摘要:

Three-dimensional models of glutamate receptors Michael J. SutcliÜe,*a Allister H. Smeeton,a Z. Galen Wob and Robert E. Oswaldb a Department of Chemistry University of L eicester L eicester UK L E1 7RH b Department of Pharmacology College of V eterinary Medicine Cornell University Ithaca New Y ork 14853 USA Receiøed 5th August 1998 Structural models of glutamate receptors have been produced as part of a multidisciplinary study of neuronal function»both ligand/receptor interactions and ion transport»at the atomic level. The models have concentrated on the agonist binding and transmembrane domains of ionotropic (i.e. ligand-gated) glutamate receptors (iGluRs) and have aided our understanding of the molecular determinants of (1) ligand binding and (2) channel activity.The model building process involved a combination of homology modelling distance geometry molecular mechanics protein»ligand and protein»protein docking electrostatic calculations and manual adjustment in conjunction with restraints from site-directed mutagenesis ligand binding and electrophysiological studies. The initial models were used to produce hypotheses which were tested experimentally ; these models have been subsequently re–ned as part of an extremely eÜective multidisciplinary study using an iterative molecular modelling/experimental veri–cation cycle in which restraints derived from experimental studies are used at all stages and the –ndings from one round of modelling are used as restraints in the next. By studying a variety of agonists and antagonists details have been built up of (1) those residues involved in ligand binding and (2) the role of agonist binding (i.e.agonist-induced conformational change) in channel gating. The models also aid our understanding of the conductance properties of the channels. Introduction Ion channels are key components in the activity of living cells. These channels are formed from membrane-bound proteins and are commonly characterised in terms of their ionic selectivity and gating properties. Ion channels activated by the binding of a ligand»either internally (e.g. calcium ions ATP and cyclic nucleotides) or externally (e.g. the nicotinic acetylcholine receptor channel [nAChR] c-aminobutyric acid [GABA] receptor channel and the glutamate receptor channel)» are known as ligand-gated ion channels.Ligand-gated channels activated by the action of neurotransmitters are involved in fast synaptic transmission in the nervous system. These include glutamate receptors»the primary excitatory neurotransmitter receptors in the vertebrate brain» and play an important role in a wide variety of neuronal functions.1 Classi–cation of glutamate receptors is based on their signal transduction mechanism» metabotropic glutamate receptors (mGluRs) are linked to GTP binding proteins and thus operate through second messengers,2 whilst ionotropic glutamate receptors (iGluRs) function as ligandgated cation channels. iGluRs are cation-selective channels and are classi–ed according to the 259 Faraday Discuss. 1998 111 259»272 Fig.1 Overview of model building procedure. agonists by which they are selectively activated (1) a-amino-3-hydroxy-5-methyl-4-isoxazolepropionate (AMPA) receptors (GluR1 to 4) (2) kainate receptors (GluR5 to 7 and KA1 to 2) and (3) N-methyl-D-aspartate (NMDA) receptors (NMDA-R1 to NMDA-R2D). In addition kainate receptors of lower molecular mass (40»50 kDa also known as kainate binding proteins) have been cloned from non-mammalian vertebrates [e.g. frog (fKBP),3 chick (cKBP),4 and gold- –sh (GFKARa and GFKARb)5] ; these exhibit considerable sequence homologies with the Cterminal half of the 100 kDa mammalian AMPA/kainate receptors. We have produced structural models of the ligand-bound and ligand-free forms of the iGluRs GFKARa GluR1 and GluR6 and the NMDARs NMDA-R1 and NMDA-R2C.These have aided our understanding of the molecular determinants of (1) ligand binding and (2) channel activity. Our initial models of iGluRs6 were used to produce experimentally testable hypotheses; these models have been subsequently re–ned7,8 using an iterative molecular modelling/experimental veri–cation cycle. There are essentially 4 stages in our modelling procedure (Fig. 1) (1) Determine the transmembrane topology of iGluRs. (2) Divide the amino acid sequence into smaller ììmodulesœœ based on step 1. (3) Generate 3-D structural models of each individual ììmoduleœœ. (4) Combine the 3-D models of the diÜerent ììmodulesœœ to form a single structure. Transmembrane topology Knowledge of the transmembrane topology of a membrane-bound protein gives an indication of which regions of the sequence are likely to be close together in the three-dimensional structure» an important prerequisite in any modelling exercise.Thus the –rst step in modelling iGluRs was to determine the transmembrane topology. Although originally thought to be structurally similar Faraday Discuss. 1998 111 259»272 260 to other ligand-gated ion channels,9 experimental analysis of the position of native and engineered N-glycosylation sites protease sites and epitopes (ref. 6 and references therein) demonstrated that the transmembrane topology of iGluRs diÜers signi–cantly from other ligand-gated ion channels. Three membrane-spanning regions (denoted M1 M3 and M4) are present the locations of which were predicted using a consensus of the results from the programs MEMSAT,10 TMAP,11 and PHD topology,12 with the N-terminal region located extracellularly and the C-terminal region located intracellularly.The region denoted M2 originally thought (based on hydropathy pro–les) to cross the membrane does not span the membrane but based on its role in ion conduction is most likely inserted into the ion conduction pathway in a manner similar to the P-segment of K` channels.13 Identi–cation of smaller ììmodulesœœ The production of a structural model of an iGluR in a single step is currently unrealistic. Modelling is made much more tractable by division of the amino acid sequence into a number of smaller ììmodulesœœ based on those regions of the sequence which are predicted to be close together in the three-dimensional structure.The iGluRs are composed of several evolutionary distinct modules:14 (1) An extracellular portion that seems to have arisen from two diÜerent classes of bacterial protein»the N-terminal half is homologous to the leucine isoleucine valine-binding protein (LIVBP; absent from the 50 kDa kainate binding proteins) and the C-terminal half is homologous to the lysine arginine ornithine-binding protein (LAOBP).15,16 The LAOBP-like domain in iGluRs consists of two distinct regions of amino acids separated in sequence (but not in threedimensional structure) by a large insert which includes two of the membrane spanning regions (M1 and M3). (2) A membrane-spanning ion conduction pathway lining region that appears similar in topology to the pore lining region of K` channels.(3) A variable C-terminal regulatory domain of unknown origin which exhibits considerable diversity amongst subtypes. Hitherto we have concentrated our eÜorts on modelling two of the domains found in iGluRs» the LAOBP-like domain and the K` channel-like domain. We should stress that as in any modelling study these models represent a guess at the structure»the result is no substitute for a high-resolution experimentally derived three-dimensional structure but nevertheless is extremely useful as a guide for subsequent experimental work particularly when used as part of an iterative molecular modelling/experimental veri–cation cycle. Modelling the LAOBP-like domain Identifying suitable 3-D templates Once a structural module has been identi–ed a search is made for a similar protein(s) of known three-dimensional structure (determined either by X-ray crystallography or NMR spectroscopy).This is based on the observation that a good correlation exists between the level of similarity in the amino acid sequence and the level of similarity in the three-dimensional structure. The amino acid sequences of the LAOBP-like domain were searched against a database containing the amino acid sequences of proteins of known three-dimensional structure using the program BLAST17 (this scans the LAOBP-like sequence against each sequence in the database in turn returns a score for each pairwise comparison ranks them and displays the most signi–cant ones). However no similarities of any signi–cance were identi–ed.An alternative approach was therefore adopted which is based on the premise that proteins with no obvious sequence similarity can still show remarkable similarities in their topologies (or ìì folds œœ) although fold similarities are not identi–ed as reliably as sequence similarities. Both the TOPITS18 and the UCLA-DOE Structure Prediction Server19 returned histidine-binding protein (HBP) as the only signi–cant hit (these scan the predicted secondary structure solvent accessibility and amino acid environment of the LAOBP-like sequence against a database containing a representative set of known protein folds rank the scores and display the most signi–cant ones). Thus HBP was identi–ed as a suitable template structure to use as the basis for modelling.The three-dimensional structure of HBP deposited in the Brookhaven Protein Data Bank20 (PDB; an international archive of experimentally determined threedimensional structures of protein molecules) is in the holo (ligand-bound) form (PDB accession 261 Faraday Discuss. 1998 111 259»272 number 1HSL21). Once a suitable fold is found other proteins of known three-dimensional structure with the same fold are identi–ed. Scanning this structure against DALI22 identi–ed the holo form of lysine arginine ornithine-binding protein (LAOBP; 1LST23) and the holo form of glutamine binding protein (QBP; 1WDN24) as additional template structures for modelling the holo form of the LAOBP-like domain. It also identi–ed the apo (ligand-free) form of LAOBP (2LAO23) and the apo form of QBP (1GGG25) as suitable templates for modelling the apo form of the LAOBP-like domain of the iGluRs.Sequence alignment Once the structural templates are identi–ed the particular amino acids in the sequence of the iGluR which correspond to the amino acids in the template structures are identi–ed. Multiple sequence alignment [using Clustal W26 and CAMELEON (Oxford Molecular Ltd Oxford UK)] was used to align the amino acid sequences of 19 members of the iGluR superfamily with the amino acid sequences of LAOBP HBP and QBP. The –nal consensus sequence alignment (Fig. 2) was produced manually within the program CAMELEON subject to the constraints that (1) wherever possible no insertion or deletion occurred within the crystallographically determined secondary structural elements of the templates ; (2) the known N-glycosylation sites (ref.6 and references therein) corresponded to surface positions [to be glycosylated an amino acid (in this case an asparagine residue) position must be accessible to the incoming sugar molecule]; (3) all those residues thought to be involved in ligand binding27h29 were positioned in the binding site ; (4) the disul–de bridge which we showed by experiment is present,30 could be formed; (5) some account was taken of the predicted secondary structure (determined using PHDsec31). Mapping amino acid sequence onto 3-D templates A set of three-dimensional models was then produced based on this alignment using the technique of homology modelling. The results of homology modelling are critically dependent upon (1) the choice of structural template(s) and (2) the sequence alignment used emphasising the need for care in the preceding stages.Programs for homology modelling use one of two approaches»either a fragment-based stepwise approach (COMPOSER;32,33 used for our early models6) or a single step approach (MODELLER;34 used for our more recent models). Although both methods produce models of equally good quality the latter is now our method of choice because it enables additional restraints»hydrogen bonds charge»charge interactions and hydrophobic interactions»to be added in the form of distance restraints during the modelling process rather than in a post hoc fashion as with the fragment-based approach. The ligand binding site of the three-dimensional models was characterised using InsightII (an interactive molecular graphics program; MSI San Diego USA) GRID35 (used to search for energetically favourable positions for functional groups in the ligand) DELPHI36 (electrostatics calculations used with ììfocussingœœ to enhance accuracy by using results from a low resolution calculation as boundary values for a higher resolution calculation) and GRASP37 (for visualising the results from DELPHI).Once possible protein»ligand interactions are identi–ed these are then included in the homology modelling process in the form of distance restraints. For example for a hydrogen bond the donor … … … acceptor distance (e.g. oxygen … … … nitrogen) is restrained to a range of 2.5»3.3 ”.Because of the less well de–ned nature of restraints between atoms thought to be involved in hydrophobic interactions such restraints are initially entered relatively loosely the resulting models analysed and the hydrophobic restraints updated accordingly. Ligand binding site We have modelled the agonist glutamate into the binding site of the holo form of GFKARa GluR1 and GluR6 NMDA-R1 and NMDA-R2C and the agonists kainate and domoate into the binding site of the holo form of GFKARa. The modelling positioned the three ììhot spotsœœ known initially to aÜect the binding of agonist in both non-NMDA and NMDA receptors in the binding site (residues 27[411] VTTIKE 32[416] 69[461] DGRYG 73[465] and 260[662] GTIKDS 265[667] in GFKARa; here and below the corresponding residue number for GluR1 is given in Faraday Discuss.1998 111 259»272 262 Fig. 2 Our current sequence alignment used for modelling the iGluRs. The region shown includes the LAOBP-like and the K` channel-like domains of the iGluRs. The positions of M1 M2 and M3 are shown on a dark background; those regions involved in ligand binding (S1»S6) are shown on a light gray background; glycosylation sites are denoted by vertical boxes; and the two half-cystines are shown on a dark background. The consensus secondary structure observed in the crystal structures of the bacterial periplasmic amino acid binding proteins is also shown. (This –gure was generated using ALSCRIPT66.) 263 Faraday Discuss. 1998 111 259»272 Fig. 3 (a) Schematic representation of a model of the NMDA-R2C»glutamate complex showing the positions of glutamate ligand (Glu) and ligand binding regions S1»S6 in one of the LAOBP-like domains (note that residues 425»454 have been omitted as these cannot be built with any con–dence since they do not correspond to any amino acid residues in the template structures).(b) Schematic representation of the glutamate binding site in a model of the NMDA-R2C»glutamate complex showing one of the LAOBP-like domains. Hydrogen bonds and charge»charge interactions are shown as green dashed lines hydrophobic interactions as red spoked arcs carbon atoms as black balls oxygen atoms as red balls and nitrogen atoms as blue balls. The purple ìì ? œœ signi–es the uncertainty in how the interaction between the two carboxyl groups is mediated.(This –gure was generated using LIGPLOT.67) Faraday Discuss. 1998 111 259»272 264 square brackets for reference»numbering begins at the start of the signal sequence) and gave insight into those amino acid residues in GFKARa likely to interact with the ligand. Our initial hypothesis based on these models was that E32[416] K71[463] Y72[464] R106[499] and R302[706] were involved in ligand binding (alongside other residues to be investigated later). These suggestions formed the basis of experimental studies which partly con–rmed protein»ligand interactions suggested by our initial models (E32[416] Y72[464] and R106[499]) and partly showed that some of the interactions had been modelled incorrectly (K71[463] and R302[706]). The orientation of the glutamate ligand was a concern at this stage»subsequent data7 suggest that the glutamate ligand binds in the same orientation as the arginine histidine and glutamine ligand in LAOBP HBP and QBP respectively (Fig.3) rather than in the ììupside downœœ orientation of our initial models6 (see ref. 7 for a more detailed discussion). However we are uncertain how the interaction between two negatively charged carboxyl groups (one from E32[416] [E413 in NMDA-R2C; note that this residue is a negatively charged glutamate in the majority of iGluRs but a polar glutamine in NMDA-R1 and GFKARb] the other from the ligand) could be mediated. We have discounted mediation by either a water molecule or a proton on the sidechain of E32[416] (i.e. E32[416] would have an unusually high pK because an E at this position binds a) glutamate ligand approximately 5 times more strongly than does Q.Also we have found recently that divalent cations do not aÜect the binding of kainate suggesting that a divalent cation very likely does not mediate the interaction between the ligand carboxyl group and E32[416]. One remaining possibility is that the carboxyl group on the ligand becomes protonated. Re–nement using iterative modelling/experimental veri–cation cycle Subsequent to our initial modelling of iGluRs,6 three additional ììhot spotsœœ for ligand binding have been identi–ed giving a total of six (Fig. 2 and 3; those that correspond to the ììhot spotsœœ included in the original modelling are denoted S1 S2 and S4). The current sequence alignment (Fig.2) corresponds to the original sequence alignment6 in 2 of these (S2 and S4) and diÜers by three residues in S1. Of the remaining three ììhot spotsœœ the current sequence alignment corresponds to the original in one (S3) and diÜers quite signi–cantly in the remaining two (S5 and S6). Interestingly all but one of these ììhot spotsœœ (S5) correspond to regions in the ligand binding site in the LAOBP-like crystal structures.21,23,25 This further justi–es the use of these crystal structures as structural templates. The absence of S5 in the bacterial periplasmic amino acid binding proteins can based on our sequence alignment be explained in terms of the insertion (2»6 residues) in the iGluRs lengthening this region of the structure thereby allowing contact with the ligand.Positions of conserved amino acids in the alignment play an important role in this realignment and all of the changes discussed leave the N-glycosylation sites accessible. A model of the LAOBP-like domain of the NMDA-R2C»glutamate complex generated using our current sequence alignment (Fig. 2) is shown in Fig. 3. Glycine binding versus glutamate binding NMDA-R1 is thought to bind glycine and NMDA-R2 subunits are thought to bind glutamate. 38h40 One of these studies39 suggests that this diÜerence could arise from the presence of more bulky aromatic sidechains in the a-amino binding region of NMDA-R1. Our study which is based on a diÜerent amino acid sequence alignment to this previous work suggests a possible alternative explanation.Inspection of our models does not reveal any diÜerences in the agonist binding site between these two types of subunit which would explain this likely diÜerence in speci–city. However inspection of the sequence alignment (Fig. 2) reveals that there is a two residue insertion in a loop in S2 in NMDA-R1 (T486 to Q487) with respect to NMDA-R2 (there is also a three residue insertion in NMDA-R1»T486 to E488»with respect to the non-NMDA receptors). Although this does not manifest itself as a diÜerence between our models of NMDA-R1 and NMDA-R2C the two additional residues in NMDA-R1 (T486 and Q487) could result in a conformational change in this loop thereby changing the position of the residue F484[464] (an important determinant in ligand binding) to form a constriction in the agonist binding site where an amino acid sidechain would be positioned thereby preventing glutamate but not glycine from binding to NMDA-R1.265 Faraday Discuss. 1998 111 259»272 Modelling the Kë channel-like domain Ab initio modelling The amino acid sequences of the K` channel-like domain of the iGluRs showed no signi–cant sequence homology to any protein of known three-dimensional structure nor did they show a signi–cant propensity to adopt any of the known protein folds. The complete ab initio prediction of protein three-dimensional structures is not possible at present and a general solution to the protein folding problem is not likely to be found in the near future. However if some knowledge is available of (1) the secondary structural elements in the module being modelled and (2) how these secondary structural elements pack together the modelling exercise becomes far more tractable despite remaining highly speculative.Using the similarity between M2 in the K` channel-like domain and the P-segment (or H5 segment) of K` channels,14 we were able to produce initial models of the K` channel-like domain using an ab initio approach6»distance geometry followed by simulated annealing and energy minimisation (XPLOR;41 built initially as an antiparallel bbarrel with appropriate distance restraints for hydrogen bonds [NH ” … … …O 1.8»2.3 and N… … …O 2.5»3.3 ”] and dihedral restraints for b-strands [/[120° to [160° and t 115° to 155°] and restrained to be symmetrical) and subsequent manual re–nement6 (InsightII [MSI San Diego USA] and SCULPT42) to ensure that all residues were positioned consistent with experimental data.These original models of GFKARa GluR1 and GluR6 were built with stoichiometries of 4 5 and 6. The size of the pore in our early models lends support to either a pentameric stoichiometry or a hexameric stoichiometry (see ref. 6,7 for a more detailed discussion) ; our more recent models of NMDA receptors also suggest that a tetramer is too small (see below). Satisfaction of experimentally derived restraints An important consideration in building the models is the position of the crucial residue known to aÜect conductance in a number of iGluRs. This position in GluR2 and GluR6 (residue 198[600]; denoted the ììQ/R/N site œœ) can be changed from Q to R by RNA editing43 and is known to be involved in the blockade of NMDA receptors by Mg2` (N598 of NMDA-R144).Electrostatic calculations (using DELPHI36 and GRASP37) suggest that the Q form provides a good cation binding site whereas the R form does not. This in turn suggests why the Q form is doubly rectifying (the D/E site is free to bind Ca2`) whereas the R form is not (the predicted Ca2` binding site is removed by formation of a salt bridge between the R and the D/E site ; see ref. 6,7 for a more detailed discussion). Site-directed mutagenesis data45h47 were also used for modelling M2.6 In addition to M2 M1 and M3 are also likely to form part of the conductance pathway. M1 was modelled as an a-helix lining the ion conduction pathway6»consistent with the results of RNA editing of residues in M1 of GluR6.48 M3 was also modelled as an a-helix lining the ion conduction pathway6»consistent with the suggested role of M3 in the channel blocking action of MK-801.49 NMDA receptor models Our recent models of the K` channel-like domain correspond to a re–nement of our initial models,6 into which we have incorporated the available experimental data for M2 from NMDA receptors.46,50h53 Although the channels are known to exist as heteromers they were modelled initially as homomers and then the homomers combined to form a series of heteromers»partly for simplicity of interpreting the models and partly because the number of each receptor type present in an active channel is unknown.The starting point for modelling the M2 region of both NMDA-R1 and NMDA-R2C was our recent model of the the P-segment (or H5 segment) of the inward recti–er K` channel Kir2.154»although the crystal structure of a K` channel has been determined recently.13 The starting structure of M2 was re–ned automatically using XPLOR41 and distance restraints derived from the results of a block arising from applying methanethiosulfonate (MTS)-based thiol reagents to a NMDA-R1/NMDA-R2C channel following cysteine scanning mutagenesis46»if the block was observed at a particular position the maximum diameter of the pore (measured using gamma atoms of the respective residue) was set at 20 ”; if the block did not occur a minimum diameter of 15 ” was used.Channels were generated with Faraday Discuss.1998 111 259»272 266 Fig. 4 Electrostatic potential on the surface of the narrowest part of the pore for (a) NMDA-R1 and (b) NMDA-R2C. Red corresponds to an electrostatic potential of O[10 kBT /e white an electrostatic potential kBT /e k and blue an electrostatic potential of P]10 In both cases the view is from the side of the of 0 BT /e. channel through the membrane with two of the four subunits cut away leaving only those two ììbehindœœ the centre of the pore. Only the M2 region is shown although calculations were performed with M1 and M3 also present. The yellow dashed line illustrates the location of the centre of the pore and the light blue circle represents a cation. both tetrameric and pentameric stoichiometries»residue accessibilities to the pore in the resulting models were consistent with experimental observations.Selectivity –lter The cation selectivity –lter (i.e. the pore lining component of M2) in these models consists of carbonyl oxygens from the polypeptide backbone (particularly G617[602] I618[603] and G619[604] in NMDA-R1) the sidechain amide oxygen of asparagine and aromatic groups 267 Faraday Discuss. 1998 111 259»272 (W608[592] and F609[593] in NMDA-R1). The presence of carbonyl oxygens is in agreement with the recent crystal structure of the KcsA K` channel,13 in which the authors suggest that a carbonyl tunnel alone forms the atomic basis of the cation selectivity –lter. Indeed the predicted presence of carbonyl oxygens in the selectivity –lter could explain why NMDA-R1 can form functional homomeric channels and the predicted inaccessibility of the corresponding carbonyl oxygens could explain why NMDA-R2 cannot form functional homomeric channels (see below).Other experimental studies on K` channels suggest that an aromatic tyrosine residue lines the selectivity –lter.54,55 Thus the selectivity –lter in our models»involving both cation»oxygen and cation»p interactions»is consistent with experimental observations in other cation channels. NMDA-R1 but not NMDA-R2 form functional homomeric channels Our models also suggest why NMDA-R1 can form functional homomeric channels but NMDA-R2 cannot.56 Electrostatic calculations on our models of homomeric NMDA-R1 and NMDA-R2C channels (Fig. 4) show that the pore lining region of the NMDA-R1 channel particularly in the region of the selectivity –lter is signi–cantly more negative than NMDA-R2C.Analysis of the models reveals that this may result from the conserved proline in NMDA-R2 receptors (P619[604] in NMDA-R2C; this is a glycine in NMDA-R1) the restricted conformational freedom of which (due to its sidechain bonding to its amide nitrogen) appears to prevent exposure of carbonyl oxygens to the pore in this region. In NMDA-R1 however the high level of conformational freedom of the two glycine residues (glycine has no sidechain) in this region could further contribute to this diÜerence. Mg2ë permeation We have mentioned above the role of the Q/R/N-site in block by magnesium.44 Additional sites also exist. Mutation of W607[592] to non-aromatic residues in NMDA-R2A and NMDA-R2B greatly increased permeation of extracellular magnesium,52 suggesting that this residue lines the pore.This is consistent with our models and also with cysteine scanning mutagenesis studies.46 The same authors note that the equivalent mutation in NMDA-R1 (mutation of W608[592] to non-aromatic residues) did not aÜect Mg2` permeation and concluded that this residue does not line the pore. This conclusion is at odds with both the cysteine scanning mutagenesis studies46 and our models. However these models oÜer a possible alternative explanation. In NMDA-R1 the residue adjacent to W608[592] is F609[593] which is exposed to the pore whereas in NMDAR2A and NMDA-R2B this residue is a leucine. Thus NMDA-R1 contains a second potential Mg2` aromatic binding site (F609[593] which remains when W608[592] is mutated) whereas NMDA-R2A and NMDA-R2B do not.The models also present a possible explanation for the higher affinity for Mg2` of the M1 region in NMDA-R2A and NMDA-R2B than either in NMDA-R2C or NMDA-R2D.57 S561[551] in NMDA-R2A and NMDA-R2B is V561[551] in NMDA-R2C and NMDA-R2D. This residue is exposed to the pore and therefore could provide an additional Mg2` binding site in NMDA-R2A and NMDA-R2B (due to its sidechain hydroxy group) but not in NMDA-R2C and NMDA-R2D (due to its hydrophobic nature). Stoichiometry Experimental determination of the stoichiometry remains somewhat ambiguous suggesting either a tetramer58h60 or a pentamer.61 The models give insight into possible stoichiometry but by no means provide a de–nitive answer.The narrowest diameter in the literature for NMDA channels is 5.5 ”53»this is consistent with our pentameric (D5.3 ”) but larger than our tetrameric channels (D4.3 ”) suggesting that perhaps the stoichiometry is pentameric. Combining diÜerent modules Once models had been produced for the diÜerent modules the –nal stage (Fig. 1) was to combine them to create a model of the iGluR. This was achieved by ensuring that the model was consistent with available experimental data. A module comprising M1 M2 and M3 (i.e. the membranespanning region which lines the pore) was used as the starting point since this already contained Faraday Discuss. 1998 111 259»272 268 Fig. 5 Schematic representation of the tetrameric structure of the NMDA-R2C»glutamate complex viewed from (a) outside membrane and (b) along membrane [with the front (green) subunit removed for clarity].Each subunit is in a diÜerent colour glutamate ligand is in orange space –lling representation and the conserved acidic amino acid and D668[652] and the adjacent conserved acidic/polar residue S667[651] are in red space –lling representation. the correct symmetry. The appropriate number of copies of the LAOBP-like domain (e.g. four for a tetramer) were added to this using interactive molecular graphics (InsightII). The model of the LAOBP-like domain was positioned empirically with respect to both the membrane and its symmetry related copies so that (1) the consensus glycosylation sites were solvent accessible (2) the agonist binding site was accessible (3) the distance between the end of the N-terminal of the LAOBP-like domain and M1 was in a reasonable range (approximately 17 residues were missing 269 Faraday Discuss.1998 111 259»272 from our models in this region) (4) the distance between the end of M3 and the start of the C-terminal section of the LAOBP-like domain was within a reasonable range (approximately 12 residues were missing from our models in this region) and (5) the domain was as close to the pore as possible without overlapping sterically with its symmetry related copies in both the apo and holo forms. In the resulting orientation the long axis of the LAOBP-like domain was roughly parallel to the surface of the membrane. This positioning of the LAOBP-like domain although not a unique solution (due to the limited experimental data) is not inconsistent with the currently available data and is in fact constrained to a large extent by the experimental results.An assembled tetrameric model of NMDA-R2C is shown in Fig. 5. Binding and signal transduction Once the models of the diÜerent modules had been combined the molecular basis of the transduction of binding energy to ion channel opening could be investigated. The models were analysed paying particular attention to those residues close to the ion conduction pathway in the ììchannel openœœ (ligand-bound) and ììchannel closedœœ (ligand-free) states of the model. If particular residues appeared to be involved in ion conduction and they are conserved across the iGluRs then these are possible candidates.It should be noted that the conformational change in the ligand binding domain assumed to take place in the transition between the ligand-bound and the ligand-free states is likely to be propagated to other parts of the model and in particular the membrane-spanning portion of the structure. The transition from the agonist-bound to the agonist-free states in our models results in a change in the orientation of a pair of conserved negatively charged amino acids. These correspond to residues S667[651] (which is a negatively charged residue in the non-NMDA receptors) and D668[652]. In the agonist-bound form but not the agonist-free form our models suggest that S667[651] and D668[652] are positioned near the opening thus electrostatically attracting cations into the channel in a manner analogous to the ring of negative charges in nicotinic acetylcholine receptors.62 This proposed role of S667[651] and D668[652] is supported by studies of NMDA-R1 which implicate the involvement of D669[652] in voltage-dependent spermine block.63 So far the assumption has been made that the LAOBP-like domains exist in one of two states»either the ligand-free or ligand-bound states»an over-simplistic picture.In fact there are at least two agonist-bound states»an initial channel open state and a subsequent desensitised (channel closed) state ; the structural reasons for this remain unclear. It has been proposed64 that the mechanism of agonist binding is akin to the ììvenus —y trapœœ mechanism observed in the bacterial periplasmic amino acid binding proteins»agonist binds –rst to Lobe 1 alone to give the ììchannel openœœ state and subsequently binds also to Lobe 2 to give the ììdesensitisedœœ state.However in the case of the bacterial periplasmic amino acid binding proteins in the initial ligandbound ììopen cleft œœ form the relative orientations of the two lobes remain unchanged (see ref. 65 for a discussion of the diÜerent forms). Therefore it is difficult to rationalise how ligand binding in such a manner could produce a sufficiently large conformational change to bring about channel opening. This is even more difficult to understand if our proposed role of S667[651] and D668[652] (residues in Lobe 2) in the channel open state is indeed correct (which it may not be) predicting a movement in Lobe 2 upon agonist binding.Our models suggest the following possibility which is speculative but nevertheless not inconsistent with the experimental results.64 Initial agonist binding causes the lobes to close together. This movement is impeded initially by the loop we denote S5 (Fig. 2 and 3) but which subsequently changes conformation to allow the two lobes to come closer together. This second change in the relative positions of Lobes 1 and 2 results in desensitisation with the agonist now bound more tightly. Our structural models can neither prove nor disprove this notion but merely suggest it as a potential mechanism that can be tested. Conclusion Structural models of iGluRs can be produced using molecular modelling techniques.This is achieved by dividing the amino acid sequence into a number of smaller ìì structural modulesœœ producing models for these and then combining the models of the ììmodulesœœ to produce a model Faraday Discuss. 1998 111 259»272 270 of the channel. The process is best achieved using an iterative modelling/experimental veri–cation cycle. As in any modelling exercise the results are no substitute for a high resolution threedimensional structure. However the resulting insights suggest directions for further study. The power of our approach is thus the iterative interplay between modelling and experiment. Acknowledgements M. J. S. is a Royal Society University Research Fellow. A. H. S. is funded by an EPSRC studentship.This work was supported in part by an National Science Foundation grant (IBN-9309480) to R. E. O. References 1 D. T. Monaghan R. J. Bridges and C. W. Cotman Annu. Rev. Pharmacol. T oxicol. 1989 29 365. 2 M. Hollmann and S. Heinemann Annu. Rev. Neurosci. 1994 17 31. 3 K. Wada C. J. Dechesne S. Shimasaki R. G. King K. Kusano A. Buonanno D. R. Hampson C. Banner R. J. Wenthold and Y. Nakatani Nature (L ondon) 1989 342 684. 4 P. Gregor I. Mano I. Maoz M. McKeown and V. Teichberg Nature (L ondon) 1989 342 689. 5 Z. G. Wo and R. E. Oswald Proc. Natl. Acad. Sci. USA 1994 91 7154. 6 M. J. SutcliÜe Z. G. Wo and R. E. Oswald Biophys. J. 1996 70 1575. 7 M. J. SutcliÜe A. H. Smeeton Z. G. Wo and R. E. Oswald Methods Enzymol. 1998 293 589. 8 M. J.SutcliÜe A. H. Smeeton Z. G. Wo and R. E. Oswald Biochem. Soc. T rans. 1998 26 450. 9 S. Nakanishi and M. Masu Annu. Rev. Biophys. Biomol. Struct. 1994 23 319. 10 D. T. Jones W. R. Taylor and J. M. Thornton Biochemistry 1994 33 3038. 11 B. Person and P. Argos J. Mol. Biol. 1994 237 182. 12 B. Rost P. Fariselli and R. Casadio Protein Sci. 1996 7 1704. 13 D. A. Doyle J. M. Cabral R. A. Pfuetzner A. L. Kuo J. M. Gulbis S. L. Cohen B. T. Chait and R. MacKinnon Science 1998 280 69. 14 Z. G. Wo and R. E. Oswald T rends Neurosciences 1995 18 161. 15 N. Nakanishi N. A. Schneider and R. Axel Neuron 1990 5 569. 16 P. J. OœHara P. O. Sheppard H. Th‘gersen D. Venezia B. A. Haldeman V. McGrane K. M. Houamed C. Thomsen T. L. Gilbert and E. R. Mulvihill Neuron 1993 11 41.17 S. F. Altschul T. L. Madden A. A. Schaé Üer J. Zhang Z. Zhang W. Miller and D. J. Lipman Nucleic Acids Res. 1997 25 3389. 18 B. Rost in T OPIT S T hreading one-dimensional predictions into three-dimensional structures ed. C. Rawlings D. Clark R. Altman L. Hunter T. Lengauer and S. Wodak Menlo Park CA 1995. 19 D. Fischer and D. Eisenberg Protein Sci. 1996 5 947. 20 E. E. Abola F. C. Bernstein S. H. Bryant T. F. Koetzle and J. Weng in Protein Data Bank ed. F. H. Allen G. BergerhoÜ and R. Sievers Data Commission of the International Union of Crystallography Bonn/Cambridge/Chester 1987. 21 N. H. Yao S. Trakhanov and F. A. Quicho Biochemistry 1994 33 4769. 22 L. Holm and C. Sander Proteins 1994 19 165. 23 B. H. Oh J. Pandit C. H. Kang K. Nikaido S.Gokcen G. F. L. Ames and S. H. Kim J. Biol. Chem. 1993 268 11348. 24 Y. J. Sun J. Rose B. C. Wang and C. D. Hsiao J. Mol. Biol. 1998 278 219. 25 C. D. Hsiao Y. J. Sun J. Rose and B. C. Wang J. Mol. Biol. 1996 262 225. 26 J. D. Thompson D. G. Higgins and T. J. Gibson Nucleic Acids Res. 1994 22 4673. 27 A. Kuryatov B. Laube H. Betz and J. Kuhse Neuron 1994 12 1291. 28 F. Li N. Owens and T. A. Verdoorn Molecular Pharmacology 1995 47 148. 29 S. Uchino K. Sakimura K. Nagahari and M. Mishina FEBS L ett. 1992 308 253. 30 Z. G. Wo and R. E. Oswald Mol. Pharmacology 1996 50 770. 31 B. Rost and C. Sander J. Mol. Biol. 1993 232 584. 32 M. J. SutcliÜe F. R. F. Hayes and T. L. Blundell Protein Eng. 1987 1 385. 33 M. J. SutcliÜe I. Haneef D. Carney and T. L. Blundell Protein Eng.1987 1 377. 34 A. Sali and T. L. Blundell J. Mol. Biol. 1993 234 779. 35 P. J. Goodford J. Med. Chem. 1985 28 849. 36 A. Nicholls and B. Honig J. Comput. Chem. 1991 12 435. 37 A. Nicholls K. Sharp and B. Honig Proteins 1991 11 281. 38 M. Honer D. Benke B. Laube J. Kuhse R. Heckendorn H. Allgeier C. Angst H. Monyer P. H. Seeburg H. Betz and H. Mohler J. Biol. Chem. 1998 273 11158. 39 B. Laube H. Hirai M. Sturgess H. Betz and J. Kuhse Neuron 1997 18 493. 40 H. Hirai J. Kirsch B. Laube H. Betz and J. Kuhse Proc. Natl. Acad. Sci. USA 1996 93 6031. 41 A. T. Brué nger X-PL OR Manual Yale University New Haven 1997. 271 Faraday Discuss. 1998 111 259»272 42 M. C. Surles J. S. Richardson D. C. Richardson and F. P. J. Brooks Protein Science 1994 3 198.43 B. Sommer M. Koé hler R. Sprengel and P. H. Seeburg Cell 1991 67 11. 44 N. Burnashev R. Schoepfer H. Monyer J. P. Ruppersberg W. Gué nther P. H. Seeburg and B. Sakmann Science 1992 257 1415. 45 R. Dingledine R. I. Hume and S. F. Heinemann J. Neurosci. 1992 12 4080. 46 T. Kuner L. P. Wollmuth A. Karlin P. H. Seeburg and B. Sakmann Neuron 1996 17 343. 47 N. Burnashev A. Villarroel and B. Sakmann J. Physiol. L ondon 1996 496 165. 48 M. Koé hler N. Burnashev B. Sakmann and P. H. Seeburg Neuron 1993 10 491. 49 A. V. Ferrer-Montiel W. Sun and M. Montal Proc. Natl. Acad. Sci. USA 1995 92 8021. 50 J. Chao N. Seiler J. Renault K. Kashiwagi T. Masuko K. Igarashi and K. Williams Mol. Pharmacol. 1997 51 861. 51 K. Kashiwagi A. J. Pahk T. Masuko K.Igarashi and K. Williams Mol. Pharmacol. 1997 52 701. 52 K. Williams A. J. Pahk K. Kashiwagi T. Masuko N. D. Nguyen and K. Igarashi Mol. Pharmacol. 1998 53 933. 53 L. P. Wollmuth T. Kuner P. H. Seeburg and B. Sakmann J. Physiol. L ondon 1996 491 779. 54 C. Dart M. L. Leyland P. J. Spencer P. R. Stan–eld and M. J. SutcliÜe J. Physiol. L ondon 1998 511 25. 55 Q. Lu and C. Miller Science 1995 268 304. 56 C. J. McBain and M. L. Mayer Physiol. Rev. 1994 74 723. 57 T. Kuner and R. Schoepfer J. Neurosci. 1996 16 3549. 58 C. Rosenmund Y. Stern-Bach and C. F. Stevens Science 1998 280 1596. 59 I. Mano and V. I. Teichberg Neuroreport 1998 9 327. 60 B. Laube J. Kuhse and H. Betz J. Neurosci. 1998 18 2954. 61 L. S. Premkumar and A. Auerbach J. Gen. Physiol. 1997 110 485. 62 K. Imoto C. Busch B. Sakmann M. Mishina T. Konno J. Nakai H. Bujo Y. Mori K. Fukuda and S. Numa Nature (L ondon) 1988 335 645. 63 K. Kashiwagi J. Fukuchi J. Chao K. Igarashi and K. Williams Mol. Pharmacol. 1996 49 1131. 64 I. Mano Y. Lamed and V. I. Teichberg J. Biol. Chem. 1996 271 15299. 65 B. H. Oh C. H. Kang H. De Bondt S. H. Kim K. Nikaido A. K. Joshi and G. F. Ames J. Biol. Chem. 1994 269 4135. 66 G. J. Barton Protein Eng. 1993 6 37. 67 A. C. Wallace R. A. Laskowski and J. M. Thornton Protein Eng. 1995 8 127. Paper 8/06183A Faraday Discuss. 1998 111 259»272 272

ISSN:1359-6640    

DOI:10.1039/a806183a

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Functional immobilization of biomembrane fragments on planar waveguides for the investigation of side-directed ligand binding by surface-con¢�ned ¢©uorescence Michael Pawlak,*¡×a Ernst Grell,b Eginhard Schick,b Dario Anselmettic and Markus Ehrata a Novartis Pharma AG Preclinical Safety DMPK CH-4002 Basel Switzerland. E-mail michael.pawlak=pharma.novartis.com b Max-Planck Institute of Biophysics D-60596 Frankfurt Germany c Novartis Services AG Scienti¡©c Services Physics CH-4002 Basel Switzerland Recei©ªed 26th August 1998 A method for the functional immobilization of Na,K-ATPase-rich membrane fragments on planar metal oxide waveguides has been developed. A novel optical technique based on the highly sensitive detection of surface-con¡©ned ¡ªuorescence in the evanescent ¡©eld of the waveguide allowed us to investigate the interactions of the immobilized protein with cations and ligands.For speci¡©c binding studies a FITC-Na,K-ATPase was used which had been labelled covalently within the ATP-binding domain of the protein. Fluorophore labels of the surface-bound enzyme can be selectively excited in the evanescent ¡©eld. A preserved functional activity of the immobilized enzyme was only found when a phospholipid monolayer was preassembled onto the hydrophobic chip surface to form a gentle biocompatible interface. In situ atomic force microscopy (AFM) was used to examine and optimize the conditions for the lipid and membrane fragment assembly and the quality of the formed layers. The enzyme©«s functional activity was tested by selective K` cation binding interaction with anti-¡ªuorescein antibody 4-4-20 phosphorylation of the protein and binding of inhibitory ligand ouabain.The comparison with corresponding ¡ªuorescence intensity changes found in bulk solution provides information about the side-directed surface binding of the Na,K-ATPase membrane fragments. The affinity constants of K` ions to the Na,K-ATPase was the same for the immobilized and the non-immobilized enzyme providing evidence for the highly native environment on the surface. The method for the functional immobilization of membrane fragments on waveguide surfaces will be the basis for future applications in pharmaceutical research where advanced methods for exploring the molecular mechanisms of membrane receptor targets and drug screening are required.Introduction Molecular interactions on biomembranes play a prominent role in the communication between cells and in signal transduction pathways.1 Thereby membrane receptors serve as the main ¡× Present address Zeptogens AG Benkenstr. 254 CH-4108 Witterswil Switzerland. 273 Faraday Discuss. 1998 111 273¡í288 targets,2 able to recognize speci–c ligands selectively which can trigger a cascade of functional cell responses such as the regulation of ion channel activity3,4 or induction of secondary messengers e.g. release of intracellular calcium upon G-protein coupled receptor activation.5,6 Because of their regulatory mechanisms and their relevance in drug»target interactions membrane receptors are currently the focus of detailed biophysical and biochemical investigations to elucidate the relation between ligand binding and functional properties or to resolve structure»activity relationships.In pharmaceutical research membrane receptors are used as targets in the drug discovery process where large numbers of new chemical entities are screened at an early phase.7 For that purpose a number of physical sensing techniques preferentially based on —uorescence detection and designed for high throughput screening have been developed recently.8,9 Therefore and due to the trend of integrating a larger and larger number of assays on a single detection plate,8 leading to increased surface/volume ratios in an assay well surface-sensitive —uorescence techniques10h12 become increasingly important.However detection of direct ligand»receptor interactions and of related protein functions is often limited not only by the physical sensing technology but also by the assay architecture the non-speci–c binding the loss of functionality upon immobilization of cells/receptors on surfaces and strong background signals from the bulk solution. Accordingly the interfacing of biologically active receptor systems with such sensing surfaces with preservation of the full functional protein activity and access to an optimum number of binding sites under controlled conditions becomes more and more important but still remains a major challenge. Biosensor technology Here we present the advantageous combination of a highly sensitive technology for the detection of surface-con–ned —uorescence excited in the evanescent –eld of planar waveguides,13,14 and a new method for the functional immobilization of biomembrane fragments which contain the receptor target on such sensor surfaces under de–ned conditions.The high assay performance of evanescent wave sensors is achieved by (i) the strong spatial discrimation by the evanescent –eld between speci–c binding signals and bulk —uorescence causing background signals (penetration depth of the evanescent –eld is typically of the order of a wavelength); (ii) highly oriented and functional biomolecules (receptor targets) on the surface and (iii) the use of —uorescent labels for signal generation independent of molecular weight. Assays demonstrating the real-time monitoring of e.g.immunorecognition at working concentrations as low as several fM have recently been published.13 Here we report for the –rst time applications using an integral membrane protein which was functionally immobilized in its native membrane environment on waveguide sensor surfaces. Na,K-ATPase was chosen for a systematic investigation of the immobilization process as well as for probing molecular interactions on an immobilized protein. Na,K-ATPase The enzyme Na,K-ATPase is a protein that exists in the plasma membrane of all higher organisms. Its principal function as an alkali-metal ion dependent ATPase was discovered by Skou.15 The free energy resulting from the hydrolysis of an intracellular ATP molecule is converted by this enzyme into the transport of 3 Na` ions out of and 2 K` ions into the cell.Both active cation transport processes occur against the existing alkali-metal ion concentration gradients of the plasma membrane. Furthermore Na,K-ATPase exhibits electrogenic properties and thus contributes signi–cantly to the regulation of the membrane resting potential. The protein itself is a heteromer and consists of the catalytic a-subunit characterized by a molecular weight of ca. 100 kDa and of the glycosylated b-subunit with a molecular weight of ca. 50 kDa. Besides its transporter function Na,K-ATPase acts as the receptor for cardiac glycosides such as ouabain which are bound to the extracellular side of the protein at very high affinity and lead to the inhibition of enzymatic activity.Since the ionic composition of the medium diÜers between the cytoplasmic and the extracellular side and because the functional properties of the enzyme on both membane sides are diÜerent the structure of the protein must also be asymmetric. Consequently the binding properties of ligands and cations are expected to be side-directed. For example the binding affinity of a ligand or of an Faraday Discuss. 1998 111 273»288 274 alkali-metal ion to a site on one side of the membrane will depend on whether this ligand or cation is absent or present on the other side. Thus if one wishes to understand the basic functional properties of the cation pump mechanism such as that of Na,K-ATPase the aspect of the sidedness is essential and calls for advanced investigation methods.Na,K-ATPase was isolated as the major protein in nanoparticulate membrane fragments (discs of ca. 250 nm mean diameter) from specialized tiss such as kidney or salt glands. In contrast to solubilized systems the protein is in its native membrane environment and retains its original biomembrane orientation (cf. Fig. 1). The isolated membrane discs however are surrounded by the same aqueous medium on both sides and are no longer capable of separating the two diÜerent aqueous cell compartments. Although such a preparation no longer allows the study of the aspect of sidedness many relevant interactions partial reactions and mechanistic aspects have been investigated. For example a —uorescent marker such as —uorescein-5-isothiocyanate (FITC) can be speci–cally bound to a single lysine residue located within the ATP binding domain for monitoring binding events by —uorescence quenching.16h18 The —uorescence of the labelled Na,KATPase changes characteristically upon binding of diÜerent ligands and cations and was also used here for the analytical detection.General working concept Our concept is based on the formation of a biocompatible sensitized waveguide chip surface and subsequent assembly of membrane fragments on this support under de–ned conditions. The combination with surface-con–ned —uorescence detection by planar waveguides allows the simultaneous investigation of side-directed ligand binding and of functional properties of for example Na,K-ATPase (Fig. 1). The initial goal of this work was the stable and functional immobilization of ATPase-rich membrane fragments using surface-assembly techniques.The on-line monitoring capability of the planar waveguide (PWG) sensor allowed us to carefully examine each step of surface preparation. Additional knowledge on the surface structure was obtained by in situ AFM which delivered important 2D-resolved information about surface coverage and surface morphology of the sensitized chip. After successful immobilization of the membrane fragments the functional activity of the immobilized protein was probed by monitoring the —uorescence changes upon speci–c side-directed binding of cations and ligands using the FITC-labelled ATPase. The described approach oÜers great advantages because no detergent solubilization of the protein is Fig.1 Schematic illustration of a cross-section through a disc-shaped membrane fragment (diameter ca. 250 nm) containing FITC-labelled Na,K-ATPase prior to adsorption onto a planar waveguide chip. The enzyme molecules consist of an a and b subunit. The whole protein is asymmetric with respect to function and protein moiety. The ATP binding side conserves a larger protein content and is located on the former native inside of the biomembrane. Protein-to-lipid ratio of a membrane disc is typically ca. 1 250 (as determined for pig kidney membranes). 275 Faraday Discuss. 1998 111 273»288 necessary as is the case for other reconstitution procedures because the assays can be optimized efficiently and because fast and reproducible measurements can be performed due to the easy exchange of the aqueous media for the diÜerent assays.The orientation of the immobilized membrane fragments on the surface was –nally concluded from the comparison of the —uorescence changes measured with the PWG sensor with those found with the same membrane sample in bulk solution. Materials and experimental methods Chemicals Salts such as NaCl choline chloride imidazole and phosphates were purchased from Merck (Darmstadt Germany) and Fluka (Buchs Switzerland) and were of the highest purity obtainable (Ultrapure or Microselect). Organic solvents such as propan-2-ol were also from Merck and of UV-spectroscopy grade (Uvasol). 1-Palmitoyl-2-oleoyl-glycero-phosphocholine (POPC) and —uorescein-lipid (PE) were from Avanti Polar Lipids (USA).Mouse monoclonal anti-—uorescein antibody (4-4-20) was from Molecular Probes (Netherlands). Lipid vesicle preparation Lipid vesicles were prepared by extrusion (100 nm polycarbonate –lters Avestin Corp. USA) with 0.75 mM 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC Avanti Polar Lipids) hydrated in 150 mM NaCl 10 mM sodium phosphate buÜer 0.02% sodium azide at pH 7.5. The vesicle mean diameter was ca. 110 nm as determined by dynamic light scattering (Zeta Plus Brookhaven Instruments USA). Na,K-ATPase preparation Na,K-ATPase-containing membrane discs were isolated from the dissected red outer medulla of pig kidney and from the rectal gland of dog–sh (Squalus acanthias) in a two-step procedure. A microsomal fraction was prepared by tissue homogenization followed by diÜerential-velocity centrifugations.19 This fraction consisted essentially of closed membrane particles where the majority of the ATP binding sites were not accessible from the external medium. The microsomal fraction was activated by a short sodium dodecyl sulfate (SDS) treatment which led to an opening and partial solubilization of the microsomal membranes. After this treatment the Na,K-ATPase discs remained intact but the enzymatic activity of the preparations increased markedly. The discs were separated by density-gradient centrifugation using sucrose.19 The pig kidney enzyme was kept in 25 mM imidazole in HCl pH 7.5 containing 0.1 mM EDTA and the dog–sh enzyme in 30 mM histidine in HCl pH 6.8 containing 25 wt.% glycerol at concentrations of ca.2 mg ml~1. The average Na,K-ATPase activity at 37°C was 30 lmol mg~1 min~1; it was reduced to less than 1% in the presence of ouabain. Protein content was determined according to a modi–ed procedure of Lowry et al.20 FITC-Na,K-ATPase was prepared as described in ref. 16 employing extended washing procedures21 and was kept in 10 mM imidazole in HCl at pH 7.5 on ice. Upon labelling the sample the Na,K-ATPase activity decreased to \1 lmol mg~1 min~1 but the 2,4-dinitrophenylphosphatase activity at 37 °C was still about 19 lmol mg~1 min~1. The degree of labelling was calculated from the absorbance in the visible range assuming that the molar absorption coefficient of FITC remained unchanged at the same pH. The concentrations of the enzyme given here were calculated on the basis of a molecular weight of 150 kDa.Planar waveguide chips and surface preparation Planar waveguide chips (16 mm]48 mm) consisted of a 150 nm thin waveguiding surface layer of Ta (refractive index\2.3) deposited on 0.5 mm thick glass substrates (Balzers AG 2O5 Liechtenstein). The self-assembling chemistry chemical metal oxide modi–cation and physicochemical analysis of the modi–ed sensor surfaces is reported elsewhere.22 Brie—y the metal oxide surfaces were routinely cleaned by ultrasonication in organic solvent followed by a UV ozone treatment. The cleaned hydrophilic surfaces were further modi–ed to reach a stable hydrophobic Faraday Discuss. 1998 111 273»288 276 character by self-assembling monolayers of mono-C -alkyl phosphates (Novartis Pharma AG) 16 onto the freshly cleaned surfaces (3 day assembly at 1 mg ml~1 in propan-2-ol).Chips were thoroughly washed in propan-2-ol sonicated dried and stored under nitrogen. Contact angles of water on the hydrophobic surfaces were determined to be in the range of 100»110°. AFM Each step of our surface preparation was carefully examined by in situ AFM. For AFM experiments under de–ned buÜer environment we used a commercial instrument (Nanoscope IIIa Bioscope Digital Instruments USA) with Si3N4-cantilevers (Digital Instruments or Olympus) with nominal spring constants of 0.02»0.58 N m~1. The AFM was operated in the tapping mode in order to reduce unwanted tip»sample interactions due to the imaging process.Scanning was typically done at an imaging line frequency of 1 Hz. The surfaces carrying the lipid layers and biomembrane fragments were always kept under buÜer during mounting onto the AFM sample stage and were prepared under conditions comparable to those used in the sensor experiments. For the chip preparation a cell made from silicon allowed careful rinsing of excess lipid and/or membrane material as well as a reproducible exchange of the buÜer solution. Fluorescence measurements with planar waveguide sensor An in-house developed waveguide sensor instrument was used for the detection of surfacecon –ned chip —uorescence measurements.13 The schematic optical set-up is shown in Fig. 2. Spolarized (\transversal electric TE-mode) excitation light at 488 nm from an Ar`-ion laser (model 5490ASL 0»30 mW Ion Laser Technology USA) was coupled into the waveguiding layer by means of a diÜractive grating at a distinct angle of incidence.Since excitation light is efficiently guided in the waveguiding layer excitation light intensities could be kept low in the range of 10»100 lW. Part of the emitted surface-con–ned —uorescence excited by the propagating wave was collected at the substrate side with a 1 1 imaging optics (numerical aperture\0.19) and spectrally discriminated by means of two six-cavity interference –lters transmitting in the range Fig. 2 Evanescent wave —uorescence measurement surface-immobilized —uorophores are excited by laser light in the evanescent –eld of a planar Ta2O5 waveguide (propagating excitation mode).Fluorescence is detected to high sensitivity by a photon-counting photomultiplier (PMT) system. Fluorescence signals were referenced by measuring the excitation light within the detection area by means of a photo diode. BuÜer and sample solutions were injected into the —uid cell with an automated —uid handling system (not shown). 277 Faraday Discuss. 1998 111 273»288 515»545 nm (530DF30 Omega Inc. USA). Part of the surface-scattered light was used as a reference for the intensity of incoupled excitation light in the imaged area of interest. The –eld of detection was typically 5 mm]1 mm placed in the centre of the propagating beam by means of an aperture. A photon-counting PMT (H6240 Hamamatsu Japan) in combination with a 225 MHz counter (model 53131A Hewlett-Packard USA) was used for the detection of —uorescence.Excitation and signal read-out was taken every 30 s with a sampling time of 1 s. BuÜers and samples were applied to the sensor chip via a —ow-through cell made from silicone elastomer (Sylgard 184 Dow Corning USA). The cell volume was 6 mm3 (15 mm]2 mm]0.2 mm). Fluid supply was performed automatically using a dispensing pump»six-way valve combination (Cavro USA) for handling diÜerent working buÜers and a sample injector (231XL Gilson USA). Injection volumes were in the range 50»100 mm3 for stopped —ow and up to 5 cm3 for continuous —ow applications. The —uorescence detection and sample handling was fully automated and controlled by a PC. The chip holder and optical unit were temperature controlled.All experiments were performed at 20 °C. The experimental buÜer conditions were buÜer I (150 mM NaCl 10 mM sodium phosphate buÜer 0.02% NaN3 pH 7.5) buÜer II (100 mM cholineCl 10 mM imidazole pH 7.5) and buÜer III (buÜer II]3 mMMgCl2). Errors given for results are determined from at least three repeated measurements. Fluorescence measurements in bulk cuvette Bulk —uorescence measurements were performed on a Spex Fluorolog 222 instrument equipped with a thermostatted cuvette holder. Measurements were performed under conditions comparable to those in the waveguide sensor using identical protein preparations. The temperature was 20 °C in all cases. Results The main goal was to establish an experimental procedure for the functional immobilization of protein-containing membrane fragments in a de–ned and controlled manner.The concept of surface immobilization was based on a two-step preparation protocol shown schematically in Fig. 3 the –rst step (left) is the transfer of the bare non-sensitized physical transducer into a biocompatible interface applying a self-assembly of a lipid monolayer to the hydrophobic metal oxide surface. The second step (right) comprises the stable physisorption of the membrane fragments onto the lipid monolayer interface exploiting the negative surface charges and strong dipolar contributions21 of the biomembrane particles in a controllable physico-chemical environment. It is important to control the salt content and pH of the surrounding media as tools for optimizing this process.The intention was to establish the basis for future assay applications by the development of a well de–ned and optimized surface-preparation protocol. Lipid monolayer formation on waveguides The process of a de–ned lipid monolayer formation (–rst step) was optimized by characterizing the surface topography with AFM. The non-sensitized waveguide chips (bare as well as hydrophobic) showed a homogeneous and —at surface with very little roughness (0.2»0.3 nm rms) thus being well suited for potential high-resolution studies. Lipid monolayer formation on the hydrophobic metal oxide chips was performed via lipid vesicle spreading. The spreading behaviour of POPC vesicles on the hydrophobic metal oxide surfaces was investigated by varying the preparation techniques of vesicles (extrusion dilution) and buÜer composition (ionic strength diÜerent inorganic/organic salts).The whole process of optimizing lipid monolayer formation on planar waveguides is reported elsewhere.23 Under optimum spreading conditions almost complete surface coverage with lipid monolayers could be realized (Fig. 4) using vesicles prepared by extrusion (diameter ca. 110 nm as determined) at relatively high inorganic salt conditions (150 mM NaCl 10 mM sodium phosphate buÜer 0.02% NaN3 pH 7.5). Lipid (POPC) concentration was always 0.5 mg ml~1. Up to 10% of the surface area was occupied by small monolayer defects homogeneously distributed over the chip as single spots of 10»20 nm in diameter (Fig. 4). From these defects the mean thickness of the lipid monolayer could be easily determined (1.8^0.6 nm) 278 Faraday Discuss.1998 111 273»288 Fig. 3 Working concept for the functional immobilization of membrane fragments –rst step (left) formation of a biocompatible interface by self-assembly of a lipid monolayer after vesicle fusion onto a hydrophobic waveguide surface ; second step (right) controlled and stable physisorption of membrane fragments onto the preformed lipid interface on the waveguide. The penetration depth of the evanescent –eld of propagating excitation light is of comparable length to the size of the membrane discs. Upon surface-immobilization the membrane discs can adopt two diÜerent orientations former (cytoplasmic) inside surface up (as illustrated here) or former outside surface up.Drawn objects do not scale to real size. corresponding well with the thickness of a single monolayer as determined by other methods.24 Since the diameter of the membrane discs for the second association process was found to be ca. 250 nm 10»20 times larger than the size of the single defects (hydrophobic spots) the probability for contacts between biomembranes and such defects (leading to protein denaturation) in an otherwise closed lipid layer was considered to be negligible. Membrane fragment immobilization Having achieved a lipid-coated biocompatible chip surface surface-association of membrane discs containing Na,K-ATPase was carried out. Surface coverages and the morphology of the immobilized membrane fragments were investigated again by in situ tapping mode AFM.The kinetics of surface-association was monitored in real time with the —uorescence sensor using the FITC-labelled enzyme analogue. Typically lipid monolayer formation and subsequent membrane fragment immobilization on a waveguide chip were performed in the PWG instrument and monitored in real time. The kinetics of these two processes is shown in Fig. 5 as —uorescence responses. After an initial phase of buÜer equilibration (5 min at 0.5 ml min~1 —ow) lipid vesicles (–rst step) or membrane discs (second step) were injected and incubated under stopped-—ow conditions for 1 h. In a subsequent buÜer wash (10 min at 0.5 ml min~1 i.e. almost a thousand times exchange of the cell volume) excess material was efficiently removed. For the lipid assembly no —uorescence increase was observed as expected.Interestingly however the process of lipid spreading could nevertheless be followed by a clearly detectable increase of the reference signal (excitation light). In the case of membrane disc association a strong and fast increase of —uorescence intensity was observed due to the association of the labelled membrane discs onto the surface. After 1 h the signal slowly approached a stationary emission at ca. 106 cps. Only a minor portion of this signal disappeared on washing. This signal decrease was interpreted as the removal of loosely bound membranes in the vicinity of the surface. Final signal-to-noise ratios of the net —uorescence were 4000^500. The established —uorescence signal F0 in buÜer II was checked for its long-term stability a prerequisite for subsequent assay 279 Faraday Discuss.1998 111 273»288 Fig. 4 Quality control of surface lipid assembly by in situ AFM a lipid monolayer formed after spreading of hydrophobic surface consisted of a self-assembled monolayer of mono-C -alkyl phosphate. Up to 10% of the POPC vesicles prepared by extrusion (vesicle diameter 110 nm) to a hydrophobic Ta2O5 waveguide. The total surface is covered by small-sized single defects. Preparation and imaging of the lipid monolayer were 16 performed in 150 mM NaCl 10 mM sodium phosphate buÜer 0.02% NaN at pH 7.5. The gray scale of the image (black to white) corresponds to a height of 10 nm. 3 applications. Signals were stable as measured for up to 3 h in running buÜer with signal readouts for 1 s every 1 min.Long-term drifts due to photo-bleaching were minimized using low illumination intensities (typically 40»100 lW before chip coupling). The concentration of membrane discs corresponded to a protein content of ca. 0.25 mg ml~1. Assembly experiments have been carried out successfully with Na,K-ATPase prepared from both pig kidney and from dog–sh rectal gland. However in the following the reported results refer only to the dog–sh enzyme. Membrane»surface association depends critically on the salt conditions of the surrounding buÜer. In order to avoid an early loading of the cation binding sites of the enzyme the use of Na` and K` cations in the assembly buÜer was excluded. Therefore surface immobilization was performed in 100 mM ChoCl 10 mM imidazole pH 7.5 (buÜer II).High ionic strength strongly assisted the association of membrane particles to the surface and in addition provided a stable contact of the discs on the surface. Fig. 5 shows a representative experiment of lipid monolayer formation and membrane disc immobilization on the preformed lipid layer. Surface coverage and topology of such preparations were investigated under otherwise comparable conditions with tapping mode AFM in solution. Fig. 6 shows a representative AFM picture of membrane fragments immobilized —at on a lipid monolayer. The average coverages of such preparations ranged between 25% and 50% depending on the scanning position on the chip and including chip to chip variation. In addition coverages determined from —uorescence signals with assembled membrane discs or with a phospholipid monolayer doped with —uorescein-labelled lipid (one label per 800 lipids) were in good agreement with the AFM results.This indicated the robustness of our preparation procedure. Concerning the topology and stability of surface-associated membrane particles membrane discs preferentially associated —at on the surface. The stability of the physisorption of the membranes was concluded from the fact that the disc-like structures could not be removed with the AFM tip from their positions during the process of scanning. At comparatively high forces the Faraday Discuss. 1998 111 273»288 280 Fig. 5 Kinetics of lipid monolayer assembly (Ö) and subsequent surface-association of dispersed membrane fragments (L) containing FITC-Na,K-ATPase (protein concentration 0.22 mg ml~1; prepared from dog–sh rectal gland) on a waveguide chip.Measurements were performed in 10 mM imidazole/HCl 100 mM choline chloride pH 7.5 (at 20 °C). Fluorescence (in cps) and reference signals (in mV) were monitored after an initial 5 min of buÜer exchange (under continuous —ow) a 60 min sample incubation (under stopped —ow) and –nal 10 min of buÜer washing (under continuous —ow). F corresponds to the asymptotically reached emission intensity of surface-immobilized FITC-Na,K-ATPase prior to the supply of samples containing interacting cations 0 or ligands. Binding of the speci–c cations or ligands leads to the characteristic —uorescence changes *F. Fig. 6 AFM quality control of immobilized membrane fragments containing FITC-labelled Na,K-ATPase on sensitized waveguide surfaces.At high ionic strength (left) up to 50% surface coverage of assembled membrane discs was achieved; at low ionic strength (right) only few and loosely de–ned structures with lower contrast could be imaged. FITC-Na,K-ATPase was prepared from the dog–sh rectal gland and used in concentrations of ca. 0.22 mg ml~1. The gray scale of the image (black to white) corresponds to a height of 50 nm. 281 Faraday Discuss. 1998 111 273»288 0 discs could even be dissected with the tip (data not shown). Further evidence for a stable membrane immobilization was the observation of a constant sensor —uorescence signal upon extensive washing with running buÜer. The AFM experiments also indicated that the surface contact zone of a membrane disc was preferentially located close to the disc edge.The edges protruded an additional 4»5 nm into the solution. The central region of a membrane disc is assumed to be fairly —exible. This was concluded from the fact that the AFM resolution in the disc centre was lower than at the edge and that the height of the disc centre with respect to the substrate surface depended on the ionic composition of the buÜer solution for example by the addition of divalent cations. In the presence of 3 mM MgCl2 the disc centre height was reduced typically from 9 to 7 nm ([25%). When membrane particles were assembled at low ionic strength (10 mM imidazole pH 7.5) a much lower surface coverage was obtained (Fig.6 right). Under these circumstances it was difficult to resolve the structures with high contrast in an AFM scan indicating that a weaker interaction in terms of a reduced contact area per disc existed. This was also evident by the observation of larger —uctuations of the —uorescence sensor signal and additionally of lower base signals F than at high ionic strength or in the presence of additional divalent ions. On individual membrane preparations AFM images with high resolution were recorded (Fig. 7). Within the central part of an immobilized membrane disc single protrusions with a density of about one stucture per 1000 nm2 were imaged. The structures protruded from the surface into the solution by an average height of 5»6 nm. The lateral size of the protrusions was in the range of 15»20 nm.Taking into account the size of the scanning tip (5»10 nm) the real size of the imaged structures must be ca. 10 nm. Such a size was interpreted as the membrane-external protein moiety of the ATPase. However the lateral size of 10 nm is too large to represent an individual protein molecule. Therefore we believe that the imaged structures correspond to aggregates of protein equal or larger than a dimer. Considering aggregation the –nal protein-to-lipid (P L) ratio in the disc centre was determined to be in the range 1 100 to 1 1000. At the disc edge P L Fig. 7 High-resolution AFM picture of a membrane fragment containing FITC-Na,K-ATPase (prepared from dog–sh) immobilized on a sensitized waveguide chip. The central part of the disc shows individual protrusions with average lateral dimensions of 15»20 nm and an average height of 5»6 nm.Protrusions are interpreted as aggregated ATPase molecules (aggregation state[dimer). Protein lipid ratios in the central part of the membrane disc were of the order of 1 100»1 1000 increasing towards and at the disc edge. The gray scale of the image (black to white) corresponds to a height of 20 nm. Faraday Discuss. 1998 111 273»288 282 ratios were still larger leading to the assumption that there stronger aggregation or even –rst indications for protein crystallization occur. Functional tests of immobilized Na,K-ATPase Once the optimum conditions for a reproducible and stable immobilization of membrane fragments were identi–ed i.e.a stable and constant base —uorescence signal F was generated on the 0 chip the functional activity of the Na,K-ATPase could be tested. This was done by using a set of complementary assays to probe the diÜerent binding sites of the protein. Each of these assays alters speci–cally in a side-directed manner the relative intrinsic —uorescence F/F of the labelled 0 enzyme. For a better overview the characteristic —uorescence changes of the membrane discs in bulk solution upon variation of the buÜer composition or additions of speci–c ligands are shown in Fig. 8. The speci–c binding of K` to the alkali-metal ion binding pocket was considered to be the most critical test for a preserved functional activity of the protein since this binding is impaired with a characteristic change in protein conformation leading to a relatively large change of the —uorescein emission ([*F/F0B30%).Speci–c Kë-binding and binding isotherm. In all preparations the speci–c K`-eÜect was tested as a reference experiment. Small volumes of KCl in buÜer II (maintaining the ionic strength constant) were applied under continuous —ow. Upon K`-contact the —uorescence dropped instantly to lower stable values dependent on the KCl concentration (cf. Fig. 9). This —uorescence eÜect was fully reversible yielding reproducible —uorescence levels after washing in the absence of KCl and in subsequent repeats. Titrations over six orders of concentration (0.1 lM to 100 mM KCl) were performed. From the speci–c —uorescence decreases ([*F/F0) a quantitative evaluation according to a binding isotherm was performed (Fig.10). The data followed ideally according to a K Langmuir binding kinetics revealing a dissociation constant of D\180^20 lM. Saturation of binding was achieved above 2 mM K` but only a 60^10% fraction of the amplitude was reached when compared to the respective —uorescence changes in bulk solution (Fig. 8). On the other hand the dissociation constant of the surface-con–ned measurements compared well to that found in bulk measurements. This indicates a preserved enzyme activity in the assembled state but with a reduction of the number of K`-binding sites per assembled protein molecule. This may imply that not all available sites are accessible by the buÜer medium. F/F Fig. 8 Schematic illustration of the relative —uorescence intensity levels 0 (jexc\488 nm; jem .\520 nm) of FITC-Na,K-ATPase (prepared from dog –sh rectal gland) as measured in bulk phase containing diÜerent cations or ligands (Pi\inorganic phosphate) relative to the base level F0 in 10 mM imidazole in HCl 100 mM choline chloride (ChoCl) pH 7.5 (buÜer II) at 20 °C.The total ionic strength of all solutions was kept constant by adjustment with choline chloride (ChoCl). 283 Faraday Discuss. 1998 111 273»288 Fig. 9 Representative course of —uorescence signals in a typical assay sequence measured after preparation of surface-immobilized FITC-labelled Na,K-ATPase (dog–sh) as shown in Fig. 5. (L) speci–c K`-binding after injection of 6 mM KCl (Ö) competitive inhibition of speci–c K`-binding in the presence of high [Na`] and signal decrease upon addition of 180 nM anti-—uorescein antibody 4-4-20 [(K) measured signal ; (=) signal corrected for unspeci–c oÜset].Speci–city of Kë-binding. Further experiments were performed to check the speci–city of K`- binding. When the immobilized membranes were saturated with high concentrations of NaCl (100 mM) no further signal decrease was observed upon addition of KCl (cf. Fig. 9). Thus speci–c K`-binding is not observed in the presence of excess Na` which is consistent with bulk measurments. Furthermore with excess KCl [10 mM an additional K`-eÜect could be resolved as visible from the observed deviations from constant saturation binding (cf. Fig. 10). This eÜect obviously suggests a weaker binding and was interpreted as an unspeci–c K`-binding to the protein or the membrane.18 2O5 Kë-binding of membranes immobilized on bare non-sensitized Ta surfaces.Control experiments were performed to clarify whether or not the preparation of a biocompatible interface i.e. the chip covered with a lipid monolayer is a necessary prerequisite for preserved ATPase binding characteristics. Therefore membrane discs were immobilized on a non-sensitized hydrophilic Ta chip. In this case the typical signal decrease upon subsequent K`-addition did not occur except for a minor eÜect of the order of [2%. It is interesting to note that much larger —uorescence signals could be observed upon assembly of the membrane fragments to the chip surface indicating a ca. 10-fold higher association constant.This phenomenon was also veri–ed in AFM experiments in which ca. 10 times more diluted membrane suspensions reached comparable surface coverages (25%»50%) as in the case of the lipid-coated chips. 2O5 Binding of anti-—uorescein antibody. As an additional feature the side of the —uorescein label placed within the ATP-binding domain of the enzyme was probed by —uorescence quenching upon speci–c binding of monoclonal anti-—uorescein antibody 4-4-20 (Molecular Probes). This was done in a typical assay sequence (in working buÜer II) after the initial testing for the speci–c K`-binding as shown in Fig. 9. The antibody is able to quench the —uorescence of free —uorescein almost completely ([90%).25 In the case of the FITC-labelled ATPase-analogue —uorescence in 284 Faraday Discuss.1998 111 273»288 Fig. 10 Speci–c K`-binding of FITC-Na,K-ATPase (prepared from dog–sh) upon additions of KCl over –ve orders of concentration maintaining the ionic strength constant. Experiments were performed with surfaceimmobilized enzyme (Ö) and with non-immobilized enzyme in bulk solution (L) in 10 mM imidazole in HCl 100 mM choline chloride pH 7.5 at 20 °C. The surface-immobilized enzyme reached only ca. 70% of the relative saturation amplitude of enzyme in solution. The concentration dependence in both cases was well described according to a Langmuir binding kinetics leading to dissociation constants of KD\180^20 lM K values indicate a preserved activity of the ATPase in the surface-immobilized state. A 70% relative saturafor the immobilized enzyme (»»») and KD\270^30 lM for the enzyme in bulk (» » » »).The comparable tion amplitude under a K comparable to solution suggests a 70% 30% orientation of membrane fragments D on the surface with the speci–c K`-binding site exposed to the solution. D bulk is only quenchable to a maximum extent of 30»40% depending on the method of preparation. We found that the —uorescence in the case of the surface-immobilized protein is only one half of the quenching eÜect in solution (20% quench) upon incubation in saturating concentrations of the antibody (180 nM). An initial fast and a subsequent slower kinetic quench response could be resolved. The initial fast one was reversible upon subsequent washing with buÜer whereas the quench representative of the slower phase (within minutes) remained stable for minutes at F/F0B 0.9 (10% quench).The later eÜect was interpreted as the speci–c (direct) antibody binding to the —uorescein label (slow dissociation) whereas the fast response may represent an unspeci–c interaction to the —uorescein emission. 2 F of the sensor were higher in this buÜer by at least 5»6%. This eÜect was explained Phosphorylation and binding of inhibitory ligand (ouabain). To investigate the speci–c binding of an inhibitory ligand (ouabain) and phosphorylation the enzyme activity was studied by another set of assays. Phosphorylation probes especially the ATP-binding site of the protein. The immobilized protein was –rst equilibrated in the presence of 3 mM MgCl (buÜer III). The base —uorescence signals 0 by the induction of a closer contact of the assembled membrane discs to the chip surface in the presence of divalent cations (see AFM results).Phosphorylation after incubation of 3 mM inorganic phosphate led to a minor decrease in F in the range 2»6%. Speci–c binding of inhibitory 0 ligand was observed by measuring the decrease in the —uorescence signal upon continuous addition of 2.5 mM ouabain in the presence of 3 mM MgCl and 3 mM inorganic phosphate. The 2 kinetics of ligand binding was as slow as in bulk membrane dispersions (ca. 20 min to reach a stationary signal). The –nal signal decrease stabilized at ca. 40% of the eÜect measured in bulk solution. Together with the K`-binding studies (60% of the solution eÜect) and the antibody-induced 285 Faraday Discuss.1998 111 273»288 quenching studies (40% of the solution eÜect) the results obtained with speci–c binding of ouabain (40% of the solution eÜect) are consistent with the assumption that the two orientations of membrane fragments upon immobilization on the waveguide (former inside up and former inside down) establish almost equally on the surface (cf. Fig. 3). Discussion We present a new method showing that transmembrane proteins embedded in natural biomembrane fragments can be immobilized with well preserved activity in a rationally de–ned and controlled manner on a planar metal oxide waveguide transducer. The optimization of the individual preparation steps for a de–ned and functional immobilization of membrane fragments was the main issue of this work.First examples are presented as to how the functional activity and the speci–c binding sites of immobilized Na,K-ATPase can be probed by surface-con–ned —uorescence. A new optical technique based on the sensitive detection of —uorescence excited in the evanescent –eld of thin planar waveguides was applied for these investigations. Evanescent –eld —uorescence detection The detection of evanescent-–eld excited surface-con–ned —uorescence has the inherent advantages of high sensitivity at large signal-to-noise (S N) ratios with an almost complete suppression of background signals from the bulk environment. In the present case of immobilized —uoresceinlabelled biomembrane particles S N ratios of up to 5000 at only partial surface coverages were obtained.This implies that under optimal sensing conditions functional probing of membrane proteins in very small detection –elds approaching sizes similar to the scanning areas of the presented AFM pictures may become possible. On the other hand even in the case of macroscopic detection –elds (P1 mm2) consumption of only low volumes of receptor samples or the use of very dilute samples e.g. at low receptor yields are required. Compared to bulk measurements biological receptors immobilized on surfaces oÜer the great advantage that diÜerent buÜer media and/or speci–c ligands or inhibitors can be sequentially applied to the same preparation and subsequently exchanged in a fast and easy manner. Thereby automation enables a high degree of reproducibility.A fast assay development and optimization,26 and –nally a high assay performance are the consequences as demonstrated in the present case. Functional surface immobilization The surface immobilization of membrane proteins under preservation of full protein activity and with presentation of a maximum number of accessible binding sites is the key challenge for the investigation of protein functions with surface-sensitive detection schemes. We have pursued the concept of immobilizing intact membrane fragments where the proteins are embedded in their natural environment. There a high likelihood consists that the protein function is maintained during the whole process of surface preparation. This is diÜerent to other methods which employ the almost random immobilization of membrane receptors in puri–ed and/or detergent-solubilized form on the surface27 or in a more de–ned way the immobilization of solubilized receptors on functionalized surfaces via chemical affinity tags (His-Tag approach) positioned at distinct protein locations.10 With the latter method the ligand-binding activity was established as for the native receptor but it has to be critically monitored for each preparation step for changes in the protein environment (e.g.change of detergent) and for each reaction partner. The presented method is the preferred one to immobilize membrane receptors in their native environment under retention of the natural biomembrane orientation. However in contrast to individual receptor molecules the demands for the surface immobilization of nanoparticulate membranes in combination with surface-sensitive detection schemes are much higher.Especially the stability of surface contacts seems to be very important since the dimensions of the membrane fragments were of the same size as the penetration depth of the evanescent –eld (few hundred nm). Instabilities in the orientation of surface-associated membranes with respect to the evanescent –eld especially when exposed to —owing buÜer solution would result in large —uctuations of the sensor signal (as observed in our –rst experiments or when applying too low ionic strength). We have achieved a stable and functional membrane fragment immobilization in a —at con–guration. Faraday Discuss. 1998 111 273»288 286 Necessity for a biocompatible interface The experiments clearly showed that the presence of a gentle biocompatible interface is needed for a functional immobilization of Na,K-ATPase.The enzyme lost its speci–c cation binding activity completely when immobilized directly on the bare chip surfaces although the protein was still embedded in its native membrane environment. The biocompatible interface was formed by covering the waveguide chip with a self-assembled phospholipid monolayer. Such a biomimetic layer protects the sensitive membrane proteins from potential denaturing contact with the bare nonsensitized transducer surface. In addition the chemical composition and the physico-chemical properties of such adlayers can be well controlled since many natural and synthetic lipids and cofactors for their formation are available.The physico-chemical properties of the lipid surface and the diÜusion properties of the monolayer can easily be modi–ed and anchor groups for affinity binding can be introduced.28 In addition lipid layers are able to suppress the interaction of soluble substances with the surface thus reducing the non-speci–c binding.24 Stability of membrane fragment immobilization in the evanescent –eld The conditions for a stable immobilization of the membrane fragments on the preformed biocompatible interface have been optimized by varying the physico-chemical properties of the surrounding buÜer media. BuÜer solutions at high ionic strength as used in the course of this study are known to overcome the long-range repulsive electrostatic interactions (membranes as well as oxide surfaces are preferentially negatively charged) and to favour the strongly attractive van der Waals forces.29 In the presence of the lipid monolayer the membrane affinity to the coated surfaces was much lower compared to the non-sensitized metal oxide surfaces.However a fully preserved protein activity established in the case of the preformed biocompatible surfaces was concluded from the comparable assay characteristics and binding constants of surface-assembled enzyme and ì free œ non-immobilized enzyme in solution (see speci–c K`-binding). Aspect of sidedness Besides the aspect of establishing a general procedure for the functional immobilization of membrane fragments the other important interest in the investigation of a surface-immobilized Na,KATPase was the aspect of sidedness of the speci–c binding sites.To address this aspect there are literature reports of electrophysiological measurements carried out with Na,K-ATPase either with intact cells such as ventricular myocytes30 or oocytes,31 where only the extracellular medium can be changed extensively or the enzyme can be reconstituted in vesicular systems.32h34 Such reconstitutions imply that the membrane-bound protein has to be solubilized with a suitable detergent with retention of its full enzymatic activity. These are obviously conditions in which the original orientation of the protein molecules in the disc membrane has been lost. The solubilized enzyme which forms a mixed protein-detergent micelle is then reincorporated in the lipid membrane of the vesicular systems.Besides the fact that it is difficult to –nd suitable detergents which do not alter irreversibly the structure and functional properties of the protein the reconstituted enzyme can adopt two diÜerent orientations (inside-out and inside-in) in the vesicular lipid membrane. In addition it can be adsorbed onto the lipid surface without being incorporated. In order to study side-directed properties of such systems it is necessary to introduce discriminations biochemically for example by inactivating one state of orientation. Two recently developed techniques provide information about directed properties of electrogenic as well as of electroneutral steps that are functionally coupled.These techniques are related to electrical measurements of Na,K-ATPase discs adsorbed onto or incorporated into black lipid membranes35,36 and also employ the patch method on excised membrane fragments originating from cellular systems. Because the transport currents of transporter proteins are very small compared with those of ion channels the development of the giant membrane patch method which permits the inside-out and the outside-out membrane orientation has led to interesting studies.37,38 In the case of a surface-immobilized ATPase the sidedness of speci–c binding sites can be determined under conditions in which all membrane fragments are homogeneously oriented on the surface. Having achieved a functional immobilization of membrane fragments in a sensor 287 Faraday Discuss.1998 111 273»288 con–guration under control of membrane fragment orientation on the sensor surface the aspects of sidedness may be investigated in a comparatively simple and straightforward manner using the planar waveguide approach. Paper 8/06704J Acknowledgements This article is dedicated to Prof. J. C. Skou who discovered the Na,K-ATPase on the occasion of his 80th birthday. The authors thank Mrs. A. Schacht for skillful enzyme preparations Mr. A. Spielmann for experimental help Dr. J. Fritz for part of the AFM studies Mr. E. Lewitzki for numerous control measurements and Dr. H. Ruf for a dynamic light scattering measurement. Many stimulating discussions with Dr. G. Duveneck Dr. P. Oroszlan Dr. G. Kraus Dr. A. Abel Dr.D. Neuschaé fer Dr. B. Klee Dr. A. Cudd and Prof. Dr. H. Vogel are acknowledged. In addition we are grateful to Prof. Dr. H. Vogel and Dr. A. Cudd for carefully reading the manuscript. References 1 R. B. Gennis in Biomembranes ed. Ch.R. Cantor Springer Verlag New York 1989. 2 D. Bray Annu. Rev. Biophys. Biomol. Struct. 1998 27 59. 3 N. Unwin Cell 1993 72 31. 4 J. P. Changeux Sci. Am. 1996 6 499. 5 R. J. Lefkowitz S. Cottechia P. Samama and T. Costa T rends Pharmacol. Sci. 1993 14 303. 6 A. G. Gilaman Angew. Chem. Int. Ed. Engl. 1995 34 1406. 7 J. R. Broach and J. Thorner Nature (L ondon) 1996 384 (6604 Suppl.) 14. 8 J. J. Burbaum and N. H. Sigal Curr. Opin. Chem. Biol. 1997 1 72. 9 P.Fué rst and J. Heim BIOforum Int. 1998 2 64. 10 E. L. Schmid A.P. Tairi R. Hovius and H. Vogel Anal. Chem. 1998 70 1331. 11 J. Hodgson Biotechnology 1994 12 31. 12 A. G. Frutos and R. M. Horn Anal. Chem. 1998 70 449A. 13 G. L. Duveneck M. Pawlak D. Neuschaé fer W. Budach and M. Ehrat SPIE Proceedings of Biomedical Systems and T echnologies 1996 2928 98. 14 G. L. Duveneck M. Pawlak D. Neuschaé fer E. Baé r W. Budach U. Pieles and M. Ehrat Sens. Actuat. B 1997 38»39 88. 15 J. C. Skou Biochim. Biophys. Acta 1957 23 394. 16 S. J. D. Karlish in Na,K-AT Pase Structure and Kinetics ed. J. C. Skou and J. G. Norby Academic Press New York 1979 p. 115. 17 C. Hegyvary and P. L. J‘rgensen J. Biol. Chem. 1981 256 6296. 18 E. Grell R. Warmuth and H. Ruf Acta Physiol. Scand. 1992 146 213; 1993 147 343. 19 P. L. J‘rgensen Biochim.Biophys. Acta 1974 356 36. 20 O. H. Lowry N. J. Rosenbrough A. L. Farr and R. J. Randall J. Biol. Chem. 1951 193 265. 21 D. Porschke and E. Grell Biochim. Biophys. Acta 1995 1231 181. 22 D. Brovelli L. Ruiz G. Kraus G. Haé hner R. Hofer A. Waldner J. Schloé sser P. Oroszlan M. Ehrat and N. D. Spencer L angmuir 1999 submitted. 23 M. Pawlak E. Grell D. Anselmetti and M. Ehrat in preparation. 24 S. Terretaz T. Stora C. Duschl and H. Vogel L angmuir 1993 9 1361. 25 E. Lewitzki E. Schick and E. Grell J. Fluoresc. 1998 8 113. 26 M. Pawlak E. Schmid R. Hovius E. Grell H. Vogel and M. Ehrat in IBC Conference Proceedings ìHigh T hroughput Screeningœ IBC Southborough USA 1998. 27 K. R. Rogers J. J. Valdes and M. E. Eldefrawi Biosens. Bioelectron. 1991 6 1. 28 C. Duschl A. F. Seç vin-Landais and H. Vogel Biophys. J. 1996 70 1985. 29 D. J. Mué ller M. Amrein and A. Engel J. Struct. Biol. 1997 119 172. 30 D. C. Gadsby M. Nakao A. Bahinski G. Nagel and M. Suenson Acta Physiol. Scand. 1992 146 111. 31 W. Schwarz and L. A. Vasilets Cell Biol. Int. 1996 20 67. 32 S. D. J. Karlish and U. Pick J. Physiol. 1981 312 505. 33 A. Rephaeli D. Richards and S. D. J. Karlish J. Biol. Chem. 1986 261 6248. 34 F. Cornelius in T he Sodium Pump Structure Mechanism and Regulation ed. J. H. Kaplan and P. De Weer Rockefeller University Press New York 1991 p. 267. 35 K. Fendler E. Grell M. Haubs and E. Bamberg EMBO J. 1985 12 3079. 36 A. Eisenrauch E. Grell and E. Bamberg in ref. 34 p. 317. 37 D. W. Hilgemann Ann. N.Y . Acad. Sci. 1997 260. 38 U. Eckstein-Ludwig J. Rettinger L. A. Vasilets and W. Schwarz Biochim. Biophys. Acta 1998 1372 289. Faraday Discuss. 1998 111 273»288 288

ISSN:1359-6640    

DOI:10.1039/a806704j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Analysis of membrane protein cluster densities and sizes in situ by image correlation spectroscopy Nils O. Petersen Claire Brown Anna Kaminski Jonathan Rocheleau Mamta Srivastava and Paul W. Wiseman Department of Chemistry T he University of W estern Ontario L ondon ON N6A 5B7 Canada Received 25th August 1998 Communication between cells invariably involves interactions of a signalling molecule with a receptor at the surface of the cell. Typically the receptor is imbedded in the membrane and it is hypothesized that the binding of the signalling molecule causes a change in the state of aggregation of the receptor which in turn initiates a biochemical signal within the cell. Subsequently many of the occupied receptors bind to membrane-associated structures called coated pits which invaginate and pinch oÜ to form coated vesicles thereby removing the receptors from the cell surface.The state of aggregation of membrane receptors is obviously in constant —ux. Any useful approach to measuring the state of aggregation must therefore allow for dynamic measurements in living cells. It is possible to use —uorescently labelled signalling molecules or antibodies directed at the receptor of interest to visualize the receptor on the cell surface with a —uorescence microscope. By employing a laser confocal microscope high resolution images can be produced in which the —uorescence intensity is quantitatively imaged as a function of position across the surface of the cell. Calculations of autocorrelation functions of these images provide direct and accurate measures of the density of —uorescent particles on the surface.Combined with the average intensity in the image which re—ects the total average number of molecules it is possible to estimate the degree of aggregation of the receptor molecules. We refer to this analysis as image correlation spectroscopy (ICS). We show how ICS can be used to measure the density of several receptors on a variety of cells and how it can be used to measure the density of coated pits and the number of molecules per coated pit. We also show how the technique can be used to monitor fusion of virus particles to cell membranes. Further we illustrate that by calculating cross-correlation functions between pairs of images we can extend the analysis to measurements of the distributions as a function of time on the second timescale as well as to measurements of the movement of the receptor aggregates on the surface.Finally we illustrate that by this approach we can measure the extent of interaction between two diÜerent receptors as a function of time. This represents the most quantitative measurement of the extent of co-localization of receptors available and is independent of the spatial resolution of the confocal microscope. The theory of ICS and image cross-correlation spectroscopy (ICCS) focussing on the interpretation of the data in terms of the biological phenomenon being probed is discussed. 289 Faraday Discuss. 1998 111 289»305 Introduction Structurally the cell surface is a heterogeneous mixture of lipids and proteins in a constant state of —ux.It contains a number of transmembrane proteins imbedded in a —uid lipid bilayer as well as a large number of membrane-associated proteins and lipid-anchored proteins. Many of these in turn bind to extracellular matrix proteins or intracellular networks of cytoplasmic proteins. All lipid components and lipid-anchored proteins exhibit a high degree of lateral mobility in the membrane. Transmembrane proteins however are subjected to several restrictions in their mobility. First they move more slowly than anticipated from theory ;1 second they are restricted by the cytoskeleton to move in sub-lm domains;2 and third they tend to aggregate with other proteins in specialized structures such as the coated pits.3 Recently special organizations of lipids and proteins have been identi–ed in cells (cavaolae)4 and there is mounting evidence that glycolipids and lipid-anchored proteins can form detergent resistant membrane domains.5h7 Functionally some lipids and many proteins in the membrane are involved as receptors to ligands that are in the medium attached to the growth substrate or on neighbouring cells.In most cases it is believed that the immediate consequence of binding of a ligand to a receptor is a redistribution of the proteins into aggregates (ranging from dimers to oligomers).8 This association of receptors allows for biochemical interactions such as trans-phosphorylation,9 or physical interactions such as enhanced binding.10 Subsequently the ligand»receptor complexes can be cleared from the surface by internalization through special structures such as the clathrin-coated pits11 or the caveolin-associated invaginations.12 The complex structure and function of the cell surface components demands increasingly sophisticated tools to probe the speci–c intermolecular interactions that are coupled to speci–c functions.For example it is argued that in order for a trans-membrane protein to be associated with a coated pit it must have a speci–c amino acid sequence exposed in a particular fold that will allow it to bind to the adaptor protein in the coated pit.13 The evidence for these and similar examples of speci–c intermolecular interactions is for the most part based on measurement of these protein»protein interactions ex situ in extracted and puri–ed preparations.14,15 Clearly it is desirable to be able to measure these interactions quantitatively in situ and in real time.A number of techniques based on electron microscopy coupled with immunological labelling techniques have provided detailed pictures of molecular associations and structural organizations. 16 Prime examples are the clathrin-coated pits17 and the caveolin-coated rosettes.6 The high resolution and the speci–city of antibody binding gives exquisite detail of organization. There are two drawbacks however it is very difficult to ensure and verify that all receptors are visualized and the nature of the specimen preparation precludes certain kinetic experiments. Immuno- —uorescence experiments provide the same speci–city but at a much reduced resolution even allowing for resolution enhancement techniques.18 Laser scanning confocal microscopy introduces additional resolution in the third dimension and permits truly quantitative measurements of —uorescence intensity as a function of position.19 This has permitted studies of the distribution and redistribution of receptor molecules at the sub-lm level in living cells with a time resolution of seconds.20,21 More recently techniques have been developed for studying the co-localization of two diÜerent receptors based on pixel-by-pixel mixing of colours representative of the —uorescence intensity green for one receptor red for the other receptor and yellow if both receptors provide intensity in the same pixels and by using point-by-point correlation analyses.22,23 We have introduced an autocorrelation function analysis of confocal images which allows us to estimate both the density of receptor clusters and the number of receptors per cluster irrespective of the actual physical dimensions of the clusters.20,24 We have also introduced a cross-correlation function analysis of pairs of images which allows us to estimate the kinetics of redistribution of the receptor clusters (if the images are obtained at diÜerent times)21 or the fraction of one receptor co-localized with another receptor (if the images are obtained for diÜerent probes).25,26 The key advantage of these analyses is that they reduce the measurement from many observations within a single image to one number with a particular understandable signi–cance.This allows for statistically meaningful investigations of large populations of cells giving rise to a more accurate measure of the relevant intermolecular interactions. Combination of the correlation analysis with more traditional image processing can also give quantitative information about the distribution of cluster sizes or about which sub-population of a receptor is interacting with other receptors. Faraday Discuss. 1998 111 289»305 290 Experimental Cell preparations Cells were grown in a T-25 —ask in medium supplemented with fetal calf serum and antibiotics in a 5% CO atmosphere. When needed for experiments the cells were plated on glass coverslips at a 2 low density. At the appropriate time in the procedures the cells were –xed in 4% aqueous paraformaldehyde at pH 7 at room temperature for 30 min or by immersion for 5 min in cold methanol ([20 °C) followed by 5 min in cold acetone ([20 °C).At the appropriate time in the protocol the cells were labelled with a primary antibody at saturating concentrations followed by a secondary antibody at saturating concentrations. The actual concentrations and times varied with the antibody and the cell type. Experiments performed with gangliosides were conducted on live cells contained in a temperature-controlled unit on the stage. Sendai Virus was labelled with the lipid dye diQ (Molecular Probes) by simple mixing and separation on a size-exclusion column. Microscopy Laser scanning confocal microscope images were obtained on a Biorad 600 attached to a Nikon inverted microscope and equipped with an argon»krypton mixed gas laser.Each image was collected at Zoom 10 providing pixel resolutions of 0.03 lm and image dimensions of 15.5 lm in each direction. The images consisted of 515x512 pixels with 8 bit integer intensity data in each. The laser beam was focussed to ca. 0.35 lm at the e~2-waist and was imaged through an expanded pinhole to give an eÜective depth of focus of ca. 1 lm. The —uorescein or nitrobenzoxadiazole dyes were excited at 488 nm while the Red-X and Di-Q dyes were excited at 568 nm. In dual-labelling experiments images were collected separately at each wavelength to minimize cross-over —uorescence between the two channels. The images were collected in the photon-counting mode to ensure a linear intensity response.In all cases the background level was set to deliver a non-zero count in each pixel and the gain was set to avoid saturating any of the pixels. This protocol ensures that no information is lost at either end of the intensity scale. As few as a single scan and as many as 25 scans were averaged depending on the experiment but for a given set of experiments the same conditions were used for all samples. Images were collected periodically during an experiment to establish the background (the sample was not illuminated but the laser was allowed to re—ect oÜ all mirrors and lenses). Unlabelled cells were imaged to obtain images representing auto—uorescence and where possible cells were labelled with the second antibody only to obtain images representing the sum of auto—uorescence and non-speci–c labelling.Image processing All images were transferred to a MasPar 2 with 2048 parallel processors. The correlation calculations were performed using 2D FFT routines available on the MasPar. The resulting correlation functions were –t to a 2D three-parameter Gaussian function using non-linear regression analysis (Marquandt Algorithms). The data output was processed using commercial software (QuatroPro SigmaPlot Photoshop CorelDraw) as appropriate. (1) g(m g)\ S(i(x y)[Si(x y)T)(i(x]m y]g)[Si(x y)T)T Si(x y)T2 Theory ICS Let the intensity in each pixel in the image be i(x y). The spatial correlation function for the image is then given by eqn. (1) where the angular brackets indicate the average over all spatial coordinates.27 It is known28 that n i(x y)] the variance of the limit of the correlation function as m and g vanish gives g(0,0)\var[d the normalized intensity —uctuations 291 Faraday Discuss.1998 111 289»305 dn i(x y)\ i(x y)[Si(x y)T Si(x y)T When the intensity is a true re—ection of the concentration this variance is also the variance in the concentration —uctuation which equals29 the inverse of the average occupation number N1 p (2) N1 g(0 0)\var(dn i(x y))\var(dn c(x y))\ 1 p The occupancy number represents the average number of —uorescent particles present in the observation volume de–ned in the confocal microscope by the convolution of the laser beam and the point spread function for the image pin hole.For molecules on thin —at membranes this corresponds to an area given by nw2 where w is the e~2 beam waist. Thus the cluster density (CD) is de–ned as27 (3) CD\ g(0 0) 1 nw2 \ n N w 1 p 2 Moreover the intensity is proportional to the average number of —uorescent molecules so that Si(x y)T\cN1 m where c is a constant re—ecting optical parameters (molar absorption coefficients quantum yields and the efficiency of collection of the confocal microscope).30 Thus we can de–ne a degree of aggregation (DA) as31 (4) N1 DA\Si(x y)Tg(0 0)\c N1 m p The constant c can be determined experimentally32 by assuming that the non-speci–c binding of —uorescently labelled secondary antibodies occurs as monomers i.e.DA ns\c. ICCS Consider the intensity distribution for two images possibly collected at diÜerent times i(x y t) and j(x y t]q). It is possible to calculate cross-correlation functions as21 (5a) gij(m g q)\ S(i(x y t)[Si(x y t)T)( j(x]m y]g t]q)[Sj(x y t]q)T)T Si(x y t)TSj(x y t]q)T (5b) gji(m g q)\ S( j(x y t]q)[Sj(x y t]q)T)(i)(x]m y]g t)[Si(x y t)T)T Sj(x y t]q)TSi(x y t)T where we expect gij(m g q)\gji(m g q) in most cases. If images are collected as a function of time q from the same cell observing the same chromophore throughout then the cross-correlation functions will report the ensemble-averaged dynamics of the aggregation pattern on the surface. In the limit where the number of aggregates remains –xed the zero-lag amplitudes of the cross-correlation functions are simple functions of time i.e.gij correlation spectroscopy [ f (q)\exp(q/q In these cases g(0 0 0) gives the cluster density as above.21 (0 0 q)\g(0 0 0) f (q) where f (q) depends on the dynamics as in normal —uorescence f)2 for 2D —ow and f (q)\(1[q/qd)~1 for 2D diÜusion]. If images are collected at the same time from the same cell observing two diÜerent chromophores then q\0 and the zero-lag amplitudes of the cross-correlation functions re—ect the extent of co-localization of the two chromophores on the cell surface at a particular time t.21,33 Speci–- cally,25 (6) gij(0 0 t)\ (N1 N1 i]N1 ij)(N1 j]N1 ij) \N1 ij gi(0 0 t)gj(0 0 t) ij where N1 represents the average number of aggregates in which there are both chromophores ij present.Accordingly we can de–ne a cluster density of co-localized chromophores as Faraday Discuss. 1998 111 289»305 292 (7) g (0 0 t) (0 0 t)nw2 \ N1 ij nw2 CDij(t)\ g (0 0 ij t)gj i Note that in this case the cluster density is proportional to the amplitude of the cross-correlation function. The cluster density of co-localized chromophores can now be calculated from the crosscorrelation and the autocorrelation functions of a pair of images. By collecting pairs of images as a function of time the evolution of co-localization can be followed. At this time we do not have a parameter corresponding to the degree of aggregation and hence the direct calculation of the number of each chromophore present in a co-localized aggregate is not available.However we can calculate the fraction of clusters of one chromophore which participates in the co-localized aggregates. This we de–ne as (8) \ gij(0 0 t) or F( j o i)\ CDij\ gij(0 0 t) F(i o j)\ CDij g (0 0 t) i CDj g (0 0 t) j CDi which we read as the fraction of clusters of type i that are co-localized with clusters of type j and vice versa. Other considerations Fitting. Confocal images of aggregates with dimensions smaller than that of the laser beam will consist of circular intensity spots with dimensions given by the size of the laser beam at the focal plane. If the confocal microscope is properly adjusted the laser beam is in the TEM mode with 00 a Gaussian transverse intensity pro–le.34,35 In this case the auto- and cross-correlation functions will have a 2D Gaussian decay centred at the origin and with a decay width given by the laser beam width w.This can be measured independently.30,34,35 Accordingly we –t the calculated correlation functions as a three parameter function27 (9) g(m g)\g(0 0)e~(m2`g2)@w2]g0 where the zero-lag amplitude g(0 0) the beam width w and the oÜset g are the –tted param- 0 eters. The amplitude is the desired parameter. The width can be compared to the known width as a control for the quality of the –t. If the –t and known beam widths diÜer by more than 30% the data normally should be rejected. In some cases all the data show larger beam widths suggesting that the clusters are comparable in dimension to the laser beam. Then the width from individual correlation functions is compared to the average width from all the relevant data.Individual data can be rejected if they fail the z-score test for outliers.36 The oÜset in eqn. (9) is included to account for the empirical observation that the correlation function will not always decay to zero. This arises from the fact that the images have –nite sizes and at long distances the information is correspondingly less accurate. For example a length of 15.5 lm will only allow for ca. 50 characteristic —uctuations of a 0.35 lm laser beam across each dimension. The uncertainty in the oÜset is therefore of the order of J50/50\0.14 corresponding closely to the uncertainty in the data from the best samples we can obtain. It has been shown that the optimum –t includes only the data in the correlation function to about three times the width.37 This is implemented by using only data for which (m2]g2)\(3w)2 with 3w being ca.32 pixel dimensions (ca. 1 lm). The autocorrelation function will always –t to a maximum amplitude at the zero-lag coordinates. The cross-correlation function need not be maximum at the zero-lag coordinates since there may be systematic shifts between the images collected at two diÜerent excitation and emission wavelengths. Hence the cross-correlation functions are –t to a maximum at g(m0 g0) and the coordinates at the maximum are recorded. Systematic variations in these coordinates can be noted and accounted for. In fact if there is a —ow process the changes in the coordinates of the maximum can be used to determine the —ow velocity.21 Random variations in the coordinates may result from small instabilities in the microscope and we accept data as being valid whenever these random variations are less than about one beam width (10 pixels) in either direction.If the cross-correlation is very small or absent the –t will be to the largest correlation in the whole 293 Faraday Discuss. 1998 111 289»305 function. This may be at large lag values since the information content is poorer because of the limited image dimension. We record the fraction of images for which the coordinates of the maximum of the cross-correlation function is greater than ca. 10 pixels and interpret this as the fraction of images which show no cross-correlation or cross-correlation below the random correlation level.The fraction of cross-correlation described above [F(i o j)] is calculated only from those images for which the cross-correlation is maximum close to the zero-lag coordinates. Spot densities. Many cells have between 100 000 and 2 000 000 receptors on their surface.32,38h40 Typical surface areas are of the order of 5000»10 000 lm2 suggesting densities of 10»400 receptors per lm2. Our images (15.5 lm]15.5 lm\240 lm2) would allow as many as 625 individual receptors to be resolved as distinct —uorescent spots [n(0.35)2 lm2\0.38 lm2]. For most receptor systems the monomer density is therefore high enough for even high-resolution confocal images to show a very homogeneous distribution of —uorescence. In fact if a monomer can be detected it can be resolved in these images only if density is less than ca.3 per lm2. If the contrast and the signal-to-noise is good it is possible to count the number of spots in the image and calculate a spot density de–ned as SD\ìthe number of spots per lm2. This is difficult or impossible in cases where the signal is weak. The advantage of the correlation analysis is that it provides an approach to estimating the density even in a noisy background. Multiple populations. Whenever the spot density can be calculated we expect agreement with the cluster density estimated from the correlation calculations. We have never seen SD[CD but we routinely observe that CD[SD. We interpret this as an indication that there are at least two populations of receptor clusters large clusters easily identi–able in the image and contributing to the spot density count and monomers or small clusters not discernible visually above the background but contributing to the cluster density through their contribution to the correlation function.In general it is possible to account for contributions of multiple populations of receptors to the amplitude of the correlation function30 since (10) g(0 0)\ 1 N1 m 2 ;i N1 i(pi2]ki2) where for the ith population k is the mean number of monomers per aggregate (degree of i aggregation) N p 1 is the mean number of such aggregates and is the standard deviation about the i i mean in the distribution of the number of monomers per aggregate. As before N1 is the mean m number of monomers.In the simplest case we assume that there are two populations with vanishingly small standard deviations so that eqn. (10) simpli–es to (11) g(0 0)\ 1 N1 m 2 (N1 1 k12]N1 2 k2 2 ) Clearly one measurement of g(0,0) cannot provide sufficient information to calculate these –ve parameters but there are often additional constraints depending on the model for aggregation in a particular system. Conservation of mass suggests that to a reasonable approximation:N1 m\N1 1k1 ]N1 2 k2 . Often N1 m is known from other sources such as biochemical binding data for a ligand to a receptor. In a model where there are small and large aggregates as suggested if CD[SD then N1 can be estimated from the spot density assuming that this is the large cluster population 2 and the other population could be assumed to be monomers so that k1\1.With this particular model we get (12) g(0 0)\ 1 N1 m 2 [N1 1](SDnw2)k2 2] and N1 M\N1 1](SDnw2)k2 which uniquely de–ne all the parameters. The interpretation is very sensitive to the model chosen. Even so the detailed calculations can provide some insight especially when changes are observed as a consequence of treatment of the cells. At this time we have not extended the consideration of multiple populations to the crosscorrelation calculations. This is important but is likely to be complex. Corrections. In general several sources of —uorescence will contribute to the measured signal Faraday Discuss. 1998 111 289»305 294 (m).32,38 These include those of interest in the sample (s) those from non-speci–c —uorescence (ns) those from auto—uorescence (a) and those from the background irradiation in the microscope (wn).The –rst three will exhibit —uctuations with the characteristic spatial dimension determined by the laser beam whereas the last will have characteristics of white noise with zero amplitude in the correlation functions away from the cental pixel. If we assume that these are independent sources (a good assumption given that the origins of the signals are from diÜerent components of the sample) then gm(m g)Sim(x y)T2\gs(m g)Sis(x y)T2]gns(m g)Sins(x y)T2 (13) ]ga(m g)Sia(x y)T2]gwn(m g)Siwn(x y)T2 For true white noise gwn(m g)\0 except at the origin where it can be very large. Hence all –ts exclude the data at the origin.Measurements on cells labelled with the secondary antibody only provide an estimate of the combined contributions from non-speci–c —uorescence and auto—uorescence. These data can be used to make a simultaneous correction for the second and third terms in eqn. (13). If necessary measurements on unlabelled cells provide an independent estimate of the contribution from auto—uorescence which can be used to extract the contribution for the nonspeci –c —uorescence. This is useful for estimating the constant c used in calculating the absolute degrees of aggregation as discussed above. We perform the correction shown in eqn. (14) to get the proper estimate of the correlation function but as implied we apply the correction only to the zero-lag amplitude of the correlation function after it has been calculated from the –t of the raw correlation functions.(14) s g (0 0)\ gm(0 0)SimT2[[gns(0 0)SinsT2]ga(0 0)SiaT2] (SimT[[SinsT]SiaT][SiwnT)2 The best experiments require measurements of instrumental background unlabelled cells nonspeci –cally labelled cells and speci–cally labelled cells under the same microscope conditions. To obtain statistically meaningful estimates we routinely collect 40 images or more for each experimental condition. Frequently the numbers are based on hundreds of images. Results Several applications of image correlation spectroscopy and image cross-correlation spectroscopy have been published.20,21,25,27,31,38 Here we provide new data generated to illustrate the scope and limitations of these techniques for making measurements of cluster densities degrees of aggregations and extent of co-localization of cell membrane associated proteins and lipids.Cluster density of growth factor receptors Platelet derived growth factor receptors. Table 1 shows the data obtained on a number of AG1523 cells –xed with 4% paraformaldehyde prior to the addition of monoclonal antibodies directed at the platelet-derived growth factor receptor (PDGF-R) and labelling the primary antibody with goat-anti-mouse IgG antibodies speci–c for either the Fc portion of the primary antibody or the Fab portion of the primary antibody. The former is labelled with —uorescein isothiocyanate the latter with tetramethyl rhodamine isothiocyanate. The average intensity is much less in the —uorescein experiments since the chromophore bleaches rapidly and it is therefore only possible to collect and sum a few (here four) images on a given cell area.The degree of aggregation [eqn. (4)] is measured under conditions where every receptor is saturated by bound primary antibody and every primary antibody is saturated by bound secondary antibody. These values are corrected for non-speci–c —uorescence and auto—uorescence. The degree of aggregation is also measured under non-speci–c labelling conditions and corrected for auto—uorescence to give an estimate of the optical constant c for these illumination conditions. From these we can then estimate the mean number of —uorescent antibodies per cluster. This is calculated from eqn. (4) and (10) allowing for a Poisson distribution of cluster sizes about a single mean value (k) with a variance equal to the mean (p2\k) so that k\[Si(x y)Tg(0 0)/c][1\DA/DAns[1.It is clear that when the antibody is speci–c for the Fab portion of the primary antibody k is about twice the value obtained when the antibody is speci–c for the Fc portion of the primary antibody. 295 Faraday Discuss. 1998 111 289»305 Table 1 PDGF-receptors on AG1523 cells TMRITC-IgG (Fab-speci–c) FITC-IgG (Fc-speci–c) Antibody Si(x,y)T DA DAns\c 4.4 4.2 0.48 7.8 3.9 2.3 0.2 0.25 0.057 3.3 3.3 2.2 k(antibody) k(receptor) CD Accounting for the fact that there are two Fab and one Fc segment per primary antibody we calculate 3.3 and 3.9 receptors per cluster.There is an excellent agreement between the two labelling approaches. Binding data show that on average AG1523 cells express as many as 150 000 copies of the platelet derived growth factor receptor (PDGF-R) on the cell surface.39 Given an average cell surface area38 of ca. 11 000 lm2 this corresponds to an upper limit of 14 receptors per lm2. The ICS data yield as many as 2.3 clusters per lm2 with 3.9 receptors per cluster corresponding to 9 receptors per lm2. This is in reasonable agreement with the binding data and illustrates that the technique is able to detect most if not all of the receptors on the surface. Also the small numbers of receptors per cluster clearly show that the confocal microscope combined with the correlation analysis enable us to image and characterize single receptors on the surface of adherent cells.Similar data can be obtained on living cells.20 Epidermal growth factor receptors. Table 2 shows the data for experiments on the mouse cell line A431 labelled with a primary antibody directed at the epidermal growth factor receptor (EGF-R) and a secondary antibody directed at the Fc segment of the primary antibody. The experiments were performed on cells which were incubated at diÜerent temperatures prior to –xing with paraformaldehyde. At 37 °C the cluster density value is ca. 20 so we see ca. eight times more EGF-R clusters per unit area on A431 cells than for PDGF-R on AG1523 cells. There are on average 11 receptors per cluster ca. 3 times as many for EGF[R on A431 compared with PDGF-R on AG1523.This corresponds to ca. 220 receptors per lm2 for EGF-R compared to ca. 9 for PDGF-R It is known that A431 cells overexpress EGF-R and as many as 2»3 million receptors are present per cell.40 These cells are smaller than the AG1523 cells with an average surface area of ca. 3000 lm2 so we expect ca. 600»1000 receptors per lm2. The ICS data con–rm the excess expression. These measurements are made with saturating conditions as illustrated in Fig. 1 yet only about one third of the receptors are detected. Nevertheless the data show clearly that the overexpression of EGF-R on A431 cells leads to both a larger number of clusters and a larger number of receptors per cluster when compared to the normal expression of PDGF-R on AG1523 cells.The receptors to EGF are highly aggregated prior to exposure to the antibody and prior to exposure to the growth factor (these cells have been starved i.e. deprived of any exposure to growth factors in the growth medium for 24 h prior to the experiment). The average beam size measured by the autocorrelation functions is larger by ca. 33% than that expected from the laser beam (0.45 cf. 0.35 lm). This is consistent for all the samples at 37 and 22 °C and indicates that the dimension of the cluster is as large as or slightly larger than the laser beam. This suggests that these receptors may not be in molecular contact in the clusters but they could simply be con–ned to a domain in the membrane. Pre-existing clusters of receptors may be functionally important since they may enhance the rate of activation by the growth factor since Table 2 EGF-receptors on A431 cells 4 °C 22 °C 37 °C CD k(receptor) w 0.45^0.03 19^4 13^4 10^2 11^3 17^3 33^2 0.46^0.03 0.40^0.01 Faraday Discuss.1998 111 289»305 296 Fig. 1 CD of epidermal growth factor receptors on A431 cells measured as a function of concentration of the primary antibody in the incubation solution. Clearly a maximum number of clusters is detected at concentrations greater than ca. 0.1 lg mL~1. Further addition of the antibody does not reveal more clusters but could reveal a larger number of receptors per antibody. In these experiments the DA reaches a maximum corresponding to ca. 11 receptors per aggregate at the same primary antibody concentrations.the mean free path for collision would be signi–cantly reduced. It appears to be a valid concept for EGF-R in A431 cells and to a more limited degree for PDGF-R in AG1523 cells. How general this observation is remains to be seen. When the A431 cells are cooled to 4 °C for 30 min prior to –xation the cluster density decreases by a factor of two and the mean number of receptors per cluster increases by a factor of three indicating that cooling causes a further aggregation of the receptors. At the same time the average dimension of the clusters also decreases signi–cantly. While these data are still preliminary they do suggest that the receptors associate more tightly at lower temperatures. A similar trend was observed previously with PDGF-R on AG1523 cells.20 Surface labelling.Note that the primary and secondary antibodies label only the top surfaces of the cells. This is illustrated in Fig. 2 which shows three confocal images at diÜerent heights through a cell. Some bright spots are clearly in focus on the top surface of the cell over the nucleus [Fig. 2(a)]. In contrast there are no —uorescent spots in focus on the membrane under the nucleus [Fig. 2(b) and (c)] indicating that there is no labelling on that surface. This is shown more clearly in Fig. 2(d) which is a cross-section of the same cell that shows the —uorescence only on the top contour of the cell. This is a characteristic of all the experiments we have conducted with antibodies that are directed at the exterior surface of the membrane.Our estimates of cluster densities in those systems re—ect this surface only and we have assumed that there is an equal expression of receptors on the bottom surface but that the antibody cannot access them within the time frame of the labelling procedure. In experiments where the antibody is directed at the cytoplasmic part of the membrane proteins both the top and the bottom membrane of the cell is labelled. In the —at peripheral regions of the cell both membranes will be imaged simultaneously [they are separated by less than the focal depth of the confocal microscope (ca. 1 lm)]. This means that the cluster density re—ects clusters on both surfaces and it is overestimated by a factor of two. We have made some measurements of cluster densities in regions under the nucleus where only the bottom surface should contribute.These are complicated by the enhanced auto—uorescence in the nuclear area but qualitatively the data are consistent with comparable cluster densities on each surface. Thus for cytoplasmic proteins (see below) we divide the cluster density values by a factor of two to get a true density of clusters per lm2 of membrane surface. 297 Faraday Discuss. 1998 111 289»305 Fig. 2 Confocal images of A431 cells labelled to reveal EGF-R on the surface. (a) Section at the top of the cell (b) section close to the top of surface of the cell at the periphery (c) section at the bottom of the cell. The dark spots represent —uorescence (in a reversed image) and the light areas show lack of —uorescence labelling.The absence of —uorescence is evident under the nucleus in (b) and over a large region of the cell in (c) indicating that the antibody only labels receptors exposed on the top surface. (d) z-scan through the middle of the cell con–rming that the labelling is con–ned to the top surface. Distribution of coated pit proteins Coated pits contain at least three protein components membrane receptors targeted for internalization adaptor proteins which bind to the receptors and clathrin which binds to adaptor proteins and form the structural framework of the coat in the pit and in coated vesicles.41 The adaptor protein which is found at the plasma membrane is designated AP-2 and diÜers structurally and immunologically from AP-1 which is found exclusively at the Golgi and AP-3 which is the least understood member of the family.42 Clathrin is found at the plasma membrane in some cytoplasmic vesicles (before they get uncoated) and at the Golgi and as a free component in the cytoplasm.Clathrin and adaptor proteins. Table 3 summarizes some of the data obtained for the distribution of clathrin and the adaptor protein AP-2 on monkey kidney –broblast cells (CV-1 cells). Both clathrin and AP-2 are proteins associated with the cytoplasmic side of the membrane so they can only be labelled after the cells are made permeable in this case through a methanol»acetone –xation at [20 °C. The antibody will label both the top and the bottom membrane and the data in Table 3 have been adjusted by the factor of two as discussed above. The cluster density is signi–cantly larger than the spot density for both proteins.This suggests that there may be two populations of proteins large aggregates that are clearly discernible as individual spots and small aggregates or monomers that blend with the background. Since the data have been corrected for the contributions from auto—uorescence and non-speci–c labelling we believe that the more disperse population is real. It could correspond to a monomer population in the thin layer of cytoplasm between the two membranes or small aggregates associated with the membrane. To test these possibilities cells were treated with saponin prior to –xation and labelling for clathrin or AP-2. Saponin is supposed to remove cholesterol from the membrane rendering it permeable and permitting cytoplasmic components to escape the cell.Table 3 shows Faraday Discuss. 1998 111 289»305 298 Table 3 Coated pits on CV-1 cells Adaptor protein (AP-2) Clathrin CD SD CD after saponin 1.4^0.20 0.28^0.04 1.5^0.05 1.32^0.08 0.55^0.24 0.50^0.10 that in the case of clathrin the CD value is reduced to the same value as the SD following saponin treatment con–rming that the discrepancy between these values arises from a cytoplasmic component of the clathrin. In contrast saponin does not aÜect the cluster density for the AP-2 suggesting that the small aggregate component of this protein is membrane associated. The cluster density of the large aggregate population of clathrin is ca. 0.5 which compares favourably with the density of coated pits estimated by electron microscopy methods.25,43 The ICS data therefore suggest that clathrin exists in two populations coated pits and free cytoplasmic protein.Using eqn. (11) or (12) we estimate that ca. 70% of the clathrin is in the coated pits with ca. 120 clathrin triskelia per coated pit and 30% is in the cytoplasm as triskelia in these peripheral regions of the membrane. The spot density for AP-2 is less than that of clathrin by close to a factor of two. We expect that these spots correspond to AP-2 associated with clathrin in coated pits. The smaller value for SD might then indicate that only half of the coated pits contain large amounts of AP-2. Alternatively the spot density may be underestimated because a smaller number of AP-2 proteins per coated pit would give a poorer signal to background.It is proposed that the ratio of AP-2 to clathrin triskelia is one-third. The number of binding sites for antibodies on AP-2 may then be almost 10-fold less making the spots less visible and harder to count. The cross correlation data discussed below support the latter alternative. Model receptor proteins. In collaboration with Professors Henis (Tel Aviv University) and Roth (Southwestern Medical Centre) we have used the model receptor system obtained by transfecting the hemagglutination protein (HA) from the in—uenza virus into CV-1 cells.44,45 The wild-type protein does not have the cytoplasmic signal sequence which allows for binding to the adaptor protein. It is therefore expected to be distributed homogeneously across the surface.The HA]8 mutant has been created by adding the eight amino acid sequence YDYKSFYN which introduces the YXX' signal sequence.45 This mutant will bind to the adaptor protein and is expected to be more aggregated as a result. Since the protein is transfected into the CV-1 cells the cluster density may be expected to depend on the eÜectiveness of the expression. For comparable intensities suggesting comparable surface expressions the cluster density of the HA-wt is consistently six times or more than that of HA]8. The cluster density of the HA]8 is ca. 3.2^0.7 for cells with good levels of transfection.31 This value of the cluster density is signi–cantly greater than the cluster density of either clathrin (1.5) or AP-2 (1.5) and the density of coated pits (0.5).This suggests that only a fraction of these receptors interact with coated pits at any given time. Intermolecular interactions. Fig. 3 shows the auto- and cross-correlation functions from images of HA-wt and clathrin obtained from the same sections of a cell. It is evident that while the autocorrelation functions have large amplitudes [Fig. 3(a) for HA-wt and 3(b) for clathrin] there is no amplitude in the cross-correlation function [Fig. 3(c)] indicating a minimal co-localization. Fig. 4 shows the auto- and cross-correlation functions from images of HA]8 and clathrin. This time the amplitudes of the autocorrelation functions [Fig. 4(a) for HA]8 and 4(b) for clathrin] and the cross-correlation function [Fig.4(c)] are all signi–cantly above background [Fig. 4(c)] indicating at least some co-localization. Table 4 summarizes the values of the cluster densities of co-localized aggregates (CDgr) the fraction of receptors interacting with clathrin and AP-2 and the fraction of clathrin and AP-2 interacting with the receptor. The amplitudes of the cross-correlation functions between the HA-wt and the clathrin or AP-2 were so small that they exceeded the random correlation function —uctuations in only ca. 20% of the measurements and in those cases the amplitudes were very small. In fact less than 2% of the receptors co-localize with either clathrin or AP-2 and vice versa. 299 Faraday Discuss. 1998 111 289»305 Fig. 3 Auto- and cross-correlation functions for images of HA-wt and clathrin.(a) Autocorrelation function for HA-wt. This is a small amplitude function since the HA-wt protein is relatively disperse. (b) Autocorrelation function for clathrin. This is a larger function since the protein is highly clustered. (c) Cross-correlation function for these two images. There is no detectable amplitude at zero-lag and the largest correlation occurs 5»6 lm from the origin. The HA]8 system diÜers from the HA-wt since all the measurements yield cross-correlation functions with signi–cant amplitudes at the zero-lag position. The density of clusters which contain both receptor and clathrin (CDgr\0.59) is the same as the density of clusters which contain both receptor and AP-2 (CDgr\0.57). These values correspond in turn very well to the density of coated pits determined above (Table 3) from the spot density and cluster density of the individual proteins.We therefore surmise that all the coated pits contain HA]8 clathrin and AP-2. Table 4 shows that only ca. 25% of the HA]8 co-localizes with either clathrin or AP-2 suggesting that 75% exists outside the coated pits. This is consistent with previous diÜusion data which suggest that ca. 25% of HA]8 is immobile whereas 75% moves freely and rapidly.46 This Table 4 Intermolecular interactions Protein Receptor N F(C o H) or F(A o H) F(H o C) or F(H o A) CDgr clathrin AP-2 clathrin AP-2 40 40 175 127 \0.02 \0.02 0.72^0.06 0.69^0.07 \0.02 \0.02 0.25^0.02 0.25^0.03 NAa NAa 0.59^0.06 0.57^0.06 HA-wt HA-wt HA]8 HA]8 a NA\not applicable.Only ca. 20% of the cells imaged showed a cross-correlation at the zero-lag indicating that there is very little co-localization. Of those that did have a crosscorrelation function at the origin the amplitudes were very small leading to very small Fvalues. N represents those images that showed a cross-correlation. Faraday Discuss. 1998 111 289»305 300 Fig. 4 Auto- and cross-correlation functions for images of HA]8 and clathrin. (a) Autocorrelation function for HA-wt. This is now a large amplitude function since the HA]8 protein is partly clustered. (b) Autocorrelation function for clathrin. This is a larger function since the protein is highly clustered. (c) Cross-correlation function for these two images.There is a signi–cant and measurable amplitude at zero-lag showing that there is at least some co-localization in these images. is also consistent with the conclusion that only a fraction of the HA]8 interacts with the coated pits derived above from the cluster density data for the HA]8 (CD\3.2). At this point we do not know the state of aggregation of the HA]8 outside the coated pits since we do not know the level of surface expression. In fact preliminary data suggest that the fraction associated with the coated pits varies with the level of surface expression in a manner predicted from a simple binding model where monomers associate with a complex with many equivalent binding sites. Approximately 72% of all the clathrin clusters co-localize with HA]8 suggesting that ca.28% of these protein clusters exist outside the coated pits. This –ts very well with the cluster density and spot density data from which we concluded that ca. 30% of the clathrin was in the cytoplasm as individual triskelia. Similarly ca. 69% of all the AP-2 clusters co-localize with HA]8 suggesting that ca. 31% of these proteins exist outside the coated pits. We have previously proposed that the AP-2 exists in coated pits and in smaller clusters which could be nucleation sites.47 If this model is correct then the cross-correlation data obtained here would be consistent with the AP-2 in coated pits interacting with HA]8 and the AP-2 outside the coated pits not binding to HA]8. This may however mean that the interpretation of the AP-2 outside the coated pits as nucleation sites is less certain.Distribution of gangliosides Gangliosides have been implicated in the formation of specialized domains in the membranes such as caveloae and detergent-resistant membranes.5h7 We have incorporated a —uorescent derivative of the ganglioside GD1a (NBD-GD1a) into CV-1 cells and studied the distribution on the surface of live cells before and after exposure to Sendai Virus at low temperature where no fusion 301 Faraday Discuss. 1998 111 289»305 Table 5 Ganglioside distributiona CD SD Treatment Probe 24 9 0.6 no virus virus virus 0.08 0.09 0.3 NBD-GD1a NBD-GD1a DiQ-virus a The absolute numbers depend on the concentration of ganglioside applied. occurs (Table 5).The ganglioside is found in both the top and bottom membranes. To some extent it is in internalized vesicles close to the nucleus but this is distant from the regions imaged in these cells. Clearly the spot density is much lower than the cluster density once again suggesting that the ganglioside exists in two populations an aggregated one seen in the spots and a dispersed one in the background. When the virus is added it labels only on the top surface. The spot density for the NBD-GD1a remains the same but the cluster density decreases by more than a factor of two. This suggests that the virus binds to and aggregates the disperse population of gangliosides without aÜecting the distribution of the highly aggregated ganglioside population or that it binds to the highly aggregated population and recruits gangliosides from the disperse population.In these experiments the spot density and the cluster density of the virus are comparable but both are highly variable and depend on the amount of virus added (these are not saturating conditions). There are more virus particles than large aggregates of the ganglioside providing initial evidence that the large aggregates are not the only sites of binding for the virus. Table 6 shows the co-localization data for the ganglioside and the virus. Less than half the cells measured show a signi–cant co-localization of ganglioside and virus and generally the amplitude of the cross-correlation function is low. Nevertheless in those cases where the amplitude can be measured ca.24% of the ganglioside is associated with virus and ca. 37% of the virus is associated with gangliosides. Visual inspection of the overlayed images shows that virtually none of the large ganglioside clusters overlaps with the virus (not shown). Thus the co-localization is con–ned to the disperse distribution of the ganglioside in agreement with the change in cluster density shown in Table 5. For comparison Table 6 also shows the co-localization of NBD-GD1a with the bsubunit of choleratoxin labelled with biotin. Choleratoxin is known to bind to the ganglioside GM1 and has been used as a marker for caveloae on the surface of cells.49 It can be seen that over 80% of the cells measured show co-localization of NBD-GD1a with choleratoxin. The fraction of ganglioside co-localized with choleratoxin is still low (28%) but the fraction of choleratoxin clusters that contain NBD-GD1a is over 80%.This is consistent with the choleratoxin binding to the same domains where NBD-GD1a coexists with GM1. Inspection of the images also reveals a signi–cant number of yellow spots in the overlapped images (not shown). For comparison the lipid probe NBD-PE and choleratoxin show little cross-correlation with no evidence for signi–- cant extent of co-localization. In other work,48 we have shown that the fusion of Sendai Virus to the cell membrane can be monitored by image correlation spectroscopy at the single cell level by following the dispersal of a dye initially in the virus membrane. This is an example of the reverse of an aggregation process.Fraction –t Protein Ganglioside F(P o G) F(G o P) CDrg 0.37 0.82 0.08 0.24 0.28 0.05 0.62 2.8 0.11 0.41 0.82 0.35 virus cholera toxin cholera toxin NBD-GD1a NBD-GD1a NBD-PE Discussion ICS is a simple image analysis tool which permits quantitative estimates of the density of —uorescent molecules on the surface of cells. When the images are collected with a confocal microscope Table 6 Ganglioside co-localization Faraday Discuss. 1998 111 289»305 302 and when proper care is taken with control experiments it is possible to detect receptors present in very low quantities. When the intensity information is used properly it is also possible to estimate quite accurately the average number of —uorescent molecules per aggregate.This was demonstrated with the PDGF-Receptors on AG1523 cells which exist on the average in clusters of 3»4 proteins at a density of ca. 2»3 clusters per lm2. It is possible in these systems to detect single receptors however so far none of the systems studied has consisted of monomers only. The technique is precise enough that small and possibly signi–cant changes in cluster densities can be determined when the cell system is perturbed. The study of EGF-Receptors on A431 cells shows that overexpression of the receptor in these cells leads to more and larger aggregates. The density and size of these aggregates is sensitive to temperature and it is possible to monitor changes by factors of two and less even when the aggregates are large.When the physical dimensions of the clusters increase the correlation functions broaden. Deconvolution calculations can in principle be used to extract information ca. the actual dimensions of the clusters. In many cases the cluster density data calculated from ICS are inconsistent with the density of bright spots observed in the image. We argue that when the experiments are performed with the proper controls this re—ects two or more populations of the protein in question. The distribution of the proteins among these populations cannot be determined uniquely from these measurements. Nevertheless system speci–c models can be introduced and tested. For example clathrin should exist in at least two populations coated pits and cytoplasmic monomers (triskelia).This model then provides enough information to give detailed estimates of the number of coated pits and the fraction of the clathrin they contain. A similar situation obtains with the adaptor protein but here the data suggest that there are large and small aggregates present. The data are also compatible with a broad distribution of aggregates but a single aggregate population is ruled out. The third bimodal system is represented by the HA]8 model receptor where some of the receptors are associated with the coated pits while others are free in the membrane. Interactions between proteins can be studied eÜectively by using the cross-correlation approach. The amplitude of the cross-correlation function at zero lag is now proportional to the extent of co-localization.If the fraction of images in which the amplitude of the cross-correlation function is signi–cantly greater at zero lag than elsewhere is close to one then most cells exhibit a signi–- cant co-localization of the proteins. In these cases it is possible to calculate the density of clusters in which the proteins interact and the fraction of the clusters of either protein which form the co-localized aggregates. The density of clusters containing both HA]8 and clathrin is found to be the same as the density of clusters containing both HA]8 and AP-2. The fraction of HA]8 clusters associated with clathrin clusters is also the same as the fraction associated with AP-2 proteins. Finally the fraction of clathrin clusters associated with HA]8 clusters is the same as the fraction of AP-2 associated with HA]8 clusters.The –rst two observations are consistent with the proposal that the HA]8-clathrin clusters are the same as the HA]8-AP-2 clusters i.e. they are the coated pits which contain all three proteins. The third observation is probably a coincidence. For technical reasons,§ we have not yet succeeded in measuring the cross-correlation between clathrin and the adaptor protein. Given the data obtained above we would anticipate a density of co-localized clusters between 0.55 and 0.60 and we expect each of the two proteins to be ca. 70% co-localized with the other. The ganglioside derivative NBD-GD1a is found either dispersed or in large bright clusters. The disperse population appears to be a better receptor for Sendai Virus which binds and causes a change in the disperse population but not the aggregated one.The extent of co-localization between the ganglioside and the virus is low and there are very few of the virus particles which are co-localized with the large NBD-GD1a clusters. On the other hand NBD-GD1a is co-localized to a much greater extent with the b-subunit of choleratoxin and there is signi–cant co-localization in the bright spot. This suggests either that NBD-GD1a is a receptor for the toxin or that it is co-localized with GM1 which is known to be a receptor for the toxin. In either case this colocalization is predominantly in the large aggregates. These may be either the caveolae or the § This experiment requires simultaneous labelling intracellularly so the primary antibodies cannot be from the same source (mouse).While antibodies from other sources are available we have not yet been able to establish the correct conditions. Hopefully the problem will be solved soon. 303 Faraday Discuss. 1998 111 289»305 putative detergent resistent membranes. Future cross-correlation experiments with caveolin and lipid anchored proteins may help distinguish these possibilities. Conclusions Careful measurements of —uorescence emission from cell surfaces using confocal microscopes can provide high quality images of the distribution of molecules in cells or on their surface. Calculation of the autocorrelation function of a single image yields quantitative estimates of the density of clusters and the number of molecules per cluster.Calculation of the cross-correlation function between a pair of images of diÜerent proteins yields an estimate of the fraction of co-localization and the density of mixed protein clusters. Measurements on a large number of cells give good statistical information and hence reliable conclusions about the protein distributions and the intermolecular interactions. Since the experiments can be performed on a rapid timescale (seconds to minutes) kinetics of redistribution of proteins and rates of intermolecular associations are accessible through these calculations. Complex systems such as receptors interacting with coated pits or virus particles binding to and fusing with cells can be investigated systematically to provide better insight into the mechanism of intermolecular associations.ICS is an ìaveragingœ technique in that all the —uorescence information in an image is reduced to a single number. This means that details are lost and speci–c information about the protein distributions is difficult to extract. However the averaging provides great sensitivity to weak signals that may otherwise be overlooked. The ease of collecting images on modern confocal microscopes and the relatively simple calculations involved allow for a vast amount of data to be collected and processed in a short time. This means better experiments and better controls. Some argue that a single picture is worth a thousand words. ICS can reduce a thousand pictures to a single number. In many systems this is important and worthwhile.Acknowledgements We greatly appreciate the many fruitful interactions with our collaborators Karl-Eric Magnusson and Birgitta Rasmusson at Linkoping University in Sweden; Richard Epand at McMaster University in Canada; Tom Flanagan at the State University of New York at BuÜalo in the United States ; Yoav Henis at Tel Aviv University in Israel and Michael Roth at Southwestern Medical School in the United States. References 1 D. Sheets R. Simson and K. Jacobson Curr. Opin. Cell Biol. 1995 7 707. 2 A. Kusumi and Y. Sako Curr. Opin. Cell Biol. 1996 8 556. 3 S. L. Schmid Annu. Rev. Biochem. 1997 66 511. 4 R. G. Parton and K. Simons Science 1995 269 1398. 5 D. A. Brown and J. K. Bose Cell 1992 68 533. 6 R. G. Parton J. Histochem. Cytochem.1994 42 155. 7 T. Harder and K. Simons Curr. Opin. Cell Biol. 1997 9 534. 8 A. Sorkin and C. M. Waters Bioessays 1993 15 375. 9 J. Schlessinger T rends Biochem. Sci. 1998 13 443. 10 T. Kirchhausen J. S. Bonifacino and H. Riezman Curr. Opin. Cell Biol. 1997 9 488. 11 T. Kirchhausen Curr. Opin. Cell Biol. 1993 3 182. 12 C. Lamaze and S. L. Schmid Curr. Opin. Cell Biol. 1995 7 573. 13 I. V. Sandoval and O. Bakke T rends Cell Biol. 1994 4 292. 14 W. Boll H. Ohno Z. Sangyang I. Rapoport L. C. Cantley J. S. Bonifacino and T. Kirchhausen EMBO J. 1996 15 5789. 15 H. Ohno J. Stewart M. Fournier H. Bosshart I. Rhee S. Miyatake T. Saito A. Gallusser T. Kirchhausen and J. S. Bonifacino Science 1995 169 1872. 16 J. E. Heuser J. Cell Biol. 1980 84 560. 17 J.E. Heuser and J. H. Keen J. Cell Biol. 1988 107 877. 18 D. Gross and W. W. Webb Biophys. J. 1986 49 901. 19 J. B. Pawley Handbook of Biological Confocal Microscopy Plenum Press New York 2nd edn. 1995. 20 P. W. Wiseman P. Hoé ddelius N. O. Petersen and K. E. Magnusson FEBS 1997 401 43. 21 M. Srivastava and N. O. Petersen Meth. Cell Sci. 1996 18 47. Faraday Discuss. 1998 111 289»305 304 22 B. van Steensel E. P. van Binnendijk C. D. Hornsby H. T. M. van der Voort Z. S. Krozowski E. R. de Kloet and R. van Driel J. Cell Sci. 1996 109 787. 23 D. Demandolx and J. Davoust J. Microsc. 1997 185 21. 24 N. O. Petersen Can. J. Biochem. Cell Biol. 1984 62 1158. 25 C. M. Brown PhD Thesis The University of Western Ontario London Canada 1998. 26 C. M. Brown Biochim.Biophys. Acta 1998 submitted. 27 N. O. Petersen P. L. Hoddelius P. W. Wiseman O. Seger and K. E. Magnusson Biophys. J. 1993 65 1135. 28 B. J. Berne and R. Pecora Dynamics L ight Scattering with Applications to Chemistry Biology and Physics Wiley New York 1976 pp. 10»22. 29 N. Davidson Statistical Mechanics McGraw-Hill 1962. 30 N. O. Petersen Biophys. J. 1986 49 809. 31 E. Fire C. M. Brown R. G. Roth Y. I. Henis and N. O. Petersen J. Biol. Chem. 1997 272 29538. 32 P. R. St-Pierre and N. O. Petersen Biochemistry 1992 31 2459. 33 Z. Foeldes-Papp A. Schnetz and R. Riegler Biophys. J. 1998 74 A184. 34 N. O. Petersen S. Felder and E. L. Elson in Handbook of Experimental Immunology ed. D. M. Weir L. A. Herzenberg C. C. Blackwell and L. A. Herzenberg Blackwell Scienti–c Edinburgh 1985 ch. 24. 35 N. O. Petersen and E. L. Elson in Methods in Enzymology ed. C. H. W. Hirs and S. N. TimasheÜ Academic Press New York 1985. 36 W. Mendenhall Introduction to Probability and Statistics 7th edn. 1987 pp. 64»65. 37 A. G. Benn and R. J. Kulperger Environmetrics 1996 7 167. 38 P. W. Wiseman PhD Thesis The University of Western Ontario London Canada 1995. 39 R. A. Seifert C. E. Hart P. E. Phillips J. W. Forstrom R. Ross M. J. Murray and D. F. Bowen-Pope J. Biol. Chem. 1989 264 8771. 40 T. Kawamoto J. D. Sato A. Le J. PolikoÜ G. H. Sato and J. Mendelsohn Proc. Natl. Acad. Sci. USA 1983 80 1337. 41 M. S. Robinson Curr. Opin. Cell Biol. 1994 6 538. 42 G. Odorizzi C. R. Cowles and S. D. Emr T rends Cell Biol. 1998 8 282. 43 R. G. W. Anderson M. S. Brown and J. L. Goldstein Cell 1977 10 351. 44 J. Lazarovits and M. G. Roth Cell 1988 53 743. 45 D. E. Zwart C. B. Brewer J. Lazarovits Y. I. Henis and M. G. Roth J. Biol. Chem. 1996 271 907. 46 E. Fire O. Gutman M. G. Roth and Y. I. Henis J. Biol. Chem. 1995 270 21075. 47 C. M. Brown J. Cell Sci. 1998 111 271. 48 B. J. Rasmusson T. D. Flanagan S. J. Turco R. M. Epand and N. O. Petersen Biochim. Biophys. Acta 1998 14357 1. Paper 8/06677I 305 Faraday Discuss. 1998 111 289»305

ISSN:1359-6640    

DOI:10.1039/a806677i

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Direct measurement of recognition forces between proteins and membrane receptors Paul F. Luckham* and Kate Smith Department of Chemical Engineering Imperial College of Science T echnology and Medicine Prince Consort Road L ondon UK SW 7 2BY . E-mail p.luckham01=ic.ac.uk Receiøed 9th September 1998 The interactions between the protein cholera toxin B subunit attached to an atomic force microscope AFM cantilever CTB and its receptor the ganglioside GM have been 1 measured in a dilute electrolyte solution pH 5.5. Although there is variation in the force separation data obtained particularly on approach of the AFM tip to the GM surface 1 where usually but not always an attraction is noted an adhesion is always noted on separation of the surfaces. The strength of this adhesion varies from experiment to experiment but appears to be quantised at a value of around 90 pN.Addition of cholera toxin to the aqueous electrolyte solution completely removes the attractive interaction and adhesion. This gives us con–dence that in the earlier experiments a speci–c interaction between the CTB and GM was measured. 1 Introduction Proteins are large structurally diverse and complex molecules. Their interactions are governed by the subtle interplay of a variety of forces. Successful exploitation of the properties of proteins important to –elds as diverse as therapeutic drug development and washing powder technology depends on a full knowledge of their behaviour. Many diÜerent methods have been employed in an attempt to measure the interactions between proteins directly.However only one of these has come close to recording the interaction between two discrete molecules. This technique has ultimately evolved from the scanning tunnelling microscope (STM).1 Binnig et al. published their –rst STM data in 1982. The technique was purportedly capable of producing constant charge images of electronically conducting sample surfaces on an atomic scale. The work was ground-breaking and earned them the Nobel Prize for Physics in 1986. However Binnig et al. were not satis–ed with the limitations of the STM imposed by its inability to image electrically insulating samples such as biological materials. In fact in the same year that they were awarded the Nobel Prize Binnig et al. published the sequel to the STM; the atomic force microscope (AFM).2 This improved version utilised a diÜerent tip»sample interaction from the STM and produced constant tip»sample force images of the sample surface.The change in the nature of the tip opened up two main avenues of research ; a star burst of related force microscopes all relying on a diÜerent type of tip»sample interaction and adaptation of the function of the AFM to form a sensitive force sensor. This modi–cation allowed the measurement of the interaction on the nanometre and sub-nanoNewton scale between any two materials that could be made to form a tip and a sample. Thus the modi–ed AFM was born. In this study we have used a modi–ed AFM to measure the interaction between two molecules of biological importance.The molecules chosen were the ìcholera toxin B subunitœ (CTB) which is 307 Faraday Discuss. 1998 111 307»320 responsible for infection of the human body by the disease cholera and ìganglioside GM1œ (GM which is known to be the natural receptor on human intestinal cells. 1) Cholera is a pandemic and epidemic disease that has afflicted the human population for over 2 millennia.3 The disease can strike without warning and may kill within 6 h although it requires a prior incubation period of between 12 and 28 h. The –rst symptom is abrupt painless and copious ìrice-waterœ diarrhoea ; up to 20 l in 24 h.4 This debilitating diarrhoea is later joined by vomiting causing severe dehydration. Further symptoms are withered skin falling blood pressure and severe muscular cramps.The corpses of cholera victims have a characteristic blue»black desiccated appearance.5 It is now known that V . cholerae adhere to and colonise the small intestine,3,6 where they secrete an exotoxin known as cholera toxin. The toxin consists of two polypeptide subunits denoted cholera toxin A subunit (CTA) and cholera toxin B subunit (CTB). The B subunit binds to receptors on the epithelial cell membrane of the small intestine particularly the jejunum and duodenum. CTA does not bind to and on its own is therefore not toxic to intact cells. It is transferred across the cell membrane where it triggers a biological cascade that ultimately results in cell death. Once inside the cell however the CTA can not infect further cells and is therefore lost when the cell dies.Thus cholera is self-limiting continuing only as long as the V . cholerae bacteria remain in the gut secreting replacement toxin.5 The accepted structure of the B subunit of cholera toxin (CTB) also known as ìcholeragenoidœ is pictured in Fig. 1.7,8 CTB is a homopentameric protein i.e. it consists of –ve identical monomers. It assumes the appearance of a doughnut-shaped ring with a central pore formed by the long a-helices on each monomer. Viewed from the side see Fig. 1 CTB has a smooth upper surface and a rough lower surface. The roughness of the lower face is the result of the extension of the long central helices beyond the monomer b-sheets. Synthetic peptides with the same structure as this rough region elicit production of protective antibodies in vivo.This ability implies that it is the rough surface that is responsible for binding target cells. CTB stands between 3 and 4 nm high and has a quoted diameter ranging from 5.2 to 7.0 nm. The width of the central pore ranges from 1.1 nm at its smooth surface to 1.5 nm at the rougher side and the length of the pore has been quoted as 3.0 nm.7,9,10 The inner surface of the pore is highly hydrophilic possessing no less than 25 positive and 15 negative charges. The pentavalent nature of CTB is achieved by the non-covalent interlinking of b-sheets on neighbouring monomers. As mentioned previously it is CTB that is responsible for the binding of cholera toxin to membranes targeting those of the mucosal cells of the small intestine. Alone CTB is biologically inactive and will not induce symptoms associated with cholera.CTB then must act as a key to unlock the target cell for the toxin and eÜect entry of the active CTA. However the exact mechanism by which CTA enters the cell is still a mystery. The gangliosides are a structurally diverse group of lipids found in the outer cell membrane of almost all vertebrate cells and tissues. They are a minor component in most cells,11 except for Fig. 1 Cholera toxin B subunit. Faraday Discuss. 1998 111 307»320 308 Fig. 2 Structure of ganglioside GM1. those found in nervous tissue where they are more common. Although they have been implicated in various cell recognition and signalling phenomena their exact role in recognition remains a mystery. GM is one of a family of sialic acid-containing glycosphingolipids.Its structure is 1 depicted in Fig. 2. It is a ceramide-based glycolipid that possesses N-acetylneuraminic acid (NANA) in common with most gangliosides.12 A ceramide is a sphingosine base that is amidelinked to a fatty acid as indicated in Fig. 2. GM is large in size relative to most other membrane 1 lipids. It has a negatively charged head-group some 3.0 nm long which extends out from the membrane surface by 2.5 nm. Experimental The GM1 dipalmatoylphosphatidylcholine DPPC and the cholera B subunit were obtain from Sigma chemicals all other chemicals were obtained from Aldrich Chemicals. Having decided to study these two materials the next task is to –nd a way in which they may be attached to surfaces in the correct orientation for a recognition interaction to occur.There are two criteria to be met; the surface»tip connection must be strong and the molecule must be oriented correctly. Careful inspection of the cholera toxin-B subunitœs structure reveals that there is a preponder- A method is therefore ance of primary amine groups on the side that is not speci–c to GM1. required that will link a silicon substrate speci–cally to an amine group. There are many diÜerent ways by which proteins can be immobilised to a surface one which meets the criteria was reviewed in 1994 by Williams and Blanch.13 The method used see Fig. 3 employs gluteraldehyde as a covalent cross-linker between the primary amine on a surface and a primary amine on the protein. This method was used to attach the CTB subunit to the silicon tip of the AFM cantilever used in these experiments.Since the B-subunit molecules form a necessarily dense coverage on the cantilever tip the sample coverage must be sparse in order to increase the likelihood of the detection of a discrete B-subunit»GM interaction. A method that is perfect for the control of the spatial concentration 1 of a molecule is Langmuir deposition.14 309 Faraday Discuss. 1998 111 307»320 Fig. 3 Protocol adopted to attach the cholera toxin to a silicon atomic force microscope tip. The GM molecule has a hydrophilic head group and a hydrophobic tail. When embedded in a 1 cell membrane in vivo the tail group anchors the molecule in the lipid bilayer coating of the cell and it is the head group that interacts with extracellular moieties such as the B-subunit of cholera toxin.The sample surface must therefore present the GM head groups to the tip. This means that the 1 sample substrate must be hydrophobic. Since glass is hydrophilic some coating must be used to make the glass sample surface hydrophobic. Such a coating may be provided by silanisation. Faraday Discuss. 1998 111 307»320 310 Fig. 4 Schematic representation of the way the mixed monolayer of GM and DPPC molecules may appear 1 on a hydrophobic glass surface. As mentioned previously the B-subunit-coated tip has a high density coverage of molecules. Thus the sample surface must have a low density coverage of GM molecules. To ensure an even 1 distribution of molecules on the substrate however the dipped monolayer must also be densely packed.This con—ict may be overcome by diluting the GM molecules in the monolayer with 1 other noninteractive molecules. Such a molecule is dipalmitoylphosphatidylcholine DPPC. Its hydrocarbon chains are very similar in length to those of GM1 but its head group is shorter and narrower. When a monolayer of a mixture of the two molecules is deposited onto a substrate such that the head groups of the molecules are uppermost the GM molecules tower above the DPPC mol- 1 ecules. This situation is represented schematically in Fig. 4. Fig. 5 Schematic representation of the way the modi–ed AFM tip and the lipid monolayer may appear close to contact of the AFM tip to the lipid surface. This diagram is approximately to scale.311 Faraday Discuss. 1998 111 307»320 1 Fig. 6 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and lipid surface of GM and DPPC (molar ratio 1 4) in water. The open symbols correspond to approach of the surfaces and closed symbols to separation. 1 Fig. 7 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and lipid surface of GM and DPPC (molar ratio 1 4) in water. The open symbols correspond to approach of the surfaces and closed symbols to separation. 312 Faraday Discuss. 1998 111 307»320 1 Fig. 8 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and lipid surface of GM and DPPC (molar ratio 1 4) in water. The open symbols correspond to approach of the surfaces and closed symbols to separation.The GM molecules were therefore diluted to one –fth of the total number of molecules of GM 1 1 and DPPC in the spreading solution a ratio of 1 4 GM1 DPPC. Isotherms of mixtures of the two species have been studied extensively and have been found to exhibit features intermediate to those of pure GM and pure DPPC. This indicates that the two molecules form a miscible mono- 1 layer.15 The solvent for the spreading solution was itself a mixture of 1 4 methanol chloroform as GM and DPPC are not very soluble in chloroform alone. The substrate used was a glass micro- 1 scope coverslip small enough to be submerged fully into the trough. The substrate was dipped at a pressure of 30 N m~1 corresponding to an area per molecule of 0.40 nm2.15 Once the substrate had been dipped the surface of the trough was sucked clean and the AFM —uid cell was introduced into the trough.The substrate and Te—on holder were then transferred under water to the —uid cell of the AFM. Thus using these methods it was possible to orient both the CTB and the GM in a manner similar to that which occurs in vivo a schematic of the way these macro- 1 molecules are oriented drawn to scale is shown in Fig. 5. The modi–ed AFM used for these measurements was a purpose built force sensing piece of equipment similar to that described by Braithwaite et al. ;16 it is based around the principles of most commercial AFM equipment. Results and discussion The interaction between CTB molecules attached to a silicon tip by gluteraldehyde cross-linkers and GM molecules diluted by DPPC molecules and deposited onto a silanized glass substrate 1 was measured repeatedly at various sample sites and at various sample speeds.In each experiment over 100 compression and decompression pro–les for the interactions between the modi–ed AFM tip and the ganglioside GM have been taken. The rate at which the 1 pro–les were obtained varied between 2 and 0.015 Hz. As will be seen when we examine the data there is some diÜerence between the compression/decompression force pro–les however there was no observable diÜerence between the rates at which the compressions were obtained. Basically three diÜerent forms of compression pro–les could be obtained. In Fig. 6 for example we may see that there is no interaction between the modi–ed tip and the GM surface until the tip is some 10 1 nm from the ganglioside surface whereupon an attractive interaction is observed.On separation an adhesion is noted but more of this later. In Fig. 7 we note that there is a repulsion between the two surfaces commencing as a surface separation of around 30 nm but that at shorter separations 313 Faraday Discuss. 1998 111 307»320 an attraction is again noted. Once more an adhesion is noted on separation. In Fig. 8 there is no observable attractive interaction on approach of the surfaces until the surfaces come close to an intimate contact whereupon a strong hard wall repulsion is observed and again an attraction is noted on separation. In Fig. 9 four compression/decompression pro–les at the same site are plotted where we can see the variation that is observed in the data more clearly.There are some features that are common to all the data sets shown here and that have been obtained through the course of these experiments. Firstly on separation an attraction is always noted this adhesion may vary between 0.15 and 0.9 nN. Also a hard wall type repulsive interaction at short separations less than 2 nm is always observed. In roughly 3/4 of the data sets an attractive interaction on approach of the surfaces was observed. It would appear likely that this attractive interaction is due to the interaction between the cholera toxin B subunit and the GM1 surface and that the adhesion observed is due to the breaking of this interaction.Let us then consider the origin of the variation in the data observed in these experiments. The repulsion on approach of the cholera toxin coated AFM tip toward the ganglioside surface is likely to be steric in origin and be due to the cholera toxin B subunit molecules on the AFM tip being out of alignment with the GM molecules on the sample surface. Since the CTB molecules 1 are eÜectively tethered onto the AFM tip by molecular ììropesœœ they will possess a certain degree Fig. 9 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and lipid surface of GM and DPPC (molar ratio 1 4) in water. The open symbols correspond to approach of the surfaces and closed symbols to separation. Here four repeat experiments at the same contact position on the surface are 1 shown.Faraday Discuss. 1998 111 307»320 314 of rotational freedom. Clearly the extent of this interaction will be diÜerent depending on the orientation of the molecule. Three mechanisms are proposed here for the AFM tip interaction with the GM1 surface and these are illustrated in Fig. 10. Each interaction in Fig. 10 depends on the initial orientation of the CTB molecule. An initial side-on approach by the CTB molecule will result in a steric repulsion. If the CTB molecule is made to rotate by this repulsive force then the CTB and GM molecules will be favourably 1 oriented and an attraction may occur. Otherwise the entire approach cycle will appear to be repulsive. Alternatively if the CTB molecule happens to be oriented correctly on approach of the sample surface then no initial repulsion will be experienced and only an attraction will be evident.It is important to remember that the force»separation curves actually measured will not be as simple as those described in Fig. 10. In reality the tip surface will be covered with CTB molecules Fig. 10 Schematic representation of the diÜerent ways the CTB subunit and the GM ganglioside can interact together with the predicted force curve for the approach of the surfaces. Note that in every case following 1 full compression an adhesion would be expected on separation. 315 Faraday Discuss. 1998 111 307»320 1 and the resultant approach portion of the force»separation curve will most likely be a random convolution of more than one of the CTB»GM interactions in Fig.10. Let us now look in more detail at the force pro–les for separation of the AFM cantilever from the AFM tip. In most instances a clean separation such as that observed in Fig. 6 is observed. However on occasions a more ripping like separation is noted see for example Fig. 7 and 9 graph 1. It is likely that in these cases the cantilever has initially separated from some of the speci–cally interacting sites but not all of them and further separation of the AFM tip is required to fully separate the surfaces. The likelihood of achieving the measurement of a single CTB»GM interaction depends on the 1 distribution of the GM molecules on the sample surface. This may be calculated approximately 1 from the isotherm of the DPPC»GM mixture used.The dipping pressure used (25 mN mv1) 1 corresponds to an average area per molecule of 0.40 nm2. Since the head groups of the GM1 molecules tower above the DPPC molecules eÜectively dominating the sample surface the area occupied by the DPPC molecules may be added to the area available to the GM head groups. 1 The ratio of GM to DPPC molecules is 4 1. Thus the average area available to each GM 1 1 molecule is –ve times the average area per molecule i.e. 2 nm2. The approximate area occupied by a single CTB molecule is equal to the square of its radius i.e. 49 nm2. Assuming that each subunit of the CTB pentamer occupies a sixth of the total area allowing space for the central barrel the area commanded by each monomer in the CTB pentamer is around 8 nm2. Thus the density of the sample surface coverage by GM molecules is such 1 that not only is the tip»sample interaction likely to correspond at least to the interaction of an entire CTB pentamer it is also very unlikely to correspond to the interaction of a single CTB molecule.Since it is not possible to view the tip»sample interaction and measure it simultaneously there is therefore no direct way of determining the number of interacting CTB»GM pairs. This 1 problem may be overcome by repeating the measurement of the interaction many times. Collation of the interaction data should reveal a frequency distribution in which the interactions may occur in discrete groups. The spacing between these groups should be equal to the magnitude of the interaction of a single CTB»GM pair. Fig.11 reveals that the adhesion values occur in multiples 1 of approximately 0.09 nN as marked. 1 If the interaction measured is indeed that between CTB and GM molecules as postulated 1 above then it should be possible to block the interaction by adding free CTB molecules in solution to the sample chamber.17,18 The free CTB molecules should cover the GM -covered surface 1 completely leaving the CTB molecules on the tip to interact with the backside of a CTB monolayer. Thus the initially attractive interaction of the CTB»GM system should be replaced by a 1 repulsive CTB»CTB interaction. Such a change in the measured interaction may be taken to imply that the initial CTB»GM interaction has been blocked. Free CTB molecules were injected into the sample chamber and the same tip»sample measurement was repeated.If the interaction measured above is indeed that between CTB and GM1 Fig. 11 Adhesion for the interaction between CTB and GM1 plotted as function of the frequency of occurrence. Note that the adhesion seems to be roughly quantised in values of 0.09 nN. Faraday Discuss. 1998 111 307»320 316 Fig. 12 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and a lipid surface of GM and DPPC (molar ratio 1 4) in water. In this experiment a dilute solution of free CTB was added to the solution. Two force pro–les are shown. The open symbols correspond to approach of the surfaces 1 and closed symbols to separation. Fig. 13 Schematic representation of the way CTB in solution can block the interaction between the CTB coated AFM tip and the ganglioside GM1.317 Faraday Discuss. 1998 111 307»320 molecules then the subsequent tip»sample interactions should display no attractive interaction. Fig. 12 contains the force»separation curves of just such a situation. As can be seen the attraction is replaced by pure repulsion. This measurement was repeated 47 times. Not one of these measurements produced attractive force»separation curves. This implies that the free CTB is adsorbing on the GM surface and blocks any attractive interaction between the CTB molecules attached to the 1 AFM tip and the GM surface as shown in Fig. 13. It is therefore contended that the earlier data 1 correspond to the interaction between CTB and GM molecules. 1 To further validate the above conclusion as control experiments the interaction of a bare silicon tip with a GM -covered surface and between a CTB-covered tip and a silanized glass 1 surface were performed.The results are shown in Fig. 14 and 15 respectively. Fig. 14 contains two typical data sets obtained from the interaction of a bare silicon tip with a GM -covered surface. In 1 each case the interaction is purely repulsive. Thus the eÜect of incomplete coverage of the tip with CTB molecules may cause excessive measured repulsion but cannot account for the attractive interactions measured in the CTB»GM system. Fig. 15 displays data typical of the interaction 1 between a CTB-covered tip and a silanised glass surface. Some small attraction is observed on approach of the tip and sample and a large adhesion is seen on retraction of 320 pN.This adhesion is much stronger than any observed in the experiments with GM and CTB. 1 Fig. 14 Forces of interaction between a bare AFM tip and a lipid surface of GM and DPPC (molar ratio 1 4) in water. The open symbols correspond to approach of the surfaces and closed symbols to separation. 1 Faraday Discuss. 1998 111 307»320 318 Fig. 15 Forces of interaction between an AFM tip bearing attached cholera toxin subunit B and a silanised glass in water. The open symbols correspond to approach of the surfaces and closed symbols to separation. It is interesting to compare the strength of the interaction between GM1 and CTB of around 90 pN with intermolecular adhesions measured for other speci–c interactions. These are summarised in Table 1.It must also be mentioned though that it is possible that the value of 90 pN may correspond to the interaction between GM and the whole pentamer of the cholera toxin. We 1 cannot be unambiguous with the current data. Table 1 Detachment forces for various intermolecular adhesions Interaction Reference 19 20 21 Avidin»biotin Avidin»iminobiotin Streptavidin»biotin Avidin»desthiobiotin Streptavidin»iminobiotin Cell adhesion proteoglycans Adenine»thymine Faraday Discuss. 1998 111 307»320 Detachment force/pN 160^20 85^15 257^25 94^10 135^15 40^15 54 319 Paper 8/07048B References 1 Ch. Binnig G. Roher C. Gerber and E. Weibel Phys. Rev. L ett. 1982 49 57. 2 G. Binnig C. F. Quate and C. Gerber Phys.Rev. L ett. 1986 56 930. 3 P. Shears Ann. T rop. Med. Parasitol. 1994 88(2) 109. 4 J. Holmgren Nature (L ondon) 1981 292 413. 5 B. D. Spangler Microbiol. Rev. 1992 56(4) 622. 6 W. Curatolo Biochim. Biophys. Acta 1987 906 137. 7 T. K. Sixma S. E. Pronk K. H. Kalk E. S. Wartna B. A. M. van Zanten B. Witholt and W. G. J. Hol Nature (L ondon) 1991 351 371. 8 W. I. Lencer C. Constable S. Moe P. A. Rufo A. Wolf M. G. Jobling S. P. Ruston J. L. Madara R. K. Holmes and T. R. Hirst J. Biol. Chem. 1997 272(24) 15562. 9 J. Yang L. K. Tamm T. W. Tillack and Z. Shao J. Mol. Biol. 1993 229 286. 10 J. Mou J. Yang and Z. Shao J. Mol. Biol. 1995 248 507. 11 P. H. Fishman T. Pacuszka and P. A. Orlandi Adv. L ipid Res. 1993 25 165. 12 W. Curatolo Biochim. Biophys. Acta 1987 906 111. 13 R. A. Williams and H. W. Blanch Biosensors Bioelectronics 1994 9 159. 14 J. Marra J. Colloid Interface Sci. 1985 107(2) 446. 15 P. F. Luckham J. Wood S. Froggatt and R. Swart J. Colloid Interface Sci. 1993 156 164. 16 G. J. C. Braithwaite P. F. Luckham and A. M. Howe L angmuir 1996 12 4224. 17 V. T. Moy E-L. Florin and E. Gaub Colloid Surf. A 1994 93 343. 18 A. Chilkoti T. Boland B. D. Ratner and P. S. Stayton Biophys. J. 1995 69 2125. 19 V. T. Moy E-L. Florin and H. E. Gaub Science 1994 266 257. 20 U. Dammer O. Popescu P. Wagner D. Anselmetti H-J. Gué ntherodt and G. N. Misevic Science 1995 267 1173. 21 T. Boland and B. D. Ratner Proc. Natl. Acad. Sci. USA 1995 92 5297. Faraday Discuss. 1998 111 307»320 320

ISSN:1359-6640    

DOI:10.1039/a807048b

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Use of a laminar ¢©ow chamber to study the rate of bond formation and dissociation between surface-bound adhesion molecules E¢­ ect of applied force and distance between surfaces Anne Pierres Anne-Marie Benoliel and Pierre Bongrand L aboratoire d©«Immunologie INSERM U 387 Ho¢� pital de Sainte-Marguerite BP 29 13274 Marseille Cedex 09 France. E-mail bongrand=inserm.fr Received 11th August 1998 It has recently been shown that much information on the behaviour of surface-bound adhesion molecules could be obtained by monitoring the motion of receptor-coated particles along ligand-derivatized surfaces in the presence of a hydrodynamic force of a few pN. This procedure is expected to allow direct monitoring of the formation and dissociation of individual bonds.We present experimental results on the interaction between streptavidin-coated spheres (1.4 lm diameter) and control or biotinylated mica surfaces in a laminar ¡ªow chamber. Moving spheres are found to display numerous arrests whose frequency is markedly increased (5¡í13-fold) in the presence of biotin groups. For a given shear rate the binding frequency is strongly dependent on the sphere¡ísurface separation. Indeed this frequency displayed a 14-fold decrease when the velocity increased from 7 to 15 lm s~1 for a wall shear rate of 20 s~1. Furthermore the lifetime of observed arrests was of the order of several seconds i.e. 5¡í50-fold higher than previously determined on models such as selectin¡íligand CD2¡íCD48 or cadherin¡ícadherin. Finally this lifetime did not decrease when the wall shear rate was increased from ca.10 to 40 s~1. Introduction Cell adhesion is a process of prominent biological importance. It is usually mediated by the interaction of dedicated receptor and ligand molecules bound to interacting surfaces. It is therefore of obvious interest to develop suitable methods for predicting whether a cell-to-cell encounter will result in durable adhesion. The initial steps of cell attachment may be viewed as follows :1 ¡©rst as a consequence of active or passive deformation and/or displacement two biological surfaces are brought into binding distance. They are separated by a gap of ¡ªuctuating thickness. Second when a ¡©rst bond is formed the distance between surfaces is likely to be decreased and maintained constant as a consequence of the balance between bond properties and repulsive forces.2 Third the lifetime of the ¡©rst bond subjected to a variety of disruptive forces is therefore a critical parameter.Indeed if this bond is maintained until a second bond is formed it is likely that the number of bonds will be progressively increased. Alternatively if the ¡©rst bond is ruptured before it is reinforced by additional interactions membranes will become separated. In conclusion the biologically relevant properties of a pair of membrane-bound ligand¡íreceptor molecules are3 the rate of bond formation as a function of the distance between the anchoring points of these molecules and the rate of bond rupture as a function of the intensity of disruptive force.321 Faraday Discuss. 1998 111 321¡í330 While an impressive amount of information has been gathered on the rate of bond formation and dissociation between soluble reagents little attention has been given to the properties of surface-bound molecules until Bell4 elaborated a theoretical framework to relate these properties to the behaviour of soluble forms of these molecules. During the following years Bellœs ideas were subjected to direct experimental tests by several authors who studied the separation of surfaces bound by a few or even a single molecular bonds. Experimental methods included the use of hydrodynamic —ow,5 soft vesicle micromanipulation,6 atomic force microscopy7 or optical tweezers8 (see ref. 9 for a review). These experiments led to results of immediate biological signi–- cance.Thus it was shown that the remarkable capacity of —owing white blood cells to roll on the surface of blood vessels was mediated by a speci–c interaction between selectin molecules and their receptors,10,11 and this interaction was characterized by a short lifetime12,13 and high tensile strength.13 While the size and complexity of cell structure sets a limit to the accuracy of information that can be extracted from cellular models it was recently found that much detailed information on the kinetics of bond dissociation14,15 and formation16,17 could be obtained by studying the motion of receptor-coated spheres along ligand-derivatized surfaces in a laminar —ow chamber under very low hydrodynamic —ow. The aim of the present paper is to illustrate the potential of this methodology by presenting preliminary experimental data on the interaction between streptavidin-coated spheres and biotinylated surfaces.The interest of this model is that it was extensively studied with powerful experimental methods such as the surface forces apparatus,18 atomic force microscopy, 7,19 soft vesicle techniques,20 as well as by theoretical approaches based on computer simulation. 21,22 The potential and limits of the —ow chamber technology are also discussed on the basis of these and previous results. Materials and methods Molecules and surfaces Streptavidin-coated beads of 2.8 lm diameter and 1.300 kg m~3 density (Dynabeads M280) were supplied by Dynal France (Compie` gne). Freshly cleaved mica surfaces (Muskovite mica Meta–x Montdidier France) were sequentially incubated with 1mM NiCl2 ,23 then biotinyl-(gly)12(His)6- NH (supplied by Neosystem Strasbourg France).The strong interaction between hexahistidine 2 groups and nickel-treated mica surfaces was expected to allow the formation of a smooth surface of biotin groups. Using —uorescence determination,14 the surface density of streptavidin groups on spheres was estimated at ca. 3460 molecules lm~2. Further mica surfaces treated with 10 lg ml~1 biotinylated peptide were found to bind 15 000 streptavidin molecules lm~2. Flow chamber Our apparatus was described in a previous paper.17 Brie—y the chamber was obtained by mechanical –xation of biotinylated mica surfaces against a drilled plexiglas block bearing a cavity of 0.1]6]20 mm3.The —ow was generated by a syringe mounted on an electric syringe holder. The chamber was deposited on the stage of an inverted microscope (Olympus IX with a long distance 40]objective) bearing a CCD camera connected to a videotimer and a tape recorder for delayed analysis. In a typical experiment the bead suspension (3]106 ml~1 in pH 7.2 phosphate buÜer supplemented with 1 mg ml~1 bovine albumin) was driven through the chamber with a wall shear rate ranging between ca. 10 and 40 s~1. Particle tracking The basis of our method has been described previously.15 Brie—y an image processing system allowed real-time determination of the position of the centroid of the images of —owing beads with ca. 0.025 lm spatial accuracy while the pixel size was 0.23 lm.Since the odd and even frames of each interlaced video image were analysed separately the temporal resolution was twice the video rate sequential positions were thus separated by a time interval of 20 ms. In a typical experiment ca. 10 000 positions were stored by tracking 50»100 individual beads. Data were processed with dedicated software written in our laboratory. Faraday Discuss. 1998 111 321»330 322 Principle of data analysis Since the details of our procedure have been described in previous papers,14,17 we shall only outline the general strategy. We shall describe sequentially the study of bond formation and bond rupture. Bond formation. The frequency of particle arrests was obtained by monitoring the motion of a series of particles and counting the binding events.The arrest frequency was simply equal to the number of counted arrests dived by the observation time. However a number of steps were required to extract the on-rate of bond formation First although particles were spheres of similar radius their velocity displayed notable heterogeneity. This was ascribed to variations of sphere»surface distance due to e.g. incomplete sedimentation. Indeed basic results of —uid mechanics24 show that the velocity of a free sphere moving in a laminar shear —ow near a wall should vanish very slowly when the thickness of the gap between the sphere and the surface decreases. Thus the average velocity of monitored spheres was determined on the preceding 160 ms interval after each step of 20 ms.The data recorded on a typical population of ca. 100 particles (yielding several tens of thousands of positions) were used to build the velocity frequency histograms of (i) all moving particles and (ii) all steps immediately followed by an arrest. The arrest frequency could thus be determined as a function of particle velocity by calculating the ratio of event numbers in both histograms for each velocity class. Second it would be tempting to use the known relationship between sphere velocity and distance to the surface24 to obtain a relationship between particle arrest frequency and distance,16 then to derive the rate of molecular association as a function of intermolecular distance d [i.e. k`(d)] by determining the average density of binding sites on interacting surfaces.16 However the validity of this procedure is hampered by the occurrence of Brownian motion that is expected to result in marked variations of sphere»surface distance during a 20 ms interval.This motion is considerably complicated by (i) perturbation of the sphere motion by the eÜect of the wall on hydrodynamic drag and (ii) the possibility of long-range electrodynamic (van der Waals) forces between spheres and surfaces. This difficulty was overcome as follows :17 (i) in each series of experiments the acceleration of monitored particles was determined on periods of 2]160 ms and the average acceleration was plotted versus velocity. Experimental curves were compared with theoretical curves obtained by computer simulation which allowed accurate derivation of the wall shear rate G and the Hamaker constant A25 for the interaction between spheres and the chamber —oor.(ii) The rate of bond formation was approximated as a step function k (1) 0 `(d)\k`0 (dOR) (d[R) where R is the interaction range. Binding curves were obtained by computer simulation using the following formula:17 (2) 1 F(d)\2n2k`0 ap1p2R3[2/3[d/R](d/R)3/3] where F(d) is the frequency of attachment between a sphere of radius a and a planar surface at distance d and p and p are the surface densities of binding molecules on the sphere and surface. 2 Bond rupture. The basic assumption12 is that when receptor-coated beads are driven along ligand-derivatized surfaces with a force of the order of 1 pN single ligand»receptor interactions should be able to generate detectable particle arrests.Further in the absence of additional interaction the probability P(t) that a particle arrested at time zero remains bound at time t should be:26 (3) P(t)\exp([k~t) where k~ is the rate of bond dissociation (oÜ-rate). Thus k~ was determined by measuring the duration of a sufficient number of arrests (typically 100 events) and plotting the number of particles remaining bound at time t after arrest versus t on a semi-logarithmic scale. The initial slope 323 Faraday Discuss. 1998 111 321»330 of the curve was used as an estimate of the oÜ-rate. The experimental curve may not be a straight line if particle attachment is strengthened after arrest due to the formation of additional bonds or non-speci–c interactions.12,26,27 Further the dependence of k~ on applied force may be studied by varying the wall shear rate.13h15 The validity of these basic assumptions will be discussed below.Results Flowing particles display numerous arrests of varying duration First we studied the motion of streptavidin-coated spheres along mica surfaces that had been treated with various amounts of biotinylated peptide. As exempli–ed in Fig. 1 the beads displayed a number of arrests of widely varying duration. Whereas most of these binding events lasted more than 1 s arrests shorter than 0.1 s were occasionally detected (Fig. 2). In order to allow computerassisted processing of binding events a particle was de–ned as arrested at some point when it moved by less than 0.46 lm (i.e.two pixel units) during a time interval of 0.32 s. Visual examination of a number of trajectories showed that only a few very transient arrests were missed and a minimal amount of false arrests were found. Surface biotinylation markedly increases interaction with streptavidin-coated beads In a –rst series of experiments we compared the motion of spheres along surfaces treated with 100 10 1 or 0 lg peptide in the presence of a wall shear rate of ca. 20 s~1. In all cases short or durable binding events were detected. However several –ndings suggested that surface biotinylation substantially increased sphere»surface interaction (i) When streptavidin-coated spheres were driven along heavily biotinylated surfaces only a very low proportion of these particles reached the microscope –eld near the centre of the chamber.(ii) As shown in Fig. 3 the average velocity of particles monitored on biotinylated surfaces was markedly higher than on the controls. Since particle velocity is known to be a strongly increasing function of distance from the surface,24 this observation suggested that particles adhered to biotinylated surface as soon as they completed sedimentation and observed particles were mostly incompletely sedimented spheres in contrast to the controls. Fig. 1 A typical trajectory. The motion of a streptavidin-coated bead along a surface treated with 0.1 lg biotinylated peptide is shown. The wall shear rate is 17.2 s~1. The position of the centre of gravity of the particle image was determined every 20 ms.A very short (arrow) and a longer (double arrow) arrest are shown. 324 Faraday Discuss. 1998 111 321»330 Fig. 2 Short arrests. An enlarged fraction of the trajectory shown in Fig. 1 is displayed. The short arrest shown in Fig. 1 is clearly visible (arrow). An artefactual discontinuity of the particle velocity is also shown (double arrow). Indeed in the latter case there is no displacement of the tangent to the trajectory. (iii) Thus in order to achieve a meaningful study of the relationship between surface biotinylation and binding efficiency we calculated in each series of experiments the number of 20 ms intervals where the particle velocity was less than the product of the particle radius and the wall shear rate.The fraction of these intervals immediately followed by an arrest was then calculated this fraction was respectively 0.0049 0.0039 0.0104 and 0.00077 when the mica surfaces were treated with 100 10 1 and 0 lg of biotinylated peptide. Thus surface biotinylation increased the particle arrest by a factor ranging between 5 and 13. Fig. 3 Distribution of particle velocity. In two representative experiments the motion of spheres on control (]) or biotinylated (=) surfaces was studied. The particle velocity U was determined every 20 ms and the frequency distribution U/aG is shown (as % of total number of intervals). Clearly particles —owing on the biotinylated surfaces display higher U/aG (i.e. higher separation from the surface) than the controls. 325 Faraday Discuss.1998 111 321»330 Fig. 4 Example of a clearcut arrest. A typical example of a clearcut arrest is shown. The particle velocity remains constant up to the binding event. Arrest frequency is a strongly decreasing function of particle velocity The particle velocity was determined for 160 ms periods preceding each arrest. As exempli–ed in Fig. 4 this was quite easy in many cases where the particle velocity was fairly constant immediately before arrest. However in some cases (Fig. 5) the particle velocity progressively decreased before the observed stop and it was felt that the de–nition of the velocity before arrest was somewhat arbitrary. The results obtained in four series of experiments (wall shear rate 20 s~1 10 lg peptide per mica slide) were then pooled in order to obtain the experimental dependence of arrest frequency on particle velocity.As shown in Fig. 6 the binding frequency decreased by 93% when the ratio U/aG increased from 0.25 to 0.55. This result demonstrates that binding frequency is indeed Fig. 5 Example of a progressive arrest. The motion of a particle binding to a mica surface is shown. The sphere displays progressive slowing making it difficult to de–ne unambiguously a ììvelocity before arrestœœ. Faraday Discuss. 1998 111 321»330 326 Fig. 6 Dependence of binding probability on particle velocity. In a typical series of experiments the arrest of streptavidin-coated spheres on surfaces treated with 10 lg biotinylated peptide was studied (=). The histogram of particle velocity was built and the arrest frequency was calculated for classes of increasing dimensionless velocity U/aG.The vertical bar length is a theoretical standard deviation calculated by assuming Poisson distribution. Theoretical curves corresponding to a bond range of 5 nm (K) and 40 nm (|) are also shown. strongly dependent on sphere»surface distance. Further due to the wide heterogeneity of particle velocities (Fig. 3) it does not seem warranted to de–ne a mean binding frequency by retaining particles whose velocity falls below some arbitrary threshold. Finally we attempted to –t experimental data to theoretical results that were previously shown to account for the homophilic adhesion between cadherin-coated surfaces.17 We used acceleration curves to estimate the Hamaker constant in ten separate series of experiments no signi–cant interaction between spheres and surfaces could be detected.As shown in Fig. 6 the slope of the theoretical curves (built with a zero Hamaker constant) was lower than the experimental one by a factor of two. Binding efficiency is decreased when the shear rate is increased The eÜect of the wall shear rate on the binding efficiency was studied. In three series of four separate experiments each the interaction between spheres and surfaces treated with 10 lg biotinylated peptide was studied with a wall shear rate of ca. 10 20 and 40 s~1. The arrest frequency was determined for particles with an instantaneous velocity U lower than the product between the wall shear rate (G) and particle velocity (a) this frequency was 0.23 0.15 and 0.10 s~1 respectively.The binding frequency was thus a slowly decreasing function of particle velocity. The lifetime of the interaction between streptavidin-coated beads and biotinylated surfaces is not markedly higher than that observed on controls The lifetime of sphere»surface associations was studied by recording the duration of arrests and building detachment curves as exempli–ed in Fig. 7. Assuming –rst-order detachment kinetics the rate of particle detachment was derived from the fraction of particles remaining bound for less than 2 s. In a preliminary series of experiments the in—uence of surface biotinylation was studied. The rate of bond dissociation was respectively 0.20 (40 arrests) 0.16 (39 arrests) 0.28 (56 arrests) and 0.40 s~1 (44 arrests) when the mica surfaces were treated with 10 1 0.1 and 0 lg of biotinylated peptide.327 Faraday Discuss. 1998 111 321»330 Fig. 7 In four series of experiments streptavidin-coated beads were driven along surfaces treated with 1 lg biotinylated peptide. The wall shear rate was ca. 20 s~1. A microscopical –eld was selected and the durations of observed binding events recorded. Results were used to plot the variations of the number of particles remaining bound at time t. The slope of the curve was de–ned as the detachment rate. There is no obvious increase in the detachment rate when the shear rate is increased The eÜect of the wall shear rate on the experimental detachment rate was also studied. When mica surfaces were treated with 10 lg biotinylated peptide and the wall shear rate was 10 20 and 40 s~1 the particle detachment rate was respectively 0.33 0.22 and 0.09 s~1.Discussion The aim of this work was to study the avidin»biotin interaction with the laminar —ow chamber. It seemed of interest to compare the data concerning bond rupture with published reports based on other experimental techniques. Also we wished to present new information on the rate of bond formation. A –rst conclusion is that the —ow chamber methodology may be more suitable to the study of weak and transient interactions such as CD2»CD48,15,16 selectin»ligand12,13 or cadherin» cadherin17 association than the stronger avidin»biotin attachment. Indeed in the present study measured detachment rates were 5»50-fold lower than measured on the aforementioned systems and presumably many spheres displayed irreversible attachment to the surface as soon as contact occurred.Further when binding site density was decreased a substantial proportion of arrests were mediated by poorly de–ned non-speci–c interactions which made it more difficult to relate the obtained results to the properties of avidin»biotin interaction. However it seems reasonable to assume that interactions between single streptavidin and biotin groups were actually observed since our methodology allows the monitoring of weak individual bonds between diÜerent adhesion receptors. Also since interacting molecules are substantially shorter in the present work than in the aforementioned study the occurrence of multiple binding events should be rarer than in those other cases.Another possible difficulty with our system may stem from the relative shortness of the adhesion molecules. Indeed in previous studies,15,17 adhesion sites were connected to streptavidin groups through a fairly —exible molecular link of ca. 40 nm length. In contrast in the present work the length of biotinylated peptides was less than 5 nm and no spacer was added to increase the distance between streptavidin groups and the particle surface. Faraday Discuss. 1998 111 321»330 328 Despite the aforementioned difficulties biotinylation of the surface of the —ow chamber markedly increased the efficiency of sphere to particle attachment strongly suggesting than avidin» biotin interactions indeed occurred.However a substantial fraction of observed attachments lasted only a few seconds while the force exerted on the bonds did not exceed a few pN.14 Several explanations may be suggested to account for this result. (i) Since particle»surface separation results from the rupture of the weakest link between these structures sphere detachment is perhaps due to the rupture of the interaction between biotinylated peptides and mica surfaces or streptavidin»sphere association. However this interpretation is not fully consistent with the observation that the particle detachment rate decreased when the hydrodynamic force was increased. Indeed there is little experimental evidence that a disruptive force can increase the lifetime of a ligand receptor interaction although in principle this might happen.28 (ii) Another possibility would be that avidin»biotin association might involve intermediate binding steps as previously reported in a study made on the antigen»antibody interaction with the —ow chamber.14 Indeed since as a rule of thumb the relative velocity between the —ow chamber and the sphere surfaces is about half the sphere velocity,24 if the total length of interacting molecules is of the order of 5 nm the time allowed for interaction is only 1 ms for a sphere of 10 lm s~1 velocity.Perhaps the paradoxical increase in bond lifetime observed when the wall shear rate was increased was due to the disappearance of putative intermediate binding steps that might become undetectable at higher shear rate. This possibility might be addressed by increasing the time resolution of our analysis and/or adding a spacer to increase the length of the interacting structures.A –nal point of concern is the discrepancy between the theoretical and experimental dependence of adhesion frequency on particle velocity whereas our theoretical framework successfully accounted for the interaction between cadherin-coated surfaces.17 Three points must be suggested to deal with this problem. First it would be useful to check the smoothness of the peptide-coated mica surfaces in order to exclude the possibility that we measured the frequency of surface defects rather than molecular associations. However if this were the case the binding frequency should increase when the shear rate was increased.Second as illustrated by Fig. 5 there is sometimes a problem with the determination of particle velocity immediately before arrests. Perhaps our analysis might be improved by increasing the shear rate and temporal resolution. Third the theoretical relationship between sphere velocity and distance to the surface24 might no longer hold at low distance. This difficulty might be alleviated if we increase the length of receptors bound to the interacting surfaces. In conclusion the laminar —ow chamber is a powerful means of studying the formation and dissociation of molecular bonds with high temporal and spatial resolution. This seems ideally suited to the study of transient interactions. More work is needed to obtain new and accurate information on high affinity ligand»receptor association.Acknowledgements The expert technical assistance of Ms Dominique Touchard is gratefully acknowledged. This work was supported by a grant from the A.R.C. References 1 A. Pierres A. M. Benoliel and P. Bongrand Curr. Opin. Colloid Interface Sci. 1998 3 525. 2 G. I. Bell M. Dembo and P. Bongrand Biophys. J. 1984 45 1045. 3 A. Pierres A. M. Benoliel and P. Bongrand J. Immunol. Methods 1996 196 105. 4 G. I. Bell Science 1978 200 618. 5 S. P. Tha J. Shuster and H. L. Goldsmith Biophys. J. 1986 50 1117. 6 E. Evans D. Berk and A. Leung Biophys. J. 1991 59 838. 7 E. L. Florin V. T. Moy and H. E. Gaub Science 1994 264 415. 8 H. Miyata R. Yasuda and K. Kinosita Jr. Biochim. Biophys. Acta 1996 1290 83. 9 A. Pierres A.M. Benoliel and P. Bongrand Cell Adhesion Commun. 1998 5 375. 10 M. B. Lawrence and T. A. Springer Cell 1991 65 859. 11 U. H. Von Andrian J. D. Chambers L. M. McEvoy R. F. Bargatze K. E. Arfors and E. C. Butcher Proc. Natl. Acad. Sci. USA 1991 88 7538. 329 Faraday Discuss. 1998 111 321»330 12 G. Kaplanski C. Farnarier O. Tissot A. Pierres A. M. Benoliel M. C. Alessi S. Kaplanski and P. Bongrand Biophys. J. 1993 64 1922. 13 R. Alon D. A. Hammer and T. A. Springer Nature (L ondon) 1995 374 539. 14 A. Pierres A. M. Benoliel and P. Bongrand J. Biol. Chem. 1995 270 26586. 15 A. Pierres A. M. Benoliel and P. Bongrand Proc. Natl Acad. Sci. USA 1996 93 15114. 16 A. Pierres A. M. Benoliel and P. Bongrand FEBS L ett. 1997 403 239. 17 A. Pierres H. Feracci V. Delmas A.M. Benoliel J. P. Thieç ry and P. Bongrand Proc. Natl. Acad. Sci. USA 1998 95 9256. 18 C. A. Helm W. Knoll and J. N. Israelachvili Proc. Natl Acad. USA 1991 88 8169. 19 G. U. Lee D. A. Kidwell and R. J. Colton L angmuir 1994 10 354. 20 E. Evans K. Ritchie and R. Merkel Biophys. J. 1995 68 2580. 21 H. Grubmué ller B. Heymann and P. Tavan Science 1996 271 997. 22 S. Izrailev S. Stepaniants M. Balsera Y. Oono and K. Schulten Biophys. J. 1997 72 1568. 23 C. R. Ill V. M. Keivens J. E. Hale K. K. Nakamura R. A. Jue S. Cheng E. D. Melcher B. Drake and M. C. Smith Biophys. J. 1993 64 919. 24 A. J. Goldman R. G. Cox and H. Brenner Chem. Eng. Sci. 1967 22 653. 25 J. N. Israelachvili Intermolecular and Surface Forces Academic Press New York 1991 p. 288. 26 A. Pierres O. Tissot and P. Bongrand in Studying Cell Adhesion ed. P. Bongrand P. Claesson and A. Curtis Springer Verlag Heidelberg 1994 p. 157. 27 A. Pierres A. M. Benoliel and P. Bongrand in Cell Mechanics and Cellular Engineering ed. V. C. Mow F. Guilak R. Tran-Son-Tay and R. M. Hochmuth Springer Verlag New York 1994 p. 145. 28 M. Dembo D. C. Torney K. Saxman and D. Hammer Proc. R. Soc. L ondon Ser. B 1988 234 55. Paper 8/06339G Faraday Discuss. 1998 111 321»330 330

ISSN:1359-6640    

DOI:10.1039/a806339g

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

General Discussion Dr Gon8 i opened the discussion of Mrs Woodhouseœs paper Could you tell us something about the stability of these tethered bilayers ? Ms Woodhouse responded Our tethered bilayers are remarkably stable compared with black lipid membranes and supported bilayers. We have demonstrated that tethered bilayers comprising hydrated di-phytanyl derivatives are stable for at least 12 weeks at room temperature and longer when frozen. This has been characterised by measuring the capacitance at 1 Hz. Additionally fully assembled biosensors have been dried using glycerol trehalose or dextran and have been demonstrated to respond to analyte following reconstitution. Dr Pawlak commented Impedance measurements on isolating layers (here the lipid membrane) are very critically dependent on even a small number of layer defects leading to unspeci–c leakage currents.The layer capacitance is hardly aÜected by these defects. How stable is your membrane preparation on the surface with time with potentials applied and with biochemicals exposed to the membrane? How long is the preparation time of the system until stable sensor signals can be measured? How reproducibly can the system be prepared (chip-to-chip variation impedance stability amount of leakage current) ? Ms Woodhouse responded The membrane stabilises very rapidly»less than one minute. The eÜect of potential on stability has not been quanti–ed as such although we have performed multiple experiments on single membranes over hour time frames without apparent degradation of response.More than 100 biochemicals and chemicals of many classes have been exposed to the membrane these range from blood and sera to organic solvents and detergents. In many cases the eÜect is concentration dependent. For example the membrane preparation is stable in 100% v/v serum and blood the eÜect of detergent depends on the protein/buÜer matrix used. The bilayer self-assembles very rapidly onto the monolayer. A stable signal can be measured immediately. The variation in leakage current between batches of chips is about 15% coefficient of variation in terms of impedance at 1 Hz. However reproducibility in terms of other parameters such as conduction and analyte response is much better. Dr Morigaki asked Is the role of membrane-spanning anchor lipids to stabilize the supported bilayer system studied ? How much eÜect if any do they have on stabilizing the membrane matrix? Ms Woodhouse responded The primary role of the membrane-spanning lipid is to provide a tether for attaching analytes to the membrane surface.The stability of the bilayer can be attributed to many parts of the biosensor design and it is difficult to elucidate which parts are more signi–cant. Dr Sansom asked Does your sensor design allow you to distinguish between the double helix vs. helix dimer models of gramicidin channels/pores ? Ms Woodhouse responded Absolutely the data show the conduction measured for various sensor constructs is negligible when they are formed with gramicidin in either one layer or the other but not both.Additionally the model –ts with the notion that the gramicidin in each layer is in equilibrium between the monomeric and dimeric forms. 331 Faraday Discuss. 1998 111 331»343 Prof. Bartlett asked In the work presented in your paper you used a –xed bias of [300 mV. Have you looked to see if any of the rate constants derived from your analysis of the experimental data show a potential dependence? In addition from the text it is not clear to me exactly how the experimental impedance measurement was carried out. Is this a three electrode measurement using a potentiostat with the gold sensor electrode held at a bias of [300 mV with respect to the Ag reference electrode ? Is this a real Ag/AgCl reference electrode or a Ag wire pseudo-reference electrode ? Ms Woodhouse responded We have not investigated the potential dependence on rate.The potential applied improves the reservoir capacity and therefore the conduction and dynamic range of response. To answer your second point this is a three terminal measurement with 300 mV applied to the reference electrode which is a Ag wire pseudo-reference electrode. Measurements have been made with a real Ag/AgCl reference electrode these have been indistinguishable from those employing a pseudo-reference electrode. Dr Lakey asked Does the data in Fig. 5c represent the Fab»digoxin affinity or are other factors determining this result ? By manipulation of the concentrations of certain components is it possible to alter the sensitivity of the biosensor to explore wider ranges of ligand concentration ? Ms Woodhouse responded The data does represent the fast antibody oÜ rates both in 3D and 2D as well as the surface concentration of MSLSA b-Fab@ g gA-dig b-Fab@ andADIMER b-Fab@.Both the sensitivity and the dynamic range of the biosensor can be altered by the manipulation of the sensor components. High surface concentrations of MSLSA compared with gA-dig results in an excess of surface bound b-Fab@ (MSLSA b-Fab@) compared with surface analyte digoxigenin so that the fraction of cross-linked b-Fab@ of the total is lower resulting in a lower sensitivity. Dr Pawlak commented You investigated the kinetics of the competitive channel response. Since a multiple set of processes is involved in your sensing mechanism you have to know carefully the individual number of molecules which contribute to the signal.What is the limit of quantitation (LOQ) of your sensor system in terms of the number of channels involved assuming a 1 1 stochiometry of analyte-to-be-detected channel? Will ì single channel measurementsœ on supported lipid membranes be achievable in the future ? How easily and generally can your sensing principle be adapted to the detection of other analytes e.g. antigens ? Is it possible to employ other channels e.g. peptide bundles or larger receptor channels? Ms Woodhouse responded Theoretically we should be able to measure single channel electrical activity»practically we currently measure the total electrical response of an electrode containing 1 channel lm~2.Through our model we can mathematically determine the number of channels involved in the gating response once we determine the actual surface concentration of MSLSA b- Fab employed. Theoretically our technology could be scaled to achieve single channel measurements we do not foresee any practical limitations. However we are not pursuing this goal presently. This sensing principle has been used for detecting many diÜerent analytes including thyroid stimulating hormone ferritin thyroxine and theophylline as described by Cornell et al.1 Raguse and co-workers has shown that valinomycin can be incorporated into the tBLM (referenced in the paper) and Yin has incorporated alamethicin and shown its selectivity to amiloride based inhibitors into the tethered bilayer.We are currently investigating methods for incorporating larger physiological channels. 1 B. Raguse V. Braach-Maksvytis B. A. Cornell L. G. King P. D. J. Osman R. J. Pace and L. Wieczorek L angmuir 1998 3 648. Dr Atay asked How do the binding constants measured in 2- and 3D compare? And have you looked at the energetics of the reaction at all ? Faraday Discuss. 1998 111 331»343 332 Ms Woodhouse responded The 2- and 3D kinetics are difficult to compare because the dimensions give diÜerent units of measurement. Also as described by Hardt1 the change in dimerisation is predictable and in agreement with our observation. We have not looked at the energetics of the reaction at all. 1 S. L. Hardt Biophys. Chem. 1979 16. 239. Dr Atay communicated I realise that the association constants mentioned in the paper were constants in 2- and 3D.Constants measured (or calculated) in 2- and 3D are rather difficult to compare due to their rather diÜerent units. We have however in a previous study shown a method of comparing these two. I suggest that the authors see the paper by Albery et al.1 They might –nd the reference useful. 1 W. J. Albery R. A. Choudery N. Z. Atay and B. H. Robinson J. Chem. Soc. Faraday T rans. 1 1987 83 2407. Prof. Smith opened the discussion of Dr SutcliÜeœs paper The recently crystallographically determined GLU-receptor is although related not quite your target. Are there any important sequence diÜerences between the two that might hamper your present validation of your model? Dr SutcliÜe responded The sequence identity is ca.40% between the S1S2 domain of GluR2 (ref. 1 for which the crystal structure has been determined) and the gold–sh kainate binding proteins I presented. Thus we cannot be totally certain of the validity of our models based on this crystal structure. However as a conservative estimate we would suggest that well over 50% of our predicted protein»ligand interactions were identi–ed correctly. Of the six regions involved in ligand binding (S1 through S6 in Fig. 2) the major diÜerence with GluR2 comes in S5 where there are two additional residues in GluR2»thus we cannot validate the presence or absence of the proposed hydrogen bond involving S5 and ligand. 1 N. Armstrong Y. Sun G. Q. Chen and E. Gouaux Nature (L ondon) 1998 395 913.Dr Smart asked What is the sequence homology between the recently published potassium channel structure (ref. 13 in your paper) and your ììK` channel domainœœ ? In addition can I comment that your success in fold identi–cation and homology modelling for a target domain with 20% homology to its nearest template is most impressive. Please could you state the overall root mean squared (rms) a-carbon deviation from the subsequently determined crystal structure of a closer homologue and what parts of the structure were best modelled. Dr SutcliÜe responded The KcsA crystal structure and the NMDA receptors have a sequence identity of ca. 20%»i.e. a similar level to that between the S1S2 domain and the bacterial amino acid binding proteins which we have modelled to what you call an ììimpressiveœœ degree of accuracy.The crystal structure which is of the S1S2 domain of GluR2 complexed with kainate has not undergone full domain closure»therefore a comparison of the rmsd is perhaps not that meaningful. The crystal structure does reveal mistakes in our sequence alignment both in Lobe 1 and Lobe 2. Lobe 2 due to fewer misalignments was ìì better modelledœœ than Lobe 1. With the exception of the region denoted S1 in Fig. 2 these do not aÜect the ligand binding site to any great extent»resulting in a reasonable model of the binding site. Dr Sansom said Would you like to comment on the NMR studies of Opella and colleagues which suggest that the NMDA receptor pore-lining M2 segment adopts an a-helical conformation when in a membrane-mimetic environment rather than the loop-like structure in your topological diagram? Dr SutcliÜe responded The glycosylation/epitope mapping/proteolytic data show that the S1S2 domain is indeed extracellular thus M1 M2 and M3 must comprise of 2 membrane spanning and one non-membrane spanning components»site directed mutagenesis and constructs which delete M1 M2 or M3 suggest that M1 and M3 are transmembrane helices and M2 a ììre-entrant loopœœ similar to the P-segment in potassium channels.Thus the diÜerent environments»isolated 333 Faraday Discuss. 1998 111 331»343 peptide vs. peptide in a protein environment»likely lead to the diÜerent conformation observed by Opella. Prof. Holzwarth asked Can your approach be described by using the known molecular composition of a functional polypeptide as bricks to form a house by making sensible assumptions about their arrangement to rooms and arranging the rooms to the house? After the –rst house was built by assumptions you test if the results from experiments can be veri–ed ; then you try to change your arrangement into the direction of better agreement until your circle results in an arrangement which could explain the experimental –ndings available so far.Is there not the danger that you will go in the wrong direction if the experimental –ndings are not accurate ? In other words you strongly depend on the quality of experiments. Dr SutcliÜe responded This is a good analogy of the ìì iterative modelling/experimental veri–cation cycleœœ we use as shown in Fig.1. The iterative nature of our approach means that misleading experimental information if not contradicted by other experimental data would lead to us moving away from rather than towards the ììcorrectœœ answer. Prof. Smith commented The probability of obtaining the correct tertiary fold in a homology model increases with the sequence identity between the aligned template and target structures. Can you give –gures for this in your extra- and intra-membrane domains and maybe comment on their signi–cance ? Dr SutcliÜe responded The sequence identity for our LAOBP-based S1S2 domain models was ca. 20%. There is therefore some uncertainty in the identi–cation of the fold. However the recent crystal structure of the S1S2 domain of GluR2 has shown that use of the LAOBP-like fold was indeed correct.Similarly our most recent models of the transmembrane domain are based on the crystal structure of the KcsA potassium channel with which they share ca. 20% sequence identity. As with the S1S2 domain we are not certain this is the correct fold but our models are consistent with all available experimental data»suggesting again that our models are roughly correct. Prof. Evans asked Although many things interfere with crystallization of proteins could the inability to crystallize the structure surrounding the binding domain re—ect ì —exibility œ and if so how would this aÜect your modeling concept? Dr SutcliÜe responded So far although crystallisation trials have been run on similar constructs from glutamate receptors other than GluR2 only the S1S2 domain of GluR2 has been crystallised.As you say many things aÜect crystallisation and it is impossible to identify —exibility as the factor preventing crystallisation. At 300 K all proteins undergo motion to some degree. It is likely that the ligand-free form of the receptors undergoes more motion that the ligand-bound form as the latter is more ììclamped togetherœœ. Nevertheless the presence of such motions does not change the interpretation of our models. Prof. Roux asked Based on your modelled structures of the ligand-bound and ligand-free form how do you explain the gating mechanism? Dr SutcliÜe responded Ligand binding results in a conformational change of the S1S2 domain and this likely propagates into the transmembrane domain.This conformational change causes the channel to gate. Our modelling suggests that channel opening positions conserved negatively charged residues in the extracellular entrance to the channel thus electrostatically ììopeningœœ the channel. Our models also suggest that the region denoted S5 might be involved in desensitisation of the channel. (For a more detailed discussion of this see the section of our paper entitled ììBinding and signal transductionœœ.) Prof. Laggner opened the discussion of Dr Pawlakœs paper In your paper you have addressed the problem of ensuring the correct sidedness of the membrane patches when they are deposited Faraday Discuss. 1998 111 331»343 334 on the surface but not really answered it. What are the actual right-side-out ratios reached in your present studies ? Dr Pawlak responded In the described studies we have –rst concentrated on –nding the optimum experimental conditions for yielding stable preparations of surface-immobilized membrane patches.Up to now we have achieved the situation of a 50% to 50% distribution on average of the two possible orientations of membrane patches. Single experiments however suggest that we have achieved up to 70% of the speci–c K` binding side exposed assuming the binding of K` exclusively to one side of the membrane. We have to admit that the success of membrane orientation is at the moment much limited by the quality of the membrane patch preparation. The size distribution is broad and a varying portion of small particles probably less de–ned associated with the surface is always present.We expect a far more de–ned membrane orientation on the surface upon using membrane patches prepared with less variation in size. These studies will be undertaken. Prof. Holzwarth asked How good are your immobilised Na,K-ATPases containing fragments in comparison with the same fragments in solution ? Do you not introduce additional problems by attaching the fragments to a surface ? In solution the fragments are accessible from both sides as in their natural surrounding in your case you block one side. What are the consequences of immobilisation for the activity of the pump? Dr Pawlak responded What do you mean ìgoodœ ? We found comparable affinity constants for speci–c binding of potassium to the Na,K-ATPase pump on the surface and in solution.Because K`-binding is a very critical test for the protein conformation in its active form (a K`-binding related structural change of the enzyme results in a large up to 30% reversible change of —uorescein emission) we regard the surface-immobilized enzyme as being of comparable activity as in solution. However you can only reach such a preserved activity by appropriate ìshieldingœ of the protein from the denaturating forces of the bare surface support. This we have achieved by introducing the biocompatible lipid layer. In answer to your second question to reach a controlled access only to one side of the membrane fragments is exactly our goal. Thus you can directly probe individual binding sites without a signal interference of binding sites on the other side of the membrane which is not possible in solution.To answer your third question we do not know at the moment since we havenœt yet tested the pump activity. Experiments are planned in this direction. Dr Keller asked How do you orient the membrane fragments on the surface ? Dr Pawlak responded The membrane fragments orient themselves on the surface by means of a self-assembly process in the presence of attractive forces between the strong dipole moments and/or charges of the membranes and the supported planar lipid monolayer. Prof. Evans said Following on from Prof. Holzwarthœs question may I oÜer a brief comment on the sensitivity of evanescent-wave induced —uorescence techniques.Such techniques do indeed select strongly for the —uorescence of groups close to the interface (usually within 100 nm or so) but the practical problem is that there is always some scattered light from surface defects and this light can excite —uorescence in the bulk solution. Such —uorescence can dominate and needs to be carefully accounted for. 1 L. Fisher in Surface Analytical T echniques for Probing Biomaterial Processes ed. J. Davies CRC Press Boca Raton 1996 pp. 43»66. Dr Pawlak responded You are right that such —uorescence ìcanœ (but must not my comment) dominate leading to an interfering bulk —uorescence background which will limit the sensitivity of a system. Thereby the roughness of the chip surface plays a crucial role in determining the 335 Faraday Discuss.1998 111 331»343 strength of this eÜect. In the case of our sensor the material and surface properties of the waveguiding layer have been optimized over years and in many applications (see ref. 13 14 and 26 in our paper). The roughness of our chip surfaces is very low only 2»3 ” as measured by AFM under high resolution (ref. 23). Therefore the eÜect of bulk —uorescence induction by surfacescattered light is minimal and negligible for our chips. We could not observe such an eÜect in our applications. Especially no shifts in —uorescence signals were detected when comparing signals in the presence and absence of —uorophores in chip rinsing solutions. Prof. Bohne asked Following the previous question could you comment on the requirement for the absolute quantum yield of the —uorophore and what intensity change you can detect.Are there any limitations on the excitation wavelengths that can be used. You commented that the technique can be time-resolved. Are there any limitations for exciting the sample with a pulsed excitation source? Dr Pawlak responded I do not see a limitation in the requirement of the absolute quantum yield of a used —uorophore because we detect relative changes of an established base —uorescence signal which is generated by the surface immobilization of a certain number of —uorophores on the surface. As you see from Fig. 5 in the paper we have a dynamic range of –ve to six orders of magnitude. If the quantum yield is too low then you compensate by the number of —uorophores.We can detect a 1»2% change of established —uorescence. This is limited by signal noise. Excitation in the far blue is limited by an increasing absorbance of the layer material. To answer your –nal question I donœt see any limitations at the moment. Dr Gon8 i commented I am puzzled by the nature of your ìì discs œœ. They appear to be intermediate structures between the intact vesicles and the lipid»detergent»protein mixed micelles. This is rather uncommon in membrane solubilisation. How have these discs been characterised in terms of size composition and detergent concentration ? Also a comment Once the ATPase has been treated with SDS however short it will never go back to the native situation. Some SDS molecules will be irreversibly bound to the enzyme.This is something that should be taken into account when analysing the results. Dr Pawlak responded The discs have been characterized by electron microscopy and dynamic light scattering for the size and size distributions (mean diameter\250 nm) and by lipid analysis for lipid composition. A considerable amount of cholesterol is present in the membrane. The SDS content of the preparation has been determined employing 14C-labeled detergent as reported by J‘rgensen.1 SDS is thought to be preferentially located at the edge of the membrane fragments in order to protect the hydrophobic section of the lipid bilayer. The membrane discs are stable in solution and cannot be regarded as intermediate structures between intact vesicles and detergent» protein mixed micelles.The short treatment of SDS in the preparation only breaks oÜ unwanted cell membrane material around the typical island-like ATPase accumulations as they appear in the natural membrane. It is not a process of protein solubilization. In response to your comment we are aware of this eÜect and will consider it in the data interpretation. We note that if the membrane discs are isolated upon a deoxycholate and saponin treatment according to Skou and Esmann2 instead of SDS the same enzymatic properties are found. This indicates some independence of detergent treatment. 1 P.L.J‘rgensen Biochim. Biophys. Acta 1974 356 34. 2 J. C. Skou and M. Esmann Biochim. Biophys. Acta 1979 567 436. Dr Morigaki asked The native membrane fragments seem to have some bending.What is the size distribution of them? Is it possible to access the curvature of them both in the suspended and the surface bound states ? If the curvature (or undulation) remains at the surface bound state it will aÜect the —uorescence quanti–cation since the evanescent light intensity decreases very sharply with the distance from the substrate. What is generally the driving force for the membrane binding to the ììbio-compatibleœœ surface ? How stable are the bound membrane fragments on the surface ? Faraday Discuss. 1998 111 331»343 336 Dr Pawlak responded The size distribution is broad. The particle diameters range from 50 to 600 nm. The bending in the surface-associated state must be minimal and hence will not aÜect the quanti–cation very dramatically (the exponential decay length of the evanescent –eld is around 150 nm).In AFM images the central part of a disc seems to be held at a certain but more or less constant distance (10 nm) from the surface. Interestingly this distance could be slightly reduced by addition of Mg2` ions which we interpret as the presence of a certain aqueous reservoir between surface and membrane. Indeed this distance reduction could be actually measured by an increase of —uorescence signal (see Fig. 9 in our paper). Regarding the driving force for membrane surface association please see the previous question and answer. We believe the contacts must be very strong since the adsorbed discs could not be rinsed oÜ in the sensor for hours and under strong hydrodynamic —ow nor be removed by the AFM tip applying increasing force.The contact regions seem to be preferentially located at the edge region of the discs. Prof. Petersen asked How will the topography aÜect the estimate of density in Fig. 7? Membranes on an angle will look in projection as if the density is greater (see Scheme 1). Dr Pawlak responded Height diÜerences between edge and central regions are also small and in the range of several nm. We should also note that only the extramembranous protein parts are imaged. In summary a strong in—uence of topography on the density estimate is not expected. Prof. Neumann opened the discussion of Prof. Petersenœs paper Do you control the number of —uorophores per protein molecule? This should matter in quantifying the number of proteins from the —uorescence intensity.Prof. Petersen responded No. We estimate the number of —uorescent probes per protein molecule by standard methods (absorbance ratios). If the number of probes per protein is uniform the eÜect is similar to having a larger quantum yield of —uorescence and there is no eÜect on the measurement of cluster densities. If there is a distribution of the number of probes per protein this will add to the distribution of —uctuations but it will not seriously aÜect the average cluster density estimate. Prof. Barclay asked Formaldehyde –xation can aÜect antigenic sites. Have you observed any loss of antigenic activity ? I have seen reports where formaldehyde –xation might aÜect the distribution of cell surface antigens. As far as I am aware it is still not clear how formaldehyde –xation works.Have you been able to con–rm your data on fresh cells ? Prof. Petersen responded We have no direct evidence whether –xation aÜects our antigenic sites. We always perform titration curves so that we ensure that we work under saturation conditions while minimising non-speci–c —uorescence. Under these circumstances the total number of Scheme 1 337 Faraday Discuss. 1998 111 331»343 receptors detected (for EGF- and PDGF-receptors) match nicely the numbers determined by standard biochemical techniques. This suggests that the eÜect is minor. We also know in the case of PDGF-receptors that the results are comparable for –xed and live cells. I agree that formaldehyde –xation in principle is undesirable.We have observed the same results with formaldehyde –xation and methanol/acetone –xation in the case of adaptor protein distributions so if there are artifacts they are similar for the two procedures. We have seen similar results on –xed and live cells for PDGF-receptors so to this point we have no evidence that –xation is a real concern. Prof. Svetina asked Is clustering of membrane proteins an eÜect of protein»protein interaction or possibly of their links to the cytoskeleton ? Prof. Petersen responded Our measurements do not speak directly to the mechanism of formation of the clusters. I suspect that the answer depends on the system. For the association of HA]8 to the adaptor protein it is clearly an amino acid sequence that is allowing for the interaction i.e.it is a protein»protein interaction. In the other cases (EGF- and PDGF-receptors) the clustering could arise from interactions with the coated pits with cytoskeletal proteins or through the con–nement into domains by the cytoskeletal ìfencesœ. Dr Amblard commented Clusters seen by your image correlation spectroscopy (ICS) methods are only resolved as such if their mean separation distance is larger than the optical resolution of the microscope. Because subwavelength resolution can be achieved by digital processing of images that contain enough signal I am wondering how good the spatial resolution is for distinguishing clusters ? Another question is related to this one because photon statistics is involved namely how precisely can you resolve the average number of molecules inside the clusters.Prof. Petersen responded The spatial resolution of the ICS technique and in particular of the image cross-correlation spectroscopy (ICCS) technique is related to how well the peaks of two adjacent —uctuations are resolved. If they are not arising from the same spatial feature the —uctuations will aÜect the width of the correlation functions. The resolution therefore is a question of how accurately the width can be determined. Our experience with a large volume of data is that we can determine the width w to within about 10%. Thus in principle the spatial resolution for the co-localisation is on the order of one-tenth of the laser width or about 35 nm. Your second question is a difficult question to answer in general.We believe that the cell-to-cell variation is the largest source of uncertainty and this generally works out to about a 30% uncertainty. In systems where there is a single population of receptors (e.g. PDGF-receptors) we have determined by two approaches an average of four receptors per cluster with an uncertainty of about one»allowing us to distinguish dimers from tetramers from hexamers. However there is clearly a distribution of sizes of clusters so that monomers co-exist with oligomers. This complicates a complete analysis. For highly aggregated systems (e.g. EGF-receptors) we have determined an average of 10 to 20 receptors per cluster depending on the temperature. We expect that these numbers are accurate to 2»4 receptors per cluster as indicated in Table 2 in the paper.Prof. Holzwarth asked Which mode of the laser did you use (TEM00) and what is the power density in the centre of the beam? I ask this question in order to judge if there is any danger of damage to your sample caused by the laser light. Secondly can you gain any molecular information about the conformation of the sugar heads and their function for surface recognition ? Prof. Petersen responded Indeed the laser is used in the TEM00 mode to ensure a Gaussian transverse intensity pro–le. We use a 25 mW laser (total power at the source) which after attenuation and losses in the system provides a power at the sample in the microwatt region. There is always a photobleaching process with all of the possible side eÜects associated with generation of free radicals.Experience in other types of photomicroscopy suggests that exposure to microwatts of power for periods of seconds lead to minimal if any damage. The simplest answer to your second question is no. It is possible that in particular instances the association of molecules is conformation dependent but then the evidence would be very indirect. Faraday Discuss. 1998 111 331»343 338 Prof. Neumann asked Can you identify rotational displacements of the chromophore residues on the macromolecules upon complex formation? Prof. Petersen responded It is possible to perform correlation analysis on —uorescence which is analyzed by polarization. This can yield rotational information. We have not attempted any of these experiments. Dr Amblard said You told us that the peak value of the spatial autocorrelation function g(0,0) gives directly the average number of molecules in the clusters.But the raw information given by your images are relative chromophore numbers on each pixel. To interpret such data I understand that some independent information is necessary about the stoichiometry of the receptor» antibody»chromophore interactions. This sort of calibration leads to relative receptor numbers on each pixel. How do you then extract absolute receptor numbers for the cluster size without additional hypothesis ? Donœt you need for instance to assume that the background staining is due to receptor monomers? Prof. Petersen responded The peak value of the autocorrelation function is a measure of the number count within the integrated area (volume) of the laser beam.Thus while there is an intensity count associated with each pixel in the image this intensity re—ects the intensity contributions from molecules with the entire area exposed by the laser beam when it is in that particular location. This is an absolute number without any assumptions needed. To extract the information about the number of monomers within each cluster we need to know more either we need to know the exact quantum efficiency of the microscope the chromophores and other optical parameters or we need to determine the autocorrelation function for a known monomer distribution. We have so far done this by assuming that the non-speci–c staining by the antibody will re—ect a monomeric distribution.This is a big assumption but the best we can do at present. This issue is addressed in detail in a paper by Wiseman and Petersen to be published in the Biophysical Journal in February 1999.1 1 P. W. Wiseman and N. O. Petersen Biophys. J. 1999 76 963. Prof. Barclay asked Does the addition of ligand such as PDGF aÜect the distribution of receptors and clusters ? As triggering through many receptors such as PDGF receptor is thought to involve bringing together receptor molecules and associated kinase domains it is surprising that you –nd the receptors already clustered. This of course has functional implications. Would you like to comment? Some of the cell times you describe have very large number of receptors. Have you looked at normal cells (untransformed) ? Prof.Petersen responded We expect it to but in our work on PDGF-receptors to date we have not seen any eÜect. This is the only system we have had a chance to look at so far. In response to your second question –rst it should be clear that our technique cannot establish whether the proteins within the clusters are in physical contact or are simply con–ned (by whatever mechanism) to a small domain. While we were surprised at the observation that receptors are already clustered we can speculate that pre-existing but inactive complexes or domains of receptors may control the activation process particularly at low ligand concentrations. The probability of encounter of two occupied receptors may be low if they are free to move over the entire cell but it can eventually occur.If the proteins are con–ned to domains two proteins in a domain must be occupied at the same time. This will not likely occur at very low concentrations. Hence the preexisting clusters can lead to a threshold below which activation will not occur. Some of the cell types you describe have very large numbers of receptors. Have you looked at normal cells (untransformed) ? In response to your third question we have only looked at tissue culture cells so these are all transformed cells. The A431 cells overexpress the EGF-receptor and therefore have an abnormally high number of receptors. The expression levels in the AG1523 CV-1 and other cells we have 339 Faraday Discuss. 1998 111 331»343 looked at are normal. We know we can detect as few as 100 000 receptors and believe that this can be extended to about 20 000 per cell but probably not fewer than that will our current technology.Prof. Morantz opened the discussion of Prof. Luckhamœs paper Have you seen any eÜects of varying the withdrawal rate ? It there a possibility of re-grabbing between the surfaces during withdrawal? Prof. Luckham responded We have done some experiments at diÜerent approach rates the results quoted in our paper correspond to approach rates between 2»0.02 Hz i.e. over two orders of magnitude. We did not obviously observe any diÜerence in the results over this somewhat limited range. We shall investigate this further. Yes there is the possibility of re-grabbing the surfaces during withdrawal this may explain the two minima in some of the withdrawal pro–les e.g.Fig. 9.1 in the paper. Prof. Evans commented In order to be con–dent that youœve tested single molecular connections of any type the frequency of ììforce eventsœœ has to be reduced (by dilution) to 1 out of 5»10 touches to a surface. Moreover the force needed to break a multiple-bonded contact does not simply scale in proportion to number nor can the statistics of force events be analyzed to derive the strength of a single molecular bond. The reason is that it is not possible to know how the force is distributed amongst the multiple bonds. The best chance of accessing properties of single bonds is to use a very soft cantilever dilute the surface concentration of reactive sites touch the surface gently and perform experiments over many orders of magnitude range in rate of force/time.Prof. Luckham responded I have no real comment to make other than to thank Prof. Evans for these comments and to point out that these comments largely relate to the detachment of the surfaces. In this paper we also report on the interactions as the cholera toxin approaches a GM1 coated surface. Dr Fisher asked Why did you choose to work in pure water rather than to suppress doublelayer interactions by adding salt ? Secondly it is possible to integrate the stepwise measurement of force during separation to obtain the work done. It would seem from the comments made during this conference that work may be a more useful measure that detachment force. Could you comment on this both in general and in relation to your own measurements? Prof.Luckham responded No reason really I hoped not to be asked that one! In retrospect it wasnœt very clever but initially we had problems with drying out and salt crystal formation on the cantilever and so Kate Smith the co-author of the paper stuck to water. To answer your second question conceptually this can be done rather straightforwardly. The problem being the value one should take for the distance over which the interaction is occurring on approach of the surfaces. As you can see from our data sometimes we do not seem to get an attraction on approach for the reasons outlined in our paper. However if we say that the interaction is occurring over 1 nm then the energy of interaction or work done E will be given by E\force]distance\20]10~12]10~9\2]10~20 J or 2]10~20 1.3]10~23]300 \5kT This is the order of the strength of a single hydrogen bond.Clearly there have been many gross assumptions particularly in the separation that one should take but it is still instructive. These comments apply to both our data and the work of others. Dr Lee commented The atomic force microscope promises to bring a molecular level of understanding to the physical properties of biological interfaces because (1) the microscope is capable of producing physical and chemical images of a surface with nanometer scale resolution (2) the Faraday Discuss. 1998 111 331»343 340 microfabricated cantilever is a sensitive force transducer with ultrahigh temporal resolution.Progress in measuring single molecule interaction with the AFM has been slowed by the difficulty of mastering the many techniques necessary to make these measurements. That is the measurement requires the mastery of surface chemical microfabrication optics and biological techniques. We have recently made two breakthroughs that should greatly accelerate these measurements. First we have developed an assay that will determine the activity of a biological molecule as a function of load. The upper limit of force that can be applied to streptavidin and a DNA oligonucleotide was approximately 1.5 nN. Using this force as a threshold of the upper limit of force that should be applied between the probe and surface we have made up to more than 6000 force measurements with a single experiment.Second we have developed a microfabricated probe array which can be used with probeless cantilevers to measure hundreds (if not thousands) of diÜerent interactions in a single experiment. This is a critical breakthrough because it allows us to make many measurements quickly with the appropriate controls. In conclusion as the key technical issues have been resolved and many groups have perfected their surface chemistries we feel there will be a rapid expansion in the number of studies made of intermolecular interactions with AFM and their quality. Prof. Luckham responded These tips would certainly make life considerably easier in performing these experiments although we found that using the procedure outlined in the paper the success of each experiment was roughly 50%.Prof. Holzwarth asked How are the molecules attached to the tip (covalently or electrostatically) and how is the coverage veri–ed ? Prof. Luckham responded The protein is attached to the tip covalently as described in the paper. Basically it is through a glutaraldahyde linkage a common means of attaching proteins to surfaces. As this is a common means of performing protein attachment we have not veri–ed that there is any protein present apart from through the results presented. Any analysis of the tip is likely to destroy the tip so we thought it best to just try it out. Roughly 50% of the tips show the results presented in the paper the rest just show repulsive interactions. We have performed a whole series of control experiments as shown in the paper to support our thesis that we are studying the interaction between cholera toxin and GM1.Dr Amblard commented To study the strength of intermolecular forces by AFM the molecules must be surface-grafted in a very controlled way. To this end silanisation techniques are very popular on glass surfaces. They are mostly carried out in liquid phase and a distillation procedure is often required to avoid the condensation of silane compounds. In our hands silanisation in gas phase by vapor deposition gives much better results. Ultrasensitive dark-–eld microscopy reveals a very homogeneous coverage which was systematically devoided of molecular aggregates very often seen after liquid phase silanisation. Prof. Luckham responded OK but in AFM experiments one has to be careful not to damage the cantilever.I agree that the procedure may not actually graft the coupling agent directly to the glass surface the silane may actually be polymerised by the presence of surface water on the silicon/silicon nitride surfaces. This may not be all bad though as this would give some —exibility to the protein that ultimately is grafted to the AFM cantilever. Dr Fielden asked Have there been measurements made in the macroscopic limit of radius on this type of system i.e. using the surface force apparatus? Prof. Luckham replied No not really. We1 have studied the interactions between GM in 1 DPPC bilayers. It must also be realised that it is much more difficult to modify mica surfaces than silicon surfaces so the immobilisation of the cholera toxin on mica would be very difficult.1 P. F. Luckham and J. Wood J. Colloid Interface Sci. 1993 156 173. 341 Faraday Discuss. 1998 111 331»343 Prof. Evans opened the discussion of Prof. Bongrandœs paper How do you determine the ìhidden moment armœ or tether that transmits hydrodynamic force to the bond? What is the force history (over time) experienced by the bond up to rupture? Can you be sure that only single bonds are formed while a sphere is brie—y captured by the surface ? Once arrested the sphere will be pushed against the surface as a reaction to the bond force which along with tangential slip at the surface will greatly increase the likelihood of forming additional bonds. Prof. Bongrand responded The force F experienced by a single bond was calculated with standard mechanical reasoning (assuming an undeformed spherical bead with a bond under tension and using known values of the force and torque exerted by hydrodynamic forces).As previously reported1 the force is inversely proportional to the square root of the bond length. For a typical bond of 20 nm length F (in pN) is about 0.5G (where G is the shear rate in s~1). Note that we reported an experimental way of estimating the eÜective length of receptors bound to the sphere surface.2 Under our experimental conditions the bond lifetime is usually much higher than the passage time of the sphere surface along a binding site (a typical value of the relative velocity of the sphere surface is 5 lm s~1 the length of a 20 nm bond is therefore spanned within 4 ms).Thus it is probably warranted to assume that bond rupture occurs under constant load. Clearly it is very difficult to prove formally that arrests involve single molecular bonds. The following arguments may support this assumption (a) when binding sites are diluted on the spheres the arrest frequency is proportional to the –rst power of the site density ; (b) the arrest duration is unaltered when binding sites are diluted. Now suppose the receptor density is say 100 lm~2; the distance between two neighbouring sites would be about 0.1 lm. Since a 0.1 lm displacement of a bead should be detected the shift from a binding site to another should be detectable. We altogether agree that once a sphere is arrested it will be pushed against the surface thus favouring the formation of additional bonds.In any case we repeatedly found that once a cell or particle stopped it displayed increased probability of stopping again. 1 A. Pierres A. M. Benoliel and P. Bongrand J. Biol. Chem. 1985 270 26586. 2 A. Pierres A. M. Benoliel and P. Bongrand C. R. Acad. Sci. Ser. D 1995 318 1191. Prof. Neumann asked The rate constant kon for association steps or bond formations were given once in the units s~1 another one in mol~1 s~1 yet another one in lm4 s~1 as compared to the unit l mol~1 s~1 for classical bimolecular association steps. Could you clarify the diÜerence in the units ? Prof. Bongrand responded If we consider two surfaces maintained parallel over a contact area A with freely diÜusing receptors and ligands of surface concentrations [R] and [L] the initial rate of bond formation (absolute number of bonds formed per unit of time) should read k[R][L]A \number of bonds per second.Constant k should thus be expressed as m2 s~1 which is equivalent to the 3-dimensional form (expressed as m3 s~1). Now if we lump the product kA into a single parameter this will be expressed in m4 s~1. However we feel that this expression is not convenient when receptors or ligands are not freely diÜusing (as occurs on the tip of an atomic force microscope or in our —ow chamber). The intrinsic parameter is thus the frequency of bond formation between two adhesion molecules whose tails are maintained at distance d. This frequency should be expressed in s~1.Dr L. Fisher asked Your method is clearly a very useful addition to the armoury of receptor force measurement techniques but seems to be restricted to spherical hard particles because of the difficulty of calculating the force and force distribution in the case of a deformable particle such as a living cell. Does this mean that there is no value in laminar —ow studies on living cells or is there something still to be gained from such measurements? Prof. Bongrand responded The laminar —ow chamber proved very suitable to study cell adhesion (e.g. ref. 12 by Kaplanski et al. as shown in our paper or Pierres et al.,1) subject to some limitations If our aim is to study the ììnatural lifetime œœ of bonds there is no need to know the Faraday Discuss.1998 111 331»343 342 precise value of the applied force. Thus we found it convenient to study the arrests of cells driven by very low shear rate»say less than 4 s~1 along ligand-coated surfaces. This approach recently allowed us to analyze the eÜect of modulating antibodies on the lifetime of bonds involving integrins. 2 Alternatively we may let cells adhere to a surface for a few minutes then exert increasing forces to study detachment under microscopic control. In this case cell deformability is certainly as important as bond density to determine detachment properties,3 but the measured force is physiologically relevant as shown e.g. in the recent paper by Palecek et al.4 who demonstrated the relevance of cell-substratum binding strength to migration capacity.Second the signi–cance of binding frequencies measured on cellular systems is probably quite diÜerent from that measured on particles and results seem to be dependent on the localization of receptors on the cell surface (e.g. tip of microvilli vs. cell body) as well as environment (e.g. thickness of cell coat). However we feel that physiologically relevant information can be obtained with this technique. 1 A. Pierres O. Tissot B. Malissen and P. Bongrand J. Cell Biol. 1994 125 945. 2 B. Masson-Gadais A. Pierres A. M. Benoliel P. Bongrand and J. C. Lissitzky J. Cell Science submitted. 3 J. L. Me` ge C. Capo A. M. Benoliel and P. Bongrand Cell Biophys. 1986 8 141. 4 S. P. Palacek S. C. Loftus M. H. Ginsberg D. A. LauÜenburger and A. F. Horwitz Nature (L ondon) 1997 385 537.Prof. Holzwarth asked Can you comment on the in—uence of the walls on your experiments? I have in mind the –rst two layers of molecules which are strongly attached to the wall and are not moved with the —ow. Can these layers not in—uence the mobility of particles next to them? Prof. Bongrand responded This is a very important point. Indeed the limitation of the —ow chamber method is probably set by the quality of surfaces rather than measurement accuracy. There are two diÜerent problems when we measure the bond lifetime and force dependence it is important that rupture occurs at the ligand/receptor interface rather than on the wall. This may be a problem for strong interactions since the only check for a given system is that higher forces and lifetimes are found when the ligand»receptor couple is replaced with a stronger one without changing the coupling procedure.In any case as stated in the conclusion of our paper the —ow method seems to us better suited to weak interactions as compared with atomic force microscopy. Secondly when we are interested in bead-to-surface distance it is important to know where the ììhydrodynamic boundaryœœ is located since a low density of surface-bound molecules may be sufficient to drag a water layer near the bead or chamber surface. The main problem lies on the bead surface since mica is expected to be very smooth and coated with a regular array of short oligopeptides. On the contrary electron microscopy showed that the bead surface was fairly fuzzy (unpublished) and the supplier informed us that the binding sites were scattered in a region of about 6 nm depth. We are presently studying the eÜect of the length of receptor structures bound to the bead on adhesion efficiency. Indeed in a previous paper (ref. 17 in our paper) we found that the range of interaction between beads and surfaces was about 10 nm while binding sites were linked to the bead through a spacer comprising two immunoglobulin molecules of about 20 nm length each. Probably the diÜerence between the eÜective and geometrical range is due to the location of the hydrodynamic surface. Prof. Neumann asked Is there an upper limit value for your kon as there is a limit for diÜusion controlled associations with kon\109 l mol~1 s~1? Prof. Bongrand replied The upper limit of 109 l mol~1 s~1 that is reported for soluble molecules is set by diÜusion (it is dependent on the viscosity of the solvent). This is not relevant to the binding frequency we estimate with the —ow chamber. Dr Fielden asked Do you think it would be possible to make a complementary measurement with total internal re—ection microscopy on this system? Prof. Bongrand responded It would certainly be very useful to combine total internal re—ection microscopy and the —ow chamber in order to study the initial steps of cell adhesion. 343 Faraday Discuss. 1998 111 331»343

ISSN:1359-6640    

DOI:10.1039/a900702d

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Concluding remarks and the challenge from the immune system A. Neil Barclay MRC Cellular Immunology Unit Sir W illiam Dunn School of Pathology University of Oxford South Parks Road Oxford UK OX1 3RE. E-mail barclay=molbiol.ox.ac.uk Received 25th January 1999 This meeting has covered a wide range of approaches to the study of interactions of biomembranes and one is struck by the progress in their analysis and the application of many new methods. The main theme of the meeting has been the structure of biomembranes and their components. Rather than pick highlights of these or attempt to summarise the –ndings in these Concluding remarks I will instead summarise some of the interactions of the cells of the immune systems for which immunologists would like explanations at a molecular level.One of the features of the immune system is that it involves a variety of populations of cells that have complex migratory patterns and interactions that occur throughout life. The surfaces of these cells»the leukocytes»mediate interactions that are essential for the –ne control of the immune system that ensures the rapid but controlled rejection of foreign materials such as viruses and bacteria. At the same time it must ensure that reactivity against self is prevented otherwise autoimmune diseases such as rheumatoid arthritis and multiple sclerosis may result. Some of the features of the interactions involved are outlined in this short overview with more detailed analysis of the leukocyte cell surface given in ref. 1. The role of the small lymphocyte It is only around 40 years since the pioneering work of James Gowans –rst showed that the small lymphocyte was a key cell type in the immune response and that it migrated from blood to lymph through specialised ììhigh walledœœ endothelial cells in the lymph nodes.2,3 Fig.1 shows a scanning electron microscope of a lymphocyte that has adhered to the endothelium prior to migrating between the endothelial cells and into the lymph node where it may encounter antigen and diÜerentiate or leave by the lymphatics and back to the bloodstream.4 Later it was established that each lymphocyte carried one receptor for antigen and the concept that lymphocytes patrol the body on the look out for pathogens was established. We are now beginning to understand the complexity of the interactions involved in terms of the proteins participating even if not the biophysics.What is at the surface of the lymphocyte? Interactions at the surface of lymphocytes are central to mediating immune reactions. These interactions include the recognition of foreign antigens interaction with specialised cells that ììpresent antigensœœ such as macrophages and dendritic cells that are involved in the control of the immune response and interactions that control the migration of lymphocytes. Each of these events Faraday Discuss. 1998 111 345»350 345 Fig. 1 Scanning electron micrograph showing adherent B lymphocytes binding specialised high walled endothelial cells of mesenteric lymph node (reproduced with kind permission from ref.4). is highly regulated and this is illustrated by the complex patterns of subpopulations of lymphocytes in secondary lymphoid organs such as lymph nodes and spleen resulting from their speci–c patterns of migration.5 Lymphocytes are one type of white blood cell which are collectively known as leukocytes and include macrophages neutrophils mast cells and dendritic cells. However lymphocytes and their progeny are the only cell types which have speci–c receptors for foreign antigen although they share many other cell surface proteins with other leukocyte populations. The leukocyte cell surface has been studied extensively for two main reasons –rst because of its importance in medicine and secondly because of the ready availability of populations of cells for biochemical analysis»compare the difficulty in obtaining isolated viable neurons! In a recent review we noted around 250 diÜerent proteins which were expressed on leukocyte surfaces but not widely on other cell types.1 These are candidates for mediating leukocyte speci–c functions such as those described above.One would expect proteins required for general housekeeping metabolism and ion transport might be shared with other cell types. Frequency (%) 49 10 13 25 3 1 The functions of leukocyte surface proteins The types of functions mediated by the extracellular parts of leukocyte surface proteins are illustrated in Table 1. The repertoire of proteins re—ects the types of functions expected with many Table 1 Frequency of functions associated with extracellular parts of leukocyte surface proteins Activity Unknown Receptors for cell surfaces or extracellular matrix Receptors for cytokines Receptors for other soluble proteins (e.g.complement receptors) Enzymes Others (e.g. transporters) Data from ref. 1. Faraday Discuss. 1998 111 345»350 346 Table 2 Binding constants for some leukocyte cell»cell interaction molecules Ref. Interaction K T /°C K K on b/M~1 s~1 d a/M off c/s~1 37 37 37 CD2 with CD48 (rat) CD2 with CD58 (human) CD48 with 2B4 (mouse) CD48 with 2B4 (human) CD80 with CTLA4 (human) CD62L (rat) with glyCAM-1 [105 [4]105 [8]105 [105 [9]105 [105 D75]10~6 D15]10~6 16]10~6 8]10~6 D0.4]10~6 100]10~6 [10 7 [4 19 3 20 [7 20 21 [0.4 [10 1.4 37 37 25 37 22 23 24 0.1 25 CD62P with PSLG-1 (human) LFA-1 with ICAM-1 (CD54) (human) 4]106 2]105 0.32]10~6 0.5]10~6 Binding constants for some other macromolecular interactions Avidin to biotin Interleukin-1 to its receptor Antibody Fab to CD2 protein Concanavalin A to trisaccharide 25 D10~15 10~10 5]10~8 6]10~6 D107 D10~8 25 106 D10~4 25 4]105 » » 25 2]10~4 25 8 37 25 a Kd\Equilibrium constant.b Kon\Association rate constant. c Koff\Dissociation rate constant. proteins involved in recognition events e.g. binding of antibodies complement components and lymphokines. No clear functional data are available for about half the known proteins but many of these are expected to interact with other proteins in solution or at the surface of cells (see below).The structures of leukocyte surface proteins Leukocyte surface proteins are a very heterogeneous population of molecules. Their abundance varies from barely detectable levels (less than 1000) to 1 000 000 molecules per cell (CD90 on rodent thymocytes). The size of their extracellular regions varies from 12 amino acids (CD52) to 4400 (CD91). Most of the proteins are glycoproteins but the degree of glycosylation varies from 0% by weight to over 70%. Estimates of their size from electron microscopy indicated many small molecules in the range 10»15 nm but also proteins in the range 40»80 nm (e.g. CD43 CD45 and CD21). The amino acid sequences of the proteins contain regions that show similarity to other proteins and one can predict that many of these proteins are organised into arrays of domains.The most common domain type is the immunoglobulin-like (Ig-like) domain which is known to be particularly suited to recognition events.6 Ig-like domains are present in around one third of leukocyte surface proteins.1 Many interactions of leukocyte surface proteins are of low affinity As the interactions between leukocytes and other cells are usually transitory in nature it is therefore not unexpected that the interactions are of low affinity. The –rst low affinity interaction between cell surface proteins to be characterised in detail was between CD2 and CD48 using new methods utilising the phenomenon of surface plasmon resonance and recombinant proteins corresponding to the extracellular regions of these proteins.This method allowed protein interactions to be followed in real time and was particularly suited to following weak interactions like that between CD2 and CD48 which has a K of around 75 lM at 37 °C with a particularly fast d dissociation rate of greater than 6 s~1 (ref. 7). Thus the monomeric interaction has a half-life of a fraction of a second. Of course when the cells make contact the concentrations of the interactants are relatively high. There are extensive data to suggest that these weak interactions are of functional signi–cance from using monoclonal antibodies to cell surface proteins in functional assays and methods involving genetic manipulation.1 Interactions of this type and even weaker interactions could have major eÜects on the alignment of cells.8 One of the best-characterised leukocyte adhesion systems is the binding of neutrophils to endothelium (reviewed in ref.9). This adhesion is triggered by changes in the endothelium which lead –rst to the weak attachment of the neutrophil which ì rolls œ along. Some cells detach but others —atten and then the neutrophil can migrate between the endothelial cells into the tissue. One of Faraday Discuss. 1998 111 345»350 347 Fig. 2 Schematic view of some of the membrane proteins involved in interactions between a T lymphocyte and an antigen presenting B lymphocyte. The dimensions are based on the approximate sizes of the proteins determined by electron microscopy and X-ray crystallography.Ig-like domains are indicated by shaded ovals domains in CD45 by clear ovals including three –bronectin type III domains; the CD40/CD154 dimensions are based on the structure of the TNF/TNFR. The peptide being presented to the TCR is indicated by a black spot between the MHC Class II and the TCR. N-linked carbohydrates are not indicated but regions with a high content of O-linked sugars are shown by solid bars in approximate accordance with site density. The carbohydrates are major features because of their bulk and this is discussed in more detail in ref. 1 and 18. Adapted from ref. 1. the proteins involved in the rolling step is P-selectin (CD62P) and this is an example of a weak interaction (Table 2). Other proteins involved in later stages of adhesion include integrins.For example LFA-1 (aLb2) interacts with ICAM-1 (CD54); this has a somewhat higher affinity that the selectins (Table 2). One of the consequences of the low affinity interactions is that it is difficult to identify novel interactions. For example the half lives of several of the interactions in Table 2 are in the order of a fraction of a second so that binding studies with puri–ed proteins will not withstand a washing step. It seems likely that many of the proteins for which functions are not known (Table 1) will fall into this category of mediating weak interactions between cells. However it is possible to make the proteins multivalent by a variety of methods and identify and characterise weak interactions e.g.ref. 10»12. The distribution of leukocyte surface proteins Fig. 1 shows that the lymphocytes are not simple spheres but have ruffles and microvilli. This adds another degree of complexity as local concentrations of proteins may vary in diÜerent regions of the cell surface. Thus immuno-electron microscopy has shown that CD62L was clustered at the tips of microvilli or membranes ruffles but was largely absent from the membrane of the cell body. In contrast the integrin aMb2 was present mainly on the membrane of the cell body and seldom on the microvilli.13 As discussed below this is consistent with the known roles of these proteins in that the CD62 is involved in the –rst interactions between the neutrophil in the blood stream and Faraday Discuss.1998 111 345»350 348 the endothelium that lead to rolling and tethering whilst the integrin is involved in later adhesion events once the cell has stopped rolling. Concluding remarks The immune system utilises complex interactions at cell surfaces in a highly controlled manner. Many of the proteins involved are now identi–ed and some of their interactions characterised in isolation. It is clearly more difficult to understand the events that occur when cells come into contact but in this meeting we have heard several systems that can be or have been applied to the particular challenges of the immune system. Acknowledgements I am grateful to support from the Medical Research Council and the EU Biotechnology programme grant for work on protein modules.The size and shape of leukocyte surface proteins The diÜerent sizes of proteins at the surface of cells have implications as to which ones are available to interact when cells come together. One of the most extensively studied cell interactions is that between T lymphocytes and those cells able to present antigen such as dendritic cells and B lymphocytes. This interaction is crucial in determining the initiation of the immune response and hence is of major interest in both understanding the immune response itself and also in –nding ways to manipulate it for medical bene–t e.g. in the treatment of autoimmune diseases and to facilitate organ transplantation without rejection. Some of the proteins involved are illustrated schematically in Fig. 2. It can be seen that several of the proteins involved are relatively small and the distance between opposing membranes when these interact is of the order of 15 nm.1,14,15 These include the important interaction between the receptor for antigen on T cells (TCR) and the antigen peptide presented on major histocompatability antigens (MHC).When one considers that the large proteins such as CD45 and CD43 are not only 2»3 times this length but are also much more abundant than say CD4 or the TCR it is evident that one cannot consider that speci–c interactions of the TCR without also considering what happens to the large abundant proteins. There is now clear evidence for the occurrence of redistribution of some of the membrane associated proteins16,17 during cell contact.Faraday Discuss. 1998 111 345»350 References 1 A. N. Barclay M. H. Brown S. K. A. Law A. J. McKnight M. G. Tomlinson and P. A. van der Merwe L eucocyte Antigens Factsbook 2nd edn. Academic Press London 1997. 2 J. L. Gowans Immunol. T oday 1996 17 288. 3 J. Gowans and E. Knight Proc. R. Soc. L ondon Ser. B 1964 159 257. 4 W. van Ewijk N. H. Brons and J. Rozing Cell. Immunol. 1975 19 245. 5 A. N. Barclay Immunology 1981 42 593. 6 A. F. Williams and A. N. Barclay Annu. Rev. Immunol. 1988 6 381. 7 P. A. van der Merwe M. H. Brown S. J. Davis and A. N. Barclay EMBO J. 1993 12 965. 8 M. L. Dustin D. E. Golan D. M. Zhu J. M. Miller W. Meier E. A. Davies and P. A. van der Merwe J. Biol. Chem. 1997 272 30 889. 9 N. Hogg and C. Berlin Immunol. T oday 1995 16 327.10 S. Preston G. J. Wright K. Starr and A. N. Barclay Eur. J. Immunol. 1997 27 1911. 11 J. D. Altman P. Moss P. Goulder D. H. Barouch W. M. McHeyzer J. I. Bell A. J. McMichael and M. M. Davis Science 1996 274 94. 12 I. Stamenkovic D. Sgroi and A. AruÜo Cell 1992 68 1003. 13 S. L. Erlandsen S. R. Hasslen and R. D. Nelson J. Histochem. Cytochem. 1993 41 327. 14 P. A. van der Merwe P. N. McNamee E. A. Davies A. N. Barclay and S. J. Davis Curr. Biol. 1995 5 74. 15 K. C. Garcia M. Degano R. L. Stan–eld A. Brunmark M. R. Jackson P. A. Peterson L. Teyton and I. A. Wilson Science 1996 274 209. 16 A. I. Sperling J. R. Sedy N. Manjunath A. Kupfer B. Ardman and J. K. Burkhardt J. Immunol. 1998 161 6459. 17 C. R. Monks B. A. Freiberg H. Kupfer N. Sciaky and A. Kupfer Nature (L ondon) 1998 395 82. 349 18 P. Rudd M. Wormald D. Harvey M. Devasahayam M. McAlister M. Brown S. Davis A. Barclay and R. Dwek Glycobiology 1999 9 in the press. 19 P. A. van der Merwe A. N. Barclay D. W. Mason E. A. Davies B. P. Morgan M. Tone A. K. Krishnam C. Ianelli and S. J. Davis Biochemistry 1994 33 10 149. 20 M. H. Brown K. Boles P. Anton van der Merwe V. Kumar P. A. Mathew and A. N. Barclay J. Exp. Med. 1998 188 2083. 21 P. A. van der Merwe D. L. Bodian S. Daenke P. Linsley and S. J. Davis J. Exp. Med. 1997 185 393. 22 M. W. Nicholson A. N. Barclay M. S. Singer S. D. Rosen and P. A. van der Merwe J. Biol. Chem. 1998 273 763. 23 P. Mehta R. D. Cummings and R. P. McEver J. Biol. Chem. 1998 273 32 506. 24 Y. Tominaga Y. Kita A. Satoh S. Asai K. Kato K. Ishikawa T. Horiuchi and T. Takashi J. Immunol. 1998 161 4016. 25 P. A. van der Merwe and A. N. Barclay T rends Biochem. Sci. 1994 19 354. Paper 9/00659A Faraday Discuss. 1998 111 345»350 350

ISSN:1359-6640    

DOI:10.1039/a900659a

出版商:RSC    

年代:1999

数据来源: RSC