combinatorial chemistry & high throughput screening, 000 ... › d98e › b45269da3...2...

Combinatorial Chemistry & High Throughput Screening, 2009, 12, 000-000 1

1386-2073/09 $55.00+.00 © 2009 Bentham Science Publishers Ltd.

QPatch: The Missing Link Between HTS and Ion Channel Drug Discovery

Chris Mathes*,1

, Søren Friis 2, Michael Finley

3 and Yi Liu

3

1Sophion Bioscience, Inc., 675 US Highway One, North Brunswick, NJ 08902, USA

2Sophion Bioscience A/S, Baltorpvej 154, Ballerup DK-2750, Denmark

3Johnson & Johnson Pharmaceutical Research and Development, L.L.C., P.O. Box 776, Welsh & McKean Roads,

Spring House, PA 19477-0776, USA

Abstract: The conventional patch clamp has long been considered the best approach for studying ion channel function

and pharmacology. However, its low throughput has been a major hurdle to overcome for ion channel drug discovery. The

recent emergence of higher throughput, automated patch clamp technology begins to break this bottleneck by providing

medicinal chemists with high-quality, information-rich data in a more timely fashion. As such, these technologies have the

potential to bridge a critical missing link between high-throughput primary screening and meaningful ion channel drug

discovery programs. One of these technologies, the QPatch automated patch clamp system developed by Sophion Bio-

science, records whole-cell ion channel currents from 16 or 48 individual cells in a parallel fashion. Here, we review the

general applicability of the QPatch to studying a wide variety of ion channel types (voltage-/ligand-gated cationic/anionic

channels) in various expression systems. The success rate of gigaseals, formation of the whole-cell configuration and us-

able cells ranged from 40-80%, depending on a number of factors including the cell line used, ion channel expressed, as-

say development or optimization time and expression level in these studies. We present detailed analyses of the QPatch

features and results in case studies in which secondary screening assays were successfully developed for a voltage-gated

calcium channel and a ligand-gated TRP channel. The increase in throughput compared to conventional patch clamp with

the same cells was approximately 10-fold. We conclude that the QPatch, combining high data quality and speed with user

friendliness and suitability for a wide array of ion channels, resides on the cutting edge of automated patch clamp technol-

ogy and plays a pivotal role in expediting ion channel drug discovery.

Keywords: Ion channels, automated patch clamp, QPatch, gigaseal, whole-cell, secondary screening.

INTRODUCTION

Until the recent advent of automated patch clamp, a real gap existed in the time-line for ion channel drug discovery. The distance between high-throughput binding or fluorescent assays and conventional patch clamp seemed insurmount-able. This discouraged many pharmaceutical companies from investing vast resources into ion channel drug discovery de-spite the fact that many ion channels are druggable targets. Instead, they focused more on targets such as kinases and G-protein-coupled receptors (GPCRs) for which secondary screening assays with reasonable throughput were readily available [1, 2]. Automated patch clamp arrived on the scene in steps from 2002 – 2004 and promised to bridge the gap between high throughput screening (HTS) and conventional patch clamp [3].

HTS labs test ~10,000 – 30,000 compounds/day at single concentrations. With ion channel targets, these screens typi-cally involve binding or fluorescence. The obvious advan-tage of these methods, high throughput, allows affordable screening of large compound libraries within a relatively short period of time. A major downside of these HTS meth-ods lies in the lack of accurate voltage control (particularly important for voltage-gated ion channels) and the high in-formation content (e.g., fast speed of compound application

*Address correspondence to this author at the Sophion Bioscience, Inc., 675

US Highway One, North Brunswick, NJ 08902, USA;

E-mail: [email protected]

to circumvent rapid desensitizing/inactivating channels) that patch clamp offers. In addition, binding assays are non-functional assays, requiring a known high-affinity, radiola-beled ligand in the first place, and in general only pick up hits that compete for binding with the ligand. Although func-tional in nature, fluorescence-based assays give indirect measurements of channel function due to the involvement of a dye. These disadvantages associated with HTS methods often result in false positives and/or false negatives.

Conventional patch clamp measures ion channel function directly. With the most common configuration, whole-cell voltage-clamp, a specialized amplifier accurately clamps the membrane potential and measures current flow across the membrane of an entire cell. The accurate and direct nature of these measurements, combined with the high information content they provide, has earned it the honor of “the gold standard” for studying ion channel function and pharmacol-ogy. However, a well-trained patch clamper typically can only test a small number of compounds per day, making it impractical for use as a secondary screening assay.

Automated patch clamp technologies bridge the gap be-tween HTS and conventional patch clamp by affording sig-nificantly increased throughput over conventional patch clamp (for example, the QPatch by Sophion Bioscience, can test 100s of compounds per day), while preserving the same features and high quality as conventional patch clamp. As a result, an increasing number of ion channel drug discovery programs incorporate automated patch clamp as a critical step in secondary screening, and in some cases, even primary

2 Combinatorial Chemistry & High Throughput Screening, 2009, Vol. 12, No. 1 Mathes et al.

screening of focused libraries. This review focuses on the QPatch automated patch clamp system. The unique features of the QPatch and its place in the history of automated patch clamp have been reported previously [3, 4]. Here, results from assay development projects for a wide variety of ion channel types (voltage-gated and ligand-gated cationic or anionic channels) in various expression systems (HEK293, CHO, native cell lines, etc.) are described and reviewed with an emphasis on success rates. More detailed analyses of some QPatch features and results are further highlighted in two case studies. Results from this review were obtained from both QPatch-16 (16-channel) and QPatch HT (48-channel) systems.

VOLTAGE-GATED ION CHANNELS

Nav1.2 Channels

Historically, Na currents achieved much attention be-cause of their vital role in driving neuronal action potentials [5]. Na

+ currents peak within ~ 1 ms, presenting a significant

challenge for automated patch clamp systems. Therefore, we begin this overview with QPatch data from experiments with neuronal Nav1.2a channels (rat brain). These Na channels are TTX-sensitive. Fig. (1) shows a family of currents ob-served in response to increasing voltage steps and the corre-sponding Current vs Voltage (I-V) relationship.

To explore the state-dependency of channel block, cells were depolarized twice with a temporal separation of 115 ms (Fig. 2A). The current-time relationship for the peak Na cur-rents recorded in response to the first (blue) and the second (magenta) depolarization is shown in Fig. (2B). As shown in this figure, the employed compound (lidocaine) inhibited the channel during the second voltage step (second lane; labeled “(1) Compound”), but not the first one. The effect of the compound was reversible (third lane; labeled “(2) Saline”).

Subsequent addition of TTX (fourth lane; labeled “(3) Refer-ence”) blocked both currents equally. The current amplitudes partly recovered from the TTX block (fifth lane; labeled “(4) Saline”).

Ten compounds were tested for potential state-dependent effects on Nav1.2a channels (Fig. 3). The compound, NS-X, exhibited the highest degree of state-dependent block on Nav1.2a channels (and also on Nav1.5a channels; data not shown). Generally, the state-dependency of blocker action was more pronounced for Nav1.5a than for Nav1.2a (Nav1.5 data not shown). Only the effect of TTX was completely state-independent. The amplitudes of Na

+ currents immedi-

ately prior to compound application were set to 100%. The rightmost bars represent control (no drug).

In this set of experiments the rate of gigaseals (i.e. > 1 giga-ohm or G ) was 53 out of 64 cells or 83%. The rate for achieving the whole-cell configuration (> 100 M and > 4 pF) was 66% (42/64) and this was similar to the number of completed experiments (40/64 = 63%).

Cav3.2 Channels

Compared to rapid Na currents, relatively slow Ca chan-nel currents are easier to measure. However, these channels are often subjected to run-down, which represents a major challenge for both conventional and automated patch clamp. Fig. (4) shows a family of currents observed in response to increasing voltage steps and the corresponding I-V relation-ship from a QPatch experiment with cells expressing Cav3.2 channels (also known as neuronal T-Type or Low-threshold Ca channels). The maximal current was observed at –20 mV which is consistent with the literature describing the Ca

V3.2

whole-cell currents obtained from HEK293 cells [6]. In an-other set of experiments with HEK cells expressing Cav3.2

Fig. (1). Biophysical characterization of rat Nav1.2 currents measured with the QPatch 16. (A) A family of Na currents elicited by 10 ms

voltage steps of 10 mV from -90 mV to +80 mV. The extracellular solution contained 140 mM Na, whereas the intracellular solution was

Na-free, the major cation being Cs at a concentration of 120 mM. Experiments were conducted at room temperature (~21°C). The holding

potential was –90 mV. Currents were filtered at 4 kHz and sampled at 50 kHz. The currents in (A) were not leak subtracted (although fast

and whole-cell capacitance were corrected digitally during experiments. (B) The resulting I-V relationship for the peak Na+

channel currents

shown in (A). The activation threshold was –40 mV, and the maximal current amplitude was obtained at -10 mV. At large positive mem-

brane potentials the current amplitude is reduced as the electrochemical driving force vanishes.

QPatch: The Missing Link Combinatorial Chemistry & High Throughput Screening, 2009, Vol. 12, No. 1 3

Fig. (2). A paired-pulse QPatch experiment to determine state-dependent block of test compounds. (A) Two voltage steps to 0 mV separated

by 115 ms. The amplitude of the second current (magenta) is slightly less than the amplitude of the first current (blue). (B) I-t plot showing

the peak amplitudes from the first pulse (blue) and the second pulse (magenta). Lane 1 shows the peak amplitudes in response to saline (i.e.

the control currents shown in (A)). Lane 2 shows that 100 M lidocaine inhibited the second pulse amplitude by ~ 50% without affecting the

amplitude of the first pulse. TTX (100 nM) blocks both currents equally (Lane 4).

Fig. (3). Results from a paired-pulse screen against rate Nav1.2a currents to determine the state-dependent block of 10 compounds. The per-

cent residual current (current normalized to the first peak amplitude are plotted on the y-axis versus the 10 compounds. Peak 1 amplitudes are

shown in black and Peak 2 amplitudes in red (means ± SEM error bars). Only NS-X and lidocaine show evidence of state-dependent block.

These experiments were done with four 16-channel QPlates.


channels we observed a gigaseal rate of 56% (27/48) and a whole-cell rate of 64% (31/48). This particular QPatch HT experiment lasted ~ 30 min. A more detailed Ca channel QPatch case study is described below.

hERG K Channels

According to FDA guideline ICH S7B (http://www.fda.gov/cder/guidance/5533dft.htm), every drug going to market must be tested against hERG K channels

using the patch clamp method. This guideline came about because certain pharmaceuticals (i.e. Seldane—a.k.a. thapsi-gargin) caused potentially fatal arrhythmias in some patients in the late 1990’s [7, 8]. At the time, this guideline seemed impossible for pharma and biotech companies to fulfill be-cause the patch clamp method was slow and required spe-cialized patch clamp electrophysiologists. With the advent of automated patch clamp, however, pharma and biotech com-

Fig. (4). QPatch 16 whole-cell recording from a HEK cell stably expressing Cav3.2 channels. Voltage steps of 10 mV were applied from -80

to +40 mV (10 ms duration) from a holding potential at -90 mV. In these experiments the following solution recipes were used: Internal solu-

tion (in mM): CsF (140), EGTA (1), HEPES (10), NaCl (10) and external solution (in mM): TEACl (157), MgCl2

(0.5), CaCl

2 (5), HEPES

(10). (A) Typical family of whole-cell current traces. Amplitudes were measured in the shaded green cursor areas. (B) The average I-V curve

from three voltage-step stimulations of the cell shown in (A). For the current amplitude analysis the left cursor interval depicted in (A) (17-

18 ms) was used.


panies can now keep up with demand, especially during early pre-clinical research.

Fig. (5A) shows raw hERG K currents and the corre-sponding amplitudes in the presence of increasing concentra-tions of the blocker, Verapamil. The peak amplitude of the tail current at -50 mV was determined by the QPatch Assay Software (shaded green cursor area) and plotted versus time (Fig. 5B). The average peak hERG tail current amplitude at -50 mV was 495 pA ± 80 pA (mean ± SEM; n=50). The QPatch assay software automatically graphs the dose-response for each compound tested. Fig. (5C) shows the Hill plot for Verapamil in this set of experiments The averaged IC50 value for Verapamil was 344 nM (n=26).

In general, CHO-hERG success rates fall within a range of 50 – 80%. For example, in a representative set of two 48-channel QPatch HT experiments (on different machines and days), the gigaseal success rate was 61% (59/96 cells) and the whole cell success rate was 73% (70/96). The whole cell success rate was higher than the gigaseal success rate primar-ily because some cells go immediately to the whole cell con-figuration without first reaching the gigaseal level. These cells often become useful whole cell recordings (and in many cases have membrane resistance, Rm, values > 1 G ).

LIGAND-GATED ION CHANNELS

P2X3 Receptor Channels

P2X receptors are ligand-gated ion channels with time constants of 10-200 ms for activation and 100-200 ms for desensitization [9, 10]. The fast kinetics present a challenge for patch-clamp recordings of physiologically realistic P2X receptor currents, and a fast solution exchange technique is required.

This section shows QPatch data from one of the fastest P2X receptors, the P2X3 receptor. Results are shown from experiments using the perforated-patch whole-cell method to measure currents from a rat lung epithelial (RLE) cell line stably expressing the human P2X3 receptor (AstraZeneca, Södertalje, Sweden). All measurements were recorded with the QPatch 16.

Exposure to the agonist CTP (20 M) led to typical tran-sient P2X3 receptor currents characterized by a fast activation and a slower subsequent desensitization (Fig. 6). The QPatch current recordings are similar to current recordings obtained with manual patch-clamp (inset in Fig. 6). Subsequent to each agonist application, two wash steps (15 μL extracellular solution) were performed. Fig. (6) shows that the response to 20 M CTP was stable throughout the experiment period indicating that the agonist was completely washed out be-tween the applications.

GABA Channels

This section presents QPatch data from whole-cell cur-rent recordings from the ligand-gated ion channel hGABAA receptors ( 1 3 2), expressed in HEK-293 cells (Millipore). GABAA receptors are anionic channels that upon activation under physiological conditions, mediate currents carried by Cl

-. Fig. (7A) shows GABAA currents in response to six in-

creasing concentrations of -amino-butyric acid (GABA), a GABAA receptor agonist, each followed by two saline washes. Fig. (7B) shows the complete concentration-

response relationship as determined with the QPatch Assay Software for this experiment. The mean EC50 was 5.9 ± 2.9 μM (n=5). The mean Hill slope = 1.6 ± 0.3. The mean rise-time for GABAA receptor currents measured at 10 μM GABA was 167.1 ± 69.5 msec (mean ± S.D.; n=6).

The average gigaseal success rate for this set of experi-ments was 40% (32/80 HEK-GABA cells). The average whole-cell success rate was 65% (52/80) and the average percentage of usable cells was 54% (43/80).

OTHER ION CHANNELS

Icrac

Rat basophilic leukaemia (RBL) cells endogenously ex-press calcium-release-activated-calcium (CRAC) channels [11]. CRAC channels are activated by depletion of intracel-lular calcium stores, probably via the involvement of STIM-1 (stromal interaction molecule) sensing the depletion of the stores and travelling to the cell membrane to activate the channel [12, 13]. There are several approaches that can be used to deplete calcium stores ultimately leading to activa-tion of Icrac and most of these have been tried on the QPatch [14].

For these experiments the suction protocol used resem-bles the one normally used for CHO cell lines (i.e. high pres-sure for positioning the cell, gigasealing and for rupturing the membrane for the whole-cell configuration). The CRAC channel has an extremely small unitary conductance esti-mated from noise analysis to be ~ 10 fS under physiological conditions [15]. The resulting whole-cell current (Icrac) is also very small (2.5 - 3 pA/pF at -80 mV) so it is of pivotal importance that the seal resistance is very high (> 1 G ) to minimize noise and get good recordings. With the RBL-2H3 cells gigaseals > 1G are easily established. In general, gi-gaseal, whole-cell and usable experiment success rates are ~ 80% with these cells.

Fig. (8) shows an example of Icrac recording by QPatch using inositol trisphosphate (IP3) (20 M) in the intracellular solution to deplete the calcium stores. The voltage protocol used was a 100 ms ramp going from -100 to +100 mV with an interval of either 3 or 6 s. The cell was held at 0 mV be-tween the sweeps.

Fig. (8A) shows raw data plots from the QPatch Assay Software showing the leak subtracted sweep response (cur-rent before subtracted from current after induction of Icrac). The average membrane resistance just after establishment of the whole-cell configuration (Rwc) was 1728 ± 226 M (n = 44, data from four QPlates from a typical experimental day). I/t plots were made with the QPatch Assay Software by in-serting a cursor at -80 mV and plotting the current as a func-tion of time (Fig. 8B).

CFTR Channels

Cystic fibrosis transmembrane conductance regulator (CFTR) functions as a chloride channel and controls the regulation of other transport pathways. Mutations in the CFTR gene have been found to cause cystic fibrosis (CF) and congenital bilateral aplasia of the vas deferens (CBAVD). This section describes QPatch experiments with NIH 3T3 cells expressing CFTR channels.


Fig. (5). The QPatch software quickly displays raw data for many experiments and converts the peak values to I-t plots and dose-response

curves. Standard CHO-hERG solutions were used for these experiments. (A) Screen shot from the QPatch Assay Software showing raw

hERG traces elicited by voltage steps from -80 to -50 (20-100 ms) for off-line leak subtraction, from -50 to +50 mV (5 s), and then back to -

50 for 5 s. Pulses were applied to cells every 15 s. Traces show currents in response to saline (control) and increasing concentrations of the

hERG blocker verapamil (30 nM – 30 M). The shaded green cursor region shows where the peak amplitude was measured by the QPatch

Assay Software for further analysis. (B) I-t plot from the experiment shown in A. (C) Dose response graph from 26 different cells. The cal-

culated IC50 value from the Hill Plot is ~ 350 nM (n=26 cells).


The sealing rate for this study (> 100 Mohms) on the QPatch was ~ 70%. The cells were generally large (15-25 pF) and gave a stable current response when the CFTR channels were closed. CFTR channels are not voltage sensi-tive and are open when exposed to Forskolin/Genistien. This property makes it difficult to compensate for Cfast, Cslow, Rser-

ies and determinate Rmembrane on QPatch, when performing experiments with different extracellular solutions. In ex-periments performed with the same solution pairs throughout the experiment, it was possible to have a holding potential at ENernst(Cl) and thereby estimate Cfast, Cslow, Rseries and Rmem-

brane during the experiment.

Forskolin activates adenylate cyclase which uses ATP to make cAMP and in turn activates PKA. PKA phosphylates the CFTR channel and thereby opens these channels. Genis-tien prolongs opening bursts and shortens CFTR channel closings.

Outward chloride (Cl) currents were observed during these QPatch experiments. Beside Cl, CFTR channels also

conduct a large variety of anions such as fluoride, glutamate, aspartate and gluconate. This makes it difficult to perform ion substitution experiments. By substituting chloride with aspartate (data not shown), gluconate or glutamate (Fig. 9), which all have a different permeability than chloride, the results suggest that the inward current was mediated by CFTR channels.

The inward and outward currents are seen in Fig. (9) (blue symbols). In the first period a stable current is ob-tained. In the second period the forskolin/genistein (10/20

M) mix is added in nearly symmetric chloride Ringer pairs (155 mM extracellular, 135 mM intracellular). In the third period the chloride is partly substituted (30 mM extracellu-lar), which reduces the driving force and as expected the obtained current amplitude. The magenta symbols show the reversal potential. In the plot to the left a fast and transient change in reversal potential can be seen. Interestingly, this change is toward ENernst (K), which could suggest that a cAMP dependent potassium channel opens.

Fig. (6). Ligand-gated channel QPatch experiment with P2X3 receptors stably expressed in RLE cells. The figure shows a screen shot from

the QPatch Assay Software with four successive and reproducible current traces of P2X3 receptor current activated by additions of the ligand

CTP (20 M). The inset shows a conventional patch clamp recording in response to 10 M -meATP for comparison. Cells were cutured in

media consisting of Williams’ Medium E (supplemented with GlutaMAX), 10% fetal bovine serum (heat-inactivated) and geneticin G-418

(all from Life Technologies). Cells were harvested in HEK SFM medium. Prior to use, the cells were maintained in the QPatch cell storage

facility in suspension (HEK SFM medium). Shortly before the experiment the cells were automatically transferred to the QPatch mini centri-

fuge, spun down and washed once, before being suspended in the extracellular solution and transferred to the pipetting wells in the QPlate.

Agonists were applied in 5 μL aliquots to the pipetting wells and reached the cell via the integrated glass microfluidic flow channels. Cell

positioning at the patch-clamp site and subsequent gigaseal formation was achieved by suction. P2X3 whole-cell currents were measured in

perforated-patch configuration. Patch perforation was accomplished by the use of amphotericin B. The solution exchange time constant has

previously been determined to be in the 50-100 ms range. Extracellular solution consisted of (in mM): 154.5 Na+, 1.5 K

+, 1.4 Ca

2+, 143.5 Cl

-,

1.5 H2PO4-, HPO4

2-, 0.3 pyruvate, 5.6 glucose, hexokinase (30 U/mL) (pH 7.4). Intracellular solution consisted of (in mM): 140 Cs

+, 10 Na

+,

140 F-, 10 Cl

-, 1 EGTA, 10 HEPES (pH 7.3). Amphotericin B (40 μg/mL, Sigma) was added before use.


Fig. (7). QPatch 16 experiment with HEK cells stably expressing GABA channels. (A) Screen shot of raw current traces. Six additions of

GABA were applied (500 nM – 100 M). After each 3 s addition of GABA, two saline washes were used to remove the ligand. (B) The

dose-response curve and Hill plot from the experiment shown in (A). The IC50 value for this experiment was ~ 10 M. Cell culture: The

HEK GABA cells were grown according to the SOP supplied by the vendor, Millipore. The cells were seeded out two days before reaching

70% confluency. Upon harvest, the cells were washed in PBS and detached from the flask using Detachin as described in the Sophion SOP

for HEK cell harvest. Planar patch-clamping: HEK-293 cells expressing the GABAA receptors ( 1 3 2) were kept in culture medium in the

stirred reservoir for up to 4 h. Prior to testing, the cells were transferred to an on-board mini centrifuge (QFuge), spun down and washed in

Ringer’s solution twice before being applied to the pipetting wells in the QPlate. Gigaseals were formed upon execution of a combined suc-

tion/voltage protocol. Further suction lead to whole-cell configuration. Solutions and compounds were applied through the glass flow chan-

nels in the QPlate. All currents were recorded at a holding potential of -80 mV. Extracellular solution: consisted of (in mM): 145 Na+, 4 K

+, 2

Ca2+

, 1 Mg2+

, 154 Cl-, 10 HEPES (pH 7.4). Intracellular solution consisted of (in mM): 120 K

+, 1.8 Mg

2+, 123.6 Cl

-, 10 EGTA, 31.3 KOH,

10 HEPES (pH 7.2).


QPatch Case Studies

This section describes secondary screening assays devel-oped using the QPatch for two ion channel targets: a ligand-gated transient receptor potential (TRP) channel and a volt-age-gated calcium channel (VGCC). Both channels were stably expressed in HEK293 cells. While both channel types permeate calcium ions when activated, they belong to dis-tinctly different gene families and have very different bio-physical and pharmacological properties. Some general con-siderations common to developing both assays are described below followed by sections more specific for each of the two assays.

A first, very important step was cell preparation. Typi-cally, cells were cultured in a T225 cell culture flask to reach 70-90% confluency on the day of use and then incubated with 5-10 mL Detachin (Genlantis) for 5-10 min at 37°C. The loosely dissociated cells/cell clusters were subsequently

triturated (gently) using a 10 mL serological pipette to break up most of the clumps and then spun down at 800 rpm in a clinical centrifuge. Cells were then resuspended in 5-10 mL 293 SFM II serum-free media (Invitrogen) and allowed to recover for 20 to 60 min in the QStir cell preparation station on the QPatch-16 before the experiment. This was probably the most critical step that affected the outcome of experi-ments (success rate). The objective here was to obtain an almost completely isolated cell suspension (i.e. few or no clumps) while minimally affecting cell viability (i.e. few or no dead/damaged cells or pieces of debris). For the TRP channel, the final cell density, within a range of 1-15 10

6

cells/mL, did not appear to significantly affect the number of cells achieving the desired seal resistance threshold (usually set at 100 M ). We typically used 2-4 10

6 cells/mL for each

experiment. For VGCCs, the relationship between cell den-sity and success rate was less clear, perhaps in part due to complications by some cell clumping that was often difficult to eliminate. However, we found that 1-2 10

6 cells/mL rep-

Fig. (8). Small Icrac currents recorded with the QPatch 16. (A) Leak subtracted ramp current showing the inwardly rectifying Icrac current,

which is highly selective to Ca2+

. The first ramp current after whole-cell break-in was subtracted from each subsequent trace in this experi-

ment. A leak subtracted current at the peak of Icrac activation is shown. (B) A screen shot from a different Icrac experiment showing the

time course of Icrac development. The figure shows typical activation of Icrac with 20 uM IP3 in the intracellular solution. Toward the end

of this experiment, 30 uM 2-APB was added to block the current. Note the lack of the delay phase and the fast activation of the current.

Cells: RBL-2H3 cells were bought from ATCC (www.ATCC.org) and grown according to standard procedures. Solutions: The intracellular

solution contained (in mM): 145 Glutamate, 8 NaCl, 1 MgCl2, 10 HEPES, 10 BAPTA (or EGTA). pH was adjusted to 7.2 with CsOH and

osmolarity was adjusted with sucrose to 320 mOsm. When IP3 was used as the activator 20 uM was added to the intracellular solution on the

day of experiments. The extracellular solution contained (in mM): 140 NaCl, 2.8 KCl, 10 CaCl2, 2 MgCl2, 10 CsCl, 10 HEPES, 10 mM glu-

cose. pH was adjusted with NaOH to 7.4 and osmolarity was adjusted with sucrose to 350 mOsm.


resented a good working density. Following recovery, cells were washed, spun down and resuspended twice in an ap-propriate extracellular buffer (this step was automatically performed by QPatch-16 when using the automation mode).

The second consideration was the osmolarity of experi-mental solutions. For both channel types, it was important to maintain a higher level of osmolarity for extracellular solu-tions than for intracellular solutions. For TRP experiments, the extra- and intra-cellular solutions were maintained at 310-320 and 290-300 mOsm, respectively. For VGCC ex-periments, the corresponding values were 300-310 mOsm and 290-295 mOsm, respectively. Where necessary, osmo-larity levels were adjusted by adding appropriate amounts of sucrose or water prior to filtration.

Solution freshness also affected the outcome of assays. While it has not been necessary for us to make fresh solu-tions every day or even every week (we have used solutions up to a month old without significantly affecting seal forma-

tion; data not shown), we highly recommend that solutions be kept at 4°C for storage and filtered each time before use. Solutions prepared daily may be more essential for some channel and/or cell types that are not discussed here.

Lastly, to minimize non-specific binding of compounds to plastic, we used glass inserts to hold compound solutions in 96-well drug plates during experiments.

The QPatch enlists various “stage gates” to indicate the status of each of the 16 chambers in a QPlate during the course of an assay. If a hole primes properly with intra- and extra-cellular solutions and a passing cell subsequently sticks onto the hole (cell-attached), an attempt to form a tight (of-ten gigaohm) seal can be made. For both TRP and VGCC assays, if the seal resistance (Rseal) failed to reach 100 M , the QPatch software automatically discarded the cell. Other-wise, additional suction was applied to rupture the cell mem-brane and form the whole-cell configuration. The cell was allowed to proceed with the assay only if the initial whole-

Fig. (9). QPatch 16 experiment with NIH 3T3 cells expressing CFTR channels. Ramps (-100 - +100 mV) were applied every 5 s. The peak

inward (at -100 mV) and peak outward (+100 mV) current amplitudes are graphed in this screen shot (blue symbols). In the Saline period a

stable baseline current was measured in nearly symmetric chloride solutions. In the second period, the forskolin/genistein (10/20 M) mix is

added. In the third period the chloride is partly substituted (EC: 30 mM). The magenta symbols show the reversal potential determined for

each ramp. In the plot to the right a fast and transient change in reversal potential can be seen when forskoline/genistein is added. This

change is towards the reversal potential for potassium, EK. We speculate that this is caused by an opening of a cAMP dependent potassium

channel. The NIH 3T3 cells were grown according to the Sophion standard operating procedure. NIH 3T3 cells expressing deltaF508CFTR

were moved to 27°C 24 h prior experiments. Extracellular solution contained (in mM): 2 CaCl2, 1 MgCl2, 10 HEPES, 4 KCl, 145 NaCl, 10

Glucose, pH 7.4 (with NaOH) and 300 to 305 mOsm. Intracellular solution contained (in mM): 5.374 CaCl2, 1.75 MgCl2, 31.25/10

KOH/EGTA, 10 HEPES, 120 KCl, 4 Na2-ATP, pH 7.2 (with KOH) and 290 to 295 mOsm.


cell membrane resistance (Rwc_ini) was 100 M . Al-though a 100 M cutoff is significantly lower than the gi-gaohm seal target and is a somewhat arbitrary selection, it has been found empirically that most cells that met this cut-off criterion actually developed significantly higher Rwc and had a high probability of completing the assay (see Table 1 and Fig. 16).

TRP Channel Assay

Fig. (10A) compares the current-voltage (I-V) relation-ship of TRP channel currents measured with either the QPatch-16 (left panel) or conventional patch clamp (right panel). Currents were elicited by a voltage ramp protocol upon activation of the channel by an agonist. As would be expected, both methods produced similarly voltage-dependent currents, with dual rectification and a reversal potential around 0 mV. In the screening assay (Fig. 10B), the holding potential (Vh) was -80 mV and the current amplitude over a 10-ms period was recorded once every 2 s. After es-tablishing a stable baseline in the control solution, an agonist was applied to open the channels, followed by the applica-tion of a test compound (antagonist) in the presence of the agonist. Finally, a saturating concentration of a reference antagonist was co-applied with the agonist to completely block the TRP current and thereby establish the leak current level from which the steady-state current amplitudes in the previous two applications were subtracted. For compounds tested against this channel, we made double compound ap-plications for each concentration (first application of 10 L followed 100 s later by a second application of 5 L) to en-sure adequate compound exposure and included 0.1% bovine serum albumin (BSA) in all extracellular solutions (control, agonist, test compound and reference compound) in the as-say. Both procedures helped to increase the potency of test compounds and to bring the correlation with results from conventional patch clamp (no BSA) close to the unity line (Fig. 11A). A decent correlation can also be established be-tween QPatch results and results from FLIPR (also with 0.1% BSA), the primary screening assay for this channel (Fig. 11B), albeit with an average of ~10 fold potency shift in favor of the QPatch.

VGCC Assay

Fig. (12A) illustrates the similarity between currents ac-tivated by depolarizing voltage steps obtained using the QPatch (upper traces) and conventional patch clamp (CPatch; lower traces). The I-V relationships generated from these currents are also very similar for these two methods (Fig. 12B).

The screening assay was designed to capture not only compounds that inhibit the channel tonically, but also those that exhibit use-dependent block. Cells were held at -80 mV and stimulated with six trains of 15 brief (20 ms) depolariz-ing pulses (to +20 mV) at 5 Hz. The interval between each train was 30 s (Fig. 13A). Fig. (13B) shows a screen shot from the QPatch Assay Software. The current amplitude elicited by each brief depolarization for select pulses throughout the duration of the assay is plotted against time (seconds). Typically, two to three concentrations (10 L each) of a test compound were applied cumulatively fol-lowed by the application of 100 M cadmium (10 L) to establish the leak current level from which the steady-state current amplitudes in the previous applications (including control) were subtracted. There was a good, near one-to-one correlation between compound potencies obtained by the QPatch and conventional patch clamp (Fig. 14A). Good cor-relations were also observed, on an individual chemical se-ries basis, between the QPatch and FDSS (a fluorescence-based calcium influx assay developed by Hamamatsu), the primary screening assay for this channel (Figs. 14B-D).

Compounds have been identified that exhibit either no use dependence or moderate use dependence (Fig. 15A). Importantly, neither cadmium (Fig. 15B) nor a neurotoxin (data not shown) exhibited use dependence in QPatch ex-periments, an observation consistent with results from con-ventional patch clamp (data not shown).

Assay Statistics

Table 1 summarizes the results of six parameters for both TRP and VGCC experiments: Rchip (chip resistance before cell attachment), Rseal (membrane resistance in cell-attached configuration), Rwc_ini (whole-cell membrane resistance at

Table 1. Values for the Following Parameters for the TRP and VGCC Assays: Open Chip Resistance (Rchip), Membrane Resis-

tance in Cell-Attached Configuration (Rseal), Whole-Cell Membrane Resistance at the Beginning and End of an Assay

(Rwc_ini and Rwc_end), and Series Resistance at the Beginning and End of an Assay (Rs_ini and Rs_end). Three Values

are Given for Each Parameter (at 25, 50 (Median) and 75 Percentiles) for Both Completed and Usable Cells

Channel Stage Rchip

(Mohms) Rseal (Mohms) Rwc_ini (Mohms) Rwc_end (Mohms)

Rs_ini

(Mohms)

Rs_end

(Mohms)

25% 50% 75% 25% 50% 75% 25% 50% 75% 25% 50% 75% 25% 50% 75% 25% 50% 75%

completed

1.91 1.98 2.06 513 1283 3079 608 2193 3825 264 597 1890 5.9 8.6 12.1 9.7 13.8 19.3 TRP

usable 1.91 1.98 2.06 532 1357 3339 694 2219 3958 319 607 1643 5.8 8.3 11.4 9.3 13.0 17.7

completed

1.75 1.82 1.89 505 881 1464.376281 587 1225 1838.590013 1340 2103 3138.068663 3.4 4.5 6.0 4.5 5.9 7.9 VGCC

usable 1.74 1.81 1.89 577 900 1503 685 1253 1857 1599.514581 2311.2898 3265.390806 3.4 4.4 5.8 4.5 5.7 7.5


the beginning of an assay), Rwc_end (whole-cell membrane resistance at the end of an assay), Rs_ini (series resistance at the beginning of an assay) and Rs_end (series resistance at the end of an assay). For each channel, stats were compiled for both completed cells and usable cells. Completed cells are those that satisfied the following conditions: Rwc 100 M and the current amplitude 20 nA. Usable cells repre-

sent a subset of completed cells that further satisfied the fol-lowing conditions: Rwc_end 100 M , Rs_ini 30 M , Rs_end 30 M , and current amplitude 30 pA. For each parameter, values are given at 25

th percentile, 50

th percentile

(median) and 75 th

percentile to give a sense of the distribu-tion of the values. For all practical purposes, values for com-pleted and usable cells are virtually identical, which is not

(A)

(B)

Fig. (10). I-V relationships for a TRP channel recorded using QPatch (A, left panel) and conventional patch clamp (A, right panel) and time

course of a TRP-expressing cell in the QPatch screening assay (B). In (B), the holding potential is -80 mV and the current amplitude over a

10-ms period is recorded once every 2 s.


unexpected given the criteria for their selection. It is evident that even though the minimum Rwc value for inclusion of completed and usable cells was set to be only 100 M , the majority of those cells developed true gigaseals.

Fig. (16) summarizes the success rates (defined as % of cells/electrode holes that meet a set of minimum require-ments for a particular stage) at various stages of the two as-says discussed above. The definitions of the various success rates are as follows. Successfully primed holes are defined as those with an “open hole” resistance in the range of 1

Rchip 5 M . Successfully attached cells are those that experienced a 750% jump in Rseal from the Rchip value. Cells with “giga” seals include all cells with an Rseal 100 M in the cell-attached configuration (as is evident from Table 1, a vast majority of those cells that met this minimum criterion had much higher values of Rseal). Cells that suc-cessfully achieved the whole-cell configuration must have a value of Rwc_ini 100 M . It is worth noting that cells that were successful in reaching the whole-cell configuration also

Fig. (12). VGCC currents elicited by voltage steps ranging from -40 mV to +40 mV (20 mV increment) for QPatch (A, upper panel) and

conventional patch clamp (A, lower panel). The holding potential is -80 mV. I-V curves are shown in panel B for both methods.

(A) (B)

Fig. (11). Correlation plots for TRP channel. (A) QPatch with conventional patch clamp. (B) QPatch/CPatch with FLIPR.


(A)

(B)

Fig. (13). (A) Voltage protocol used in the screening assay for VGCC (upper trace) and currents elicited by the protocol (lower trace). Each

train contains 15 brief (20 ms) depolarizing steps to +20 mV at 5 Hz. The time interval between each train is 30 s. The holding potential is -

80 mV. (B) Time course of peak VGCC current amplitudes elicited at +20 mV in the absence or presence of a test compound. For viewing

clarity, only data points for pulses 1, 2, 5, 10 and 15 are displayed.


had a high probability of reaching completion. The overall success rate (% of usable cells) was 34% and 45% for TRP and VGCC, respectively. The decrease in success rate from “completed” to “usable” for both assays primarily resulted from the exclusion of those completed cells that did not have adequate functional expression.

QPATCH HT

The 48-channel QPatch HT extends the throughput of the QPatch 16 by about three fold without sacrificing data qual-ity. It uses the same QPlate technology as the QPatch 16 - just repeated three times (Fig. 17). Of the several new/improved features added to the QPatch HT design is the capability of handling multiple compound plates in an auto-mated fashion to maximize throughput. In an effort to esti-mate the maximum daily throughput, three 96-well com-pound plates were tested using one 48-channel QPlate. Mul-tiple compounds per cell were added to CHO-hERG cells at a screening concentration of 10 M. An average of 5.6 com-pounds per cell was tested at a rate of 5.8 compounds per minute. Of 286 experiments started, 251 completed success-

fully (88%). Extrapolating the rate of 5.8 compounds per minute gives a maximum capacity of ~ 6000 compounds. While it is unreasonable to assume non-stop, 24 hday opera-tions (without multiple working shifts), a 16 h day is quite attainable because of the walk-away time afforded by the QPatch in the unattended mode. Therefore, in the screening mode, the QPatch HT is capable of screening ~ 4,000 com-pounds in a 16 h day (assuming multiple compounds per cell).

In order to determine assay variability with the QPatch HT, we conducted a single concentration screening experi-ment with HEK Nav1.2 cells and computed Z’ values for the fraction of block [16]. A 96-well compound plate was set-up with 48 wells containing saline and 48 wells with 100 nM TTX to block the Na currents. The assay was setup to execu-te as many experiments per cell as possible (140 experiments were completed). Fig. (18) shows the results from this ex-periment with a plot of fractional block versus sample num-ber. The calculated Z’ value of 0.58 shows an acceptable amount of assay variability and confirms that the QPatch HT can be used as a compound screening tool.

Fig. (14). Correlation plots for VGCC. (A) QPatch with conventional patch clamp. (B-D) QPatch with FDSS for three individual chemical

series.


DISCUSSION

The QPatch has been used to develop secondary screen-ing assays to follow up on primary screens from more con-ventional HTS approaches, as exemplified in the case studies described in this review. Here, the successful development of secondary screening assays was described for two (biophysi-cally and pharmacologically) distinct ion channels: a ligand-gated TRP channel and a voltage-gated calcium channel. There was decent correlation between compound potencies measured by the QPatch on one hand, and those measured by

either conventional patch clamp or primary HTS assays on the other. The overall success rate for usable data was be-tween ~1/3 and ~1/2 for the two assays (some cells were lost due to lack of expression). Operated in the unattended, automation mode, the QPatch was capable of completing 8-10 QPlates/8-h day for either assay, which translates into 40-50 usable cells/day for these assays at the present levels of success. This represents significantly (~10 fold) higher throughput than our conventional patch clamp (as measured by the number of usable cells tested per assay day/operator). Importantly, since the automation mode of the QPatch en-

Fig. (15). Use dependence of inhibition of VGCC by the QPatch assay. (A) Examples of use-independent (Compound 1, left panel) and use-

dependent (Compound 2, right panel) inhibition. IC50 values are calculated using steady-state peak currents during the first and 15th

depolar-

izing pulses to +20 mV, respectively. (B) Correlation of IC50s between the first pulse and the 15th

pulse for a collection of compounds tested

using QPatch-16.


ables up to 4 h of unattended screening, the throughput could be increased by up to 50% if it was also set up to run during after hours. At least for the VGCC assay, we found that the overall success rate (% usable cells) was significantly higher in the unattended mode (45%) than in the manual mode (27%).

Fig. (16). Success rates at various stages of the QPatch-16 assays

for TRP (red bars; 97 QPlates) and VGCC (green bars; 89 QPlates)

stably expressed in HEK293 cells and for Nav1.2 channels (blue

asterisks; 12 QPlates) stably expressed in CHL1610 cells. Defini-

tion of success rates is given in the main text.

Fig. (17). Picture of a 48-channel QPlate electrode plate.

The two assays discussed here were both developed for ion channels expressed in HEK293 cells. With certain ion channel/cell line combinations, better results can be obtained with assays developed using Chinese hamster ovary (CHO) cells. For example, the asterisks in Fig. (16) indicate various success rates of an assay developed for Nav1.2 channels ex-pressed in CHL1610 cells, a close cousin of CHO cells. The success rates for this assay were significantly higher across the board compared to the two HEK293 assays. In addition, we also find that CHL1610 cells are viable in the QPatch cell preparation station for a longer period of time ( 4 h) than

HEK293 cells. Based on these results, it seems reasonable to, whenever feasible, develop assays using CHO or possibly other cell lines instead of HEK293 cells so as to obtain higher success rates. Nonetheless, the throughput of our HEK293 assays still represent an order of magnitude im-provement over conventional patch clamp and makes it pos-sible to provide critical and timely information for chemistry efforts.

Fig. (18). Z -Factor Determination of HEK Nav 1.2 assay. Z’-factor

was determined by measuring the current response to 100 nM TTX

or saline. A compound plate containing 48 negative saline controls

and 48 wells with TTX was used. The assay was setup to execute as

many experiments per cell as possible. Only experiments where the

membrane resistance at all times was above 700 M and the cur-

rent amplitude was above 100 pA were included in the analysis. A

total of 140 experiments was performed on one QPlateXL with a

Z’-factor determined to be 0.58 by the following equation:

where st is the standard deviation of the blocked current amplitudes

at the top (ttx block; open circles), sb is the standard deviation of the

blocked current amplitudes at the bottom (saline controls; filled

circles), mt is the mean current amplitude at the top (ttx block; open

circles), mt is the mean current amplitude at the bottom (saline con-

trols; filled circles). This particular experiment took 84 min to com-

plete.

In general, we observed that success rates with channels expressed in CHO cells generally exceeded those of channels expressed in HEK cells. We speculate that this observation results from the fact that HEK cells tend to stick or clump together in suspension. This should come as no surprise, since these cell lines were first developed for primary screen-ing in fluorescent based assays in which adhering to the bot-tom of multi-well plates was important for high quality data.

Interestingly, HEK-293 cells without any genetically engineered ion channels expressed show relatively high suc-cess rates in QPatch experiments. In one set of experiments with five 16-channel QPlates using solutions similar to those used for CHO cells, we observed a gigaseal success of 74% with native HEK cells (59/80 cells). In these experiments, the whole-cell success rate was 66% (53/80 cells). Therefore, it seems as if the combination of cell line and expressed ion


channel creates a unique cell “personality” that leads to dif-fering success rates with automated patch clamp assays.

It is worth noting that the success rate for completed cells is 57% and 48% for VGCC and TRP channels, respectively. This is substantially higher than for usable cells (45% and 34%, respectively). The same conclusion can also be drawn for the Nav1.2 assay using CHL1610 cells (79% completed vs 49% usable). This drop in success rate is primarily due to failure of some completed cells to have sufficiently high channel expression for inclusion. It is conceivable that this gap may be significantly reduced if assays are developed using cell clones that are first selected using QPatch for high channel expression. Also interesting to note is the higher success rates for VGCC than for TRP (for completed and usable cells). While many factors could have contributed to this difference, including differences in channel “personali-ties” (see above), it is likely that the significantly longer du-ration for the TRP assay (~30 min for TRP vs ~12 min for VGCC) was mainly responsible for this difference. This is supported by the data in Fig. (16) showing that there is virtu-ally no difference between the two assays in terms of whole cell success rate, whereas the difference emerges by the end of the assays (completed cells).

This review showed QPatch data from both voltage- and ligand-gated ion channels including Nav1.2, Cav3.2, hERG, P2X3, GABA, CRAC, CFTR, a TRP channel and another voltage gated calcium channel. The P2X3 assay was per-formed in perforated patch clamp mode, showing that this patch clamp configuration is possible with the QPatch. Other assays developed by Sophion (but not shown here) include alpha-7 nicotinic AChR’s, ASIC, Nav1.5, Kv1.5 and KCNQ. Typically, CHO and HEK cell lines express these channels. In some cases, other cell lines were used including GH4 (al-pha-7 nicotinic AChR’s), NIH 3T3 (CFTR), and RBL-2H3 (CRAC).

Gigaseal, whole-cell and usable recording success rates for QPatch ranged from 40-80%, depending on the cell line and ion channels expressed, as described above. Success rates at the lower end of this range typically resulted from cells lacking sufficient ion channel expression and/or inade-quate assay development time. These success rates are simi-lar to those reported for other similar automated patch clamp systems [17, 18]. It is also worth noting that gigaseal success rates can be misleading. In most experiments, many seals reach values between 500-999 Mohms. These seal values are not included in the gigaseal success rates. Most of these high seal resistance (but below 1 gigaohm) lead to high-quality whole-cell recording, as shown in the case studies.

In this review, we showed that the QPatch can be used to develop screening assays for various types of ion channels with greatly improved throughput over conventional patch

clamp while retaining the same high quality of data. Indeed, it is the authors’ belief that the QPatch throughput is suffi-ciently high (particularly with the unique automation capa-bility of QPatch and the advent of QPatch HT) to keep up with typical medicinal chemistry efforts at the lead optimiza-tion stage.

ACKNOWLEDGEMENTS

Many scientists at Sophion Bioscience contributed data for this review. We would like to especially thank the fol-lowing people for their contributions: Rasmus B. Jacobsen (CFTR), Ali Yehia (native HEK data), Hervør L. Olsen (GABA), Naja M. S Sørensen (hERG), Rikke S. Schrøder (Icrac), M. Knirke Jensen (helpful discussions). Many thanks also to Morten Sunesen for leading the biology team at Sophion and for providing thoughtful scientific input into the research and this manuscript. We appreciate expert technical assistance from Dorthe Nielsen, Nadia Larsen, Jeffrey Web-ber and Yan Wang at both Sophion and Johnson & Johnson.

Many thanks to Mads G. Korsgaard (Neurosearch, DK) for providing Nav1.2a data, bSYS for providing CHO-hERG cells, and Millipore for providing GABA and P2X3 cells.

REFERENCES

[1] Pereira D.A.; Williams J.A. Br. J. Pharmacol., 2007, 152,53.

[2] Fox S.; Farr-Jones S.; Sopchak L.; Boggs A.; Nicely H.W.; Khoury R.; Biros M. J. Biomol. Screen., 2006, 11,864.

[3] Mathes C. Expert Opin.Ther. Targets, 2006, 10,319. [4] Willumsen, N. J. Am. Biotechnol. Lab., 2006, 24,20.

[5] Hille, B. Ion Channels of Excitable Membranes, 3rd edition, Sinaur Associated:Sunderland, MA, 2001.

[6] Perchenet L.; Benardeau A.; Ertel E.A.; Naunyn Schmiedebergs Arch Pharmacol., 2000,361,590.

[7] Roden, D.M.; Balser, J.R.; George, A.L.; Anderson, M.E. Annu. Rev. Physiol., 2002, 64,431.

[8] Brown A.M. Cell Calcium, 2004, 35, 543. [9] Conley, E.C. The Ion Channel FactsBook. Extracellular ligand-

gated channels. Academic Press: London, 1996. [10] North, R.A. Physiol.Rev., 82, 1013.

[11] Hoth M.; Penner R. Nature, 1992, 355, 353. [12] Zhang S.L.; Ying Y.; Roos J.; Kozak J.A.; Deerinck T.J.; Ellisman

M.H.; Stauderman K.A.; Cahalan M.D. Nature, 2005, 437, 902. [13] Vig M.; Peinelt C.; Beck A.; Koomoa D.L.; Rabah D.; Koblan-

Huberson M.; Kraft S.; Turner H.; Fleig A.; Penner R.; Kinet J.P. Science, 2006, 312, 1220.

[14] Schrøder, R.S.; Friis, S.; Sunesen, M.; Mathes, C.; Willumsen, N.J. J. Biomol. Screen. (in review).

[15] Zweifach A.; Lewis R.S. Proc. Natl. Acad. Sci. USA., 1993, 90, 6295.

[16] Iversen P.W.; Eastwood B.J; Sitta Sittampalam G.; Cox K.L. J. Biomol. Screen., 2006, 11, 247.

[17] Dubin A.E.; Nasser N.; Rohrbacher J.; Hermans A.N.; Marrannes R.; Grantham C.; Van Rossem K.; Cik M.; Chaplan S.R.; Gallacher

D.; Xu J.; Guia A.; Byrne N.G.; Mathes C. J. Biomol. Screen., 2005, 10,168.

[18] Brueggemann A., George M.; Klau M.; Beckler M.; Steindl J.; Behrends J.C.; Fertig N. Curr. Drug Disc. Tech., 2004, 1, 92.

Received: January 24, 2008 Revised: February 2, 2008 Accepted: February 2, 2008

combinatorial chemistry & high throughput screening, 000 ... › d98e › b45269da3...2...

Documents