Determination of Dissociation Constants for Protein− Ligand

ACS Chemical Biology 2014 9 (1), 218-226 ... Mass Spectrometry Reveals That the Antibiotic Simocyclinone D8 Binds to DNA ... by Mass Spectrometric Met...
61 downloads 0 Views 87KB Size
Anal. Chem. 2004, 76, 4325-4331

Determination of Dissociation Constants for Protein-Ligand Complexes by Electrospray Ionization Mass Spectrometry Agneta Tjernberg,*,†,‡ Sofi Carno 1 ,† Frida Oliv,† Kurt Benkestock,† Per-Olof Edlund,† ‡,§ William J. Griffiths, and Dan Halle´n†

Departments of Structural Chemistry and Preclinical R&D, Biovitrum AB, SE-112 76 Stockholm, Sweden

A fully automated biophysical assay based on electrospray ionization mass spectrometry (ESI-MS) for the determination of the dissociation constants (KD) between soluble proteins and low molecular mass ligands is presented. The method can be applied to systems where the relative MS response of the protein and the protein-ligand complexes do not reflect relative concentrations. Thus, the employed approach enables the use of both electrostatically and nonpolar bound complexes. The dynamic range is wider than for most biological assays, which facilitates the process of establishing a structure-activity relationship. This fully automated ESI-MS assay is now routinely used for ligand screening. The entire procedure is described in detail using hGHbp, a 25-kDa extracellular soluble domain of the human growth hormone receptor, as a model protein. There are several biophysical techniques and methods for quantitative studies on noncovalent interactions between soluble proteins and low molecular mass molecules. For many reasons, methods that do not need labeling of any of the interacting molecules, such as surface plasmon resonance, nuclear magnetic resonance, isothermal titration calorimetry (ITC), and electrospray ionization mass spectrometry (ESI-MS), are mostly preferred. When large numbers of compounds need to be studied, as in pharmaceutical screening for binders, the method of choice needs to be simple, robust, and easily automated in order to obtain high sample throughput. The interest in using ESI-MS as a tool for detecting noncovalent interactions started with the first publications by Ganem et al. as well as by Katta and Chait.1-3 The question was whether the interactions in the gas phase resemble those formed in the solution, and whether it was possible to quantify the interactions. * To whom the correspondence should be addressed. Phone: +46 8 697 20 00. Fax: +46 8 697 23 19. E-mail: [email protected]. † Biovitrum AB, Stockholm. ‡ Department of Medical Biochemistry and Biophysics, Karolinska Institute, SE-17177 Stockholm, Sweden. § School of Pharmacy, University of London, 29-39 Brunswich Square, London WC1N1AX. (1) Ganem, B.; Li, Y. T.; Henion, J. D. J. Am. Chem. Soc. 1991, 113, 62946296. (2) Ganem, B.; Li, Y. T.; Henion, J. D. J. Am. Chem. Soc. 1991, 113, 78187819. (3) Katta, V.; Chait, B. T. J. Am. Chem. Soc. 1991, 113, 8534-8535. 10.1021/ac0497914 CCC: $27.50 Published on Web 06/12/2004

© 2004 American Chemical Society

It was later shown that the interactions could be quantified and that the affinities obtained from ESI-MS analyses reflect the situation in solution.4-11 Both direct affinity determinations and competition measurements have been used to determine affinities. A common feature has been to use the relative MS responses of the host molecule, usually a protein, and the complex, reflecting the relative concentrations in solution. A limitation with this type of approach is that it is only valid for complexes dominated by polar or ionic interactions, where the dissociation of the complex in gaseous phase can be neglected. For a low molecular mass molecule that is ionic or highly polar, the Gibbs energy favors the compound to be bound to a protein in the gaseous phase prior to be dissociated. It would energetically be too costly to dissociate a highly polar complex in the gaseous phase. On the other hand, a compound forming a complex that has significant contributions from nonpolar interactions is thermodynamically favored as a dissociated molecule in the gaseous phase. There are even complexes with significant nonpolar interaction contributions that dissociate completely in the gaseous phase. Thus, the relative MS responses do not reflect the situation in the aqueous phase for all types of systems, and the referred approach will in many cases give incorrect affinities. In this report, we demonstrate a procedure where noncovalent complexes can be investigated regardless of the type of interactions. The binding affinities between proteins and low molecular mass ligands are determined from direct binding experiments as well as from competition experiments with a reference ligand. This is possible by the introduction of an MS response factor, which reflects the MS response of a fully ligand-saturated protein. The response factor is also a measure of the stability of the complex in the gaseous phase. The response factor of a complex having a (4) Bligh, S. W. A.; Haley, T.; Lowe, P. N. J. Mol. Recognit. 2003, 16, 139-147. (5) Griffey, R. H.; Hofstadler, S. A.; Sannes-Lowery, K. A.; Ecker, D. J.; Crooke, S. T. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 10129-10133. (6) Jorgensen, T. J. D.; Roepstorff, P.; Heck, A. J. R. Anal. Chem. 1998, 70, 4427-4432. (7) Greig, M. J.; Gaus, H.; Cummins, L. L.; Sasmor, H.; Griffey, R. H. J. Am. Chem. Soc. 1995, 117, 10765-10766. (8) Lim, H. K.; Hsieh, F. Y. L.; Ganem, B.; Henion, J. J. Mass Spectrom. 1995, 30, 708-714. (9) Loo, J. A.; Hu, P. F.; McConnell, P.; Mueller, W. T.; Sawyer, T. K.; Thanabal, V. J. Am. Soc. Mass Spectrom 1997, 8, 234-243. (10) Sannes-Lowery, K. A.; Griffey, R. H.; Hofstadler, S. A. Anal. Biochem. 2000, 280, 264-271. (11) Zhang, S.; Van Pelt, C. K.; Wilson, D. B. Anal. Chem. 2003, 75, 30103018.

Analytical Chemistry, Vol. 76, No. 15, August 1, 2004 4325

value identical to the MS response of the protein when the ligand is not present shows high stability in the gaseous phase. In the present study, the MS responses of the saturated complexes are much lower than the MS response of the protein in a ligand-free sample. Hence, the complexes easily dissociate in the gaseous phase, due to their nonpolar interactions. In the competition experiments, only the intensity of the complex between the protein and the reference ligand is measured. Thus, it is only the displacement of the reference ligand from the binding site of the protein that is measured. We do this to avoid differences in response factors for different complexes. The principle of measuring MS response data of noncovalent complexes has previously been used12 for determinations of binding affinities of crown ether-alkali metal complexes, and the principle is further discussed in a review article.13 However, the procedure of calculating the dissociation constants differs from our method. In the work of Kempen and Brodbelt,12 a calibration curve for the reference ligand complex is needed. In this report, it is also demonstrated that the procedure can be applied as a robust screening method. As a model system, the extracellular soluble domain of human growth hormone receptor, hGHbp, was used. This protein forms complexes with small ligands mainly by nonpolar interactions. EXPERIMENTAL SECTION Materials. All reagents used were of the highest purity available. Ammonium acetate and DMSO were both from Merck (Darmstadt, Germany). The soluble domain of human growth hormone receptor (hGHbp) was produced in Escherichia coli essentially as described in the literature14 and purified according to Sundstro¨m et al.15 All the ligands used were synthesized at Biovitrum AB. Ligands denoted compounds 1-6 were neutral nonpolar compounds from the compound collection at Biovitrum AB. Mass Spectrometry. The ESI mass spectra were recorded on a Q-TOF Ultima API instrument (Micromass, Manchester, U.K.), operating in positive ion mode, and equipped with an extended 8K Quadrupole. The importance of optimization of the MS parameters in order to preserve noncovalent binding in the gas phase has been described for the Micromass Q-TOF 1 instrument.16,17 The optimal instrument conditions on the Q-TOF Ultima API instrument were as follows: For the ESI interface, the capillary voltage was set to 3000 V, cone voltage 100 V, source block temperature 45 °C, desolvation temperature 60 °C, desolvation gas flow 200 L/h, and cone gas flow 50 L/h. An elevated instrument pressure in the source region, as well as collisional cooling, was applied in order to preserve the noncovalent interactions in the gas phase. We found that a critical parameter for optimal preservation of the protein-ligand complex on our Q-TOF Ultima instrument was to keep a pirani pressure at 2 mbar. Ions (12) Kempen, E. C.; Brodbelt, J. S. Anal. Chem. 2000, 72, 5411-5416. (13) Daniel, J. M.; Friess, S. D.; Rajagopalan, S.; Wendt, S.; Zenobi, R. Int. J. Mass Spectrom. 2002, 216, 1-27. (14) Fuh, G.; Mulkerrin, M. G.; S., B.; McFarland, N.; Brochier, M.; Bourell, J. H.; Light, D. R.; Wells, J. A. J. Biol. Chem. 1990, 265, 3111-3115. (15) Sundstro ¨m, M.; Lundqvist, T.; Ro¨din, J.; Giebel, L. B.; Milligan, D.; Norstedt, G. J. Biol. Chem. 1996, 271, 32197-32203. (16) van Berkel, W. J. H.; van den Heuvel, R. H. H.; Versluis, C.; Heck, A. J. R. Protein Sci. 2000, 9, 435-439. (17) Tito, M. A.; Miller, J.; Griffin, K. F.; Williamson, E. D.; Titball, R. W.; Robinson, C. V. Protein Sci. 2001, 10, 2408-2413.

4326

Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

were scanned from m/z 1000 to 5000, with data accumulation of 2 s per spectrum and an interscan time delay of 0.1 s. Mass spectra were averaged over 80 scans. The spectra were deconvoluted over the mass range 23 200-25 800 Da. The deconvoluted spectra were generated using the MaxEnt1 deconvolution program. The MS response referred to in this paper was defined as the intensity of the peak height corresponding to the protein-ligand complex after deconvolution of the MS spectrum. The mobile phase used was 10 mM ammonium acetate pH 6.8, and the sample flow rate was 5 µL/min. A volume of 2 µL was injected every 6 min from a CapLC pump system (Waters, Milford, MA) equipped with an autosampler. Methods. The protein hGHbp was dialyzed off-line against 10 mM ammonium acetete pH 6.8, by means of a Slide-a-lyzer (Pierce, Rockford, IL), MWCO 10000. The reference ligand stock solution, 20 mM dissolved in DMSO, was diluted in 10 mM ammonium acetate to 250 µM. The stock solutions for the competing ligands, 20 mM dissolved in DMSO, were further diluted in 10 mM ammonium acetate to 250 µM for single point measurements and to appropriate concentrations between 250 µM and 5 mM for dose-response measurements. Final sample concentrations of the hGHbp were 1-8 µM, and the concentrations of the reference ligands were in all experiments 5 times higher than the protein concentration. All solutions were filtered through a 0.45-µm cellulose acetate filter to remove particles. The samples were prepared fresh on the day of analysis in order to avoid precipitation or degradation of the components in the sample. The incubation time was ∼15 min at room temperature. The ”blank“ samples, containing protein and reference ligand but no competing ligand, were spiked with DMSO to a concentration equal to the DMSO content in the other samples. In the examples presented, 4 µM protein and 0-60 µM ligand were used in the direct KD measurements. For the competition experiments, single point determination, the molar ratio between the protein, reference ligand, and competing ligand was 1:5:10 with a protein concentration of 6 µM. For the dose-response curve, the amount of protein was 6 µM, the reference ligand 30 µM, and the competing ligand 0-300 µM. Automated MS Data Processing. The raw data achieved from the MS analyses were automatically processed by a modified version of OpenLynx (Waters, Manchester, UK). Deconvolution, integration, and peak height determination of the peak corresponding to the protein-ligand complex were performed automatically. The processing time for each sample was 20 s. The peak heights obtained were subsequently used for the calculations of KD values in the Origin 5.0 software (MicroCal Ltd); see below. Data Analysis. Binding of One Ligand to a Protein. The stability of the complex in the gas phase varies between different nonpolar complexes, resulting in varying MS response for a given amount of complex in solution. However, the MS response is proportional to the concentration of the complex in solution. Thus, by varying the ligand concentration and keeping the protein concentration constant, a traditional binding curve can be obtained. To resolve the affinity of the ligand to the protein in the solution for a complex that easily dissociates in the gas phase (resulting in a relatively weak response), we introduce a response factor, R. The response factor is system dependent and instrument dependent. It is dependent on molecules forming the complex, instrument

geometry, pressure, cone voltage, and number of scans used for the integration. An expression describing the relationship between the MS response, S, and the complex concentration [ML] in the solution can the defined

S ) R[ML]

(1)

To obtain the affinity of the ligand L to the protein M we apply the law of mass conservation. To do this we here define the equilibrium in the aqueous phase as the dissociation of the complex ML into M and L:

ML h M + L

(2)

reference ligand, A, with a known affinity, KD,A, and a competing ligand, B, that binds to the same site as the reference ligand is injected into the MS. The change in the MS response of the protein-reference ligand complex due to competition gives information on the affinity of the competing ligand, KD,B. The procedure of obtaining the affinity of the competing ligand is based upon the law of mass conservation. There are two equilibriums in the liquid phase:

MA h M + A

(9)

MB h M + B

(10)

and

The dissociation constant, KD, is defined as

KD ) [M][L]/[ML]

(3)

The total concentrations of the reactants are

Mt ) [M] + [ML]

(4)

Lt ) [L] + [ML]

(5)

We can express the concentration of ML by substituting (3) into (5)

[ML] )

[M]Lt

Rearranging (6) and substituting into (4), we obtain a new expression for Mt,

(

Lt KD - [ML]

)

(7)

We can now solve the complex concentration, [ML], from the resulting second-degree polynomial. [ML] is a function of KD, the total concentration of the protein, Mt, and the total concentration of the ligand, Lt,

[ML] ) Mt

(

KD,A ) [M][A]/[MA]

(11)

KD,B ) [M][B]/[MB]

(12)

The total concentrations of the three reactants are

Mt ) [M] + [MA] + [MB]

(13)

At ) [A] + [MA]

(14)

Bt ) [B] + [MB]

(15)

(6)

[M] + KD

Mt ) [ML] 1 +

The dissociation constants for MA and MB are KD,A and KD,B, respectively,

)

r + X + 1 - ((r + X + 1)2 - 4r) 2

(8)

For clarity we have introduced in (8) the variables r, which is the molar ratio between the ligand and the protein in solution, Lt/Mt, and X, which is the ratio between the affinity and the total concentration of the protein in solution, KD/Mt. From ESI-MS experiments where Mt is kept constant and Lt is varied, the response factor, R, and the affinity, KD, are solved by nonlinear regression. The response factor should be regarded as a systemand instrument-dependent fitting parameter. However, it is equal to the maximum response for the complex. A complex that dissociates in the gaseous phase will have a low R while a complex that is intact in the gaseous phase will have a high R. Competition between Two Ligands for One Site. In competition binding experiments, a mixture of the host molecule, M, the

We can express the concentration of MA by substituting (11) into (13)

[MA] ) [M]At/([M] + KD,A)

(16)

and the concentration of MB in the same manner by substituting (12) into (15)

[MB] ) [M]Bt/([M] + KD,B)

(17)

Ligand A is a reference ligand for which the concentration is kept constant throughout the experiment, and B is the competing ligand. Increasing the concentration of B will partially displace A from the complex with M. The degree of displacement will be dependent on the binding constants and concentrations of A and B. The distribution of the different species of the host molecule is expressed in terms of the molar ratio of the three different host molecule species relative to the total concentration of the host molecule,

xM ) [M]/Mt

(18)

xMA ) [MA]/Mt

(19)

xMB ) [MB]/Mt

(20)

The sum of the molar ratios is by definition unity

xM + xMA + xMB ) 1 Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

(21) 4327

In the following, we introduce the molar ratio between the stoichiometric concentrations of the ligands and M, sA and sB, respectively

sA ) At/Mt

Table 1. Dissociation Constants (KD (µM) ( Standard Deviation) for hGHbp and Six Nonpolar Ligands Derived by Different ESI-MS Techniques and from ITC

(22) compd

and

sB ) Bt/Mt

(23)

For simplification we also introduce the ratio between of the affinity constants and the stoichiometric concentration of M, rA and rB, respectively,

rA ) KD,A/Mt

rB ) KD,B/Mt

(25)

Substituting (20)-(23) into (11), (14), and (15) gives

xMA ) sAxM/(xM + rA)

(26)

xMB ) sBxM/(xM + rB)

(27)

Substituting (26) and (27) into (21) results in an expression that contains xM as the only unknown. To find the root of (21), we need to solve a third degree polynomial of the form

xM3 + axM2 + bxM + c ) 0

(28)

a ) rA + rB + sA + sB

(29)

b ) rB(sA - 1) + rA(sB - 1) + rArB

(30)

c ) rArB

(31)

where

The only physically meaningful solution of the cubic polynomial is18

-2x(a2 - 3b) cos(θ/3) - a 3

in which θ is calculated from the equation

θ ) arccos

(

)

-2a2 + 9ab - 27c 2x(a2 - 3b)3

0.75 ( 0.15 0.97 ( 0.23 2.2 ( 0.50

ESI-MS competition reverse single dosedosepoint response response 0.81b 1.0a 2a 3.6b 23b,c 1.2a

0.76 ( 0.15b 0.80 ( 0.01a 1.7 ( 0.09a 3.6 ( 0.9b 20 ( 8b,* 1.8 ( 0.2a

ITC 0.7 ( 0.3

2.6 ( 0.4b 27 ( 5b

a Compound 1 used as reference ligand. b Compound 2 used as reference ligand. c Reversed mode. *The data are insufficient for calculating accurate KD.

(24)

and

xM )

1 2 3 4 5 6

ESI-MS direct

(32)

(33)

Once we have calculated xM we can calculate the concentrations of all the host molecule species by substituting (32) into (26) and (27). (18) Wang, Z.-X. FEBS Lett. 1995, 360, 111-114.

4328 Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

The procedure of the proposed method is to prepare solutions with constant concentrations of M and A and varying concentration of B. The observed MS response of the MA complex when B is present, Sobs, is compared with the MS response of the MA complex of the reference solution, Sref, which is free from B. The equilibrium constants KA or KB can be calculated from the concentration dependence of Bt of the relative responses, Sobs/ Sref, by nonlinear regression using the Marquart-Levenberg algorithm as applied by the software Origin (MicroCal Ltd). Isothermal Titration Calorimetry. ITC experiments were performed on a heat-flow twin microcalorimeter with an insertion titration unit, ThermoMetric 2277 TAM (Thermometric AB). A nonstirred titration unit charged with 900 µL of water was used on the reference side of the microcalorimeter. Prior to the experiment, the protein solution was dialyzed against 50 mM Tris pH 7.4. Compound 1 was diluted from 30 mM in DMSO stock solution to 150 µM in 50 mM Tris pH 7.4. The calorimetric vessel was loaded with 900 µL of a 12 µM hGHbp in 50 mM Tris pH 7.4 and 4.5 µL of DMSO. In the experiments, 25 aliquots of 7.2 µL of the ligand was added. Electrical calibrations were performed prior to each titration experiment to evaluate the calorimetric gain and the time constants of the instrument. RESULTS The low molecular mass inhibitors were initially dissolved in DMSO as stock solutions, which resulted in small amounts of DMSO (0.3-1.5%) in the samples. The ESI-MS measurements revealed two DMSO-dependent effects on the charge-state intensity of the protein. At low concentration of DMSO, from 0.4%, there was a shift in the charge state of the protein toward higher massto-charge values (less protonation) in the mass spectrum, while at higher concentrations of DMSO, from 2% DMSO, the chargestate intensity started to shift toward lower masses. As a consequence of these findings, the maximum DMSO content in the samples was set to 1.5%. To avoid problems with fluctuating intensities of the different charge states, all data used in this study were deconvoluted data. Experimental data for the different techniques and methods are summarized in Table 1. Direct KD Measurements. For the direct KD determination approach, the protein concentration Mt was held constant at 4 µM

Figure 1. Direct KD determination of the binding of 1 to hGHbp. The MS responses corresponding to the protein-ligand complex are plotted as a function of 1 concentration. KD was solved by nonlinear regression described in eqs 1 and 2. The protein concentration (4 µM) was constant, and the ligand concentration vas varied from 2 to 60 uM. The KD was estimated to 0.75 uM.

and the amount of ligand, Lt, was varied from 2 to 60 µM. At a certain ligand concentration the MS response of the proteinligand complex reached a plateau and no longer increased. KD was solved by nonlinear regression as described under data analysis, eqs 1 and 2. Figure 1 shows the MS response of the complex as a function of ligand concentration for one of the tested compounds (1). The KD values obtained by ESI-MS (0.75 ( 0.15 µM) and ITC (0.7 ( 0.3 µM) agreed well. Competition Experiments. Single Point Determination. Figure 2A, upper spectrum, shows a deconvoluted mass spectrum of a sample containing 6 µM hGHbp and 30 µM reference ligand A, 1. The peak labeled M corresponds to the unbound protein, and the peak labeled MA corresponds to the protein-reference ligand complex. Figure 2A, lower spectrum, shows a mass spectrum of a sample containing hGHbp, the reference ligand, and a competing ligand B with unknown KD (molar ratio 1:5:10). In the lower spectrum, peaks from both the protein-reference ligand complex MA and the protein-competing ligand complex MB were observed. As the extent of dissociation of MB in the gas phase was unknown, measurements were solely done on MA. The MS response of MA was less abundant in the lower spectrum as compared to the upper spectrum, indicating that the reference ligand had been partly displaced by the competing ligand at the same binding site. To calculate the KD of the competing ligand, the peak height of MA was measured before and after the addition of the competing ligand. The obtained peak height ratio of MA was used to calculate the KD value according to eqs 20-26; see Figure 2B. This single point measurement gave less reliable KD values compared to KD values obtained using dose-response curves. However, this method is very useful as it is faster and gives a reliable ranking of the screened ligands. The sample throughput for single point determination is in the order of 100 compounds a day. Dose-Response Curve. In the cases where more accurate KD measurements are desirable, dose-response curves should be used instead of single point measurements. Figure 3A displays five deconvoluted MS spectra from samples containing 6 µM

Figure 2. Graphical presentation of determination of KD using the competition single point assay. In the figure, the reference ligand A is 1, the competing ligand B is 6, and M is hGHbp. (A) Deconvoluted mass spectra prior and after addition of a competing ligand B. The upper spectrum represents a sample containing 6 µM hGHbp and 30 µM reference ligand A. The peak labeled M corresponds to the free protein, and MA corresponds to the protein-reference ligand complex. The lower spectrum represents a sample containing 6 µM hGHbp, 30 µM reference ligand A, and 60 µM competing ligand B. (B) The results from (A) are graphically presented in the plot. The curve shows how the relative MS response for reference ligand A is dependent on KD for a competing ligand for the used concentrations of protein, 1, competing ligand and KD for 1. The calculated relative MS response, S/Sref, of the reference complex is 0.48, represented by the horizontal line. The horizontal line coincides with the calculated curve and the KD is estimated from the vertical line to 1.2 µM.

hGHbp, 30 µM reference ligand A, 1, and increasing amounts of the competing ligand B (0-300 µM). The ratios between the heights of the protein-reference ligand peaks and the proteinreference ligand peak in the sample without competing ligand were plotted against the amount of added competing ligand (Figure 3B). From this dose-response curve, the KD for the competing ligand was calculated as described. The standard deviation of dose-response curves was shown to be less that 20%. KD Measurements of Weak/Low-Solubility Ligands. The flexibility of the ESI-MS method makes it possible to produce reliable KD values of low-affinity ligands, even if the solubilities of the ligands Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

4329

Figure 4. Graphs showing how to determine KD for weak binders. (A) A “normal” KD measurement of a weak competing ligand (5), with a fixed concentration (30 µM) of the reference ligand (1). A 10-fold excess of 5 was necessary for displacement of 1. The data are insufficient for calculating accurate KD due to the weak competition over the concentration range. (B) A “reversed” procedure for determining KD for 5. The concentration of 5 was constant (300 µM), and the concentration of 1 was varied. The KD was determined to 27 ( 5 µM.

Figure 3. Determination of KD for the competing ligand B using a dose-response curve. In the spectra, the reference ligand A is 1, the competing ligand B is 3, and M is hGHbp. (A) Five MS spectra from samples containing hGHbp (6 µM), reference ligand A (30 µM), and varying concentration of the competing ligand B (0-200 µM). (B) The relative MS responses of the reference ligand, S/Sref, are plotted against the concentration of 3 added. The line is the calculated dose-response curve. The upper pane shows the residuals obtained from the regression. The KD was determined to 1.7 ( 0.09 µM.

are poor. Figure 4 demonstrates a “normal” and a “reversed” way of measuring the KD of a weak binder. In the “normal” prodedure of KD measurement (Figure 4A), increasing amounts (0-300 µM) of the competing ligand was added to the six samples containing 6 µM hGHbp and 30 µM reference ligand. Not until there was a 10-fold excess of the competing ligand did a displacement occur and the MS response of the protein-reference ligand complex started to decrease. The KD value obtained (20 µM) from the dose-response curve had a high relative standard deviation (RSD) (41%). The solubility of the competing ligand was low, so addition of more competing ligand to the samples would increase the risk 4330 Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

of precipitation. Instead, a “reversed” procedure of KD measurement was created (Figure 4B), where the amount of reference ligand was increased (0-120 µM) in the five samples, and the amount of competing ligand was fixed at 300 µM. From the doseresonse curve, a KD of 27 µM was obtained and the RSD was down to 19%. Reproducibility of Peak Heights. No internal standard was used in any of the ESI-MS measurements. Therefore, it is important to evaluate the reproducibility of the injection system and the MS response obtained from different injections. Six samples containing 6 µM hGHbp and 30 µM of 1 were prepared individually, and aliquots of each sample were injected and analyzed. The recorded spectra were processed, and the peak height of the protein-ligand complex from each sample was measured. The RSD of the six injections was found to be 3.9%. Comparison of the Methods Described. In Table 1, data for six compounds are collected. The KD values for the different compounds agree well for all methods. For the low-affinity compounds, the statistical uncertainty in the KD values is reduced when the reversed procedure is used. ITC Measurements. Notably, for the ITC measurement, the enthalpy of binding was low, -14 kJ mol-1 for 1, which means that the measured heats are low, e30 µJ. The enthalpies of binding

for the other compounds, 2-6, were even lower. An enthalpy close to zero at room temperature is typical for hydrophobic interactions. The only way to obtain higher measured enthalpies would be to significantly increase the concentrations of hGHbp and the compounds. However, the protein would then be aggregated and the concentrations needed for the compounds exceed the solubility limit. Neither the ligands nor the binding site of the protein contains any proteolytic group. Hence, there is no proton linkage in the process and no pH dependence in the affinity. DISCUSSION We have shown that quantitative studies on complex formation by ESI-MS are not limited to polar or ionic interacting molecules, but it is also valid for molecules that interact by nonpolar binding surfaces. This is achieved by only using the MS response data of the protein-ligand complex and introducing a response factor. The response factor normalizes the measured intensities to the situation in the solution for a given experimental condition. In the analysis of competition binding experiments, the response factor is canceled out, since we use relative changes in intensities of the protein-reference ligand complex. The method is independent of variations in the ESI-MS efficiencies, or the response factor, for the different ligands, as long as the instrument parameters are constant during the experiment. This enables the use of ESI-MS as a rapid drug screening tool, measuring binding affinities for the complexes between soluble proteins and low molecular mass ligands. In Table 1, we show that the various ESI-MS methods described in this report are in good agreement. Compound 1 has been measured by direct KD determination (0.75 ( 0.15 µM), single point determination (0.81 µM), and dose-response determination (0.76 ( 0.15 µM). The three methods agree well, and even more important, they agree well with the ITC measurement (0.7 ( 0.3 µM), which is based on thermodynamic properties of binding. As for all affinity determinations the quality of the affinity data is dependent on the experimental design of the experiment. Therefore, for single point determinations in a screening situation, the quality of the estimated affinity is dependent on the affinity of the competing ligand relative to the reference ligand. For weaker binders, as well as for binders that are from 3 orders of magnitude stronger binders than the reference ligand, the quality of the affinity data will be low. This is exemplified in Table 1 by the weaker binder 4. A larger difference in the KD values was obtained from the different ESI-MS methods of 4 compared to the results of 1 and 2. The reversed dose-response method showed a KD of 2.6 ( 0.4 µM, while the single point determination and the ‘normal’ dose-response determination showed values of 3.6 and 3.6 ( 0.9 µM, respectively. (19) Swayze, E. E.; Jefferson, E. A.; Sannes-Lowery, K. A.; Blyn, L. B.; Risen, L. M.; Arakawa, S.; Osgood, S. A.; Hofstadler, S. A.; Griffey, R. H. J. Med. Chem. 2002, 45, 3816-3819.

The ESI-MS assay is generic and can be applied to most soluble proteins with minimal assay development work. The specificity of the method and the use of nonlabeled compounds make it possible to change the reference ligand in order to optimize the experimental conditions for obtaining high-resolution data. The possibility to change the reference ligand depending on the affinity of the competing compound enables us to achieve a wide dynamic range of attainable affinities. It is desirable to have high-quality affinity data for high-affinity complexes as well as for low-affinity complexes in order to establish a well-defined structure-activity relationship in the drug design process. The low hit rate in traditional high-throughput screening assays is in many cases due to difficulties in detecting and determining low affinity interactions.19 The observed DMSO effect is probably not a specific effect on our model system, but rather a general consideration to be aware of when designing experimental conditions for direct as well as for competition binding experiments. This is especially important for the ESI-MS method and may be of importance for any other method. The introduction of an automated MS data processing has reduced the data process time from approximately 4 min to 20 s per sample. To be useful as a screening tool, the sample throughput in the ESI-MS assay must be high. By using a simple sample preparation procedure and automation of the MS analysis as well as automation of the MS data processing, we were able to screen over 100 different compounds per day against hGHbp using the single point competition experiment. In these experiments, we used a protein concentration of 3 µM. CONCLUSION We have demonstrated that binding affinities can be investigated for nonpolar interactions between proteins and small molecular ligands. The complexes can be investigated regardless of the type of interaction by using MS responses of the proteinligand complex and by introducing a response factor. The presented ESI-MS assay fulfills the criteria for a generic, robust, and easily automated biophysical method for the quantitative studies of noncovalent interactions. The introduction of an automated MS data processing has increased the sample throughput significantly. ACKNOWLEDGMENT We thank Dr. Jenny Ro¨nnmark for providing the hGHbp used in this work and Professor Jan Johansson, Dr. Lars Tjernberg, and Dr. Johan Weigelt for critically reading the manuscript. We also thank Waters, Manchester, U.K., for our custom-made OpenLynx Software. Received for review February 6, 2004. Accepted May 11, 2004. AC0497914

Analytical Chemistry, Vol. 76, No. 15, August 1, 2004

4331