ANALYTICALBIOCHEMISTRY

Analytical Biochemistry 359 (2006) 94–105

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, USAb Gilead Sciences, Foster City, CA 94404, USA

c Agilent Technologies, Palo Alto, CA 94304, USAd Argonne National Laboratory, Argonne, IL 60439, USA

e Genzyme Corporation, Framingham, MA 01701, USAf Dyaxsa, B-4000 Liege 1, Belgium

g Georgia State University, Marietta, GA 30302, USAh National Cancer Institute, Frederick, MD 21702, USA

i National Institute of Allergies and Infectious Diseases, Bethesda, MD 20892, USAj Roche, Palo Alto, CA 94304, USA

k Reichert Analytical Instruments, Depew, NY 14043, USAl Adamed, 05-152 Czosnow, Poland

m Amgen, Thousand Oaks, CA 91320, USAn diaDexus, South San Francisco, CA 94080, USA

o Wyeth, Andover, MA 01810, USAp Wyeth, Cambridge, MA 02140, USA

q Abbott Laboratories, Abbott Park, IL 60064, USAr University of Tennessee, Knoxville, TN 37996, USAs University of California, Berkeley, CA 94720, USA

t Human Genome Sciences, Rockville, MD 20850, USAu Tibotec, B-2800 Mechelen, Belgium

Received 20 July 2006Available online 7 September 2006

Abstract

In this benchmark study, 26 investigators were asked to characterize the kinetics and aYnities of 10 sulfonamide inhibitors binding tothe enzyme carbonic anhydrase II using Biacore optical biosensors. A majority of the participants collected data that could be Wt to a 1:1interaction 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 variationin the reported constants was observed for compounds with exceptionally weak aYnity and/or fast association rates. The binding con-stants determined using the biosensor correlated well with solution-based titration calorimetry measurements. The results of this study

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

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

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 95

provide insight into the challenges, as well as the level of experimental variation, that one would expect to observe when using Biacoretechnology for small molecule analyses.© 2006 Elsevier Inc. All rights reserved.

Keywords: Biacore; Surface plasmon resonance; Protein–protein interaction; Kinetics; SPR

There is growing interest in applying optical biosensor study was to determine the consistency of the results for

technology, such as surface plasmon resonance (SPR)1-based Biacore instruments, in drug discovery. With thewide array of opportunities oVered by this technology,however, come some challenges. For example, based on ouryearly review of the optical biosensor literature [1–6], weWnd that there is still a need to train investigators in how todesign and perform biosensor experiments as well as in howto properly process and analyze data. And based on com-ments we receive from manuscript reviewers and partici-pants at scientiWc meetings, we see that some skepticismremains regarding the validity of the binding constantsdetermined using biosensors. For the past few years, wehave been addressing both of these issues by coordinatingintra- and inter-technology benchmark studies.

In our Wrst comparative study, 30 SPR biosensor usersillustrated that the aYnity determined from the biosensorfor a small molecule/enzyme interaction matched the valuedetermined in solution using isothermal titration calorime-try (ITC) [7]. These results conWrmed that immobilizationof an enzyme onto the sensor surface does not necessarilyaVect the binding constants as is often presumed by criticsof the biosensor approach. Then, we organized a study with36 participants who analyzed the interaction of a smallmolecule/macromolecular target system having a very fastassociation rate [8]. The results of this test establishedexperimental and data processing protocols for mass trans-port-limited reactions. Recently, we coordinated a studywith 22 users resolving the binding constants for a high-aYnity monoclonal antibody–antigen interaction [9]. Thiswork trained users in how to generate and analyze data fora slowly dissociating system. Together, these benchmarkstudies educated the users of biosensor technology, as wellas the general scientiWc community, with regard to what wecan expect to see in terms of the variability of results withina given experimental system.

The current study builds on this previous work and ismodeled on a growing biosensor application in a drug dis-covery setting: the rapid and reliable determination ofkinetic and aYnity constants for a small panel of com-pounds binding to a single protein target. This informationis becoming crucial to better rationalizing compoundbehavior in bioassays and helping to steer compound selec-tion and design strategies. The goal of the current biosensor

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

small molecule analysis and to discover ways to furtherimprove the application.

The target protein for this work was carbonic anhydraseII, an enzyme responsible for the conversion of carbondioxide to bicarbonate. Current inhibitors of this enzymeare used to treat diseases such as glaucoma and epilepsy,and future drugs targeting this enzyme may lead to treat-ments for cancer and obesity [10]. Thus, the choice of car-bonic anhydrase as the target represents a realistic modelfor small molecule drug studies.

All participants in this study were provided the samereagents and were asked to follow the same experimentalprotocol. We found that a majority of the participants wereable to resolve the binding constants for all 10 compounds.We did, however, identify a suboptimal subset of data (sen-sorgrams that were marred by features such as spikes anddrift) and we explored the causes of these features. We alsocompared the equilibrium dissociation constants obtainedfor the 10 compounds with those determined by ITC. Wefound an excellent correlation between the two methodsacross the wide range in aYnities exhibited by these com-pounds. Our results highlight the strengths of biosensoranalysis 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), N-ethyl-N�-(3-dimethylaminopropyl)carbodiimide (EDC),and ethanolamine HCl, as well as sampling vials, caps, and96-well plates, were obtained from Biacore. Carbonic anhy-drase isozyme II (CAII) from bovine erythrocytes, the 10sulfonamide inhibitors, dimethyl sulfoxide (DMSO), buVerreagents, and general laboratory supplies were purchasedfrom Sigma–Aldrich (St. Louis, MO, USA).

Instrument cleaning

Before beginning the experiment, each participant wasasked to perform a series of instrument cleaning steps.First, the instrument was primed with water, the previouslyused chip was undocked, and a maintenance chip wasdocked. The instrument was then primed Wve times using0.5% (w/v) sodium dodecyl sulfate (SDS), once using water,Wve times using 50 mM glycine (pH 9.5), and once againusing water.

96 Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105

Chip preconditioning

After docking the CM5 sensor chip and priming theinstrument with running buVer, the chip was precondi-tioned 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-buVeredsaline (PBS, 1£D20 mM Na2HPO4–NaH2PO4, 150 mMNaCl, pH 7.4) stock solution. Then 70 ml of this 10£ PBSwas added to 630 ml of degassed deionized water to make1£ PBS buVer, and this buVer was Wltered. Then 18 ml ofDMSO (also provided to the participant) was added to600 ml of the 1£ PBS buVer. DMSO is commonly used as asolvent for small molecule studies. Although the Wnal con-centration of DMSO in this buVer via this addition is actu-ally 2.97%, for simplicity this buVer is termed “runningbuVer with 3% DMSO”. The remaining 100 ml of 1£ PBSwas divided into two 50-ml portions. One of these portionswas set aside for use as “immobilization buVer”, and theother was set aside as “sample preparation buVer with noDMSO” for use in analyte preparation.

Enzyme immobilization

Participants were provided with aliquots of 100�g oflyophilized CA II that were centrifuged and dissolved in400�l of 10 mM sodium acetate (pH 4.9) after the biosensorwas primed with immobilization buVer and equilibrated to25 °C. Using a Xow rate of 20 �l/min, the surface of Xow cell1 was activated for 7 min using a mixture of 0.1 M NHS and0.4 M EDC, 0.25 mg/ml CAII was injected for 7 min, andresidual activated groups on the surface were blocked by a7-min injection of 1 M ethanolamine (pH 8.5). Users of S51platforms activated the surfaces via sequential injections ofNHS and then EDC for 7 min. This was followed by injec-tion 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 ofethanolamine at a Xow rate of 10 �l/min. On average, 6900response units (RU) of CAII were immobilized. For this setof studies, we chose to leave Xow cell 2 unmodiWed as a ref-erence surface.

Analyte preparation

For simplicity, each of the compounds analyzed in thisstudy was assigned a number (Table 1). Participants wereprovided with a stock solution of each compound dissolvedin DMSO. To prepare the samples for analysis, 30�l of eachcompound solution was added to 1 ml of sample preparationbuVer with no DMSO and mixed thoroughly. Preparation ofanalyte in this manner ensures that the concentration ofDMSO is matched with that of running buVer with 3%DMSO. Three lower concentrations of each of the 10 com-pounds were then prepared via Wvefold serial dilutions intorunning buVer with 3% DMSO. For analyses using Biacore1000, 2000, and 3000, 100�l of each compound at the highestconcentration was added to 400�l of running buVer with 3%DMSO and this dilution was repeated. Samples and buVerblanks were aliquotted in 7-mm Biacore vials, which werethen capped with soft rubber caps (BR-1005-55) that resealafter puncture to minimize evaporation. Using these samplevolumes and the resealable caps permitted the three concen-trations of each compound to be sampled twice. Biacore S51-and T100-based analyses used 96-well plates for sampling.Dilutions were performed directly in the plates by taking50�l from wells containing 250�l of the highest analyte con-centration and diluting successively into wells containing200�l of running buVer with 3% DMSO.

To correct for the excluded volume eVect, a DMSO cali-bration series was prepared by adding 15 �l of DMSO and150 �l of water to separate 1.5-ml aliquots of running buVerwith DMSO to obtain “+” and “¡” solutions. Then 400,300, 200, 100, and 0 �l of the “+” solution were dispensedinto Eppendorf tubes labeled d1 to d5, and 0, 100, 200, 300,and 400 �l of the “¡” solution were added to tubes d1 to d5so that each tube contained a Wnal volume of 400 �l. TheWve DMSO calibration solutions were either aliquotted into7-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 therunning buVer and sample solutions, water (0.4% of thetotal volume of running buVer) was added to the runningbuVer with 3% DMSO. By making the start of each injec-

Table 1Compounds examined in this study

Sample Compound Molecular mass (Da) Concentration (�M)

1 (§)-Sulpiride 341 2, 10, 502 4-(Aminomethyl)benzenesulfonamide hydrochloride hydrate 223 2, 10, 503 Sulfanilamide 172 0.4, 2, 104 Furosemide 331 0.4, 2, 105 4-Carboxybenzenesulfonamide 201 0.4, 2, 106 Dansylamide 250 0.08, 0.4, 27 1,3-Benzenedisulfonamide 236 0.08, 0.4, 28 Benzenesulfonamide 157 0.4, 2, 109 7-Fluoro-2,1,3-benzoxadiazole-4-sulfonamide 217 0.016, 0.08, 0.4

10 Acetazolamide 222 0.016, 0.08, 0.4

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 97

tion easily identiWable, this mismatch aids in data process-ing [8]. The instrument was primed three times with thisspiked running buVer and then normalized using 40% (v/v)glycerol for Biacore 1000 and 2000 or 70% (v/v) glycerol forBiacore 3000, S51, and T100.

Analyte injection order and analysis method

Data were collected at the highest collection rate possi-ble and at 25 °C. An initial series of buVer blanks wasinjected Wrst to fully equilibrate the system. The DMSO cal-ibration series was then injected, followed by two addi-tional blank injections. Finally, the individual compound

Fig. 1. High-quality kinetic data set. Duplicate analyses of 0, 0.4, 2.0, and10 �M furosemide binding to 6450 RU of immobilized CAII. Data shownare from the analysis of compound 4 by participant H and were collectedusing a Biacore 2000 instrument.

Res

pons

e (R

U)

Time (s)0 60 120 180

0

25

50

samples were tested (from lowest to highest concentrationswithin each series) and each compound series was sepa-rated 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 monitoredfor 140 s. The analyte injection was followed by an Extra-Clean wash command that automatically Xushes the sam-ple delivery system with running buVer (for Biacore 2000and 3000, a 15-�l injection of running buVer was alsoincluded as an additional wash step). Analytes were sam-pled twice using Biacore 1000, 2000, and 3000 and weresampled 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-refer-enced [11] association and dissociation phase data for com-pounds 1 to 8 were globally Wt to a simple 1:1 interactionmodel (A + BDAB). Compounds 9 and 10 were Wt to a 1:1interaction model that included a mass transport term(AoDA + BDAB) [12]. Means and standard deviationswere computed using Microsoft Excel software. The experi-mental error or coeYcient of variation (CV) was calculatedas (Standard Deviation/Mean)£ 100.

Isothermal titration calorimetry

Protein was dissolved in 1£PBS and buVer exchanged tominimize any contaminants from lyophilization. Compoundwas diluted in the same buVer as protein to minimize heat of

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

A B C

D E F

98 Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105

dilution eVects from buVer mismatch. Experiments were per-formed with the compound in the syringe and the protein inthe cell. Protein concentration was measured using an extinc-tion coeYcient at A280 of 50,070L mol¡1 cm¡1. Isothermaltitration calorimetry experiments were performed using aMicrocal VP-ITC instrument (MicroCal, Northampton, MA,USA) [13]. Experimental parameters included: 30 injections ofcompound, with an initial Wrst injection of 3�l and subse-quent 10-�l injections of compound into protein with 240sbetween each injection and a stirring speed of approximately300. Experiments were performed at 25 °C. The heat of dilu-tion was taken either from the points after saturation, controlexperiment of compound into buVer, or from a minimizationof the �2 value. The concentrations of compounds titratedwere 960, 666, 600, and 400�M for compounds 1 to 4, respec-

tively, 350�M for compounds 5–8, and 300�M for com-pounds 9 and 10. Corresponding concentrations of carbonicanhydrase were 44, 65, and 38�M for titrations of com-pounds 1–3 and 28�M for titrations of compounds 4–10.Data were analyzed with a single-site binding model using theOrigin software provided by MicroCal. To minimize thedependency of the Wtting parameters for compound 1, theweakest inhibitor, the stoichiometry value was set to 1 assum-ing a 1:1 ratio, which has been well established for this system.

Results

In this benchmark study, participants used Biacore bio-sensor technology to characterize the interactions of 10 com-pounds with the enzyme CAII. We provided each participant

Fig. 3. Data sets obtained using Biacore 1000 (A) and Biacore 2000 (B–I) instruments. Columns 1 through 10 correspond to the numbering of compoundsin 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 thisWgure legend, the reader is referred to the Web version of this article.)

A

B

C

D

E

F

G

Res

pons

e (R

U)

H

I

90450 90450 90450 90450 90450 90450 90450 9045090450 90450

Time (s)

1 2 3 4 5 6 7 8 9 10

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 99

with a detailed experimental protocol as well as stock solu-tions of the enzyme, compounds, and buVer. CAII wasimmobilized on one Xow cell surface of a CM5 sensor chipusing standard amine-coupling chemistry and an unmodiWedXow cell surface served as a reference. Immobilization densi-ties ranged from 2400 to 13,000 RU; this variation was usefulin determining whether there were any systematic trends inthe results with surface density. Three concentrations of eachcompound, prepared in a Wvefold dilution series, were ana-lyzed for CAII binding at 25°C. Each concentration serieswas bracketed by buVer blanks that were included for doublereferencing [11]. Because each compound dissociated com-pletely from the CAII surface within minutes after the end ofthe association phase, no regeneration step was required.

Participants using Biacore 1000, 2000, and 3000 platformsdid repeat injections of each analyte concentration, whereasthe Biacore S51 and T100 users performed a single injectionof each.

Good and bad sensorgrams

To illustrate how we interpreted the responses generatedby the diVerent users, we Wrst highlight examples of goodand bad data. An example of a good data set obtained forone compound is shown in Fig. 1. From inspection of thisdata set, we can tell that the experiment was designed andexecuted properly, the biosensor was performing optimally,and the obtained responses are reliable. Hallmarks that

Fig. 4. Data sets obtained using Biacore 3000 (J–S) 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.

Time (s)

Res

pons

e (R

U)

K

L

M

N

O

P

Q

R

J

90450 90450 90450 90450 90450 90450 90450 9045090450 90450

1 2 3 4 5 6 7 8 9 10

S

100 Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105

make this data set a good one include responses that areconcentration dependent, replicate injections that overlay,and clearly discernible exponential curvature during boththe association and dissociation phases. When data sets areof high quality like the one shown in Fig. 1, they often canbe described by a 1:1 interaction model. And when data setsare of poorer quality (e.g., when they exhibit unusual pro-Wles or can be described only by a complex model), theanomalies and/or complexity most often can be eliminatedby optimizing the assay design.

Although we intentionally chose well-characterized,well-behaved 1:1 interactions for this study, the quality ofsome of the data sets generated by our study participantswas not on par with the example shown in Fig. 1. Six of thecommon anomalous sensorgram features we observed insome data sets are depicted in Fig. 2.

For example, signiWcant spikes at the both the beginningand end of the association phase apparent after referencing(Fig. 2A) indicate poor-quality injections that may becaused by a fouled integrated Xuidic cartridge. Although it

is common to have a few data points producing spikes atthe transitions between running buVer and the analyte sam-ple, spikes that occur over a larger window of time (e.g.,>2 s) can result in the loss of kinetic information when thebinding responses are fast. Frequent spikes throughout asensorgram or sets of sensorgrams (Fig. 2B) often are indic-ative 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 buVerblanks shown in Fig. 2C would make it diYcult to choosewhich blanks to use for double referencing. Also, inconsis-tencies in buVer injections may indicate sample carryover.The inability to achieve steady state after a fast associationphase (Fig. 2D) in a system known to be 1:1 suggests thatthere may be problems with reagent preparation, the bio-sensor’s Xuidic system, or both. Sensorgrams that, after ref-erencing, dip down during the association phase, and insome cases drop below the baseline during the dissociationphase (Fig. 2E), may result from nonspeciWc binding to thereference surface. Sustained jumps in refractive index

Fig. 5. Data sets obtained using Biacore S51 (T–Z) and Biacore T100 (AA) instruments. Columns 1 through 10 correspond to the compound numbering givenin 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 tocollect 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.)

Time (s)

Res

pons

e (R

U)

T

U

V

W

X

Y

Z

AA

90450 90450 90450 90450 90450 90450 90450 9045090450 90450

1 2 3 4 5 6 7 8 9 10

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 101

(Fig. 2F) make obtaining reasonable Wts impossible withoutelaborate data processing. Sensorgrams that displayed anyof the characteristics illustrated in Fig. 2 were highlightedin the participants’ data sets.

Visual inspection of participants’ data

All of the data sets submitted by the 26 participants areshown in Figs. 3–5. Down each column in these Wgures, the

consistency in the binding proWles demonstrates the simi-larity between data sets obtained for a single compoundfrom diVerent participants. Across each row, the diVerentbinding proWles demonstrate that the 10 compounds bindthe target with diVerent aYnities. Sensorgrams consideredto be acceptable for kinetic analysis are shown in black,whereas those shown in red exhibited one or more prob-lems. SigniWcantly Xawed sensorgrams were omitted fromthe kinetic analysis.

Fig. 6. Kinetic analysis of data sets shown in Figs. 3–5. 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 trans-port term. Rate constants determined from this analysis are summarized in Table 2. Individual data sets omitted from the kinetic analysis are indicated byred 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 Wgurelegend, the reader is referred to the Web version of this article.)

B

C

D

E

F

G

H

I

Res

pons

e (R

U)

L

M

Q

90450 90450 90450 90450 90450 90450 90450 9045090450 90450

Time (s)

1 2 3 4 5 6 7 8 9 10

102 Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105

Data set A was obtained using a Biacore 1000 instru-ment (Fig. 3). Given the challenges in working with thisolder system, participant A focused on the analysis of thehighest aYnity compounds: 8, 9, and 10. Due to the absenceof an in-line reference Xow cell in Biacore 1000, changes inbulk refractive during the course of the injection could notbe referenced properly. In some cases, it is possible tomodel the bulk shift, but other deviations in the sensor-grams require the ability to double reference the data. It issurprising that it is possible to detect binding responses forsmall molecules using the Biacore 1000. Although the datamight not be as clean as those with more recent instrumentplatforms, it is interesting to note that even these very earlyversions of Biacore technology were capable of small mole-cule detection at the qualitative level.

Data sets B–I were obtained using Biacore 2000 instru-ments (Fig. 3). Overall, the responses are of excellent qual-ity, and 96% of these sensorgrams were included in thekinetic analysis. Data sets J–S were obtained using Biacore3000 instruments (Fig. 4). Much of the data collected fromthe Biacore 3000 instruments were compromised in quality,

and only 18% of these sensorgrams were included in thekinetic analysis. Common features include large spikes atthe beginning and end of the association phase as well asspikes throughout the data sets. Interestingly, data sets Qand S (Fig. 4) were obtained using Biacore 3000 instru-ments after servicing that included replacing the integratedXuidic cartridge, and yet the data are still of poor quality.These results suggest that the problem might lie elsewherein these systems. At this time, we do not know the cause ofthe high number of failed sensorgrams obtained from theseand the other Biacore 3000 instruments.

Data sets T–Z and AA (Fig. 5) were obtained using Bia-core S51 and Biacore T100 instruments, respectively, and90% of these sensorgrams were included in the kinetic anal-ysis.

Kinetic analysis

Data sets collected for compounds 1–8 were Wt to a sim-ple 1:1 model, and data sets for compounds 9 and 10 wereWt to a 1:1 model that included a term for mass transport

Fig. 6 (continued)

Res

pons

e (R

U)

90450 90450 90450 90450 90450 90450 90450 9045090450 90450

Time (s)

T

U

V

W

X

Y

Z

AA

1 2 3 4 5 6 7 8 9 10

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 103

[12]. The Wts of the data sets shown in Figs. 3–5 aredisplayed in Fig. 6. In general, each compound–targetinteraction was well described by a 1:1 interaction model.The range of kinetic parameters derived for the 10compounds is listed in Table 2 and illustrated in the kineticdistribution plot (Fig. 7). The spread of the data points inthe horizontal, vertical, and diagonal (perpendicular to thedashed lines) directions shown in Fig. 7 for each compoundcorresponds to the variability in ka, kd, and KD reportedby the panel of participants. In general, data points aretightly clustered and the CV in each of the bindingconstants is less than 40%: CV(ka)D34%, CV(kd)D24%,and CV(KD)D 37%.

The spread in the constants determined for compounds2 and 7 (e.g., CV(KD)D47 and 87%) resulted from one andfour outlier data sets, respectively; the cause of these dis-crepancies is not known. An above-average CV (32%) in kdfor compound 1 is not unexpected given that the dissocia-tion rate for this compound is very fast and therefore thisset of sensorgrams contains less information than those inother data sets. A larger experimental error in ka and kd forcompounds 9 and 10 relate to the fact that these com-

Table 2Rate and equilibrium constants determined from Biacore at 25 °C

Sample n ka (M¡1 s¡1) kd (s

¡1) KD (�M)

1 15 8.0§ 2.3£ 103 0.38 § 0.12 48 § 142 18 2.6§ 0.5£ 104 0.083 § 0.018 3.4§ 1.63 18 3.7§ 1.3£ 104 0.11 § 0.01 3.1§ 1.14 18 6.3§ 1.4£ 104 0.061 § 0.006 1.0§ 0.25 17 3.9§ 1.2£ 104 0.037 § 0.011 0.97 § 0.176 19 2.6§ 0.4£ 105 0.11 § 0.03 0.44 § 0.127 19 2.1§ 0.9£ 105 0.13 § 0.03 0.85 § 0.748 18 1.7§ 0.6£ 105 0.12 § 0.02 0.80 § 0.289 17 1.0§ 0.4£ 106 0.037 § 0.012 0.040§ 0.013

10 18 3.0§ 2.1£ 106 0.079 § 0.031 0.031§ 0.011

pounds have very fast association rates and are inXuencedby mass transport. However, because under transport lim-ited conditions the ka and kd values become coupled, theirratio in fact leads to similar KD values. The changes in oneparameter are compensated for by changes in the otherparameter, so the ratio of the two (i.e., KDDkd/ka) is notaVected during Wtting. The result is that the CVs for the KDof compound 9 (33%) and compound 10 (35%) comparefavorably with the average for the 10 the compounds(CVavg(KD)D 37%).

The 10 compound–target interactions were also studiedin solution using ITC. Fig. 8A shows the heat evolved ascompounds 2, 4, 6, and 8 were titrated into solutions ofCAII. The titrations shown are representative of thoseobserved for all of the compounds. The raw heats fromthese titrations were integrated to generate plots of kiloca-lories/mole of injected compound versus the molar ratio ofcompound and CAII (Fig. 8B). Importantly, all show amolar ratio of 1 at the midpoint of titration, indicating a 1:1stoichiometry for each interaction.

The KD values from the ITC analyses were plottedagainst those obtained using SPR (Fig. 9). The aYnitiesdetermined from the two methods are highly correlated(99.8%), with the greatest diVerence in KD between the twomethods being only twofold for compound 6.

Discussion

A total of 26 participants contributed to this collabora-tive study of 10 compounds binding to CAII. To balancesampling throughput and kinetic resolution, the partici-pants examined relatively few (three), but widely diluted(Wvefold), concentrations of each compound. Participantsperformed the analysis using Wve diVerent Biacore plat-forms. In our review of the participants’ data sets, we were

Fig. 7. ka versus kd plot for compounds 1–10 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) 94–105

surprised by the large number of low-quality sensorgramsobtained using Biacore 3000 instruments (Fig. 4). At thistime, we do not know the source of the poorer quality dataoriginating from Biacore 3000 instruments and whether ornot it is speciWc to this particular set of samples. Initially,we thought that this was perhaps attributable to contami-nated Xuidic cartridges. This explanation seems unlikely,however, given that two data sets were collected from Bia-core 3000 instruments before (P and R) and after (Q and S)professional servicing during which the Xuidic cartridgeswere replaced. No improvement in data quality wasobserved between data sets R and S, and only a partialimprovement in sensorgram quality was observed betweendata sets P and Q. These results suggest that the issue withrunning this assay using Biacore 3000 instruments may lieelsewhere.

The kinetic distribution plot (Fig. 7) of the compoundsprovides insight into the variation one could expect to seefor this type of analysis. We observed greater experimentalerror in the binding constants for weak interactions, andthis was expected because there is less information in thesedata sets. We also observed greater experimental error for

Fig. 8. ITC analysis of compounds 1–10 binding to CAII. (A) Representa-tive titrations of compounds 2, 4, 6, and 8. (B) Plots of kilocalories/mole ofinjectant versus molar ratio for each of the 10 compounds.

Molar Ratio0.5 1.0 1.5 2.0 2.5 3.0

kcal

/mol

of i

njec

tant

-13

-9

-5

-1

cpd data 1

2

3

4

5

6

7

8

9

10

μcal

/sec

Time (min) Time (min)0 50 75 25 100 50 75 250

2

4

6

8

A

B

100

compounds that were inXuenced by mass transport. Thiswas not unusual and helps to quantitate the types of varia-tion one would expect to see in the assay. Overall, theexperimental error in the reaction parameters for all 10compounds is small (average »30%), particularly when oneconsiders the challenges associated with analyzing smallmolecule interactions using SPR.

The dissociation constants derived from the SPR andITC studies of the 10 compounds were in excellent agree-ment (Fig. 9). This observation, along with the high degreeof correlation observed for a number of other experimentalsystems [6,14], deXates the criticism that the surface-basedSPR method alters binding constants. This is because theseexperiments, like the great majority of Biacore-based exper-iments, involved the immobilization of ligand to a Xuid-likedextran layer attached to the gold surface and not to thegold surface directly. Clearly, the aYnity of CAII for thesesmall molecules is not signiWcantly altered on immobiliza-tion in this manner. Rather than the surface itself produc-ing erroneous binding constants, our experience suggeststhat sample heterogeneity and inappropriate experimentaldesign and/or data analysis are the most common sourcesof reported discrepancies between solution- and surface-based methods.

This study also emphasizes the value of kinetic analysescompared with equilibrium-based analyses that yield onlyaYnity information; the CAII aYnities of compounds 4, 5,and 8 are statistically identical, but these compounds aredistinguishable based on their kinetic proWles. This kineticproWling provides critical information in a drug develop-ment setting, for example, in which an investigator wants tounderstand in detail how changes in a compound’s struc-ture aVect its binding activity.

In general, we are pleased to see that the variation in thebinding constants for all 10 compounds averaged approxi-mately 30%. We recognize that, to some degree, this was anidealized study because each participant was provided withreagents and the experimental protocol, and we certainly

Fig. 9. Correlation plot of aYnities determined for each of the 10 com-pounds using SPR and ITC. The dashed line depicts a correlation of 1.

10-4

10-6

10-7

KD

SP

R(M

)

KDITC (M)

KDSPR = KD

ITC

10-5 10-6 10-7

10-5

109

10-4

10-8

10-8

75

6

48

23

1

Comparative analysis using Biacore technology / G.A. Papalia et al. / Anal. Biochem. 359 (2006) 94–105 105

would expect the standard errors in the binding parametersto increase as the number of experimental variablesincreases. This benchmark study, however, serves as anexcellent tool for educating users and validating that thebiosensor technology does in fact work when it is appliedproperly. These results and discussions should stimulatefurther interest and conWdence in the use of biosensorsthroughout the drug discovery process.

References

[1] R.L. Rich, D.G. Myszka, Survey of the 1999 surface plasmon reso-nance biosensor literature, J. Mol. Recogn. 13 (2000) 388–407.

[2] R.L. Rich, D.G. Myszka, Survey of the year 2000 commercial opticalbiosensor literature, J. Mol. Recogn. 14 (2001) 273–294.

[3] R.L. Rich, D.G. Myszka, Survey of the year 2001 commercial opticalbiosensor literature, J. Mol. Recogn. 15 (2002) 352–376.

[4] R.L. Rich, D.G. Myszka, A survey of the year 2002 commercial opti-cal biosensor literature, J. Mol. Recogn. 16 (2003) 351–382.

[5] R.L. Rich, D.G. Myszka, Survey of the year 2003 commercial opticalbiosensor literature, J. Mol. Recogn. 18 (2005) 1–39.

[6] R.L. Rich, D.G. Myszka, Survey of the year 2004 commercial opticalbiosensor literature, J. Mol. Recogn. 18 (2005) 431–478.

[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-MIRG’02 study: assembly state, thermody-namic, and kinetic analysis of an enzyme/inhibitor interaction, J. Bio-mol. Tech. 14 (2003) 247–269.

[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. Dar-ling, 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. Luginbühl, 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 multi-ple users of Biacore technology, Anal. Biochem. 330 (2004) 98–113.

[9] P.S. Katsamba, I. Navratilova, M. Calderon-Cacia, L. Fan, K. Thorn-ton, M. Zhu, T.V. Bos, C. Forte, D. Friend, I. Laird-OVringa, G. Tav-ares, J. Whatley, E. Shi, A. Widom, K.C. Lindquist, S. Klakamp, A.Drake, D. Bohmann, M. Roell, L. Rose, J. Dorocke, B. Roth, B. Lug-inbühl, D.G. Myszka, Kinetic analysis of a high-aYnity antibody/antigen interaction performed by multiple Biacore users, Anal. Bio-chem. 352 (2006) 208–221.

[10] A. Cecchi, S.D. Taylor, Y. Liu, B. Hill, D. Vullo, A. Scozzafava, C.T.Supuran, Carbonic anhydrase inhibitors: inhibition of the human iso-zymes I, II, VA, and IX with a library of substituted diXuoromethane-sulfonamides, Bioorg. Med. Chem. Lett. 15 (2005) 5192–5196.

[11] D.G. Myszka, Improving biosensor analysis, J. Mol. Recogn. 12(1999) 279–284.

[12] D.G. Myszka, T.A. Morton, M.L. Doyle, I.M. Chaiken, Kinetic analy-sis of a protein antigen–antibody interaction limited by mass trans-port on an optical biosensor, Biophys. Chem. 64 (1997) 127–137.

[13] T. Wiseman, S. Williston, J.F. Brandts, L.N. Lin, Rapid measurementof binding constants and heats of binding using a new titration calo-rimeter, Anal. Biochem. 179 (1989) 131–137.

[14] R.L. Rich, D.G. Myszka, Advances in surface plasmon resonance bio-sensor analysis, Curr. Opin. Biotechnol. 11 (2000) 54–61.