The membrane anchor influences ligand binding two-dimensional kinetic rates and three-dimensional affinity of FcgammaRIII (CD16).

Kinetic rates and affinity are essential determinants for biological processes that involve receptor-ligand binding. By using a micropipette method, we measured the kinetics of human Fcgamma receptor III (CD16) interacting with IgG when the two molecules were bound to apposing cellular membranes. CD16 is one of only four eukaryotic receptors known to exist natively in both the transmembrane (TM, CD16a) and glycosylphosphatidylinositol (GPI, CD16b) isoforms. The biological significance of this anchor isoform coexistence is not clear. Here we showed that the anchor influenced kinetic rates; compared with CD16a-TM, CD16a-GPI bound faster and with higher affinities to human and rabbit IgGs but slower and with lower affinity to murine IgG2a. The same differential affinity patterns were observed using soluble IgG ligands. A monoclonal antibody bound CD16a-GPI with higher affinity than CD16a-TM, whereas another monoclonal antibody reacted strongly with CD16a-TM but weakly with CD16a-GPI. No major differential glycosylation between the two CD16a isoforms was detected by SDS-polyacrylamide gel electrophoresis analysis. We suggest a conformational difference as the mechanism underlying the observed anchor effect, as it cannot be explained by the differing diffusivity, flexibility, orientation, height, distribution, or clustering of the two molecules on the cell membrane. These data demonstrate that a covalent modification of an Ig superfamily receptor at the carboxyl terminus of the ectodomain can have an impact on ligand binding kinetics.

The structural segments of cell surface receptors consist of distinct domains as follows: a glycosylated extracellular domain linked to either a transmembrane (TM) 1 domain with a cytoplasmic tail or to a glycosylphosphatidylinositol (GPI) moi-ety without TM and cytoplasmic domains (1,2). The anchor can influence the function of a receptor. TM anchors of some receptors carry information for protein internalization and subunit association, whereas those of other receptors transduce signals (3,4). The GPI moiety consists of a glycan core sandwiched between ethanolamine and a lipid tail. Ethanolamine is covalently attached to the carboxyl terminus of the protein by an amide bond, whereas the lipid tail directly inserts into the outer leaflet of the membrane but does not cross the bilayer (5). The GPI anchor has been implicated in facilitating the lateral mobility of the protein on the cell surface (6) and enhancing receptor-mediated cell adhesion (7,8).
There are four known eukaryotic receptors that naturally exist in both membrane anchor isoforms as follows: neural cell adhesion molecule, lymphocyte function-associated antigen 3, vascular cell adhesion molecule 1, and Fc␥ receptor III (Fc␥RIII or CD16). Neural cell adhesion molecule mediates Ca 2ϩ -independent homophilic adhesion during the development of neurons with the GPI-anchored isoform being expressed later in development than the TM-anchored isoform (9,10). Lymphocyte function-associated antigen 3 is expressed on human erythrocytes as a GPI-anchored protein but on all nucleated cells as both membrane anchor isoforms (11)(12)(13). Vascular cell adhesion molecule 1 has been found with both anchors in murine but not in human cells (14,15). The TM-anchored Fc␥RIIIa (CD16a) is expressed on macrophages, natural killer (NK) cells, and subsets of monocytes and T cells. The GPIanchored Fc␥RIIIb (CD16b) is only expressed on neutrophils (16 -21). The physiological significance of this coexistence of two distinct membrane anchor isoforms for the same receptor is not clear.
CD16 is a 50 -80-kDa highly glycosylated cell surface receptor for monomeric IgG (see Ref. 22 and references cited therein). CD16b is polymorphic with the two alleles being termed neutrophil alloantigen 1 (NA1) and 2 (NA2). CD16a and CD16b are products of two highly homologous genes, and their 191 amino acid extracellular domains differ by only 6 amino acids (Fig. 1). CD16b lacks the 20-amino acid TM segment as well as the 25-amino acid cytoplasmic domain of CD16a. In addition, the surface expression of CD16a requires associated subunits, such as the ␥ chain of the Fc receptor or the chain of the T cell receptor, which form a homo-(␥-␥ or -) or hereto (␥-)-dimer in complex with CD16a ( Fig. 1). Binding of antigen-constrained IgGs brings about cross-linking of CD16, which can trigger a variety of immune functions, including immune complex clearance, phagocytosis, antibody-dependent cellular cytotoxicity, release of inflammatory mediators, and enhancement of antigen presentation.
Ligand binding of CD16 can be influenced by a number of factors. CD16a on NK cells binds monomeric human IgG (hIgG) with higher affinity than CD16b on neutrophils (23). It is not known, however, to what degree this differing affinity is due to the differences in the extracellular domain, in the membrane anchor, or in the cellular background of the two membrane isoforms. The two alleles of CD16b, which differ by four amino acids and two glycosylation sites in the ectodomain (Fig. 1), have the same affinity for hIgG1 complex but different affinities for hIgG3 complex (24). NK cell CD16a has a higher affinity for hIgG than monocyte CD16a despite the fact that the two proteins have an identical polypeptide core, suggesting that glycosylation can influence CD16a's affinity for the ligand (25). Miller et al. (26) reported higher affinity for murine IgG2a (mIgG2a) of CD16a than of CD16a-GPI, a molecule created by replacing the TM and cytoplasmic domains of CD16a with a GPI anchor (27). These authors suggested an affinity-enhancing role for the associated ␥ chain of CD16a.
In the present work we quantified the kinetic rates and binding affinity of CD16-IgG interaction using a recently developed micropipette method. We asked whether the membrane anchor (including the associated subunits) itself had any influence on these intrinsic binding parameters. We further asked what mechanisms might cause such differences and what role the associated ␥ or subunit might play.
Our micropipette method determines reaction kinetics between CD16 and IgG when the two molecules are bound to apposing surfaces as in the case of cell-cell adhesion, i.e. twodimensional or solid-state binding kinetic rates (28). Only recently have such measurements become experimentally possible, and relatively little data exist in the literature (28 -31).
Here we report the systematic measurements of forward and reverse rate constants of two human CD16 membrane isoforms interacting with human, rabbit, and murine IgGs.
The micropipette method mechanically assays the chemistry of receptor-ligand binding at the level of a single pair of cells and, most frequently, at the level of a single pair of molecules. To determine whether the membrane anchor effect, if detected, was dependent on the two-dimensional nature of the experiment or on the specific technique used, we also measured binding affinities of the same CD16-expressing CHO cells for soluble IgGs from the same species, i.e. three-dimensional affinity, by a conventional competitive inhibition method.
Our data show that, in comparison to CD16a-TM, CD16a-GPI bound with faster forward rates and higher affinities to hIgG and rabbit IgG (RbIgG) but with a slower forward rate and a lower affinity to mIgG2a. Thus, the membrane anchor of CD16 influenced its ligand binding kinetic rates and affinity, and the qualitative trend of such an effect was inverted when the ligand was changed. Furthermore, this effect exhibited the same pattern regardless whether binding was measured with membrane-bound ligands using the micropipette method or with fluid-phase ligands via competitive inhibition. The findings that the anchor effect flipped with different ligands and that the same results were observed in both two-dimensional and three-dimensional studies are important. They enabled us to rule out the following six possible mechanisms for the membrane anchor effect: differing lateral diffusivity, rotational flexibility, molecular orientation, binding site height, surface distribution, and functional clustering of the two molecules. Furthermore, no major differential glycosylation between the two CD16a isoforms was found by SDS-PAGE analysis. We hypothesize that there is a conformational difference between the two CD16a isoforms that causes the observed anchor effects. This proposed mechanism was supported by the finding that whereas an anti-CD16 monoclonal antibody (mAb) bound with higher affinity for CD16a-GPI than CD16a-TM, another mAb reacted strongly with CD16a-TM but only weakly with CD16a-GPI, suggesting that an antigenic epitope was downregulated after the membrane anchor of CD16a had been changed from TM to GPI.

EXPERIMENTAL PROCEDURES
cDNA Constructs-The cDNAs encoding human CD16A-TM in a pSVL vector and CD16A-GPI in a pCDM8 vector were provided by Dr. J. Ravetch (Sloan-Kettering Institute for Cancer Research, New York). We further subcloned the CD16A-TM cDNA into the PCR3uni vector (Invitrogen, San Diego, CA). The cDNAs encoding the two human CD16B alleles, the ␥ subunit of rat Fc⑀RI, the two chimeric CD16A-␥ and CD16B-, and the hygromycin or neomycin resistance genes have been described (24).
The anti-CD16 adhesion-blocking mAbs CLBFcgran-1 (murine IgG2a) and 3G8 (murine IgG1) as well as the irrelevant control mAb X63 (murine IgG1) were produced in-house from hybridomas as described previously (20). The anti-CD16 mAb VEP13 (murine IgM) was obtained from IV Leukocyte Workshop. The FITC-coupled F(abЈ)2 fragment of goat anti-hIgG, anti-RbIgG, anti-mIgG, and anti-mIgM poly-FIG. 1. Schematic of CD16 isoforms. a, two CD16a membrane anchor isoforms. b, two CD16b alleles. c, two CD16a-subunit chimeras. The extracellular domains, which begin at amino acid 18, are depicted as two Ig-like globules with the glycosylation sites shown as sticks. The amino acids in the ectodomain that differ among the various molecules are listed. Comparing to those in CD16a, the different amino acids in CD16b are underlined, and the differences between the two CD16b alleles are shown in italics. The lost glycosylation site in CD16b NA1 due to the change Ser-65 3 Asn-65 is indicated by Ϫ*, whereas the gained glycosylation site in CD16b NA2 due to the change Asp-82 3 Asn-82 is indicated by ϩ*. The CD16a-TM and the two CD16a-subunit chimeras differ by the TM and cytoplasmic domains but have the same ectodomain and anchor to the cell surface via the same TM mechanism. By comparison, the CD16a-GPI and the two CD16b alleles use a lipid tail to insert into the outer leaflet of the bilayer but do not cross the membrane. The GPI moiety is of similar size to an Ig globule. Hence it may extend the membrane-proximal ligand-binding Ig domain further above the membrane. It may also alter the orientation and/or conformation of the molecule. clonal antibodies used in flow cytometry were from Sigma. The cleaving of CLBFcgran-1 into Fab fragment was done by Lampire (Pipersville, PA). The dimeric soluble form of CD16a (sCD16a) and another soluble molecule sB7 that were used as irrelevant controls were produced by our laboratory and will be described elsewhere. 2 IgG Ligands and Their Membrane Coating-Total hIgG (Lampire and Jackson ImmunoResearch, West Grove, PA), RbIgG (Sigma), and mIgG2a (Sigma) were used as ligands for CD16. Before each competitive inhibition experiment, IgG ligands were centrifuged at 100,000 rpm for 1 h to remove large aggregates from solution. IgG ligands were coated onto human red blood cells via a chromic chloride method (32). This method covalently links free protein in solution to proteins expressed on the red blood cell surface via interactions at the carboxyl groups. Different coating densities on the red blood cell membrane are easily achieved by varying the soluble protein and CrCl concentrations during the procedure. Procedures for collection and isolation of red blood cells have been previously described (28).
Quantification of Surface Protein Density-Quantification of the cell surface densities of the receptor and ligand is necessary for the twodimensional kinetic measurements. This was done primarily by flow cytometry but also was cross-checked by radioimmunoassay, as described previously (28). Briefly, CD16-expressing and control CHO cells were incubated first with primary mAb CLBFcgran-1 (25 g/ml in PBS for 30 min on ice) or without primary antibody for control and then with the FITC-labeled, F(abЈ) 2 fragment of goat anti-mIgG secondary antibody (50 g/ml in PBS for 30 min on ice). The ligand-coated red blood cells were incubated directly with FITC-labeled IgG species-matched (or -mismatched for control) goat anti-IgG secondary antibody. Cells were analyzed by FACS® (Becton Dickinson, San Jose, CA), and their mean fluorescence intensities were compared with standard calibration beads (Flow Cytometry Standards Corp., San Juan, Puerto Rico) to determine the mean number of fluorophores per cell which was then converted into labeled protein per cell (33,34).
Flow cytometry was also used to determine the expression of an antigenic epitope recognized by mAb VEP13 on the CD16a molecules. In this experiment the concentrations of the primary antibody (VEP13) were titrated in serial dilutions to ensure that the comparison was not biased by insufficient mAb binding at sub-saturation concentrations.
The radioimmunoassay for determining the CD16 expression on CHO cells was done by a small alteration in the Scatchard protocol during the three-dimensional binding assays (below). Determination of receptor density by this method was less desirable than that of flow cytometry because it required the constant availability of 125 I-CLBFcgran-1 (Fab fragment); however, the correlation between the two methods is good (R 2 ϳ90%) (28) and imparts more confidence on our surface protein density estimations. Radiolabeling of proteins was done by using IODO-GEN-coated tubes (Pierce) (24).
Three-dimensional Binding Studies-The binding affinities of CLBFcgran-1 Fab for the two CD16a membrane anchor isoforms were determined by Scatchard analysis (35). The low affinity of monomeric IgG for CD16 makes direct measurement by the Scatchard method unreliable. To circumvent this difficulty, a competitive inhibition assay was used where the low affinity ligand IgG competes with the high affinity antibody CLBFcgran-1 for receptor binding (36). Briefly, CHO cells were grown in flasks until near confluence. Cells were rinsed once in PBS and then removed from the flask using PBS/EDTA (containing 5 mM EDTA). After washing, they were resuspended at 1 ϫ 10 6 cells/ml in PBS/EDTA, pH 7.5. Cells were then added to V-bottom 96-well plates at 100 l per well. The wells were precoated with 1% IgG-free BSA (Sigma) in PBS by incubating at room temperature for 2 h. They were rinsed with PBS/EDTA and kept on ice until the cells were added. After adding cells, the plates were spun at 2000 rpm for 2 min. The supernatant was removed, and a solution of 50 l of PBS/EDTA and titrated amounts of IgG was added to each well with mixing. Then, 50 l of PBS/EDTA, 0.25-0.50 g/ml 125 I-CLBFcgran-1 Fab was added to each well, followed by a 45-min incubation on a shaker at 5°C. After washing 3 times, the cell pellets were removed and counted in a gamma counter.
In the presence of increasing concentrations of the low affinity ligand (IgG, concentration c l l ), the binding of the high affinity ligand ( 125 I-CLBFcgran-1 Fab, concentration c l h ) to the cell surface receptor (CD16) is gradually reduced or displaced. The displaced fraction (F), defined as the bound fraction (f) of CLBFcgran-1 normalized by the value when no IgG was present (f 0 ), can be expressed by Equation 1.
Since the affinity to CD16 of 125 I-CLBFcgran-1 Fab (K a h ) and the receptor concentration (c r ) were predetermined from a separate experiment by Scatchard analysis, the only unknown in Equation 1 is the affinity of IgG (K a l ). Therefore, K a l can be calculated from a single measurement of F without the experimental displacement curve to include data at the IC 50 point. To increase the accuracy of the K a l value, however, the predicted displaced fraction (Equation 1) was nonlinearly fit to the entire F versus c l l data set. Two-dimensional Binding Studies-Two-dimensional kinetic rates of CD16-IgG binding with both interacting molecules being cell-bound were measured by using our recently developed micropipette method (28). Briefly, a microscope chamber filled with 3 ml of half-isotonic HBSS (Sigma) plus 1% BSA was injected with 1 ϫ 10 3 receptor-expressing CHO cells and 1 ϫ 10 4 ligand-coated red blood cells. A single CHO cell and a single red blood cell were respectively aspirated by two apposing micropipettes and aligned via micromanipulation (Fig. 2). The suction pressure was precisely controlled by a pressure regulation system such that the unaspirated portion of the red blood cell in the free state remained a spherical shape ( Fig. 2a) but was easily deformed by subpiconewton forces acting at the apex. A computer program drove the piezo translator on which the red blood cell pipette was mounted to move the two cells into contact for a pre-determined area and duration (Fig. 2b). Upon the pipette retraction, the two cells either were immediately separated (i.e. no adhesion, scored 0) or remained bound with the red blood cell being elongated (Fig. 2c) for a short while before being detached by force (i.e. adhesion, scored 1).
This adhesion test cycle was then repeated hundreds of times to obtain an estimate for the adhesion probability from the running adhesion frequency versus test cycle count data (Fig. 3a). For the reversible binding (28) observed in the present study, the total adhesion probability (P t ) can be simply estimated from the average adhesion score, or adhesion frequency, calculated at the last test (ϭ number of adhesions/number of tests). The probability of specific adhesion (P a ) was then calculated by removing that of the nonspecific adhesion (P n ) according to P a ϭ (P t Ϫ P a )/(1 Ϫ P n ). P a relates to the receptor and ligand densities (m r and m l , respectively), the contact area (A c ), and the contact time (t) via Equation 2 (28).
where m r and m l were measured independently by a separate experiment (above) and A c was kept constant throughout. The contact time was kept constant in each sequential adhesion test series conducted using a single cell pair to allow for measurement of P a that corresponded to that t. In different test series using different cell pairs, t was systematically varied over a range. These data were then fit to Equation 2 ( Fig. 3b) using a -squared error minimization method that returned the two-dimensional binding affinity (K a ) and reverse rate (k r ).

RESULTS
The Adhesion Probability Measured from the Micropipette Binding Tests Was Specific-Because the micropipette method is designed to measure adhesion mediated by a low number (may even be just one) receptor-ligand bonds, which amounts to forces as low as even a single piconewton, it is important that the nonspecific adhesion level be low. Fig. 4 exemplifies some of the control experiments performed to ensure that binding curves such as those shown in Figs. 3b and 5 were results of specific CD16-IgG interactions. When RbIgG-or mIgG2acoated red cells were allowed to contact CHO cells expressing either CD16a-TM or CD16a-GPI, frequent adhesions were detected (Fig. 4a). In contrast, contacts of red cells coated with an irrelevant protein (BSA) with the same CHO cells resulted in much less frequent adhesions. Similarly, only rare adhesions to IgG-coated red cells were detected when the CD16-transfected CHO cells were replaced by either untransfected CHO cells (Ϫ) or CHO cells transfected with an irrelevant receptor (integrin ␣ IIb ␤ 3 ). Binding was inhibited by preincubation of the CD16aexpressing CHO cells with the anti-CD16 adhesion-blocking mAb CLBFcgran-1 or by preincubation of the RbIgG-coated red cells with the soluble competitor sCD16a (Fig. 4b). In comparison, binding was unaffected by preincubation with an irrelevant mAb X63 or an irrelevant soluble protein sB7. Blocking experiments performed for the mIgG2a ligand yielded results similar to Fig. 4b (not shown). These data confirm our previous control experiment for hIgG (28) and collectively demonstrate FIG. 3. Measurement of adhesion probability. a, representative binding evolution curves. Repeated micropipette adhesion tests were conducted using a single pair of CHO and red cells. For each test, the cumulative adhesion scores up to that test were divided by the most recent test number to calculate the running adhesion frequency at that test cycle. Plotting running frequency against the cycle number gives rise to a binding evolution curve. For the reversible binding seen in the present study, a characteristic binding evolution curve fluctuates when the test cycle count is small, reflecting the inherent random nature of small number receptor-ligand binding but stabilizes as the test number becomes sufficiently large, perhaps Ͼ50. This allows the adhesion probability to be estimated from the running adhesion frequency at large test counts. Data from three sequential adhesion test series are shown with the same contact time (10 s). The bottom curve results from an IgG-free BSA-coated red blood cell touching a CD16a-TM-expressing CHO cell, which illustrates the nonspecific binding background. The other two curves result from two RbIgG-coated red blood cells respectively interacting with a CHO cell expressing either CD16a-TM or CD16a-GPI (indicated). While the corresponding densities for the receptors and ligands (m r and m l ) were different for the two curves (indicated), the m r ϫ m l products were comparable (2.4 and 2.2 ϫ 10 5 m Ϫ4 for the CD16a-TM and CD16a-GPI curves, respectively). According to Equation 2, the higher curve indicates a greater binding affinity than the lower curve, since the adhesion probability increases with K a when t and m r ϫ m l are kept constants, as were in the present case. b, measured (points) and predicted (curves) adhesion probability as a function of the contact time and ligand density. Specific adhesion probability (after background removal) P a was measured at 1, 2, 5, 10, and 20 s using CHO cells expressing 1250 molecules/m 2 of CD16a-TM. Two batches of red blood cells were used, with RbIgG-coating densities 190 (diamonds) and 360 (squares) molecules/m 2 , respectively. The dependence of P a on t and m l is apparent. Equation 2 was simultaneously fit to both data curves using a 2 minimization method that returned the values of the two fitting parameters, A c K a and k r (Table I). Data are presented as mean Ϯ S.E. of 2-4 cell pairs of 50 -200 tests for each pair to ensure the adhesion probability measured at each point is estimated from no less than 400 adhesion tests.

FIG. 2. Photomicrographs (and an overlay drawing) of a typical adhesion test.
A hIgG-coated red blood cell aspirated by a micropipette, left, and a CD16a-TM-expressing CHO cell also aspirated by another micropipette (only partially shown), right. a, the unaspirated portion of the red blood cell is shown in its free, spherical shape. b, the red blood cell was pushed onto the CHO cell with an overall apparent contact diameter of approximately 2 m. The contact area and time between the CHO cell and red blood cell were carefully controlled. c, a red blood cell that previously contacted the CHO cell was being retracted. It was elongated by the adhesion force, allowing binding to be unambiguously detected despite the fact that there was only ϳ10 pN force acting at the single attachment point at the apex. d, overlay drawing of the photomicrographs shown in b-c. The bars represent 5 m.
the specificity of the adhesion probabilities measured from the micropipette binding tests.
The Two-dimensional Kinetic Rates Depended on the CD16 Membrane Anchor and This Dependence Varied with Ligand-The micropipette method quantifies the dependence of adhesion probability on contact time and densities of the receptors and ligands, as exemplified in Fig. 3b. To allow for direct visual comparison of the kinetic rates and binding affinity, the mass action effect, manifesting itself in Fig. 3b as an upward or downward shifting of the curves depending on the receptor and ligand densities (m r and m l ), must be eliminated. This is achieved by a simple transformation of Equation 2 into Equation 3.
It is evident that the far right-hand side of Equation 3 depends only on the binding affinity (K a ) and the reverse rate constant FIG. 4. Demonstration of binding specificity. a, the adhesion probability varied with the presence or absence of the interacting molecules on the apposing cell surfaces. When CD16a-TM or CD16a-GPI was expressed on the CHO cells along with RbIgG (solid bars) or mIgG2a (hatched bars) coated on the red blood cells (RBC), high adhesion probabilities were observed. In contrast, when either no receptor (Ϫ) or an irrelevant receptor (␣ IIb ␤ 3 ) was expressed on the CHO cells, low adhesion probabilities were observed for the same IgG coating on the red cells. Similarly, when an irrelevant protein (BSA, open bars) was coated on the red blood cells, the adhesion probability was reduced to low levels for the same CD16a-expressing CHO cells. b, the adhesion could also be inhibited by treating the cells with blocking agents. The addition of the conditioned media from hybridoma secreting anti-CD16 mAb CLBFcgran-1 (ϳ10 g/ml antibody) greatly reduced the adhesion probability. Similarly, the addition of the conditioned media of soluble CD16a secreting CHO cells (ϳ10 g/ml sCD16a) decreased the adhesion probability to low levels. In contrast, conditioned media from hybridoma secreting an irrelevant mAb X63 and of CHO cells secreting an irrelevant soluble molecule B7 did not block binding. Each bar in a and b represents mean Ϯ S.E. of data from 2 to 4 series of 50 -200 tests each at a contact time of 5 s. In these control experiments the densities of CD16a-TM and CD16a-GPI were not matched, nor were those of RbIgG and mIgG2a. from human (a), rabbit (b), and mouse (c). The adhesion probability data, exemplified in Fig. 3b, were transformed according to Equation 3 (the logarithm of the reciprocal of 1 Ϫ P a was taken and divided by m r and m l ) and then plotted against the contact time t. By normalizing the ordinate with respect to m r and m l , the family of curves corresponding to the same pair of interacting molecules but different densities of receptors and ligands collapse. Furthermore, the dependence of binding affinity K a and reverse rate k r on molecular identity becomes apparent. The higher plateau level indicates a greater K a ; and a shorter half-time (t1 ⁄2 , indicated) reflects a faster k r . The theoretical predictions (Equation 3, curves) were fit to each data set (points, mean Ϯ S.E.) to evaluate the K a and k r values, which are listed in Table I. The number of cell pairs examined were 43, 17, 20, 22, 11, and 12 for the CD16a-TM-hIgG, CD16a-GPI-hIgG, CD16a-TM-RbIgG, CD16a-GPI-RbIgG, CD16a-TM-mIgG2a, and CD16a-GPI-mIgG2a curves, respectively. Each cell pair was repeatedly tested 50 -200 times to obtain a stable adhesion probability estimate.

FIG. 5. Transformed binding curves of CD16a-TM-and CD16a-GPI-expressing CHO cells interacting with red blood cells coated with IgGs
(k r ) of the interacting molecules, provided that the contact area (A c ) is kept constant, as in the present study. Thus, this transformation collapses a family of P a versus t curves for the same receptor-ligand pair into a single curve on the transformed ordinate. Only when different interacting molecules are tested will the transformed binding curves shift, allowing for a direct visualization of the distinct K a and k r values.
The transformed binding curves are shown in Fig. 5 for each of the two human CD16a isoforms interacting with IgG from each of the three species. It is evident from Fig. 5a that for hIgG the CD16a-GPI binding curve achieved a Ͼ2-fold higher level than the CD16a-TM binding curve at steady state (contact time t Ͼ 10 s). To test further this result, two additional ligands, IgGs from rabbit and murine, were examined, since they were known to interact well with human CD16. A similar trend was obtained when RbIgG was tested (Fig. 5b). When mIgG2a was tested, however, an inverted result was observed. The CD16a-TM binding curve reached a ϳ7-fold higher equilibrium level than the CD16-GPI curve (Fig. 5c). The plateau level of the transformed binding curve provides a direct measure for the binding affinity because the far right-hand side of Equation  3 approaches A c K a as t approaches infinity.
The values for the reverse rate constant can also be visually compared from the transformed binding curves. It follows from Equation 3 that k r is equal to ln2 divided by the time required for the curve to reach half-maximum, t1 ⁄2 , as shown in Equation 4.
From the data shown in Fig. 5, the t1 ⁄2 values appear to be comparable for the two CD16a isoforms interacting with the same IgG species, suggesting their similar reverse rate constants. Whereas the transformation given by Equation 3 makes the effect of the membrane anchor on kinetic rates and binding affinity readily visible without analysis, fitting Equation 3 to the data (Fig. 5) allows for quantitative evaluation of k f , k r , and K a . The values of these two-dimensional kinetic properties are presented in Table I, which quantitatively confirm the visual observations.
Thus, the micropipette data revealed the membrane anchor of CD16a as a major determinant of its kinetic properties; it altered the forward, but not the reverse, rate constant and thereby also altered the binding affinity. Significantly, this effect varied with the ligand species, with the GPI-anchored CD16a having higher affinity for hIgG and RbIgG but lower affinity for mIgG2a than its TM-anchored counterpart. This "flipping" phenomenon is interesting and may be useful in understanding the underlying mechanism (see "Discussion").
The CD16 Membrane Anchor Effect on Binding Affinity and Its Ligand Dependence Was Also Found in Three-dimensional Binding Studies-The preceding section represents the first application of the newly developed micropipette method for determining two-dimensional kinetic rates in a comparative study of the ligand-binding property of a cell surface receptor. Although our method has been carefully validated (28), it is important to further verify it by comparing its predictions with those derived from more established methodologies. For this reason, three-dimensional binding studies were conducted to address the following questions. Was the CD16 membrane anchor effect on ligand binding seen in Fig. 5 a true indication of the structure-function relationship of the interacting molecules, a real reflection of the two-dimensional experimental conditions, or a mere artifact of the micropipette method?
The affinities of an anti-CD16 mAb CLBFcgran-1 in fluidphase for both CD16a-TM and CD16-GPI on CHO cells were first determined by Scatchard analysis (Fig. 6). The membrane anchor effect (in this case using the Fab fragment of CLBFcgran-1 as ligand) again is apparent in Fig. 6, with the GPIanchored CD16a exhibiting a higher affinity than the TManchored CD16a.
Because CD16 is a low affinity receptor for monomeric IgG, measuring CD16-IgG interactions by the Scatchard method is rather difficult and results so obtained may not be reliable. We therefore adapted a competitive inhibition protocol using nonlinear curve-fitting based on Equation 1. Representative data comparing the two human CD16a membrane anchor isoforms are presented in Fig. 7, a-c, using hIgG, RbIgG, and mIgG2a as respective competitors. It is evident from Fig. 7, a and b, that CD16a-GPI had higher affinities than CD16a-TM for hIgG and RbIgG, as the same ligand concentration inhibited the CLBFcgran-1 binding to a greater extent. The fact that CLBFcgran-1 bound with a higher affinity to CD16a-GPI than CD16a-TM (Fig. 6) should lead to less, rather than more (as was the case), inhibition by hIgG or RbIgG. The situation in Fig. 7c is less obvious. The lower curve in this panel corresponds to CD16a-TM, which has a lower affinity than CD16a-GPI for CLBFcgran-1 (Fig. 6). Hence it would be easier to be inhibited even if mIgG2a were to bind both CD16a membrane isoforms with the same affinity. However, fitting Equation 1 to the data in Fig. 7c revealed that CD16a-TM indeed had a higher affinity than CD16a-GPI for mIgG2a, as was in the two-dimensional case. The three-dimensional affinity results are summarized in Table II.
CD16a-␥ and CD16a-Chimeras Behaved Similarly to CD16a-TM-In contrast to CD16a-GPI which is expressed on the CHO cell surface by itself, CD16a-TM has to be coexpressed and complexed with a ␥-␥ dimer (Fig. 1a). It has been suggested that the associated ␥ chain could enhance the ligand binding affinity of Fc␥ receptor (26). To define further the role of the ␥ chain, we performed parallel experiments to measure the three-dimensional affinities for various ligands of two chimeric molecules, CD16a-␥ and CD16a-. The chimeras were created by fusing the extracellular domain of human CD16a with the TM and cytoplasmic domains of either the rat ␥ chain or the human chain (Fig. 1c). The results of these experiments are The kinetic rates were calculated by fitting Equation 3 using a Levenberg-Marquardt nonlinear 2 error minimization method to the data shown in Fig. 5 panels a-c for hIgG, RbIgG, and mIgG2a, respectively. The S.E. were calculated from the data variance.

CD16a
IgG presented in Table II. It can be seen that the chimeric CD16a-␥ and CD16a-behaved similarly to the TM-anchored CD16a but differently from its GPI-anchored counterpart. CD16b Behaved Differently from CD16a-GPI-There have been conflicting reports on whether the small differences in the ectodomain of CD16 isoforms (Fig. 1, a and b) can influence binding (23,37). Comparison and interpretation of these data are often obscured by other differences that were also present, such as the membrane anchor and cellular background. To delineate the effect on binding of the six amino acid substitutions and the loss (or gain) of one glycosylation site in the extracellular domain of CD16b NA1 (or CD16b NA2 ), their threedimensional affinities were compared with that of CD16a-GPI. All CD16 isoforms were expressed on CHO cells to obtain the same cellular background. Comparisons were made in side-byside experiments using the same preparation of reagents to minimize inter-experimental variations. The results of these experiments are presented in Table II, which show that both alleles of CD16b bound to hIgG and RbIgG with 2 orders of magnitude lower affinities than CD16a-GPI.
CD16a-TM and CD16a-GPI Showed No Detectable Differential Glycosylation-A possible mechanism for the differing ligand binding kinetics of the CD16a isoforms may be their differential glycosylation. TM-and GPI-anchored molecules may have different resident times in the Golgi and other processing apparatus, which may affect their post-translational modifications. To investigate this possibility, various forms of CD16 purified from CHO cells were analyzed by SDS-PAGE (Fig. 8). The glycosylation differences were evident in the three GPI-anchored CD16 molecules that have the same number of amino acids in their protein backbone. The difference in the mobility of CD16b NA1 (47 kDa), CD16b NA2 (56 kDa), and CD16a-GPI (51 kDa) correlates well with the number of glycosylation sites (4, 6, and 5, respectively, cf. Fig. 1). The faster mobility of CD16a-GPI than CD16a-TM (55 kDa) can be accounted for by the deletion of the TM and cytoplasmic domains (ϳ5.4 kDa, calculated from 45 amino acids) and the addition of the GPI moiety (ϳ1 kDa, calculated from fatty acid compositions) of the former isoform. These data suggest that there is no major difference in glycosylation between the CD16a-GPI and CD16a-TM.
Anti-CD16 mAb VEP13 Reacted Strongly with CD16a-TM but Weakly with CD16a-GPI-Another possible mechanism un-  Table II. FIG. 7. Competitive inhibition curves for determining the binding affinities of CD16a-TM and CD16a-GPI for IgG of three species. CHO cells expressing CD16a membrane isoforms were allowed to bind 125 I-labeled CLBFcgran-1 Fab in the presence of varying concentrations of hIgG (a), RbIgG (b), or mIgG2a (c). Points are data presented as mean Ϯ S.D. of triplicate wells, whereas each curve is a 2 fit to that set of data using Equation 1. By fitting the entire curve, multiple data points (each of which by itself can be used to calculate a K a l ) are utilized in the calculation of the fitting parameter, the threedimensional binding affinity of CD16a for IgG, K a l . The K a l values so obtained should be more accurate and reliable, as they reflect the average prediction from all data points. This also allows for determination of K a l without competition to the IC 50 point, as in the CD16a-GPI curve in c. For each IgG ligand, at least two experiments were performed in parallel with two CD16a membrane isoforms to minimize inter-experimental variations; the data shown are representative. The mean and S.D. of the K a values of each ligand to both CD16a-TM and CD16a-GPI are listed in Table II. derlying the observed anchor effect is a conformational difference between the two molecular isoforms. To test this hypothesis we screened a panel of anti-CD16 mAbs using flow cytometry to see if any changes in expression of antigenic epitopes could be detected. Among the mAbs tested, VEP13 reacted differently to the CD16a isoforms. It exhibited a concentration-dependent and saturable binding to CD16a-TM (Fig. 9, left column), and its fluorescence histogram matched that of another anti-CD16 mAb 3G8 at the saturation concentrations (Fig. 9, 1:2 and 1:10 dilutions). In contrast, not until its concentration reached the level at which reactivity to CD16a-TM was saturated did VEP13 begin to show some weak binding to CD16a-GPI (Fig. 9, right column), and its fluorescence histograms displayed substantial leftward shifts from those of 3G8 toward much lower levels of expression even at the highest concentration tested (Fig. 9, 1:2 and 1:10 dilutions). Our limited VEP13 reagent (ascites) did not allow us to measure its binding affinity for the CD16a isoforms. Nevertheless, Fig. 9 shows a qualitative trend opposite that seen in Fig. 6. Thus, the dependence of the anchor effect on the molecule to which the CD16a isoforms bind was seen not only with IgG ligands but also with mAbs against CD16a.
Several other anti-CD16 mAbs, including CLBFcgran-1, recognized both isoforms of CD16a in a manner similar to 3G8 (data not shown). Thus, the VEP13 epitope on the ectodomain of CD16a was down-regulated after its membrane anchor had been changed from TM to GPI. The weak reactivity with CD16a-GPI was not due to the GPI anchor per se, since VEP13 is known to react with both the TM-anchored CD16a and the GPI-anchored CD16b, respectively, expressed on NK cells and polymorphonuclear leukocytes (38). It can also inhibit binding of immune complexes and 3G8 to CD16 and block rosette formation of RbIgG-opsonized ox red blood cells with CD16-expressing leukocytes (39,40). Monoclonal antibody 3G8 identifies the ligand-binding site of CD16 (41). Thus, the epitope recognized by VEP13 is likely to be proximal to the ligand bind site of CD16a-TM. DISCUSSION A major goal of the present work was to compare the twodimensional binding kinetics and affinities of two human CD16a membrane anchor isoforms for the Fc domain of IgG. Since adhesion of Fc receptor-expressing leukocytes to IgGcoated targets is an initiating step for many immune responses, the determination of CD16-IgG kinetic rate constants is important in both biological and clinical settings. Unlike typical hormone receptors that bind soluble ligands (i.e. three-dimensional binding kinetics), adhesion receptors bind membranebound ligands (i.e. two-dimensional adhesion kinetics). Clearly, it is the two-dimensional, rather than the three-dimensional, kinetic rate constants that are most relevant to physiological situations such as adhesion of a CD16a-expressing NK cell or macrophage to an antibody-coated target cell. Although twodimensional adhesion kinetic rates are of obvious importance, their determination has not been possible experimentally until very recently due to a lack of methodology. Our micropipette method is one of only three existing approaches in the literature available to measure two-dimensional kinetic rates, compares favorably to the other two flow-based methods (29 -31), and offers several technical improvements (28). The results summarized in Table I expand our original data on the twodimensional kinetic rates of CD16a-TM binding to hIgG and RbIgG (28).
By applying this newly developed method, we attempted to address the following biological questions. Did the membrane anchor (including the associated subunits) influence the kinetic rates? If so, what might be the mechanisms causing the observed changes? To address these questions, we isolated the effects of the extracellular domain and the membrane anchor using the lipid-anchored CD16a-GPI construct. CHO cell transfectants were used to obtain a uniform cellular background across the CD16 membrane isoforms expressed. Under these conditions, the membrane anchor effect was clearly revealed by the micropipette experiment (Fig. 5). To the best of our knowledge, this is the first experimental demonstration of the membrane anchor effect on the kinetic rates.
We also measured three-dimensional binding for the same interacting molecules (Fig. 7) in order to elucidate the mechanism underlying the observed anchor effect. Because much work is available in the literature on three-dimensional CD16-IgG interactions, comparison to those helps validate our two-  Fig. 6). The low affinities for IgGs of the same receptors were calculated by fitting Equation 1 to data exemplified in Fig. 7 using a Levenberg-Marquardt nonlinear 2 error minimization method. Data represent the mean Ϯ S.D. of (N) experiments. ND, not determined.

Receptor
CLBFcgran-1 hIgG RbIgG mIgG2a . Cells were stained for flow cytometry using anti-CD16 mAbs 3G8 (thick curves), VEP13 (shaded areas), or an irrelevant mAb X63 (thin curves). The concentrations for each row were the indicated dilution of culture media from hybridoma secreting 3G8 or of ascites containing VEP13. The secondary antibodies were FITCconjugated anti-mouse IgG (for 3G8) or anti-mouse IgM (for VEP13) goat polyclonal antibodies. Experiments were repeated several times, and the results shown are representative. a higher three-dimensional binding affinity for monomeric hIgG of CD16a on NK cells (3-10 ϫ 10 6 M Ϫ1 ) than of CD16b on neutrophils (value not determined). The former is therefore referred to as an intermediate affinity receptor, whereas the latter is considered as a low affinity receptor. It was not demonstrated, however, to what degree this differing affinity was due to the differences in the extracellular domain, in the membrane anchor, or in the cellular background of the two membrane isoforms. On the other hand, Tamm et al. (37) reported a similar three-dimensional avidity of CD16a-GPI and CD16b NA2 , both expressed on a human embryonic kidney cell line 293, for heat-aggregated hIgG1 (ϳ20 ϫ 10 6 M Ϫ1 ). These authors suggested that the minor differences in the ectodomains of the two molecules (Fig. 1, a and b) would not affect their affinity, although their data showed an ϳ2-fold higher avidity of CD16a-GPI than CD16b for dimeric hIgG1 (3.7 and 2.1 ϫ 10 6 M Ϫ1 , respectively). We found that in our CHO cell system using hIgG and RbIgG, CD16a-GPI showed consistently higher affinity than CD16a-TM in both two-dimensional (Table I) and three-dimensional (Table II) binding studies. Furthermore, repeated side-by-side experiments reproducibly showed that the order of three-dimensional affinities for hIgG, hIgG1, and RbIgG was CD16a-GPI Ͼ CD16a-TM Ͼ CD16b NA1 ϳ CD16b NA2 (Table II and data not shown). The discrepancies in the absolute values of three-dimensional affinities measured by the different laboratories might be due to cell type-specific glycosylation of CD16 (25). However, our data were obtained using the same CHO cells to express various CD16 membrane isoforms. Our results appeared to indicate that in comparison to CD16a, the variations in the extracellular domains of the two CD16b alleles (Fig. 1, a and b), although very small, did significantly reduce their affinities for hIgG and RbIgG (Table  II) and for hIgG1 (data not shown). This is not surprising, as two independent single nucleotide polymorphisms (resulting in amino acid changes Leu-48 to Arg-48 or His-48 and Val-176 to Phe-176) in CD16a have been reported to alter the affinity of IgG binding (42,43). The CD16a-TM and CD16a-GPI molecules used in the present study do not have these polymorphic variations (27), as confirmed by sequencing during subcloning of the CD16a-TM cDNA into the pcDNA3 vector.
Possible Role of the Associated Subunits-Miller et al. (26) suggested that the associated ␥ chain could enhance the ligand binding affinity of Fc␥R. They reported at least an order of magnitude higher three-dimensional affinity of CD16a-TM than CD16a-GPI for mIgG2a. The affinity of CD16a-TM for mIgG2a obtained by these authors (12 ϫ 10 6 M Ϫ1 ) is much higher than our value, which may be due to the differences in the cells and/or the ␥ chain used. Miller et al. (26) used monkey kidney COS cells to express transiently human CD16a-TM in association with human ␥ chain. By comparison, we used stably transfected CHO cells to express human CD16a-TM in association with rat ␥ chain. Furthermore, Miller et al. (26) employed direct Scatchard analysis, whereas we used indirect competitive inhibition to measure affinity. Nevertheless, we found the same trend as Miller et al. (26) that CD16a-TM had higher affinity than CD16a-GPI for mIgG2a ( Fig. 7c and Table II).
However, we found that this effect was ligand-dependent. For CLBFcgran-1, hIgG, and RbIgG, the trend inverted, with CD16a-TM having lower affinity than CD16a-GPI (Figs. 6 and  7, a and b). These three-dimensional results were supported by those of the two-dimensional micropipette experiments (Fig. 5), which involved direct visualization of over 20,000 controlled single cell pair adhesion tests. In addition, affinity measurements showed that the chimeric molecules CD16a-␥ and CD16abound similarly to CD16a-TM but differently from CD16a-GPI for all ligands tested, including IgGs from three species and a mAb (Table II). Thus, our findings indicate that the role of the associated subunit, if it is indeed the cause of the anchor effect, is not limited to the ␥ chain but also includes the chain. Moreover, their putative role is not to enhance, but rather to alter, the ligand binding affinity.
The Difference in Ligand Binding Kinetic Rates and Affinity of CD16 Isoforms Cannot Be Explained by Their Differing Diffusivities-The GPI-anchored CD16 molecules (including both alleles of CD16b and the CD16a-GPI) exhibit a few folds faster translational diffusion on the CHO cell membrane than TManchored CD16a as determined by preliminary fluorescence recovery after photobleaching measurements (data not shown). In addition, the GPI anchor is likely to provide more flexibility to the ectodomain of the receptor, which increases its rotational diffusion coefficient, a parameter more relevant to enhancing two-dimensional binding than the translational diffusion coefficient. One should therefore consider whether the faster binding of CD16a-GPI than CD16a-TM to hIgG and RbIgG could be explained by the diffusional difference of the two membrane anchor isoforms. However, the following three lines of reasoning argue against this explanation and allow us to rule it out as the cause for the different kinetic rates of CD16 anchor isoforms.
First, faster diffusion of CD16a-GPI cannot explain the observed effect of the GPI anchor on the forward, and the lack thereof on the reverse, rate constants ( Fig. 5 and Table I).
Existing theories have shown that diffusion influences k f and k r similarly but not K a ϭ k f /k r , since the diffusion effects on the two rates cancel each other in the ratio (44,45). Second, the anchor effect was seen not only in the two-dimensional micropipette experiment, but also in the three-dimensional binding assays (Figs. 6 and 7 as well as Table II) where identical but soluble ligands and the same CD16-expressing CHO cells were examined. Diffusion should not affect the three-dimensional results not only because it was the ratio K a , not k f or k r separately, that was measured but also because diffusion is unlikely to be the rate-limiting step in the two-step binding process (the other step being intrinsic reaction). The diffusivity of proteins in fluid phase (Ͼ10 m 2 /s) is usually orders of magnitude greater than cell surface proteins (45). Finally, the diffusion mechanism cannot explain the inversion of the anchor effect; the faster diffusing CD16a-GPI bound with slower forward rate and lower affinity to mIgG2a than CD16a-TM (Figs. 5c and 7c as well as Tables I and II). This negative correlation between diffusion coefficient and forward rate/binding affinity provides direct and definitive experimental proof for the inability of the differing diffusivities to account for the CD16a anchor effect on kinetic rates and affinity.
We should point out that in the micropipette experiment accumulation of receptors in the contact area by lateral diffusion is unlikely. The diffusion coefficients for various CD16 isoforms on CHO cells are of the order of 0.01 m 2 /s (data not shown). The longest contact time in the micropipette experiments was 20 s, which was far from sufficient for the receptors to accumulate in an apparent contact area of ϳ3 m 2 . Furthermore, only a few bonds were formed in an adhesion produced by the controlled contact in the micropipette experiments (28), which is a negligibly small number comparing to the hundreds and thousands of receptors in the contact area. This will not generate any appreciable density gradient of free receptors to drive them to diffuse into the contact area.
The Difference in Ligand Binding Kinetic Rates and Affinity of CD16 Isoforms Cannot Be Explained by Their Differing Orientations and Lengths-The lack of TM and cytoplasmic domains as well as the associated subunits of the CD16a-GPI may alter the orientation of its extracellular domain. Moreover, the GPI moiety may extend the Fc binding epitope further outward relative to the glycocalyx (Fig. 1). The length of a receptor has been demonstrated to influence its ability to support adhesion at 4°C (but the effect diminished at higher temperatures) (46) and under flow (but not static) conditions (47). It is thought that a longer and more flexible molecule can explore larger space above the membrane and assume more spatial configurations. This lengthens the interaction range of the receptor, thereby facilitating its effort to find the ligand when it is surface-linked (48). However, although both orientation and length can influence ligand binding kinetic rates and affinity, this effect should be qualitatively monotonic for all ligands. Therefore, our observation that the GPI anchor increases affinity for human and rabbit IgGs but decreases affinity for murine IgG2a allow us to exclude orientation and length as possible causes for the anchor effect.
The Difference in Ligand Binding Kinetic Rates and Affinity of CD16 Isoforms Cannot Be Explained by Their Differential Distribution and Clustering-Some GPI-anchored proteins have been suggested to be clustered in glycosphingolipid and cholesterol-enriched domains, e.g. caveolae (49). CD16a-GPI might appear to bind better than CD16a-TM in the threedimensional assay should the former isoform be functionally clustered, since binding of aggregated soluble ligands to receptor clusters might result in an apparently higher avidity.
Similarly, being distributed in different membrane domains might potentially influence the two-dimensional binding properties of CD16a-GPI. The forward rate k f and binding affinity K a measured from the micropipette assay are lumped with the contact area A c . The true or functional contact area A c was not measured but should be proportional to the apparent contact area directly visible under the light microscope (Fig. 2b) (28). The CHO cell surface displays extensive roughness; thus only the "hills," not "valleys," of the membrane folds are likely to be part of the functional A c . 3 Since the same CHO cells were used to express CD16a regardless of the anchors and the same red blood cells were used to present IgG regardless of the species, A c would be a constant if the apparent contact area was kept constant, as was the case in all of our experiments. Thus, not knowing the value of A c or of which membrane microdomains it was composed should not affect conclusions based on relative comparisons, provided that the two CD16a membrane anchor isoforms were similarly distributed in the contact area. However, while some adhesion molecules (e.g. ␤ 2 integrins) are more or less uniformly distributed on the cell surface, others (e.g. L-selectin) are known to localize on the microvillous tips (50). Because the latter molecules are present at higher densities on A c , their k f and K a values will be overestimated by a calculation that assumes a uniform molecular distribution. Thus, CD16a-GPI would appear to bind better than CD16a-TM should CD16a-GPI be differentially enriched on the hills (accessible area) or CD16a-TM be differentially enriched on the valleys (inaccessible area) of the membrane folds.
Although the arguments in the preceding paragraphs may seem consistent with the binding pattern seen in the human and rabbit IgG experiments, they cannot explain the inversion of that trend observed in the mIgG2a experiment. The differential distribution and clustering of receptors should not flip when IgGs from different species were used as ligands to assay kinetics. Moreover, differential distribution of GPI-and TManchored CD16a should not affect the three-dimensional results, since the soluble ligands should be able to access any membrane domains, and the calculation is based on average measurements over the entire cell surface, not particular compartmentalized domains. We thus conclude that the differences in ligand binding kinetic rates and affinity of CD16a isoforms are not due to their differential distribution and clustering.
No Major Differential Glycosylation of CD16a Isoforms Can Account for the Anchor Dependence of Ligand Binding Kinetic Rates and Affinity-GPI-anchored proteins may have different resident time and may hence be processed differently in the Golgi apparatus than TM-anchored proteins, resulting in different carbohydrate modifications. However, analysis of CD16 isoforms purified from CHO cells using SDS-PAGE did not reveal major differences in N-linked glycosylation. The ϳ4-kDa higher molecular mass of CD16a-TM than CD16a-GPI (Fig. 8) can be well accounted for by the 45 amino acid TM and cytoplasmic domains (ϳ5.4 kDa) of the former isoform and the GPI moiety (ϳ1 kDa) of the latter isoform. Of course, more extensive analysis is required to test whether site-specific differential glycosylation exists between the two CD16a membrane anchor isoforms and, if so, whether it is the cause of the observed anchor effects on kinetic rates and binding affinity.

Could the Difference in Ligand Binding Kinetic Rates and Affinity of CD16 Isoforms Be Explained by Their Differing
Conformations-A consequence of the possible concentration of CD16a-GPI in glycosphingolipid-enriched membrane microdomains may be its being surrounded by different neighboring molecules. It is conceivable that neighboring molecules of CD16a-GPI could affect its ligand binding. But to affect binding in one way with hIgG, RbIgG, and mAb CLBFcgran-1 but in an opposite way with mIgG2a and mAb VEP13 would most likely require CD16a-GPI to associate with the neighboring molecules. Such an association would most likely have to be in sufficiently close proximity to render a conformational change of the receptor, resulting in variable accessibility by different ligands and mAbs. So far, only the myeloid cell-specific integrin ␣ M ␤ 2 has been reported to associate with the GPI-anchored CD16b (51). CHO cells do not express ␤ 2 integrins. By comparison, in order for it to be expressed on the CHO cell surface CD16a-TM must be associated with the ␥ chain (Fig. 1a). Such an association has been shown to alter the binding of not only CD16a but also CD64 (Fc␥ receptor I) (26).
Finally, an alternative and perhaps simpler hypothesis may be that the differing membrane anchors themselves, when they are inserted into the cell surface, yield such a conformational difference. The view that different conformations of the two CD16a isoforms may be the mechanism underlying their different kinetic rates and binding affinity is supported by the following resemblance between our observations and those commonly accepted as valid evidence for conformational change in integrins. Certain "activation-reporter" mAbs bind integrins only after they have been converted from resting to activated states, which is interpreted as a conformational change that allows for expression of antigenic epitopes that are specific to the activated conformer (e.g. see Ref. 52 and references cited therein). Another characteristic of conformational changes of integrins is changes in their abilities to bind soluble ligands and to mediate cell adhesion. Here we have also identified a mAb (VEP13) that reacts strongly with CD16a-TM but only weakly with CD16a-GPI, suggesting that the epitope detected by mAb VEP13 is substantially down-regulated after the molecule's anchor has been changed from TM to GPI. In addition, the abilities to bind soluble ligands (and the mAb CLBFcgran-1) and to mediate cell adhesion are different for the two CD16a membrane isoforms, which are caused by their different kinetic rates and binding affinities for ligands (and for mAb CLBFcgran-1). Similarly, Kukulansky et al. (53) reported that following anchor cleavage by phospholipase C, the reactivity of the solubilized Thy-1 with several mAbs is lost, and its reactivity with polyclonal anti-Thy-1 antibodies is markedly decreased. These authors interpreted their finding by a GPI anchordependent conformational change of the Thy-1 molecule. Thus, our data strongly suggests that replacement of polypeptide anchor with GPI anchor resulted in a conformational change of CD16a.
The diversity of Fc receptors in both structure and function has long been appreciated (22). The ability of Fc receptors to bind ligand has been shown to be influenced by a variety of structural variations. In this study we measured both twodimensional kinetic rates and three-dimensional affinities of Fc␥RIII membrane isoforms for various ligands, and we showed that the membrane anchor had an effect on these binding properties, which is likely caused by a conformational difference between the two CD16a membrane isoforms. These findings provide insights into the biological significance of distinct anchors of cell surface proteins.