Determining the affinities of high-affinity antibodies using KinExA and surface plasmon resonance

ABSTRACT

Accurate and efficient affinity measurement techniques are essential for the biophysical characterization of therapeutic monoclonal antibodies, one of the fastest growing drug classes. Surface plasmon resonance (SPR) is widely used for determining antibody affinity, but does not perform well with extremely high affinity (low picomolar to femtomolar range) molecules. In this study, we compare the SPR-based Carterra LSA and the kinetic exclusion assay (KinExA) for measuring the affinities of 48 antibodies generated against the SARS-CoV-2 receptor-binding domain. These data reveal that high-affinity antibodies can be generated straight from selections using high-quality in vitro library platforms with 54% correspondence between affinities measured using LSA and KinExA. Generally, where there was a 2-fold or greater difference between LSA and KinExA, KinExA reported that affinities were tighter. We highlight the differences between LSA and KinExA, identifying the benefits and pitfalls of each in terms of dynamic range and throughput. Furthermore, we demonstrate for the first time that single-point screening with KinExA can significantly improve throughput while maintaining a strong correlation with full binding curve equilibrium measurements, enabling the accurate rank-ordering of clones with exceptionally tight binding properties.

KEYWORDS:

• Antibody
• drug discovery
• antibody therapeutics
• surface plasmon resonance
• kinetic exclusion assay
• antibody characterization
• antibody affinity

Introduction

Monoclonal antibodies are presently the second largest drug class (after small molecules), with an estimated market size of ~$217.3 billion USD in 2021.¹ This market is projected to grow by ~15% CAGR for at least the next 10 years, with therapeutic antibodies overtaking small molecules to become the largest drug class sometime in the early 2030s. The features contributing to their popularity as a drug class include the following: 1) expansion of the therapeutics pipeline to previously undruggable targets; 2) relative ease of generation; 3) straightforward improvement of affinities and other biophysical properties; 4) availability of many formats, reflecting different possible modes of action; 5) straightforward development and approval pathways; 6) established production methods; 7) high approval success rates;² and 8) enormous markets.

Antibodies are generated by immunization or selection from in vitro display platforms. Immunization was long considered the best way to generate antibodies,³ but we recently showed that equally, or even more, potent antibodies can be selected from well-designed antibody libraries.^4–6 Beyond biological activity, which is indispensable, high antibody affinity is usually one of the most desirable characteristics. However, this is not uniformly required, particularly in cell therapies,^7,8 or antibodies designed to modulate receptor signaling,⁹ where lower affinities may be more effective. When high affinity is required, high-throughput affinity screening is usually carried out on large antibody panels.

Surface plasmon resonance (SPR) is commonly used to determine antibody affinity,^10,11 with the Carterra LSA being especially useful in high throughput.^12,13 However, as it is extremely challenging to measure off rates slower than 10⁻⁵ by SPR, high-affinity antibodies tend to be described as having affinities “less than” rather than with more accurate values. In contrast, the solution-based kinetic exclusion assay (KinExA) can provide accurate measurements for high-affinity antibodies.¹⁴ Generally, it has been assumed there is better correlation between SPR measurements and solution kinetics at weaker affinities. Unlike SPR, which can be used in high throughput, KinExA throughput is much lower.

Here, we compare the affinity measurements for 48 antibodies generated against the SARS-CoV-2 receptor-binding domain (RBD) from an antibody library platform⁵ designed to directly yield antibodies with few biophysical developability problems. The characterized antibodies were selected from the library directly using a combination of phage and yeast display, without undergoing affinity maturation. We used random picking to isolate a first set of picked antibodies⁴ and next-generation sequencing⁶ to subsequently explore the full diversity and affinities of the selection output. When measured by SPR (LSA), the affinities of 200 characterized antibodies ranged from <26.7 pM to ≥1 µM,⁶ and we found that we could not effectively discriminate many of the antibodies in the sub-100 pM range due to flatline off-rates.^5,6 As we wanted to discover the true affinities of these antibodies under the conditions used, as well as the correlation between LSA SPR affinities and those measured by KinExA, we used KinExA to further analyze 48 of the higher-affinity antibodies (SPR measured affinities <26.7 pM to 1.20 nM). Our results show that ~54% of the affinities measured by SPR or KinExA are within 2-fold of one another, with most, but not all, of the remaining antibodies showing tighter affinities by KinExA.

Results

Affinities at the limits of the LSA

Using our Generation 3 library platform,⁴ we successfully generated a large panel of antibodies against the RBD of SARS-CoV-2.^5,6 Among the strongest binding antibodies with solid SPR profiles (Supplementary Figure S1), a subset exhibited flatline off-rates with no detectable downward slope (Figure 1a), preventing the accurate measurement of the dissociation constants. Antibodies at this lower off-rate limit are assigned an equivalent value 1 × 10⁻⁵ s⁻¹, giving rise to the aligned set of affinity points at an off-rate of 10⁻⁵ in isoaffinity plots of antibodies isolated in this campaign (Figure 1b). The differentiated on-rates give rise to the different putative affinities (Table 1). Of the antibody affinities to RBD calculated by KinExA and below 50 pM, 13/15 (87%) exhibit LSA determined off-rates of ≤1×10⁻⁵ s⁻¹. Of the high-affinity (≤50 pM by KinExA) antibodies with 2-fold affinity differences between the LSA and KinExA measurements, 8/9 (89%) exhibit off-rates of ≤1×10⁻⁵. Steady-state equilibrium KD measurements (Supplementary Table S1) deviated from kinetics measurements by 10–100 fold (in the direction of worse affinity) relative to the KinExA- and LSA-based analyses. The accurate affinities, however, could not be assessed by SPR without method modifications, such as direct coupling of the anti-RBD antibody directly to the surface or significantly extending dissociation time, which may result in heterogenous profiles or excessive resource use.

Figure 1. Inability to discriminate antibodies based on off-rate. a) Sensorgrams from antibodies with slow off-rates (≤1×10⁻⁵ s⁻¹) and strong binding affinities (<0.1 nM). b) Isoaffinity plot shows a number of clones (blue) that could not be differentiated by off-rates due to limits of the capture kinetics and current run conditions (15-min dissociation) using the HC200M chip.

Table 1. Measured affinities for LSA, KinExA, and ratio of LSA/KinExA. Antibodies within 2-fold of the two separate measurements are given in darker shade of blue.

Employing kinetic exclusion to measure affinities in the subnanomolar range

To avoid these constraints, we decided to use KinExA, a solution-based exclusion assay technology that accurately measures affinities in the femtomolar range,¹⁵ and compare the results obtained with the SPR-based Carterra LSA (Figure 2). Standard KinExA (Figure 3a) uses 12 samples of a constant amount of antibody titrated against the target to calculate KD based on the percentage of free antibody, which is determined by removing the complexed antibody. The analysis employs a dual curve, global fit, method when measuring affinities.¹⁴ Although KinExA provides a more accurate measurement at the low picomolar to femtomolar range relative to SPR, it comes at the cost of lower throughput (Figure 3c). By contrast, the LSA achieves higher throughput with the use of two microfluidic modules, the single flow cell and 96-channel printhead in which one or the other may be docked onto the chip at any given time (Figure 3b). Samples, such as the capture antibody to be coupled to the surface or from diluted antigen at a given concentration, derived from the single flow cell, can be injected via a single needle that can pass across the entire chip surface before passing to waste. Samples, such as the panel of antibodies, that pass through the 96-channel printhead are injected via a 96-needle manifold, through 96-parallel flow cells able to contact the chip in a perpendicular manner to apply antibodies to discrete spots on the chip surface. This 96-channel printhead can be repositioned to apply to four distinct areas to array up to 384 spots. Finally, the bidirectional flow of a fixed injection volume (250 µL in single flow cell or 200 µL in the 96-channel printhead in addition to dead volume, up to 50 µL) with set time ensures relatively low reagent consumption (for more details see carterra-bio.com).¹² To overcome the KinExA throughput limitation, we used a single-point screening method (Figure 3a), which uses two samples for each antibody, one with a fixed target concentration and one without, to determine the percentage of free antibody after reaching equilibrium. Our results show the KinExA-based single-point screening method can improve the throughput of relative ranking by 10-fold.

Figure 2. Sensitivity comparison of KinExA and LSA. a) Measurement method for KinExA. Signals are generated from a fluorescent anti-species antibody that detects the specific antibody captured on antigen-coated beads. Panels A1–A3 represent experiments, where the sample contains antibody only and can bind to antigen-coated beads. The resulting high signal is indicated in panel B. Panels C1–C3 represent a sample that contains an excess of antigen, which limits the available free antibody that can bind to antigen-coated beads. The resulting low signal is indicated in panel B. b) A routine LSA capture/couple kinetic protocol. A monoclonal antibody is captured or coupled directly to the gold surface with the antigen flow across the region of interest. Association of the antigen is depicted with an elevated response unit, until the point of equilibrium. c) Affinity ranges reliably, determined by different measurement platforms.

Figure 3. Throughput comparisons of KinExA and LSA. a) In the standard KinExA assay (top), 5 × 12 samples are prepared with a constant amount of antibody in each. The target is titrated through these samples and passed over the capture column after coming to equilibrium. This allows a calculation of the KD, based on the % of free antibody in each sample. The single-point method (bottom) uses two samples for each of 48 antibodies, with one sample interacted with a fixed concentration of target and the other with no target. This allows the determination of % free antibody after the sample is passed over the capture column after equilibrium is reached. b) Instrument setup of the Carterra LSA (see more details at carterra-bio.com). c) Throughput of different instruments for the characterization of monoclonal antibody affinities with the assumption of one user, for one instrument in a 24-h timeframe.

Correlations between equilibrium binding among the distinct platforms

To understand the correlations between LSA and KinExA for the higher-affinity antibodies, we measured the affinities of 48 antibodies (Supplementary Table S2) from the LSA using KinExA. We observed high-quality data collected across all the antibodies from the n-Curve analysis (Supplementary Figure S2). We found that most of the antibodies (n = 26) had affinity values within 0.5- to 2-fold for the two platforms (Figure 4a and Table 1), indicating agreement between LSA and KinExA measurements for these samples. The 95% confidence intervals supplied by KinExA encompass the LSA K_d in 17/26 antibodies (Supplementary Figure S4 and Supplementary Table S3), further supporting agreement between these values. The next largest group (n = 18) comprised antibodies where the measured KinExA affinities were more than 2-fold higher than the LSA-measured affinities (Figure 4b). The smallest sample set (n = 4) comprised those, where the measured LSA affinities were more than 2-fold higher than KinExA affinities (Figure 4c).

Figure 4. Correlation of KinExA to LSA. a–c) LSA and KinExA affinity measurements of KinExA and LSA affinities a) within 2-fold, b) LSA affinity values > 2× higher than KinExA (KinExA affinity measurements are tighter), and c) KinExA affinity values > 2× higher than LSA values (LSA affinity measurements are tighter). d) The barplot shows the ratio of LSA to KinExA affinity measurements, with the shaded (dark blue) area indicating ratios from 0.5 to 2.0 of antibodies within 2-fold affinities. e) The isoaffinity plot of the characterized antibodies showing where the improved affinities occur when considering on-rate and off-rate. f) The log–log scatterplot of LSA to KinExA affinities with the correlation coefficients (R², Pearson and Spearman) calculated using a linear regression of the LSA to KinExA affinities. Blue points and line indicate the measured affinities and regression line for all antibody affinities (blue), antibody affinities below 100 pM (orange), and those above 100 pM (dark red). The shaded area represents the 95% confidence interval, which is the range within which the regression line (or true population parameter) is expected with 95% probability, by calculating the standard error of the regression coefficients and critical value from the t-distribution. g) The log-linear scatter plot of the percent free measurement of KinExA relative to the KinExA dual curve equilibrium dissociation constant with the correlation coefficients (R², Pearson and Spearman) calculated using a linear regression of the single point percent free to KinExA dual curve affinities. Blue points and line indicate the measured affinities and regression line for all antibody affinities (blue), antibody affinities below 100 pM (orange) and those above 100 pM (dark red). The shaded area represents the 95% confidence interval, which is the range within which the regression line (or true population parameter) is expected with 95% probability, by calculating the standard error of the regression coefficients and critical value from the t-distribution.

In general, when the affinities start to deviate (e.g., when ratios are >2 or <0.5), the affinities tend to be tighter for KinExA measurements compared to LSA measurements (Figure 4d, Table 1). In Figure 4e, we observed that when KinExA-measured affinities are at least 2-fold tighter than LSA, these are distributed across the range of affinities, e.g., from low nanomolar to sub-100 pM. Of the four antibodies that showed improved affinities using LSA relative to KinExA, two had 10⁻⁵s⁻¹ off-rates (Figure 4e), at the limits of the capture kinetics protocol, while the remainder had off-rates of 10⁻⁴ and 10⁻³.

We found moderate to strong positive correlations (R² = 0.557; Pearson = 0.747 and Spearman = 0.719) between LSA and KinExA affinity measurements across all affinities from sub-100 picomolar to low nanomolar range (Figure 4f). This correlation is reduced in the sub-100 picomolar range (R² = 0.435; Pearson = 0.660 and Spearman = 0.659) (Figure 4f), and removal of antibodies with dissociation constants (KD) of ≤1×10⁻⁵ s⁻¹ from the dataset resulted in a slight increase in the R² (0.640), but with a drop in Pearson and Spearman correlations at 0.748 and 0.519, respectively. By contrast, antibody affinities ≥ 100 pM show even stronger linear relationships after log-transformation relative to those below 100 pM (R² = 0.589, Pearson = 0.767), although the drop in the Spearman coefficient (0.433) suggests that this population is less monotonic relative to the other groups, with greater fluctuations relative to when measuring all affinities or those below 100 pM.

Correlations of single-point screening with full binding curve equilibrium

In the traditional KinExA screening approach (Figure 3a), a series of samples are prepared with a constant amount of antibody in all samples. The target is titrated through these samples and passed over the capture column after coming to equilibrium. This provides a measure of free antibody in each sample, allowing a calculation of the concentration at which half occupancy occurs (KD). This extensive experimental approach is responsible for the accuracy of the measurement, as well as the low throughput. An alternative, more rapid approach is to carry out a single-point analysis, in which two samples are prepared for each antibody. The target is added to only one, allowing a calculation of the % free antibody after the sample comes to equilibrium and is passed over the capture column. This approach greatly expands the throughput of the KinExA from 5 to 48 measured interactions in a 24-h period (Figure 3b), with a strong correlation (R² = 0.809; Pearson = 0.900 and Spearman = 0.880) between the calculated % free antibody and the measured KD from the picomolar to low nanomolar range, with a slight reduction for affinities below 100 pM (R² = 0.701; Pearson = 0.837 and Spearman = 0.778) (Figure 4g) and a further reduction for values ≥ 100 pM (R² = 0.505; Pearson = 0.711 and Spearman = 0.712) with minimal impact on the slope.

Discussion

Historically, high-affinity antibodies (subnanomolar range) directly from discovery campaigns were restricted to in vivo approaches (e.g., animal immunization), with in vitro display systems requiring additional affinity maturation of the discovered antibodies. In recent years, a substantial improvement in the quality of in vitro technologies has resulted in large panels of high-affinity antibodies directly from selections. We recently showed that we could identify large panels of highly diverse (anti-SARS CoV-2) antibodies, many in the sub-100 pM range, from a carefully designed in vitro library^6,16. Considering this, and the need to further optimize affinity through affinity maturation, this study highlights the importance of capturing the accurate affinities to effectively rank order antibodies in these extremely high-affinity ranges. The present study not only shows the complementary nature of SPR and KinExA in characterizing and optimizing high-affinity antibodies, but also showcases select advantages of KinExA when affinities are lower than 100 pM, which represents the limits of SPR, particularly when off-rates are slower than 10⁻⁵ s⁻¹ (Figure 1a). The flatline off-rate is the likely cause of the poorer correlation, as one observes that 16/19 (84%) of the antibodies below 100 pM exhibit flatline off-rates ≤1.0×10⁻⁵ s⁻¹. As might be expected, steady-state kinetics (Supplementary Table S3) exhibited an even worse association with KinExA, as most concentrations will not achieve equilibrium at these picomolar affinities.

For most of the antibodies analyzed, we observed good agreement between the affinity measurements obtained from LSA and KinExA (Figure 4a, Table 1). This finding suggests that both techniques can be reliably used for affinity characterization of monoclonal antibodies, even in the subnanomolar range. Despite the general agreement between these two approaches, we observed some discrepancies, particularly for antibodies at the lower end of the LSA dynamic range (Figure 4b, c, f). Notably, when affinities start to deviate more substantially between the two assays, KinExA affinities typically appeared to be stronger. As KinExA is known to be able to measure higher affinities (down to femtomolar affinities) compared to SPR, and many of the measured affinities were assigned minimum equivalent off-rates by LSA, it is assumed that the measured KinExA affinity values for many of these tight-binding antibodies are more in line with the true affinities (under the conditions used). The loss of correlation in the sub-100 pM range, and moderate correlation of LSA to KinExA, is likely due to the assigned off-rate of ≤ 1×10⁻⁵ s⁻¹, preventing an accurate calculation of KD, which is dependent on the ratio of dissociation (k_d) and association (k_a) rate constants.

While this accuracy presents a specific advantage of KinExA over LSA (and other SPR approaches), it comes at the cost of lower throughput compared to SPR (Figure 3b). This limitation could potentially hinder the rapid screening and characterization of large panels of antibodies in a timely manner. However, our study also demonstrates that KinExA single-point screening can significantly increase the throughput by rank-ordering antibodies (based on percent free values), as these data show a strong correlation with traditional KinExA equilibrium measurements (Figure 4g). This approach greatly enhances the utility of KinExA in rank-ordering larger panels, particularly high-affinity antibody therapeutics where LSA and other SPR methodologies fall short (Figure 2). This improves the throughput of KinExA prioritization in that a larger panel of antibodies can be screened (48 as opposed to 4–5) with strong correlations to accurate affinity. Thus, if an antibody panel can be first screened by single-point screening, the time taken to measure affinities of the best antibodies is significantly less than measuring the entire panel, if these top leads are measured with traditional KinExA analysis for accurate KD information. This is particularly important in cases where high affinity is essential for therapeutic efficacy. Of course, if affinities for all antibodies are required, this approach does not confer an added throughput benefit.

In conclusion, our study demonstrates the complementary nature of SPR and KinExA in characterizing and optimizing high-affinity antibodies. The ability of KinExA to accurately measure affinities in the low picomolar to femtomolar range makes it a valuable tool in the development of high-affinity antibody therapeutics, particularly in cases where the strength of the interaction is directly correlated to efficacy. Our findings also highlight the potential of single-point screening with KinExA to increase the throughput while maintaining a strong correlation with full binding curve measurements.

Materials and methods

Kinetic exclusion KD measurements

All measurements were performed at 25°C using a KinExA 4000 installed in a TC1000 temperature control chamber (both from Sapidyne Instruments Inc., Boise Idaho). The running buffer was HBSTE (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.01% Tween 20). The samples were prepared in running buffer supplemented with 0.5 mg/ml bovine serum albumin.

To coat the solid phase (Azlactone activated beads Part#444110 Sapidyne Instruments Inc., Boise, Idaho), 10 ug of RBD (SARS-CoV-2 (COVID-19) S protein RBD, His Tag (MALS verified) Part # SPD-C52H3-100 ug Acro Biosystems, Newark, Delaware) per vial was used, following Sapidyne Instrument’s suggested protocol (Sapidyne Instruments How to Guide HG209).

For KD measurements, binding curves were generated by measuring the free antibody concentration of a series of samples containing a constant concentration of antibody with a 2-fold titration series of RBD. These mixtures were incubated at 25°C to equilibrium (time varied with antibody and concentration, minimum 2 h). For each antibody, two binding curves, at different antibody concentrations, were analyzed in Sapidyne Instruments’ KinExA Pro software (version 4.4.40) n-curve module.

To measure the free antibody concentration in the equilibrated samples, the KinExA 4000 was programmed to fill the observation column with solid phase and then sequentially flow each antibody–RBD mixture over the solid phase at a flow rate of 0.25 mL/min. The volume was varied (from a maximum of 3.0 mL to a minimum of 0.01 mL) according to specific antibody and antibody concentration to achieve signals in the optimal working range. Post sample, 500 uL of 500 ng/mL Alexa Fluor® 647 AffiniPure Goat Anti-Human IgG (H+L) (Part # 109-605-003 Jackson ImmunoResearch Inc., West Grove, Pennsylvania) was flowed through the column at 0.25 mL/min to label the captured specific antibody. Signals were read once per second and stored automatically by the KinExA 4000. The KinExA Pro software provides 95% confidence intervals in addition to the best-fit KD value and active concentration of the antibody (see Results section for discussion).

For single-point screening measurements, two samples were prepared for each antibody containing antibody only (Sig100) and antibody + RBD (inhibited). To generate the nonspecific binding signal (NSB), a sample buffer with zero antibody concentration was run. A concentration of RBD was chosen that would theoretically inhibit 50% of the antibody at the desired cutoff for the screening assay. The inhibited sample was incubated at 25°C to equilibrium. Following the above-outlined procedure, using the same capture and detection materials, the samples were measured on the KinExA 4000 instrument. Signals from the two samples and the NSB signal were used to calculate the inhibition level for each antibody. The equation used is: percent free = ((Signal-NSB)/(Sig100-NSB)) x 100. The antibodies were then placed in rank order based on the percent free calculation, with the lower percent free considered the tighter binding antibodies.

Additional details regarding the KinExA instrument, its operation, and data analysis can be found in previous publications.^14,15

LSA kinetics

All measurements were performed at 25°C. Kinetic experiments were conducted using an HC200M sensor chip (Carterra #4297). The chip was activated with a solution containing 33 mM N-hydroxysulfosuccinimide (S-NHS, Sigma #56485) and 133 mM N-(3-Dimethylaminopropyl)-N’-ethylcarbodiimide hydrochloride (EDC, Sigma #E7750) in 0.1 M MES, pH 5.5, for 5 min. The capture antibody, an anti-human Fc (Southern Biotech, #2048–01), was diluted to 50 µg/mL in 10 mM acetate, pH 4.33, and immobilized on the chip for 20 min. The chip surface was deactivated with a 1 M ethanolamine solution, pH 8.5, to inhibit further primary amine coupling. Individual antibody clones were diluted to 10 µg/mL in 1×HBSTE (Carterra #3630) and printed onto the chip for 12 min. The RBD (Acro Biosystems #SPD-C52H3) was prepared in a 7-point dilution series ranging from 300 nM to 411 pM. All data were fitted using the Kinetics software suite (Carterra) with a one-site model.

Abbreviations

SPR	=	surface plasmon resonance
KinExA	=	kinetic exclusion assay
RBD	=	receptor-binding domain

Disclosure statement

MFE, FF, SDA, LS and ARMB are employees of Specifica, a Q2 Solutions Company, which sells antibody libraries and discovery services. AAT and CLL were formerly employees of Specifica. MD and EH are employees of Sapidyne, which sells KinExA equipment and services.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2291209

Additional information

Funding

This work was not supported by any funding.