Influence of Affinity and Antigen Density on Antibody Localization in a Modifiable Tumor Targeting Model¹

Significant progress has been achieved in the use of radiolabeled mAbs³ for the imaging and therapy of malignancy. At present, four murine antibody-derived radiopharmaceuticals are Food and Drug Administration-approved for the imaging of tumors with other promising agents for radioimmunoimaging and therapy in the regulatory pipeline(1). Despite this achievement, controversy still persists regarding several fundamental issues relating to imaging and therapy with antibody-based radiopharmaceuticals. One important but unresolved question is whether increased antibody affinity is advantageous for the targeting of tumors (2, 3, 4).

Various provocative mathematical models have been used to probe specific theoretical issues relating to antibody targeting, including the influence of affinity on binding (5, 6, 7, 8, 9, 10). Fundamental behavior of antibodies has also been investigated in experimental in vitro studies (9, 10, 11). Nonetheless, the extrapolation of these findings to clinical situations has proven uncertain (12, 13, 14), and experimental validation in animal studies is needed. In fact, a limited number of in vivostudies have been performed to investigate factors influencing antibody binding (8, 13, 15, 16, 17, 18), often with conflicting or unclear conclusions. It has proved difficult to identify antibodies with significantly different affinity but with the same class and subclass that bind to an identical epitope. Another frequent limitation of in vivo models is a restricted ability to modify relevant variables. For example, the influence of antigen density on antibody targeting has not been extensively studied in experimental models,largely because it is difficult to systematically vary this parameter in tumor xenografts. This is regrettable in that the density of an antigen determines the ability of a multivalent antibody to engage multiple antigenic sites simultaneously. In contrast with“affinity,” which measures the binding of an individual Fab-binding site and antigen, the overall strength of interaction between a multivalent antibody and antigen is termed“avidity” (19). This parameter is of paramount importance in predicting antibody binding in a given context(9, 10, 11).

The goal of this paper is to investigate the relationship between antibody affinity, antigen density, and antibody targeting by applying a highly modifiable, well-controlled antibody system to in vitro and in vivo experimental models. For this purpose, we have chosen to study antibodies that bind to the well-characterized chemical hapten Ars on the basis of the spectrum of affinities available and the ability to modify the density of the antigen target. Investigations were initially performed in vitro using ELISA and bead-binding assays. This system was then extended into an in vivo targeting model using radiolabeled antibodies and an artificial “tumor” implanted s.c. in mice where variables relating to the immunoglobulin and antigen could be rigorously controlled. The efficacy of targeting was assessed by the serial imaging of live animals on a gamma-scintillation camera with quantitative image analysis and by postmortem tissue counting.

36-65 and 36-71 are IgG1 mouse mAbs that bind the chemical hapten Ars and are encoded by the same heavy- and light-chain V(D)J regions(20). mAb 36-65 uses germline V regions and has a K_a of 2.5 × 10⁵ m⁻¹ as determined by fluorescence enhancement, whereas mAb 36-71 has undergone somatic mutation and has a K_a of 4.5 × 10⁷ m⁻¹ (21). Subcloned hybridoma cell lines were used to make ascites in pristane-primed SCID mice, which lack potentially confounding endogenous immunoglobulins. To avoid bias that might be introduced if the antigen were used to purify high- and low-affinity antibodies, pooled ascites from each of the cell lines were purified by protein A or G affinity chromatography (HiTrap Protein A or G 1-ml columns; Pharmacia LKB Biotechnology, Piscataway,NJ) according to standard protocols. Purified MOPC 21, used as a murine IgG1 control, was obtained from a commercial vendor (The Binding Site,Birmingham, United Kingdom).

Attachment of the hapten Ars to carrier BSA at specific substitution ratios was performed as described previously(22), and the resultant Ars-BSA was stored at −70°C until use. For the purpose of the current studies, antigen with a ratio of 1.8 Ars molecules per BSA molecule was selected for use as a HD antigen, whereas antigen with a 0.15 Ars:BSA molar substitution ratio was used as a LD antigen. Unconjugated BSA served as an Ars-negative control target.

Interaction of antibody affinity and antigen density was illustrated by ELISA based on the variation of a technique published previously(22). Binding of the specific antibodies (mAbs 36-65 and 36-71) to HD and LD antigen plates was studied in duplicate. Unless otherwise indicated, dilutions were performed in a solution of 1 mg of BSA/100 ml of PBS, incubations were at 37°C, and volumes of 50μl/well were used. Plates were washed three times between each reagent (Microwash II; Skatron Instruments, Sterling VA) with PBS containing 0.05% Tween 20.

Antigen was diluted in PBS to a concentration of 10 μg/ml and distributed into disposable 96-well ELISA plates (Corning Glassworks,Corning NY). Two h thereafter, plates were washed, blocked with a solution of 1 mg BSA/100 ml of PBS (100 μl/well) and stored at 4°C until used. At that time, plates were rewashed and Ars-binding antibodies, at concentrations of 20 μg/ml, were serially diluted(1:2) across the plate and incubated for 1 h. Alkaline-phosphatase conjugated goat antimouse γ₁ (Southern Biotechnology Associates, Inc., Birmingham AL) at a concentration of 1μg/ml was then added to the washed plates with a repeat washing 2 h later and the addition of substrate (ρ-nitrophenyl phosphate, disodium) dissolved in a bicarbonate buffer [0.001 m MgCl₂/0.05 mNa₂CO₃ (pH 9.8)]. Absorbance of the plates was read at 405 λ, after color developed(Multiskan MS; Labsystems).

For the in vitro and in vivo bead studies,antibodies (mAbs 36-65, 36-71, and MOPC 21) were iodinated using the Iodogen method (Pierce, Rockford Il; Ref. 23) under mild conditions with resultant specific activity of between 3–5 μCi/μg. Typically, 150 μCi of [¹²⁵I]NaI were incubated with 24 μg of antibody in a volume of 100 μl for 40 min. Free iodine was then separated from the iodinated proteins by size-exclusion chromatography over a Sephadex G-25 column (Pharmacia,Piscataway NJ). Typical reaction yields of 45–50% were obtained and trichloroacetic acid precipitation revealed 93–98%protein-bound activity.

HD and LD Ars-BSA and control BSA lacking Ars were each conjugated to commercially available, chemically activated beads (Reacti-Gel HW-65F;Pierce, Rockford IL; nominal diameter of 32–63 μm) under identical reaction conditions. Twenty-one mg of protein in 5 ml of 0.1 m borate buffer (pH 9.0) were added per 5 ml of beads. After a 72-h incubation (4°C), beads were separated from solution and 1 m Tris buffer was added for 24 h to block remaining active sites. Beads were then spun down and stored in PBS and 0.02%azide at 4°C.

Binding of the three radiolabeled antibodies (mAbs 36-65, 36-71, and MOPC 21) to beads bearing HD, LD, and control antigen was studied using duplicate samples per combination of antibody and antigen. Within the upper chamber of a 0.45-μm cellulose acetate SPIN-X microtube(Costar, Cambridge MA), 0.004 μg of radiolabeled antibody, 5 μl of freshly washed antigen beads, and 200 μl of 1% BSA/PBS were combined and agitated on a rotating platform at room temperature. Seventy-two h later, supernatant was separated from the beads by spinning the microtubes at 1500 rpm for 5 min, rinsing with 300 μl of 1% BSA/PBS,and respinning. Duplicate supernatant and bead samples were counted on a γ counter (Compugamma 1282; LKB Wallac, Turku, Finland), and the mean fraction of bound antibody was calculated.

SCID mice were chosen for these studies because they lack any potentially confounding endogenous immune response (24). Fifty μl of antigen-specific beads (HD or LD) were introduced into the left inguinal region of anesthetized adult female mice by s.c. injection, whereas 50 μl of antigen-negative control BSA beads were introduced into the contralateral right side. Mice were prepared at least 1 week before targeting studies to allow any acute changes surrounding the beads to subside. Thyroid uptake of radioiodine was blocked by the addition of three drops of Strong Iodine (Lugol’s)Solution USP (Regional Service Center, Inc., Woburn, MA) per bottle of drinking water from this time and throughout the duration of the study.

Two separate investigations were performed as described below. The first study characterized the time course of antibody binding and validated choice of an optimal time point for statistical comparison of the groups. The second investigation studied a large number of mice at a single time point after binding had peaked to generate statistically valid comparisons of antibody binding in the different groups.

To determine the time course of antibody binding to antigen, the four permutations of antigen density (HD and LD) and antibody affinity (mAbs 36-65 and 36-71) were studied in groups of four mice each. Mice were injected by tail vein with 2 μg of[¹²⁵I]-labeled antibody. Over the course of 14 days, two mice per each group (one for the final image set) were periodically anesthetized with i.m. Pentobarbital, and imaged for 5 min in ventral projection using a dedicated animal gamma-camera imaging system (Picker Dynacamera 4C; Picker International, Highland Heights,OH; and NucLear MAC computer system, Scientific Imaging, Inc.,Littleton, CO). The camera was peaked on ¹²⁵I using a 20% energy window, with collimation achieved by perpendicularly placed radiography grids (Model 278032; Liebel Flarsheim, Cincinnati, OH). Identically sized ROIs were placed over the specific and control antigen beads, and a ROI-binding ratio was generated on the basis of the ratio of left to right groin counts. Values from duplicate animals were averaged for the initial six time points whereas the solitary value was used for the seventh and final point.

In addition, a single mouse per combination of antibody and antigen density was sacrificed at each of four time points over the 14-day period, and specific and control beads were dissected out of the inguinal areas, weighed, and counted by γ-counter (Compugamma 1282;LKB Wallac, Turku, Finland). A relative binding ratio was calculated based on the ratio of counts bound per g of specific Ars-beads (left groin) divided by those bound per g of nonspecific control beads (right groin).

The effect on the binding of antibody affinity and antigen density was studied in a total of 28 animals consisting of two cohorts of mice,which were pooled to increase sample size. Mice were prepared with specific (HD or LD) and control beads as above, injected with one of three radiolabeled antibodies (mAbs 36-65, 36-71, or MOPC 21), and sacrificed after 10 or 13 days, after binding had generally plateaued. Specific and control antigen beads were removed, weighed, and counted by γ-counter as above. The percent injected dose/g of beads was calculated as was the relative binding ratio between specific and control beads. Relationships between the relative binding ratio,antibody affinity, and antigen density were analyzed by two-way ANOVA.

The present studies contrast the targeting of two murine IgG1 antibodies that bind to the chemical hapten Ars. On the basis of serology and sequencing (21), both anti-Ars antibodies are encoded by the same germline V(D)J genetic elements for their heavy-and light-chain variable regions and bind to the identical epitope. mAb 36-65 is a relatively low-affinity germline-encoded antibody with a true K_a of 2.5 × 10⁵ m⁻¹(21). mAb 36-71 has a markedly higher affinity with true K_a of 4.5 × 10⁷ M⁻¹ as a result of somatic mutation and is among the highest affinity antibodies that have been elicited by this hapten (21).

Binding of mAbs 36-71 and 36-65 to HD and LD antigen is illustrated by ELISA in Fig. 1. Duplicate measurements are shown and were highly consistent. On HD antigen, both antibodies bound effectively, with the higher-affinity mAb 36-71 binding to a quantitatively greater degree than mAb 36-65. On LD antigen, a marked qualitative difference in binding between the two antibodies was noted. mAb 36-71 bound effectively to the LD antigen plate, whereas, for all intents and purposes, the low-affinity mAb 36-65 did not bind. The necessity of a HD target for effective binding of a low-affinity antibody reflects the fact that individual antigen-Fab interactions are weak and highly reversible and require the divalent attachment of both Fabs from each IgG molecule. This higher-avidity interaction can occur only if antigen density is above a certain threshold. High-affinity antibodies, in contrast, are able to bind effectively even with a single antigen-Fab interaction, and therefore are retained well on both HD and LD targets.

To develop a system for evaluating the role of affinity and avidity in the targeting of antibody, we initially coupled HD and LD Ars-BSA to microbeads, as described in “Materials and Methods” and studied antibody-binding in vitro using radiolabeled antibodies. The relative binding of mAbs 36-65 and 36-71 to the beads was compared by determining the percentage of radioiodinated antibody bound to the beads after a 72-h incubation. Mean binding results for the nine permutations of antibody affinity and the antigen target density are shown in Table 1. Duplicate measurements were all within 2.5%. The slightly higher binding of mAbs 36-65 and 36-71 to control BSA beads as compared with the MOPC 21 isotype-matched control antibody was a consistent finding, suggesting weak cross-reactivity of the specific antibodies for the BSA-beads.

Both high- and low-affinity antibodies bound effectively to HD antigen beads with retention of 75.8 and 57.8%, respectively. With respect to the LD target, a much greater fraction of high-affinity mAb 36-71 bound(56.6%) than did the low-affinity mAb 36-65 (17.6%), the latter approaching the level of background binding. These findings are similar to the ELISA observations and reflect the necessity of divalent attachment for successful binding of antibodies of lower affinity.

The ultimate goal of our study was to examine antibody behavior in vivo. The time course of binding was demonstrated over a 14-day period in adult female SCID mice. Fifty μl of antigen-specific beads (HD or LD) were introduced into the left inguinal region by s.c. injection, whereas 50 μl of antigen-negative BSA beads were similarly introduced on the right side. One week later, mice were injected with 2μg of ¹²⁵I-labeled mAbs 36-65 or 36-71. Two animals per group (one in the final session) were periodically anesthetized and imaged (Fig. 2). Identically sized ROIs were placed over the specific and control antigen beads, and a ROI binding ratio was generated based on an average value per time point (Fig. 3). On sequential gamma-camera images (Fig. 2), activity is initially noted in the vascular structures, with prominence of the heart, great vessels,and viscera, especially the liver. Over time, this activity lessens as the blood pool activity decreases. It is interesting to note a mild amount of activity which was localized to the control beads(right groin) in all groups of animals. This nonspecific accumulation may be attributable in part to increased vascular and interstitial volume surrounding the beads as well as to a mild degree of cross-reactivity of the antibodies with control BSA beads, as noted in the in vitro bead study (Table 1). The presence of nonspecific uptake highlights the need to compare specific localization of antibody to antigen-positive beads in the left groin with the mild degree of nonspecific localization in the right groin. No specific localization of low-affinity antibody to LD antigen was noted(column 1), although specific binding was apparent on HD antigen (column 3). High-affinity mAb 36-71 bound appreciably to both LD antigen (column 2) and especially to HD antigen (column 4). Visual impression was quantitated and validated by the ROI analysis (Fig. 3 A).

At each of four serial time points over 14 days, one animal per group was killed, and the extirpated beads weighed and counted. A relative binding ratio was determined based on the ratio of counts per g of specific beads (left groin) divided by counts per g of nonspecific control beads (right groin; Fig. 3,B). Binding of low-affinity mAb 36-65 to the LD antigen was similar to control antigen, with a ratio of 1, whereas this antibody did bind effectively to the HD antigen target. High-affinity mAb 36-71 bound specifically to both the LD and especially to the HD antigen. As a general rule,antibody binding appeared to plateau between 10–14 days, validating the choice of this interval for use in the quantitative analysis described below. mAb 36-71 binding to HD beads was not only the highest for each observation, but continued to increase throughout the duration of the study. This is not surprising, in that the high affinity and avidity of interaction between 36-71 and the beads would dictate that any available antibody would continue to bind to the beads, whereas effectively none would be lost by elution. Visual (Fig. 2) and ROI observations (Fig. 3,A) were consistent with the quantitative postmortem bead measurements (Fig. 3 B), though somewhat blunted in magnitude. Decreased ratios for the imaging data are not surprising based on the degrading effects of overlapping background tissues, limited camera spatial resolution, and the attenuation by overlying soft tissues. These technical factors may explain the inability to obtain correlation between absolute counts and imaging results in a previous study by Andrew et al.(16).

Binding of control, low-, and high-affinity antibodies was also compared in a larger group of 28 mice to enable valid statistical analysis. SCID mice bearing HD or LD antigen, prepared as above, were injected with 2 μg of ¹²⁵I-labeled mAbs 36-65,36-71, or MOPC 21. After 10 or 13 days, mice were sacrificed and specific and control antigen beads were dissected, weighed, and counted. The absolute binding of antibody to the antigen and control beads was expressed as a percentage of injected dose/g of beads(%ID/g; Table 2). A relative binding ratio was also calculated based on the ratio of counts bound/g of specific beads (left groin) divided by those bound to the nonspecific control beads (right groin).

Percent ID/g and absolute localization ratios for the six combinations of antibody and bead-density are listed in Table 2. Binding of MOPC 21 control antibody to the HD or the LD antigen target did not differ appreciably from binding to contralateral control BSA beads, with localization ratios statistically indistinguishable from 1(P = 0.25 and 0.50, respectively). Binding of low-affinity antibody to the LD antigen was also indistinguishable from the control groups and was statistically similar to 1, indicating a lack of effective targeting (all Ps >0.85). Binding of the high-affinity mAb 36-71 to the HD antigen was statistically superior to all other groups (P = 0.0001) with a localization ratio of 10.5 ± 3.0. Binding of high-affinity mAb 36-71 to LD antigen beads was intermediate in strength and similar to the binding of the low-affinity antibody on HD beads (P = 0.80), although different from each of the other four groups (Ps <0.05). The qualitative change in the binding of mAb 36-65 when binding LD versus HD antigen, as contrasted with the relatively similar behavior of mAb 36-71, indicates a statistically significant interaction between antibody type and antigen density.

In the present study, we have addressed questions regarding optimal antibody affinity for targeting by using a highly manipulatable, though admittedly artificial, model system. Findings made on ELISA plates, an in vitro bead assay, and an in vivo animal model consistently demonstrate that the need for a high-affinity antibody is in fact dependent on the density of the target antigen. On HD antigen, both antibodies bound effectively,whereas on the LD target, only the high-affinity antibody was able to bind, with insignificant binding of the low-affinity antibody. This relationship between the antibody affinity, the density of the target,and binding can best be understood as reflecting different minimal requirements needed for the binding of the high and low-affinity antibodies. For low-affinity antibodies, such as mAb 36-65, the strength of a monovalent interaction between a single Fab binding arm and antigen is insufficient for retention. However, if the density of the antigen is high enough so that both arms of the IgG1 molecule can bind simultaneously, even if one Fab temporarily dissociates, the antibody can again bind by both arms and is retained. With high-affinity antibodies, such as mAb 36-71, the strength of the interaction between a single Fab binding site and the antigen is sufficient to retain antibody with only monovalent binding, and the antibody is bound effectively, irrespective of the antigen density. The findings in our model system clearly reiterate the need to consider antigen density in in vivo as well as in vitroantibody binding studies.

It is interesting to consider how the present findings can be generalized to actual clinical tumor targeting. In the present paper, a difference in the ability to bind divalently was noted,although the HD and LD antigens only varied in density 12-fold. These concentrations were, in fact, chosen to demonstrate a dramatic difference in divalent binding, even within a small range of densities,without confounding secondary factors such as the total number of binding sites or mass-action. In more extreme examples of HD and LD antigen, antibody would still be only able to bind to LD antigen monovalently, whereas divalent binding could occur with HD antigen. In vivo, it is difficult to theorize what concentration of antigen would enable divalent binding. One could compare the radius of the binding arms of an IgG molecule with the average density of antigen on the cell surface, but this would ignore the ability of antigen to migrate in the fluid cell membrane. Many antigens, such as mucins, also contain repetitive epitopes, which would have to be analyzed. Furthermore, issues such as accessibility of antigen, steric hindrance,and the flexibility of the IgG molecule would be difficult to predict. Perhaps the best proof regarding applicability of these concepts to actual tumor targeting is the fact that a large array of multivalent constructs, discussed below, has been shown to be qualitatively superior to univalent molecules in clinical and experimental studies.

In fact, the present findings provide interesting insights into clinical studies and suggest potential novel strategies for improving target-to-background localization. Clearly, the critical parameter for predicting antibody binding is avidity rather than affinity, because it takes into account the effect of multivalent interaction. A decrease in binding caused by reduced valence has been shown experimentally(25) and has long been clinically recognized when targeting enzymatically generated Fab and Fab′ antibody fragments. This observation has taken on new relevance as novel immunological molecules are designed using recombinant DNA techniques. Many of the initial constructs developed, such as scFvs (26, 27) were monovalent, and consequently exhibited relatively poor binding. To compensate, divalent constructs were devised, such as the divalently linked scFvs (28), diabodies (29), and minibodies (30). In carefully controlled animal studies,the advantages of divalent constructs have been shown to be independent of the size and rate of vascular clearance (28, 31). Beyond producing small divalent constructs, other efforts have been expended to artificially increase the valence of immunologically derived molecules to supranormal values such as trimers or tetramers of Fab fragments (32, 33, 34) or multimers of intact IgG molecules (35, 36, 37). This trend, in fact, recalls an earlier observation by one of the current authors (M. D. S.)regarding mutant antibodies spontaneously derived from the 36-65 cell line which demonstrated increased antigen binding to low-density Ars-BSA. Analysis showed that the mutations were not in the variable region increasing affinity, as might have been expected, but were,rather, in the constant region, leading to polymerization(22, 38). The consequent increase in span of the immunogobulin and valence resulted in >100-fold increases in antigen binding, as a manifestation of increased avidity.

The benefit of increasing the valence of the binding molecule would only be effective in the presence of antigen density adequate for multivalent binding. In clinical practice, a low-affinity divalent antibody may be quite adequate for targeting a specific tumor expressing a high density of antigen on its surface, but totally inadequate for a different tumor with a lower density of antigen. Experimental strategies to improve antibody targeting by increasing antigen density, such as biological response modifiers (reviewed in Refs. 39 and 40), can therefore be understood not only quantitatively as a means of increasing the total number of antigen-binding sites, but also qualitatively, as a strategy to increase the avidity of binding, which is especially important when antibody affinity and baseline antigen density are low. The importance of total antigen in driving the kinetics of antibody-antigen interaction has been considered previously in theoretical discussions(8, 41) and in animal experiments (42),although the qualitative effect of antigen density on avidity has not been singled out in these discussions.

It is interesting to contemplate tailoring the antibody affinity and valence to the specific clinical usage at hand. Taking into account the relative density of antigen on target and nontarget tissues, it could be possible to select antibodies of specific affinity and valence so as to better differentiate tissues bearing HD and LD antigen. For example,if an epitope is present at a high density on the tumor target and more sparsely on benign tissues, then it would be advantageous to choose a lower-affinity multivalent antibody so that avidity can be used to discriminate between binding on the two tissues. In an analogous manner, avidity may be the key to understanding the initially surprising absence of interference by some circulating antigens when targeting tumors with radiolabeled antibodies. Because interaction between an antibody and many soluble antigens is often monovalent,overall avidity of binding is low. This allows antibodies of low-to-moderate affinity to readily dissociate from circulating antigen and to bind in a more irreversible manner to the target tumor where antigen density, and hence avidity, is greater (22). This difference in binding would be lessened in the case of high-affinity antibodies, which could presumably bind irreversibly to circulating antigen even monovalently, or when targeting with monovalent fragments,which do not bind multivalently to the tumor antigen. In these cases,circulating antigen may conceivably represent a greater problem for tumor targeting.

When targeting with monovalent antibody fragments, the present observations suggest that binding would succeed best with higher affinity antibodies. This is in accordance with the observations of Adams et al. (43), who generated a set of scFv molecules with increased affinity by site-directed mutagenesis. Using this well-controlled family of reagents to target a SCID mouse xenograft model, increased affinity resulted in improved tumor targeting. In his model system, Adams has also noted that, above a given threshold, no additional benefit is realized (44). The existence of a ceiling of affinity above which no additional advantage is obtained should be even more evident when targeting with multivalent immunological molecules. Whereas a small experimental literature comparing divalent antibodies of differing affinity has appeared (13, 14, 15, 16, 17, 18, 45, 46, 47), it is difficult to extrapolate from many of these studies because of unclear results or the presence of confounding factors, such as differences in the epitope targeted.

Much has also been made of the provocative, yet persuasive, arguments of Weinstein et al. (5) and Fujimori et al. (6), who discussed the effect of elevated affinity on homogeneity as well as on total uptake by tumor nodules. Their modeling work suggested that an overly high affinity might be detrimental in therapy applications because the antibody would tend to accumulate at the periphery of tumor deposits. This phenomenon, which they termed the “binding site barrier,”would thereby deprive the innermost tumor of therapeutic effect. In fact, this concept has been supported by experimental data(48) and certainly would be relevant in situations of elevated avidity attributable to multivalent binding.

We have investigated the relationship between antibody affinity,antigen density, and antibody targeting using modifiable,well-controlled in vitro and in vivo experimental models based on the chemical hapten Ars and isotype and epitope-matched monoclonal antibodies that vary 200-fold in affinity. In contrast to high-affinity antibody, which binds effectively to both HD and LD antigen, low-affinity antibody only binds appreciably to HD antigen because of its requirement for divalent attachment. In the rational design of immunological reagents for specific in vivoapplications, the impact of antibody affinity must be viewed in the context of the antigen density of the target and background tissues.