Muscarinic Toxin 7 Selectivity Is Dictated by Extracellular Receptor Loops*

Muscarinic toxin 7 (MT7) is a mamba venom protein antagonist with extremely high selectivity for the M1 muscarinic acetylcholine receptor. To map the sites for the interaction of MT7 with muscarinic receptors we have used chimeric M1:M3 receptors and site-directed mutagenesis of the M3 and M4 receptor subtypes. Two Glu residues in M1, one in extracellular loop 2 and one in extracellular loop 3, were found to be important for the high affinity binding of MT7. Substitution of the corresponding Lys residues in the M3 receptor with Glu converted the M3 mutant to an MT7 binding receptor, albeit with lower affinity compared with M1. A Phe → Tyr substitution in extracellular loop 2 of M3 together with the 2 Glu mutations generated a receptor with an increased MT7 affinity (apparent Ki = 0.26 nm in a functional assay) compared with the M1 receptor (apparent Ki = 1.31 nm). The importance of the identified amino acid residues was confirmed with a mutated M4 receptor constructs. The results indicate that the high selectivity of MT7 for the M1 receptor depends on very few residues, thus providing good prospects for future design and synthesis of muscarinic receptor-selective ligands.

Many components in the venom of mamba snakes (genus Dendroaspis) have been purified and characterized as highly valuable tools in biophysical, biochemical, and pharmacological research. These include ion channel blockers such as the curarimimetic ␣-neurotoxins (1), dendrotoxins (2,3), calcicludines (4) and calciseptines (5), the acetylcholinesterase inhibitors fasciculins (6), the platelet aggregation inhibitor mambin (7), and the muscarinic acetylcholine receptor ligands muscarinic toxins (MTs) 1 (8 -11). MTs are peptides of 65 or 66 amino acids including eight cysteines which form internal disulfide bridges to generate a characteristic compact protein core structure with three protruding loops (12). This so-called three-finger motif is common to many toxin molecules from snake venoms and has by evolution generated a wealth of molecules with high specificity simply by variation of mainly the loop sequences (12,13).
Since the first discovery of MT activity in chromatographic fractions of Dendroaspis angusticeps venom (8) much effort has been put into purification and pharmacological characterization of different MTs (for review, see Refs. 14 and [15][16][17]. This has led to the identification of receptor subtype non-selective MTs, such as MT␣ from Dendroaspis polylepis (18), modestly receptor subtype selective MTs, such as MT3 (19), also referred to as m4-toxin (14), and an extremely receptor subtype selective MT7 (11,20), also referred to as m1-toxin1 (9,21). MT7 binds to muscarinic M1 receptors in the pico-to nanomolar range when assayed with radioligand binding with no detectable binding to the other subtypes, except for a low affinity (micromolar) binding to M4 receptors (9, 20 -22). Despite the well characterized pharmacological profile of the MTs, virtually no information is available concerning the structural basis for interaction with the receptors. There are several reasons for this. One is that the natural source of MTs is limited, and because especially MT7 is also very scarcely expressed in the venom (11,23), the availability for research is limited. This has during the last years been partially overcome by the production of recombinant MT7 in insect cells (20) and in yeast (21) as well as by chemical synthesis (22). Another reason is the lack of high resolution structure of muscarinic receptors and, as a logical consequence, a receptor-MT complex. Considering the large size of the MT ligands and the very high affinity, one can assume that several sites of contact exist between the receptor and the toxin molecule. This has in the case of acetylcholinesterase and fasciculin, which is structurally closely related to MTs, been demonstrated by resolving the crystal structure of the bound complex (24,25). Subsequent mutagenesis studies of this protein pair also have demonstrated that rather few of the interactions are important for high affinity binding (26,27). This led us to hypothesize that the high selectivity of MT7 for M1 receptors might be due to a limited number of receptor-toxin interactions based on the following. (i) Comparison of the primary sequences of different MTs with different receptor selectivity profiles suggests that rather few amino acid differences can have a large impact on the selectivity, (ii) the close homology between muscarinic receptor subtypes, especially in the ligand binding transmembrane domains, suggests that a common binding site for all MTs may exist and that the high affinity binding and selectivity comes from additional specific interactions with more divergent receptor domains, such as the extracellular domains. Our aim was, thus, to identify interaction sites for MT7 in the receptor structure. For this we chose the MT7 non-binding M3 receptor as a target for mutagenesis. Initial screening was performed with chimeric M1:M3 receptors to identify larger domains involved in the binding. The results show that a limited number of mutations in the extracellular loops 2 and 3 of the M3 receptor converts it to a high affinity MT7 binding receptor. The generality of the results was further confirmed by mutagenesis of the M4 receptor. Chimeric Constructs and Mutagenesis-The human M1 receptor with a signal sequence-flag amino-terminal extension, which contains a cleavable signal sequence previously shown to enhance expression and a FLAG epitope (28), and the human M3 and human M4 receptors have been described earlier (20,29). All chimeric constructs and site-directed mutagenesis constructs were generated using the PCR technique with in-house-synthesized specific oligonucleotide primers with desired mutations. Subcloning was performed using the pBluescript KS vector (Stratagene) or the pFastBac1 vector (Invitrogen), and the correct sequences were verified by automated DNA sequencing.
M3-E was generated by fusing the intact M3 sequence with M1:M3-V at the unique AatII site (nt ϩ1328). For further site-directed mutagenesis of the M3 receptor, an NheI site (nt ϩ648) was generated to facilitate subcloning, leading to a change of the Pro-Pro dipeptide in the M3 sequence to a corresponding Leu-Ala dipeptide in M1 receptor (M3LA). M3LA-E was generated by fusing the M3LA sequence with M3-E at the unique AatII site and M3ELA-E by fusion of M1:M3-VI with M3LA-E at the NheI site. M3ELAQ-E, M3ELAY-E, and M3ELAQY-E were constructed by insertion of PCR fragments with desired mutations into M3ELA-E between the NheI and AatII sites.
M4ELY was constructed by fusion of two PCR fragments with an introduced XbaI site (nt ϩ540), which also generated the Pro-181 to Leu mutation. The Lys-177 to Glu mutation was included in the 5Ј fragment, and Phe-186 to Tyr was included in the 3Ј fragment.
For the fura-2-Ca 2ϩ assay with the M4 receptor constructs we generated a chimeric G␣ 11 :G␣ i protein, referred to as G 11 ␣ i5 . cDNA for bovine G␣ 11 (G␣ L2 ) in pVL1393 (30) was a kind gift from Dr. T. Haga (University of Tokyo, Tokyo). The cDNA was transferred to pBluescript KS, and the sequence coding for the last five amino acids was exchanged to ␣ i sequence by fusion of a pair of overlapping oligonucleotides to the PstI site (nt ϩ1049).
All new constructs described above, except for M1:M3-IV, were finally transferred to the pFastBac1 vector for generation of recombinant baculovirus with the Bac-to-Bac baculovirus expression system (Invitrogen). M1:M3-IV was originally constructed for another purpose in the pLucGRBac vector (31), and baculovirus was generated by cotransfection of vector and linearized baculovirus DNA (Pharmingen) into Sf9 cells.
For recombinant protein expression Sf9 cells were plated on tissue culture dishes and allowed to attach. A high titer virus stock was then added to give a multiplicity of infection of 3-5. For the functional assay the infections were allowed to proceed for 26 -28 h, and cultures were infected for radioligand binding for 48 h.
Functional Assay-Receptor-expressing cells were assayed essentially as described in Kukkonen et al. (29). Cells were detached from plates, centrifuged 150 ϫ g for 4 min, resuspended in a smaller volume growth medium, and incubated for 20 min with 4 M fura-2 acetoxymethyl ester at room temperature. Thereafter cells were centrifuged 150 ϫ g for 4 min, washed once in assay buffer, resuspended in assay buffer, and divided into aliquots. The aliquots were briefly spun down in a microcentrifuge, the medium was removed, and cell pellets were stored on ice until use. Before fluorescence recording, the cells were resuspended in assay buffer with or without MT7 for 5 min. Thereafter the cells were placed in a cuvette in a thermostatted (27°C) cuvette holder with magnetic stirring in a fluorescence spectrophotometer (Photon Technology International), and fluorescence recording at alternating 340 and 380 nm (excitation) and 510 nm (emission) was started. Different concentrations of carbachol were added in a cumulative fashion, and calibration of each experiment was performed with 0.04% Triton X-100 to measure F max and 11 mM EGTA to measure F min . Intracellular free Ca 2ϩ concentrations ([Ca 2ϩ ] i ) were calculated from fluorescence obtained at 340 nm according to [Ca 2ϩ (32)), the extracellular fura-2 fluorescence being subtracted from the F values.
Radioligand Binding-Infected cells from monolayer culture were harvested in phosphate-buffered saline solution, centrifuged 1500 ϫ g, and stored as pellets at Ϫ80°C until use. Data Analysis-Non-linear curve-fitting of the dose-response data was performed using GraphPad Prism (GraphPad Software). The apparent K i values in the functional assay were obtained from data fitted to the equation and [I] are the concentrations of agonist and inhibitor, respectively, EC 50 is the [A] producing half-maximal stimulation, and K i and K i Ј are the surmountable ("competitive") and the insurmountable ("noncompetitive") inhibition constants, respectively. The K i value gives a measure of the right shift of the dose-response curve, and the K i Ј value gives a measure of the suppression of the maximum response. The equation was applied earlier to functional antagonism analysis of muscarinic receptor-induced Ca 2ϩ increases (33,34). In this study only the apparent K i values are given because, whereas the apparent K i values were similar irrespective of MT7 concentration used, the apparent K i Ј values for several constructs varied significantly. This could be due to complex inhibition patterns like the ones displayed by atropine on the M1 and M5 receptors, where the insurmountable part of the inhibition appears to saturate, and only the surmountable antagonism can, thereafter, be increased with increasing antagonist concentrations (34). The observed insurmountable inhibition was nevertheless taken into account with all constructs in calculation of the apparent K i values. The IC 50 values from radioligand binding data was obtained by one site competition fitting using GraphPad Prism.

Measurements of [Ca 2ϩ
] i responses were done using suspensions of baculovirus-infected Sf9 cells. Activation of M1 receptors with carbachol resulted in an immediate increase in [Ca 2ϩ ] i as determined from fura-2 fluorescence (Fig. 1A). The basal Ca 2ϩ level was 101 Ϯ 7 nM (mean Ϯ S.D., n ϭ 3). In this heterologous expression system, the Ca 2ϩ responses are very stable in the continuous presence of agonist, declining very slowly toward the basal level. The addition of MT7 during the stable phase resulted in inhibition of the receptor-mediated response in the same manner as with atropine, with a restoration of cellular basal Ca 2ϩ level (Fig. 1A). Preincubation with MT7 resulted in a total blockage of response to low concentrations of carbachol but was competitively relieved at higher agonist concentrations (Fig. 1B). Using similar assay conditions, we next examined the dose-dependent inhibition of the M1 receptor with MT7 ( Fig. 2A). The inhibition appeared to be both insurmountable, in terms of depression of response maxima, and surmountable, in terms of right-shifted dose-response curves. The calculated apparent K i value for MT7 was 1.31 Ϯ 0.36 nM (mean Ϯ S.D., n ϭ 5). We used the M1 receptor construct with a signal sequence-flag amino-terminal extension that is expressed at higher levels and gives greater responses compared with the native M1 (29). The amino-terminal extension did not affect the inhibition pattern of MT7 (data not shown). The M3 receptor was not inhibited by MT7 up to 100 nM (Fig. 2B).
To get insight into the difference in selectivity of MT7, we constructed chimeric M1:M3 receptors with M1 receptor middle portion and M3 receptor sequences at either end (Fig. 3). The functional assay data for all constructs are compiled in Table I. The M1:M3-I construct did not show significantly altered MT7 affinity, but M1:M3-II and M1:M3-IV displayed an ϳ10-fold decrease in MT7 affinity and no obvious insurmountable inhibition (Fig. 4, A and B). Because the M1 and M3 receptors are very homologous in the transmembrane regions, we hypothesized that the outer loops are involved in the binding. Comparison of primary sequences of outer loops 2 and 3 showed that there are several amino acid differences in these regions between M1 and M3 receptors (Fig. 5). Quite dramatic differences are represented by the negatively charged Glu-170 in loop 2 and Glu-397 in loop 3 of the M1 receptor, the corresponding residues in the M3 receptor being positively charged lysines. Mutations of these Lys residues to Glu in the chimeric constructs M1:M3-VI and M1:M3-V rescued the M1:M3-II and M1:M3-IV, respectively, in terms of MT7 affinity (Fig. 4, C and  D). The M1:M3-VI construct displayed an even lower surmountable apparent inhibition constant as compared with the M1 receptor and a substantial insurmountable inhibition by MT7.
Mutation of Lys-523 in loop 3 of the M3 receptor to Glu  (M3-E) did not produce a receptor inhibited by MT7, nor did the M3LA or M3LA-E mutants ( Fig. 5 and Table I). Substitution of Lys-213 to Glu in loop 2 (M3ELA-E) resulted in a detectable inhibition in the functional assay with 100 nM MT7 but with a 20-fold lower affinity as compared with M1. Further comparison of sequences in outer loop 2 inspired us to mutate the Glu-220 to a corresponding Gln in M1, and Phe-222 to a corresponding Tyr in M1 (Fig. 5). Whereas the M3ELAQ-E mutant did not show increased MT7 affinity as compared with M3ELA-E, the M3ELAY-E construct was very potently inhibited by MT7 and displayed an affinity about 5-fold higher than the M1 receptor (Fig. 6A). Although the Glu-220 3 Gln mutation did not improve the MT7 affinity, it had a major influence on the response kinetic in the M3ELAQY-E construct (Fig. 6B), displaying substantial insurmountable inhibition in a fashion similar to M1:M3-VI (Fig. 4C). The basal Ca 2ϩ level in cells expressing chimeric constructs and M3 mutants was not significantly different from cells expressing M1 receptors and ranged from 85 to 134 nM (n ϭ 12).
Encouraged by the results with the M3ELAY-E construct, we examined the sequences of the other muscarinic receptor subtypes. The human M4 subtype has an Asp at the equivalent position of Lys-523 in M3 but, otherwise, the same residues as M3 at the other two critical sites in the outer loop 2 (Fig. 5). Thus, the Lys-177 in M4 was mutated to a Glu and the Phe-186 to a Tyr in M4ELY. The additional Pro to Leu mutation was a result of an introduced XbaI site to facilitate subcloning. The M4 receptor is a G i /G o protein-coupled receptor not linked to [Ca 2ϩ ] i increases when expressed in Sf9 cells (data not shown). To apply the same functional assay to the M4 receptor constructs, we generated a chimeric G 11 ␣ i5 protein, a generally used method to switch effector coupling of G protein-coupled receptors (35,36). In coexpression experiments with G 11 ␣ i5 and the native M4 receptor, receptor stimulation resulted in robust [Ca 2ϩ ] i increases, but we could not detect inhibition with 100 nM MT7 (Fig. 7A). The M4ELY construct, on the other hand, was potently and competitively inhibited by MT7 (Fig. 7B), with an apparent K i of 5.04 Ϯ 1.12 nM (mean Ϯ S.D., n ϭ 4), which is close to the apparent K i value obtained with the M1 receptor.
To confirm the findings in the functional assay, we performed [ 3 H]NMS radioligand binding to cell homogenates expressing different constructs (Table II). It was found that the expression levels varied substantially among constructs, with  some constructs at the limit of detection. However, in the functional assay these poorly expressed constructs were still functional with regard to [Ca 2ϩ ] i increases, albeit showing reduced agonist potencies (Table I). MT7 has been demonstrated to bind pseudo-irreversibly to M1 receptors (37,38) and to trap prebound ligands in the receptor (16,37). For inhibitionconstant determinations, we therefore preincubated receptors with MT7 before the addition of  (Table II)  The ability of MT7 to trap ligands in the M3ELAY-E mutant was assessed by measurement of atropine-induced dissociation of prebound [ 3 H]NMS as previously described (21,22,38). The Addition of MT7 (10 nM) to homogenates preincubated with [ 3 H]NMS did not affect the total or the specific [ 3 H]NMS binding (data not shown). On the other hand, the dissociation of [ 3 H]NMS from M3ELAY-E was decelerated 3.7-fold by MT7 compared with a 2.9-fold deceleration from M1 (Fig. 8). DISCUSSION We have identified three amino acid residues on the extracellular side of the M1 muscarinic receptor that are essential for high affinity binding and critical for subtype selectivity of the mamba venom antagonist MT7. Negatively charged residues in extracellular loops 2 and 3 appear to be the main determinants together with a tyrosine residue in loop 2.
MT7 appears to bind allosterically to M1 receptors. This is based on the fact that a prebound ligand, such as [ 3 H]NMS, is not competed away by MT7 but, rather, gets trapped in the receptor (16,37). On the other hand, prebound MT7 blocks binding of [ 3 H]NMS (14,20,38,39). The orthosteric site, to which common competitive ligands bind, is located within a pocket built up of the transmembrane helices (for review, see Ref. 40). In the rhodopsin structure, the only available high resolution structure of a G protein-coupled receptor, the outer loop 2 folds down into the ligand binding pocket and contributes to ligand binding, in this case the 11-cis-retinal chromophore (41). Although not explicitly shown for muscarinic receptors, the loop 2 has also been implicated in ligand binding in adrenergic receptors (42), in adenosine receptors (43,44), and in dopamine receptors (45), suggesting a general mode of binding for rhodopsin-like receptors. In light of the findings in this report, it is possible that MT7, by its interaction with the outer loop 2, stabilizes the roof of the ligand binding pocket and thereby also blocks the entrance to the orthosteric site. Such binding would resemble the way fasciculins interact with acetylcholinesterase, competing with allosteric ligands and also blocking the entrance to the active site (24,25,46). On the other hand, we cannot exclude that some part, most likely the central finger, of MT7 reaches down to the ligand binding pocket to interact with transmembrane helix residues, as has been suggested for MT2 based on the determined NMR resolution structure of the toxin (12).
Based on the information obtained with the chimeric M1:M3 constructs, the M3 receptor could, thus, be converted by mutations of a limited number of residues (M3ELAY-E) to a receptor with even higher sensitivity to MT7 than the M1 receptor, suggesting that additional binding sites are more favorable in the M3 receptor. There are many positively charged Arg and Lys residues in MT7, and an Arg to Ala mutation at the tip of toxin loop II reduces the toxin binding to M1 receptors about 100-fold (22). Three-dimensional modeling also indicates that MT7 has a predominantly positively charged surface (22). It is, therefore, likely that negatively charged Glu residues in loops 2 and 3 of the M1 receptor interact with positively charged Lys/Arg residues of MT7. It has also been noticed that the last amino acid residue of MT7 (Lys-65) is important for stabilizing the toxin-receptor interaction, although not obviously involved in determining binding affinity (21).
Whereas the M3-E and M3LA did not show strongly reduced expression levels nor reduced functional efficacies, the combi-nation of the mutations in M3LA-E as well as in M3ELA-E and M3ELAQ-E resulted in severely impaired expression, albeit still functional. In rhodopsin the residues corresponding to LA are located close to TM7 and loop 3 in the x-ray structure (41). Thus, it is possible that intramolecular contacts exist between extracellular loops 2 and 3 that are critical for proper folding and for the formation of the disulfide bridge between Cys-221 in loop 2 and Cys-141 at the end of outer loop 1. Disruption of this conserved disulfide in M3 has previously been shown to impair receptor surface expression but not the functional activation of G proteins (47). The introduction of the tyrosine in loop 2 (M3ELAY-E and M3ELAQY-E) appeared to rescue the mutants in terms of expression and agonist potency. This phenomenon we cannot explain with the current knowledge about extracellular domain structures.
Comparison of the M3ELA-E and M3ELAY-E constructs suggests that the hydroxyl group of the tyrosine in loop 2 is directly involved in binding MT7. It is becoming clear that the residue carboxyl-terminally adjacent to the disulfide-forming Cys residue of loop 2 is of fundamental importance for ligand binding in the rhodopsin family of receptors. In rhodopsin itself, a cluster of residues around the Cys forms part of the retinal binding (41). In ␣ 1 -adrenergic receptors, a cluster of residues, including the one corresponding to Tyr-179 in M1, is responsible for subtype-specific binding of a number of antagonists (42). Selective ketanserin binding to human 5-HT 1D receptor over the canine homologue is partially dependent on the Cys-adjacent residue (48), and the corresponding residue in the D 2 dopamine receptor is also involved in antagonist binding (45).
The functional antagonism caused by MT7 was somewhat complex. The carbachol-stimulated increase in [Ca 2ϩ ] i through the M1 receptor was rapidly reversed by MT7 in a fashion similar to the commonly used antagonist atropine. This is in agreement with a study, where Max et al. (39) measured the association kinetics of m1-toxin1 (same as MT7) to M1 receptors in a ligand binding assay and found that full binding was achieved in less than 30 s (39). It has also recently been shown that MT7 inhibits functional responses by increasing the agonist dissociation rate (49). Examination of the dose-response curves for M1-induced Ca 2ϩ responses generated in the absence or presence of MT7 indicated both a surmountable (doseresponse right shift) and an insurmountable (maximum response suppression) inhibition. At equilibrium, an allosteric   inhibitor should cause insurmountable inhibition, and a competitive inhibitor should cause surmountable inhibition. However, a strong allosteric inhibitor can produce surmountable inhibition behavior (for review, see Ref. 50), and competitive inhibitors have been shown to exhibit insurmountable inhibition, especially when measuring rapid responses under nonequilibrium conditions (33,34,51), as in the case of our functional assay. The insurmountable behavior of orthosteric ligands has been shown to correlate with the dissociation rate, slower dissociation leading to increased insurmountability, but not to correlate with the affinity of the receptor for the ligand determined at equilibrium (34,51). In this context it is interesting to compare the functional behavior of the M3ELAY-E and M3ELAQY-E constructs, which display similar affinities for MT7 but produce quite different inhibition patterns in the functional assay, with M3ELAQY-E showing strongly depressed response maxima. This suggests that the presence of a Gln or an uncharged residue at position 220 is of importance for tight interactions between the M3 receptor mutants and MT7, albeit not directly being involved in determination of binding affinity.
MT7 has also been reported to bind to M4 receptors (9,20). In an earlier report we estimated the affinity of MT7 for M4 to be 10,000 times lower than for M1, although we were not able to determine a precise affinity constant (20). Nevertheless, based on the functional assay in this report, the affinity of M4 is not high enough to show detectable inhibition of M4 responses at the highest concentration (100 nM) used. Comparison of primary amino acid sequences of muscarinic receptors showed that the human M4 receptor has an Asp residue at the critical Glu-397 in outer loop 3 of M1, whereas the M2, M3, and M5 subtypes do not provide a negative charge at this position. Assuming that the Asp residue is important for an overall detectable binding to M4 receptors, the introduction of a Glu and a Tyr in outer loop 2 would also convert this subtype to a receptor that would bind MT7 with high affinity if there is a general mode of binding of MT7 to muscarinic receptors. This we could also demonstrate with the M4ELY mutant, confirming the findings with the M3 mutants and providing promising perspectives for development of subtype-selective MTs.
In conclusion, we have identified residues in the outer loops of muscarinic receptors that are critical for determining the selectivity of the highly M1 receptor-specific MT7 antagonist. We have also demonstrated that by mutating the corresponding residues in other receptor subtypes, these receptors acquire the capacity to bind MT7 with high affinity. This study, thus, opens up promising prospects for transferring this M1 specificity of MT7 to other receptor subtypes by directed mutagenesis of the toxin molecule.