67], a high sensitivity to GABA [1, 16] and a slow rate of desensitization [16], factors that allow it to exert a tonic inhibition of neuronal excitability [19]. It also has a potential involvement in epilepsy [15, 21, 63, 68]. Consequently, it is a promising target for the development of novel drugs.

We transfected tsA 201 cells with cDNAs for α4-, β3- and δ-subunits. The α-subunit had a FLAG-His6 epitope tag, the β-subunit had a V5-His6 tag, and the δ-subunit had an HA-His6 tag. All tags were on the C termini of the subunits. Receptors were isolated from the transfected cells by the usual procedure. Fab fragments of the anti-FLAG antibody were generated using papain digestion. When receptors were incubated with these Fab fragments, receptor–Fab complexes were produced (Barrera et al., unpublished data). A frequency distribution for angles between pairs of bound Fabs had a single peak, at about 144°, indicating that the α-subunits are non-adjacent. We then incubated the receptors with both anti-FLAG Fabs and anti-HA antibodies, to decorate both α- and δ-subunits simultaneously. We identified receptors that had been decorated with two Fabs and one antibody. A representative image is shown in Fig. 5a. Note that the Fabs can be clearly distinguished from the whole antibody on the basis of their smaller size. Hence, we now have a method for finding the position of the δ-subunit relative to that of the two α-subunits.

Fig. 5
figure 5

AFM analysis of α4β3δ GABAA receptors. a Representative AFM image of a receptor (large central particle) liganded by one anti-HA antibody, directed against the δ-subunit (larger peripheral particle, arrow), and two anti-FLAG Fab fragments, directed against the two α-subunits (smaller peripheral particles, arrowheads). b Illustration of the experiment designed to determine the orientation of the receptor on the mica support. Twice as many receptors were decorated with concanavalin A (con A) or monoclonal antibody bd17 when the receptors were bound to poly-l-glutamate than when they were bound to poly-l-lysine, indicating that the receptors normally prefer to bind extracellular face down to poly-l-lysine-coated mica

Full size image

The receptors are normally bound to poly-l-lysine-coated mica, which provides a positively charged surface. We speculated that the receptors would bind to this support through a negatively charged surface on the protein. This is most likely to be the extracellular domain, which contains negatively charged oligosaccharides, in contrast to the intracellular domain, which contains many positively charged amino acids. If this is indeed the case, then the orientation of the receptors should be reversed when they are bound to poly-l-glutamate-coated mica, which provides a negatively charged surface. To test these ideas, we bound receptors to the two surfaces and incubated them with either the lectin concanavalin A, which should bind to the oligosaccharides on the extracellular face of the receptor [33], or monoclonal antibody bd17, which recognises an epitope at the N terminus of the β-subunit [60]. We reasoned that receptors bound extracellular face down would have their binding sites for both concanavalin A and antibody bd17 occluded. Receptors decorated with smaller particles were observed after incubation with either concanavalin A or antibody bd17, irrespective of the mica coating. Significantly, however, we found that for both concanavalin A and antibody bd17, there was about twice as much binding when the receptor was bound to poly-l-glutamate than to poly-l-lysine, indicating that more extracellular faces were occluded when the mica was coated with poly-l-lysine. We conclude that the receptor normally binds predominantly extracellular face down to poly-l-lysine-coated mica (Fig. 5b). Armed with this information and the ability to decorate two different subunits simultaneously, we are now in a position to determine the absolute subunit arrangement within any GABA receptor containing up to three subunits.

The use of concanavalin A to determine the orientation of a protein on the mica should be applicable to many integral membrane proteins, which tend to be glycosylated on their extracellular faces. The additional use of an antibody directed against an epitope on either the extracellular or the intracellular face will obviously depend on the availability of suitable antibodies.

Future perspectives

We have recently extended our AFM-based method to a study of the subunit stoichiometry of transient receptor potential channels (TRPC). In a preliminary series of experiments, we have confirmed that TRPC1 homomers are tetrameric [7]. It is well known that TRPC subunits assemble to form a variety of heteromeric channels. Our method provides a means of addressing the subunit arrangement within these channels. It also should be applicable to a wide variety of multi-subunit proteins, including other ionotropic receptors and ion channels.