ANALYTICAL BIOCHEMISTRY

Analytical Biochemistry 359 (2006) 94105 www.elsevier.com/locate/yabio

Comparative analysis of 10 small molecules binding to carbonic anhydrase II by diVerent investigators using Biacore technology
Giuseppe A. Papalia a, Stephanie Leavitt b, Maggie A. Bynum c, Phinikoula S. Katsamba a, Rosemarie Wilton d, Huawei Qiu e, Mieke Steukers f, Siming Wang g, Lakshman Bindu h, Sanjay Phogat i, Anthony M. Giannetti j, Thomas E. Ryan k, Victoria A. Pudlak k, Katarzyna Matusiewicz l, Klaus M. Michelson m, Agnes Nowakowski n, Anh Pham-Baginski o, Jonathan Brooks p, Bryan C. Tieman q, Barry D. Bruce r, Michael Vaughn r, Michael Baksh s, Yun Hee Cho t, Mieke De Wit u, Alexandra Smets u, Johan Vandersmissen u, Lieve Michiels u, David G. Myszka a,
a

Center for Biomolecular Interaction Analysis, School of Medicine, University of Utah, Salt Lake City, UT 84132, USA b Gilead Sciences, Foster City, CA 94404, USA c Agilent Technologies, Palo Alto, CA 94304, USA d Argonne National Laboratory, Argonne, IL 60439, USA e Genzyme Corporation, Framingham, MA 01701, USA f Dyaxsa, B-4000 Liege 1, Belgium g Georgia State University, Marietta, GA 30302, USA h National Cancer Institute, Frederick, MD 21702, USA i National Institute of Allergies and Infectious Diseases, Bethesda, MD 20892, USA j Roche, Palo Alto, CA 94304, USA k Reichert Analytical Instruments, Depew, NY 14043, USA l Adamed, 05-152 Czosnow, Poland m Amgen, Thousand Oaks, CA 91320, USA n diaDexus, South San Francisco, CA 94080, USA o Wyeth, Andover, MA 01810, USA p Wyeth, Cambridge, MA 02140, USA q Abbott Laboratories, Abbott Park, IL 60064, USA r University of Tennessee, Knoxville, TN 37996, USA s University of California, Berkeley, CA 94720, USA t Human Genome Sciences, Rockville, MD 20850, USA u Tibotec, B-2800 Mechelen, Belgium Received 20 July 2006 Available online 7 September 2006

Abstract In this benchmark study, 26 investigators were asked to characterize the kinetics and aYnities of 10 sulfonamide inhibitors binding to the enzyme carbonic anhydrase II using Biacore optical biosensors. A majority of the participants collected data that could be Wt to a 1:1 interaction model, but a subset of the data sets obtained from some instruments were of poor quality. The experimental errors in the ka, kd, and KD parameters determined for each of the compounds averaged 34, 24, and 37%, respectively. As expected, the greatest variation in the reported constants was observed for compounds with exceptionally weak aYnity and/or fast association rates. The binding constants determined using the biosensor correlated well with solution-based titration calorimetry measurements. The results of this study

Corresponding author. Fax: +1 801 585 3015. E-mail address: dmyszka@cores.utah.edu (D.G. Myszka).

0003-2697/$ - see front matter 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.ab.2006.08.021

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

95

provide insight into the challenges, as well as the level of experimental variation, that one would expect to observe when using Biacore technology for small molecule analyses. 2006 Elsevier Inc. All rights reserved.
Keywords: Biacore; Surface plasmon resonance; Proteinprotein interaction; Kinetics; SPR

There is growing interest in applying optical biosensor technology, such as surface plasmon resonance (SPR)1based Biacore instruments, in drug discovery. With the wide array of opportunities oVered by this technology, however, come some challenges. For example, based on our yearly review of the optical biosensor literature [16], we Wnd that there is still a need to train investigators in how to design and perform biosensor experiments as well as in how to properly process and analyze data. And based on comments we receive from manuscript reviewers and participants at scientiWc meetings, we see that some skepticism remains regarding the validity of the binding constants determined using biosensors. For the past few years, we have been addressing both of these issues by coordinating intra- and inter-technology benchmark studies. In our Wrst comparative study, 30 SPR biosensor users illustrated that the aYnity determined from the biosensor for a small molecule/enzyme interaction matched the value determined in solution using isothermal titration calorimetry (ITC) [7]. These results conWrmed that immobilization of an enzyme onto the sensor surface does not necessarily aVect the binding constants as is often presumed by critics of the biosensor approach. Then, we organized a study with 36 participants who analyzed the interaction of a small molecule/macromolecular target system having a very fast association rate [8]. The results of this test established experimental and data processing protocols for mass transport-limited reactions. Recently, we coordinated a study with 22 users resolving the binding constants for a highaYnity monoclonal antibodyantigen interaction [9]. This work trained users in how to generate and analyze data for a slowly dissociating system. Together, these benchmark studies educated the users of biosensor technology, as well as the general scientiWc community, with regard to what we can expect to see in terms of the variability of results within a given experimental system. The current study builds on this previous work and is modeled on a growing biosensor application in a drug discovery setting: the rapid and reliable determination of kinetic and aYnity constants for a small panel of compounds binding to a single protein target. This information is becoming crucial to better rationalizing compound behavior in bioassays and helping to steer compound selection and design strategies. The goal of the current biosensor

study was to determine the consistency of the results for small molecule analysis and to discover ways to further improve the application. The target protein for this work was carbonic anhydrase II, an enzyme responsible for the conversion of carbon dioxide to bicarbonate. Current inhibitors of this enzyme are used to treat diseases such as glaucoma and epilepsy, and future drugs targeting this enzyme may lead to treatments for cancer and obesity [10]. Thus, the choice of carbonic anhydrase as the target represents a realistic model for small molecule drug studies. All participants in this study were provided the same reagents and were asked to follow the same experimental protocol. We found that a majority of the participants were able to resolve the binding constants for all 10 compounds. We did, however, identify a suboptimal subset of data (sensorgrams that were marred by features such as spikes and drift) and we explored the causes of these features. We also compared the equilibrium dissociation constants obtained for the 10 compounds with those determined by ITC. We found an excellent correlation between the two methods across the wide range in aYnities exhibited by these compounds. Our results highlight the strengths of biosensor analysis but also point to some challenges. Materials and methods Instrumentation and reagents Interaction analyses were performed using Biacore 1000, 2000, 3000, S51, and T100 instruments (Biacore, Uppsala, Sweden). Sensor chips, N-hydroxysuccinimide (NHS), Nethyl-N-(3-dimethylaminopropyl)carbodiimide (EDC), and ethanolamine HCl, as well as sampling vials, caps, and 96-well plates, were obtained from Biacore. Carbonic anhydrase isozyme II (CAII) from bovine erythrocytes, the 10 sulfonamide inhibitors, dimethyl sulfoxide (DMSO), buVer reagents, and general laboratory supplies were purchased from SigmaAldrich (St. Louis, MO, USA). Instrument cleaning Before beginning the experiment, each participant was asked to perform a series of instrument cleaning steps. First, the instrument was primed with water, the previously used chip was undocked, and a maintenance chip was docked. The instrument was then primed Wve times using 0.5% (w/v) sodium dodecyl sulfate (SDS), once using water, Wve times using 50 mM glycine (pH 9.5), and once again using water.

1 Abbreviations used: SPR, surface plasmon resonance; ITC, isothermal titration calorimetry; NHS, N-hydroxysuccinimide; EDC, N-ethyl-N-(3dimethylaminopropyl)carbodiimide; CAII, carbonic anhydrase isozyme II; DMSO, dimethyl sulfoxide; SDS, sodium dodecyl sulfate; PBS, phosphate-buVered saline; RU, response units; CV, coeYcient of variation.

96

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

Chip preconditioning After docking the CM5 sensor chip and priming the instrument with running buVer, the chip was preconditioned with two 10- l injections each of 100 mM HCl, 50 mM NaOH, and 0.5% SDS at a Xow rate of 100 l/min. BuVer preparation Participants were provided a 10 phosphate-buVered saline (PBS, 1 D 20 mM Na2HPO4NaH2PO4, 150 mM NaCl, pH 7.4) stock solution. Then 70 ml of this 10 PBS was added to 630 ml of degassed deionized water to make 1 PBS buVer, and this buVer was Wltered. Then 18 ml of DMSO (also provided to the participant) was added to 600 ml of the 1 PBS buVer. DMSO is commonly used as a solvent for small molecule studies. Although the Wnal concentration of DMSO in this buVer via this addition is actually 2.97%, for simplicity this buVer is termed running buVer with 3% DMSO. The remaining 100 ml of 1 PBS was divided into two 50-ml portions. One of these portions was set aside for use as immobilization buVer, and the other was set aside as sample preparation buVer with no DMSO for use in analyte preparation. Enzyme immobilization Participants were provided with aliquots of 100 g of lyophilized CA II that were centrifuged and dissolved in 400 l of 10 mM sodium acetate (pH 4.9) after the biosensor was primed with immobilization buVer and equilibrated to 25 C. Using a Xow rate of 20 l/min, the surface of Xow cell 1 was activated for 7 min using a mixture of 0.1 M NHS and 0.4 M EDC, 0.25 mg/ml CAII was injected for 7 min, and residual activated groups on the surface were blocked by a 7-min injection of 1 M ethanolamine (pH 8.5). Users of S51 platforms activated the surfaces via sequential injections of NHS and then EDC for 7 min. This was followed by injection of 0.25 mg/ml CAII for 7 min at a Xow rate of 10 l/ min. Surfaces were then blocked via a 7-min injection of ethanolamine at a Xow rate of 10 l/min. On average, 6900 response units (RU) of CAII were immobilized. For this set of studies, we chose to leave Xow cell 2 unmodiWed as a reference surface.
Table 1 Compounds examined in this study Sample 1 2 3 4 5 6 7 8 9 10 Compound

Analyte preparation For simplicity, each of the compounds analyzed in this study was assigned a number (Table 1). Participants were provided with a stock solution of each compound dissolved in DMSO. To prepare the samples for analysis, 30 l of each compound solution was added to 1 ml of sample preparation buVer with no DMSO and mixed thoroughly. Preparation of analyte in this manner ensures that the concentration of DMSO is matched with that of running buVer with 3% DMSO. Three lower concentrations of each of the 10 compounds were then prepared via Wvefold serial dilutions into running buVer with 3% DMSO. For analyses using Biacore 1000, 2000, and 3000, 100 l of each compound at the highest concentration was added to 400 l of running buVer with 3% DMSO and this dilution was repeated. Samples and buVer blanks were aliquotted in 7-mm Biacore vials, which were then capped with soft rubber caps (BR-1005-55) that reseal after puncture to minimize evaporation. Using these sample volumes and the resealable caps permitted the three concentrations of each compound to be sampled twice. Biacore S51and T100-based analyses used 96-well plates for sampling. Dilutions were performed directly in the plates by taking 50 l from wells containing 250 l of the highest analyte concentration and diluting successively into wells containing 200 l of running buVer with 3% DMSO. To correct for the excluded volume eVect, a DMSO calibration series was prepared by adding 15 l of DMSO and 150 l of water to separate 1.5-ml aliquots of running buVer with DMSO to obtain + and solutions. Then 400, 300, 200, 100, and 0 l of the + solution were dispensed into Eppendorf tubes labeled d1 to d5, and 0, 100, 200, 300, and 400 l of the solution were added to tubes d1 to d5 so that each tube contained a Wnal volume of 400 l. The Wve DMSO calibration solutions were either aliquotted into 7-mm Biacore vials or, for Biacore S51 and T100 analyses, dispensed into the 96-well plate. Instrument optimization To create a slight mismatch in the refractive index of the running buVer and sample solutions, water (0.4% of the total volume of running buVer) was added to the running buVer with 3% DMSO. By making the start of each injec-

Molecular mass (Da) 341 223 172 331 201 250 236 157 217 222

Concentration ( M) 2, 10, 50 2, 10, 50 0.4, 2, 10 0.4, 2, 10 0.4, 2, 10 0.08, 0.4, 2 0.08, 0.4, 2 0.4, 2, 10 0.016, 0.08, 0.4 0.016, 0.08, 0.4

()-Sulpiride 4-(Aminomethyl)benzenesulfonamide hydrochloride hydrate Sulfanilamide Furosemide 4-Carboxybenzenesulfonamide Dansylamide 1,3-Benzenedisulfonamide Benzenesulfonamide 7-Fluoro-2,1,3-benzoxadiazole-4-sulfonamide Acetazolamide

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

97

tion easily identiWable, this mismatch aids in data processing [8]. The instrument was primed three times with this spiked running buVer and then normalized using 40% (v/v) glycerol for Biacore 1000 and 2000 or 70% (v/v) glycerol for Biacore 3000, S51, and T100. Analyte injection order and analysis method Data were collected at the highest collection rate possible and at 25 C. An initial series of buVer blanks was injected Wrst to fully equilibrate the system. The DMSO calibration series was then injected, followed by two additional blank injections. Finally, the individual compound
50

samples were tested (from lowest to highest concentrations within each series) and each compound series was separated by a blank buVer injection. During each binding cycle, analyte was injected for 1 min at a Xow rate of 100 l/min (90 l/min for Biacore S51) and dissociation was monitored for 140 s. The analyte injection was followed by an ExtraClean wash command that automatically Xushes the sample delivery system with running buVer (for Biacore 2000 and 3000, a 15- l injection of running buVer was also included as an additional wash step). Analytes were sampled twice using Biacore 1000, 2000, and 3000 and were sampled once using Biacore S51 and T100. Data processing and kinetic analysis Data sets were processed and analyzed using Scrubber 2 (BioLogic Software, Campbell, Australia). Double-referenced [11] association and dissociation phase data for compounds 1 to 8 were globally Wt to a simple 1:1 interaction model (A + B D AB). Compounds 9 and 10 were Wt to a 1:1 interaction model that included a mass transport term (Ao D A + B D AB) [12]. Means and standard deviations were computed using Microsoft Excel software. The experimental error or coeYcient of variation (CV) was calculated as (Standard Deviation/Mean) 100.

Response (RU)

25

0 0 60 120 180

Time (s)
Fig. 1. High-quality kinetic data set. Duplicate analyses of 0, 0.4, 2.0, and 10 M furosemide binding to 6450 RU of immobilized CAII. Data shown are from the analysis of compound 4 by participant H and were collected using a Biacore 2000 instrument.

Isothermal titration calorimetry Protein was dissolved in 1 PBS and buVer exchanged to minimize any contaminants from lyophilization. Compound was diluted in the same buVer as protein to minimize heat of
C

Fig. 2. Problematic sensorgrams. (A) SigniWcant spikes at the start of the association and dissociation phases. (B) Spikes within a sensorgram. (C) BuVer blank responses that do not overlay. (D) Double-referenced response not described by a simple exponential. (E) Dip below baseline for double-referenced response. (F) Refractive index jumps.

98

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

dilution eVects from buVer mismatch. Experiments were performed with the compound in the syringe and the protein in the cell. Protein concentration was measured using an extinction coeYcient at A280 of 50,070 L mol1 cm1. Isothermal titration calorimetry experiments were performed using a Microcal VP-ITC instrument (MicroCal, Northampton, MA, USA) [13]. Experimental parameters included: 30 injections of compound, with an initial Wrst injection of 3 l and subsequent 10- l injections of compound into protein with 240 s between each injection and a stirring speed of approximately 300. Experiments were performed at 25 C. The heat of dilution was taken either from the points after saturation, control experiment of compound into buVer, or from a minimization of the 2 value. The concentrations of compounds titrated were 960, 666, 600, and 400 M for compounds 1 to 4, respec1
A

tively, 350 M for compounds 58, and 300 M for compounds 9 and 10. Corresponding concentrations of carbonic anhydrase were 44, 65, and 38 M for titrations of compounds 13 and 28 M for titrations of compounds 410. Data were analyzed with a single-site binding model using the Origin software provided by MicroCal. To minimize the dependency of the Wtting parameters for compound 1, the weakest inhibitor, the stoichiometry value was set to 1 assuming a 1:1 ratio, which has been well established for this system. Results In this benchmark study, participants used Biacore biosensor technology to characterize the interactions of 10 compounds with the enzyme CAII. We provided each participant
6 7 8 9 10

Response (RU)

45 90

45 90

45 90

45

90

45 90

45 90

45

90

45 90

45 90

45

90

Time (s)
Fig. 3. Data sets obtained using Biacore 1000 (A) and Biacore 2000 (BI) instruments. Columns 1 through 10 correspond to the numbering of compounds in Table 1. Responses shown in red exhibited one or more of the characteristics described in Fig. 2. (For interpretation of the references to color in this Wgure legend, the reader is referred to the Web version of this article.)

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

99

with a detailed experimental protocol as well as stock solutions of the enzyme, compounds, and buVer. CAII was immobilized on one Xow cell surface of a CM5 sensor chip using standard amine-coupling chemistry and an unmodiWed Xow cell surface served as a reference. Immobilization densities ranged from 2400 to 13,000 RU; this variation was useful in determining whether there were any systematic trends in the results with surface density. Three concentrations of each compound, prepared in a Wvefold dilution series, were analyzed for CAII binding at 25 C. Each concentration series was bracketed by buVer blanks that were included for double referencing [11]. Because each compound dissociated completely from the CAII surface within minutes after the end of the association phase, no regeneration step was required.
1
J

Participants using Biacore 1000, 2000, and 3000 platforms did repeat injections of each analyte concentration, whereas the Biacore S51 and T100 users performed a single injection of each. Good and bad sensorgrams To illustrate how we interpreted the responses generated by the diVerent users, we Wrst highlight examples of good and bad data. An example of a good data set obtained for one compound is shown in Fig. 1. From inspection of this data set, we can tell that the experiment was designed and executed properly, the biosensor was performing optimally, and the obtained responses are reliable. Hallmarks that
6 7 8 9 10

Response (RU)

45 90

45 90

45 90

45

90

45

90

45 90

45

90

45 90

45

90

45

90

Time (s)
Fig. 4. Data sets obtained using Biacore 3000 (JS) instruments. Columns 1 through 10 correspond to the compound numbering given in Table 1. Responses shown in red exhibited one or more of the characteristics described in Fig. 2.

100

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

make this data set a good one include responses that are concentration dependent, replicate injections that overlay, and clearly discernible exponential curvature during both the association and dissociation phases. When data sets are of high quality like the one shown in Fig. 1, they often can be described by a 1:1 interaction model. And when data sets are of poorer quality (e.g., when they exhibit unusual proWles or can be described only by a complex model), the anomalies and/or complexity most often can be eliminated by optimizing the assay design. Although we intentionally chose well-characterized, well-behaved 1:1 interactions for this study, the quality of some of the data sets generated by our study participants was not on par with the example shown in Fig. 1. Six of the common anomalous sensorgram features we observed in some data sets are depicted in Fig. 2. For example, signiWcant spikes at the both the beginning and end of the association phase apparent after referencing (Fig. 2A) indicate poor-quality injections that may be caused by a fouled integrated Xuidic cartridge. Although it
1
T

is common to have a few data points producing spikes at the transitions between running buVer and the analyte sample, spikes that occur over a larger window of time (e.g., >2 s) can result in the loss of kinetic information when the binding responses are fast. Frequent spikes throughout a sensorgram or sets of sensorgrams (Fig. 2B) often are indicative of a poorly degassed buVer or a dirty injection system. Blanks should have similar proWles and generally overlay, but the not-reproducible responses from the replicate buVer blanks shown in Fig. 2C would make it diYcult to choose which blanks to use for double referencing. Also, inconsistencies in buVer injections may indicate sample carryover. The inability to achieve steady state after a fast association phase (Fig. 2D) in a system known to be 1:1 suggests that there may be problems with reagent preparation, the biosensors Xuidic system, or both. Sensorgrams that, after referencing, dip down during the association phase, and in some cases drop below the baseline during the dissociation phase (Fig. 2E), may result from nonspeciWc binding to the reference surface. Sustained jumps in refractive index
6 7 8 9 10

Response (RU)

AA

45 90

45 90

45 90

45

90

45 90

45 90

45

90

45 90

45 90

45

90

Time (s)
Fig. 5. Data sets obtained using Biacore S51 (TZ) and Biacore T100 (AA) instruments. Columns 1 through 10 correspond to the compound numbering given in Table 1. Responses shown in red exhibited one or more of the characteristics described in Fig. 2. Due to a lack of compound, participant Z was unable to collect data for compound 8. (For interpretation of the references to color in this Wgure legend, the reader is referred to the Web version of this article.)

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

101

(Fig. 2F) make obtaining reasonable Wts impossible without elaborate data processing. Sensorgrams that displayed any of the characteristics illustrated in Fig. 2 were highlighted in the participants data sets. Visual inspection of participants data All of the data sets submitted by the 26 participants are shown in Figs. 35. Down each column in these Wgures, the
1
B

consistency in the binding proWles demonstrates the similarity between data sets obtained for a single compound from diVerent participants. Across each row, the diVerent binding proWles demonstrate that the 10 compounds bind the target with diVerent aYnities. Sensorgrams considered to be acceptable for kinetic analysis are shown in black, whereas those shown in red exhibited one or more problems. SigniWcantly Xawed sensorgrams were omitted from the kinetic analysis.
6 7 8 9 10

Response (RU)

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

Time (s)
Fig. 6. Kinetic analysis of data sets shown in Figs. 35. Binding responses (black lines) are overlaid with the Wt of a 1:1 interaction model (red lines). Responses for compounds 1 to 8 were Wt to a simple 1:1 model, and responses for compounds 9 and 10 were Wt to a 1:1 model that included a mass transport term. Rate constants determined from this analysis are summarized in Table 2. Individual data sets omitted from the kinetic analysis are indicated by red Xs, whereas data panels from participants A, J, K, N, O, P, R, and S were excluded entirely. (For interpretation of the references to color in this Wgure legend, the reader is referred to the Web version of this article.)

102

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

1
T

10

Response (RU)

AA

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

45 90

Time (s)
Fig. 6 (continued)

Data set A was obtained using a Biacore 1000 instrument (Fig. 3). Given the challenges in working with this older system, participant A focused on the analysis of the highest aYnity compounds: 8, 9, and 10. Due to the absence of an in-line reference Xow cell in Biacore 1000, changes in bulk refractive during the course of the injection could not be referenced properly. In some cases, it is possible to model the bulk shift, but other deviations in the sensorgrams require the ability to double reference the data. It is surprising that it is possible to detect binding responses for small molecules using the Biacore 1000. Although the data might not be as clean as those with more recent instrument platforms, it is interesting to note that even these very early versions of Biacore technology were capable of small molecule detection at the qualitative level. Data sets BI were obtained using Biacore 2000 instruments (Fig. 3). Overall, the responses are of excellent quality, and 96% of these sensorgrams were included in the kinetic analysis. Data sets JS were obtained using Biacore 3000 instruments (Fig. 4). Much of the data collected from the Biacore 3000 instruments were compromised in quality,

and only 18% of these sensorgrams were included in the kinetic analysis. Common features include large spikes at the beginning and end of the association phase as well as spikes throughout the data sets. Interestingly, data sets Q and S (Fig. 4) were obtained using Biacore 3000 instruments after servicing that included replacing the integrated Xuidic cartridge, and yet the data are still of poor quality. These results suggest that the problem might lie elsewhere in these systems. At this time, we do not know the cause of the high number of failed sensorgrams obtained from these and the other Biacore 3000 instruments. Data sets TZ and AA (Fig. 5) were obtained using Biacore S51 and Biacore T100 instruments, respectively, and 90% of these sensorgrams were included in the kinetic analysis. Kinetic analysis Data sets collected for compounds 18 were Wt to a simple 1:1 model, and data sets for compounds 9 and 10 were Wt to a 1:1 model that included a term for mass transport

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

103

[12]. The Wts of the data sets shown in Figs. 35 are displayed in Fig. 6. In general, each compoundtarget interaction was well described by a 1:1 interaction model. The range of kinetic parameters derived for the 10 compounds is listed in Table 2 and illustrated in the kinetic distribution plot (Fig. 7). The spread of the data points in the horizontal, vertical, and diagonal (perpendicular to the dashed lines) directions shown in Fig. 7 for each compound corresponds to the variability in ka, kd, and KD reported by the panel of participants. In general, data points are tightly clustered and the CV in each of the binding constants is less than 40%: CV(ka) D 34%, CV(kd) D 24%, and CV(KD) D 37%. The spread in the constants determined for compounds 2 and 7 (e.g., CV(KD) D 47 and 87%) resulted from one and four outlier data sets, respectively; the cause of these discrepancies is not known. An above-average CV (32%) in kd for compound 1 is not unexpected given that the dissociation rate for this compound is very fast and therefore this set of sensorgrams contains less information than those in other data sets. A larger experimental error in ka and kd for compounds 9 and 10 relate to the fact that these comTable 2 Rate and equilibrium constants determined from Biacore at 25 C Sample 1 2 3 4 5 6 7 8 9 10 n 15 18 18 18 17 19 19 18 17 18 ka (M1 s1) 8.0 2.3 103 2.6 0.5 104 3.7 1.3 104 6.3 1.4 104 3.9 1.2 104 2.6 0.4 105 2.1 0.9 105 1.7 0.6 105 1.0 0.4 106 3.0 2.1 106 kd (s1) 0.38 0.12 0.083 0.018 0.11 0.01 0.061 0.006 0.037 0.011 0.11 0.03 0.13 0.03 0.12 0.02 0.037 0.012 0.079 0.031 KD ( M) 48 14 3.4 1.6 3.1 1.1 1.0 0.2 0.97 0.17 0.44 0.12 0.85 0.74 0.80 0.28 0.040 0.013 0.031 0.011

pounds have very fast association rates and are inXuenced by mass transport. However, because under transport limited conditions the ka and kd values become coupled, their ratio in fact leads to similar KD values. The changes in one parameter are compensated for by changes in the other parameter, so the ratio of the two (i.e., KD D kd/ka) is not aVected during Wtting. The result is that the CVs for the KD of compound 9 (33%) and compound 10 (35%) compare favorably with the average for the 10 the compounds (CVavg(KD) D 37%). The 10 compoundtarget interactions were also studied in solution using ITC. Fig. 8A shows the heat evolved as compounds 2, 4, 6, and 8 were titrated into solutions of CAII. The titrations shown are representative of those observed for all of the compounds. The raw heats from these titrations were integrated to generate plots of kilocalories/mole of injected compound versus the molar ratio of compound and CAII (Fig. 8B). Importantly, all show a molar ratio of 1 at the midpoint of titration, indicating a 1:1 stoichiometry for each interaction. The KD values from the ITC analyses were plotted against those obtained using SPR (Fig. 9). The aYnities determined from the two methods are highly correlated (99.8%), with the greatest diVerence in KD between the two methods being only twofold for compound 6. Discussion A total of 26 participants contributed to this collaborative study of 10 compounds binding to CAII. To balance sampling throughput and kinetic resolution, the participants examined relatively few (three), but widely diluted (Wvefold), concentrations of each compound. Participants performed the analysis using Wve diVerent Biacore platforms. In our review of the participants data sets, we were

Fig. 7. ka versus kd plot for compounds 110 binding to CAII. IsoaYnity diagonals are shown as dashed lines.

104

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

surprised by the large number of low-quality sensorgrams obtained using Biacore 3000 instruments (Fig. 4). At this time, we do not know the source of the poorer quality data originating from Biacore 3000 instruments and whether or not it is speciWc to this particular set of samples. Initially, we thought that this was perhaps attributable to contaminated Xuidic cartridges. This explanation seems unlikely, however, given that two data sets were collected from Biacore 3000 instruments before (P and R) and after (Q and S) professional servicing during which the Xuidic cartridges were replaced. No improvement in data quality was observed between data sets R and S, and only a partial improvement in sensorgram quality was observed between data sets P and Q. These results suggest that the issue with running this assay using Biacore 3000 instruments may lie elsewhere. The kinetic distribution plot (Fig. 7) of the compounds provides insight into the variation one could expect to see for this type of analysis. We observed greater experimental error in the binding constants for weak interactions, and this was expected because there is less information in these data sets. We also observed greater experimental error for
A
0 25

Time (min)
50 75 100 0 25

Time (min)
50 75 100

compounds that were inXuenced by mass transport. This was not unusual and helps to quantitate the types of variation one would expect to see in the assay. Overall, the experimental error in the reaction parameters for all 10 compounds is small (average 30%), particularly when one considers the challenges associated with analyzing small molecule interactions using SPR. The dissociation constants derived from the SPR and ITC studies of the 10 compounds were in excellent agreement (Fig. 9). This observation, along with the high degree of correlation observed for a number of other experimental systems [6,14], deXates the criticism that the surface-based SPR method alters binding constants. This is because these experiments, like the great majority of Biacore-based experiments, involved the immobilization of ligand to a Xuid-like dextran layer attached to the gold surface and not to the gold surface directly. Clearly, the aYnity of CAII for these small molecules is not signiWcantly altered on immobilization in this manner. Rather than the surface itself producing erroneous binding constants, our experience suggests that sample heterogeneity and inappropriate experimental design and/or data analysis are the most common sources of reported discrepancies between solution- and surfacebased methods. This study also emphasizes the value of kinetic analyses compared with equilibrium-based analyses that yield only aYnity information; the CAII aYnities of compounds 4, 5, and 8 are statistically identical, but these compounds are distinguishable based on their kinetic proWles. This kinetic proWling provides critical information in a drug development setting, for example, in which an investigator wants to understand in detail how changes in a compounds structure aVect its binding activity. In general, we are pleased to see that the variation in the binding constants for all 10 compounds averaged approximately 30%. We recognize that, to some degree, this was an idealized study because each participant was provided with reagents and the experimental protocol, and we certainly
KDSPR = KD ITC

B
-1

cal/sec

10-8

kcal/mol of injectant

cpd
1

data
10
-7

9 10

-5

2 4 5

KDSPR(M)

-9

6 7 8

10

-6

6 85 7 4 23

-13

9 10

10

-5

1
0.5 1.0 1.5 2.0 2.5 3.0

10-4 10-4

10-5

10-6

10-7

10-8

Molar Ratio
Fig. 8. ITC analysis of compounds 110 binding to CAII. (A) Representative titrations of compounds 2, 4, 6, and 8. (B) Plots of kilocalories/mole of injectant versus molar ratio for each of the 10 compounds.

KDITC (M)
Fig. 9. Correlation plot of aYnities determined for each of the 10 compounds using SPR and ITC. The dashed line depicts a correlation of 1.

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94105

105

would expect the standard errors in the binding parameters to increase as the number of experimental variables increases. This benchmark study, however, serves as an excellent tool for educating users and validating that the biosensor technology does in fact work when it is applied properly. These results and discussions should stimulate further interest and conWdence in the use of biosensors throughout the drug discovery process. References
[1] R.L. Rich, D.G. Myszka, Survey of the 1999 surface plasmon resonance biosensor literature, J. Mol. Recogn. 13 (2000) 388407. [2] R.L. Rich, D.G. Myszka, Survey of the year 2000 commercial optical biosensor literature, J. Mol. Recogn. 14 (2001) 273294. [3] R.L. Rich, D.G. Myszka, Survey of the year 2001 commercial optical biosensor literature, J. Mol. Recogn. 15 (2002) 352376. [4] R.L. Rich, D.G. Myszka, A survey of the year 2002 commercial optical biosensor literature, J. Mol. Recogn. 16 (2003) 351382. [5] R.L. Rich, D.G. Myszka, Survey of the year 2003 commercial optical biosensor literature, J. Mol. Recogn. 18 (2005) 139. [6] R.L. Rich, D.G. Myszka, Survey of the year 2004 commercial optical biosensor literature, J. Mol. Recogn. 18 (2005) 431478. [7] D.G. Myszka, Y.N. Abdiche, F. Arisaka, O. Byron, E. Eisenstein, P. Hensley, J.A. Thomson, C.R. Lombardo, F. Schwarz, W. StaVord, M.L. Doyle, The ABRF-MIRG02 study: assembly state, thermodynamic, and kinetic analysis of an enzyme/inhibitor interaction, J. Biomol. Tech. 14 (2003) 247269.

[8] M.J. Cannon, G.A. Papalia, I. Navratilova, R.J. Fisher, L.R. Roberts, K.M. Worthy, A.G. Stephen, G.R. Marchesini, E.J. Collins, D. Casper, H. Qiu, D. Satpaev, S.F. Liparoto, D.A. Rice, I.I. Gorshkova, R.J. Darling, D.B. Bennett, M. Sekar, E. Hommema, A.M. Liang, E.S. Day, J. Inman, S.M. Karlicek, S.J. Ullrich, D. Hodges, T. Chu, E. Sullivan, J. Simpson, A. RaWque, B. Luginbhl, S.N. Westin, M. Bynum, P. Cachia, Y.J. Li, D. Kao, A. Neurauter, M. Wong, M. Swanson, D.G. Myszka, Comparative analyses of a small molecule/enzyme interaction by multiple users of Biacore technology, Anal. Biochem. 330 (2004) 98113. [9] P.S. Katsamba, I. Navratilova, M. Calderon-Cacia, L. Fan, K. Thornton, M. Zhu, T.V. Bos, C. Forte, D. Friend, I. Laird-OVringa, G. Tavares, J. Whatley, E. Shi, A. Widom, K.C. Lindquist, S. Klakamp, A. Drake, D. Bohmann, M. Roell, L. Rose, J. Dorocke, B. Roth, B. Luginbhl, D.G. Myszka, Kinetic analysis of a high-aYnity antibody/ antigen interaction performed by multiple Biacore users, Anal. Biochem. 352 (2006) 208221. [10] A. Cecchi, S.D. Taylor, Y. Liu, B. Hill, D. Vullo, A. Scozzafava, C.T. Supuran, Carbonic anhydrase inhibitors: inhibition of the human isozymes I, II, VA, and IX with a library of substituted diXuoromethanesulfonamides, Bioorg. Med. Chem. Lett. 15 (2005) 51925196. [11] D.G. Myszka, Improving biosensor analysis, J. Mol. Recogn. 12 (1999) 279284. [12] D.G. Myszka, T.A. Morton, M.L. Doyle, I.M. Chaiken, Kinetic analysis of a protein antigenantibody interaction limited by mass transport on an optical biosensor, Biophys. Chem. 64 (1997) 127137. [13] T. Wiseman, S. Williston, J.F. Brandts, L.N. Lin, Rapid measurement of binding constants and heats of binding using a new titration calorimeter, Anal. Biochem. 179 (1989) 131137. [14] R.L. Rich, D.G. Myszka, Advances in surface plasmon resonance biosensor analysis, Curr. Opin. Biotechnol. 11 (2000) 5461.