Utility of Resazurin, Horseradish Peroxidase, and NMR Assays to Identify Redox-Related False-Positive Behavior in High-Throughput Screens

O R I G I N A L A R T I C L E

Utility of Resazurin, Horseradish Peroxidase,

and NMR Assays to Identify Redox-Related

False-Positive Behavior

in High-Throughput Screens

1

,2

2,3

2

Matthew Tarnowski, Am e´ lie Barozet, Carina Johansson,

INTRODUCTION

2

Per-Olof Eriksson, Ola Engkvist,

Jarrod Walsh, and J. Willem M. Nissink

ith the advent of high-throughput screening

4

1

(

HTS), it has become possible to rapidly screen

1

,2

large libraries of compounds. However it is

now clear that with larger compound libraries,

1

Oncology IMED, AstraZeneca, Cambridge, United Kingdom.

Discovery Sciences, AstraZeneca R&D Gothenburg,

Gothenburg, Sweden.

Department of Biology, Lund University, Lund, Sweden.

Discovery Sciences, AstraZeneca, Alderley Park, United Kingdom.

W

2

the risk of identifying and pursuing results that are artifacts of

3

screening increases. Pursuing such false positives results in

3

4

wasted resource, time, and effort. False-positive results may

arise from interference with signal readout through fluores-

cence or quenching, or from an undesirable mechanism of

inhibition of the assay target. Examples of the latter include

4

,5

inhibition by aggregation and chemical interference such as

redox-mediated modification or reactive modification of the

ABSTRACT

6

Discerning false positives from true actives in high-throughput

screening (HTS) output is fraught with difficulty as the reason of

anomalous activity seen for compounds is often not clear-cut. In

this study, we introduce a novel medium-throughput NMR assay

for the identification of redox-cycling compounds (RCCs), which

is based on detection of oxidation of a reducing agent. We

compare its outcomes to those from horseradish peroxidase

target.

Redox-cycling compounds (RCCs) are known to cause false-

positive hits in HTS against redox-susceptible targets such as

cysteine proteases, protein tyrosine phosphatases, and meta-

7

–10

lloproteases.

RCCs catalyze a radical-mediated redox cycle

between a strong reducing agent such as dithiothreitol or

tris(2-carboxyethyl)phosphine (TCEP) (often added to protect

1

1,12

(

HRP)/phenol red and resazurin (RZ)-based assays that are more

protein from oxidation in assays)

and oxygen, ultimately

commonly used for triaging HTS outputs. Data from NMR, RZ,

and HRP redox assay are shown to correlate, with the NMR assay

showing the greatest accuracy. In addition, historical data

analysis was used to identify compounds frequently active in

assays for redox-susceptible targets. We provide examples of

compound classes found and conclude that the NMR redox assay

offers a novel and reliable way of identifying RCCs at a medium

throughput. The HRP and RZ assays are reasonable higher-

throughput alternatives, with both showing similar sensitivity to

redox-cycling and false-positive compounds. The RZ assay has a

higher hit rate, reflecting its ability to pick up multiple modes of

action.

resulting in the production of hydrogen peroxide, which

damages the target through oxidation. Such damage may

subsequently lead to loss of activity, and this results in an

apparent inhibition of the target.

Redox-detecting assays have been put forward in the lit-

erature, and we have implemented two commonly used var-

11

iants: a resazurin (RZ)-based generic redox assay and a

1

3

horseradish peroxidase-based assay, specifically developed

to detect RCCs. Other assays detecting reactivity have been

14

reported in the literature, but are more generic in nature.

Redox effects remain a challenge in certain target classes, and

application of redox-deconvoluting assays has been reported

recently, for example, for lysine demethylases where half or

15

Keywords: redox false positives, HTS output deconvolution,

horse-radish peroxidase/phenol red assay, resazurin-based

redox assay, NMR

more of the hits may suffer from such interference. In this

study, we compare results from two commonly used assays to

those of a novel and orthogonal NMR-based approach, which

DOI: 10.1089/adt.2017.838

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 1

TARNOWSKI ET AL.

monitors the emergence of TCEP oxidation product as a proxy

indicator of redox cycling.

filter of 16.9 ms, which effectively eliminated the exchange-

broadened signals from TCEP (Fig. 2 and Supplementary

Data). Each spectrum was collected with 8 dummy scans and

128 scans, an acquisition time of 1.36 s, and a repetition delay

of 3.2 s, resulting in a total experimental time of 10 min and

23 s. From the time point of the files in the NMR data directory,

the start and finish of data acquisition could be determined.

Together with the timestamp of the first mixing stroke in the

sample preparation obtained from the Gemini logfile, the time

between sample preparation and start of data collection was

8 min and 12 s with a standard deviation of 6 s. Thereby, the

NMR data represent a –5 min time average with a midpoint at

13 min and 24 s after sample preparation. The NMR data were

processed with 2 Hz exponential line broadening and baseline

correction around the high field TCEPO multiplet (2.1 ppm)

(Fig. 2 and Supplementary Data). Control samples containing

TCEP in the absence of compound were run after every 10

samples. TCEP oxidation was quantified by integration of the

high field peak multiplet of TCEPO (2.1 ppm) coming from the

methylene protons next to the phosphorus atom. The peak in-

tegrals were extracted using a modified TopSpinꢀ (Bruker)

multispectra integration script and normalized to a control

sample run in absence of added compound. Compounds were

regarded as redox active if they increased the amount of TCEPO

by at least 10% compared to the control. Classification of the

redox actives in categories corresponds to TCEPO integrals

(relative to control) of 1–5 for weak, 5–15 for medium, and 15–

25 for strong actives.

To further analyze the nature of the compounds that are

suggested by these assays to be false positives, we performed

retrospective data mining of HTS results for a set of targets

from classes known to be redox sensitive: cysteine proteases,

metalloproteases, and protein tyrosine phosphatases.

MATERIALS AND METHODS

NMR Redox Assay

The NMR redox assay is based on detection of the H NMR

1

spectrum of the P-oxide of TCEP (TCEPO) in an aqueous buffer

containing the compound of interest and TCEP (see Supple-

mentary Fig. S2; Supplementary Data are available online at

www.liebertpub.com/adt). The spectrum of TCEP at pH 7.5 is

16

heavily broadened by chemical exchange effects, whereas

the spectrum of TCEPO consists of two well-resolved multi-

plets from the methylene groups (Fig. 2 and Supplementary

Data). By applying a transverse relaxation filter to the 1D

NMR pulse sequence, the exchange-broadened TCEP signals

are eliminated and the appearance of TCEPO in the sample is

easy to follow.

The samples were prepared in an aqueous buffer with 50 mM

TRIS-d , (98% D; Cambridge Isotope Laboratories, Inc., #DLM-

1

814-5), 200 mM NaCl, pH 7.4, 10% D O, and 22 mM TMSP, and

2

compound of interest at 100 mM. The samples were prepared

just in time before the NMR measurements with a TECAN

EVO100 sample preparation robot running the Gemini software

(

ver 4.0) (Tecan AG Zurich, CH). Aliquots of 5 mL of cooled and

pH-adjusted stock solution of TCEP (Sigma Aldrich, MA,

C4706) (0.3 M), and of stock solutions of compounds (10 mM

RZ-Based Redox Assay

11

We refer to Lor et al. for a complete description of the assay.

The original procedure was taken and optimized for assaying in

the presence of the reducing agent TCEP (Sigma Aldrich,

#C4706). Based on titration of RZ (Sigma Aldrich, #199303) and

TCEP concentration (Supplementary Data), 5.0 mM TCEP and

5.0 mM RZ were selected for screening. These provided a com-

#

in nondeuterated DMSO) were added to the NMR tube con-

taining 490 mL. The samples were mixed mechanically by using

a pipetting needle with a teflon plug attached, acting as a pis-

ton. The samples were then transported individually to the NMR

spectrometer with a SampleRailꢀ (Bruker Biospin, GmbH,

Rheinstetten, DE) followed by spectrum acquisition.

0

promise between an acceptable signal to background ratio and Z

value, while maintaining assay sensitivity to detect RCCs. In

summary, compounds (100 nL, 10 concentrations from 100 to

0.2 mM in DMSO) were loaded into black, low-volume 384-well

1

The 1D H NMR spectra were acquired at 293 K using a

Bruker Avance III HD 800 MHz spectrometer equipped with a

cryogenically cooled probe head. The water magnetization

and the signal from DMSO were simultaneously suppressed by

ꢁ

Greiner 784076 plates using an Echo (Labcyte, Inc.). On-board

DMSO and NSC663284 (1.5 mM in DMSO; Sigma Aldrich,

#N7537) control wells were used. RZ (5.0 mM in assay buffer)

was sonicated for at least 10 min and then added to an assay

buffer (50 mM HEPES [Sigma Aldrich, #H7006] and 50 mM NaCl

[Sigma Aldrich, #S7653], pH 7.4 in milliQ water) to give a final

concentration of 5.0 mM, and TCEP (5.0 mM in assay buffer, pH

7.4) was added to give a final concentration of 5.0 mM. Then,

10 mL of this solution was added to all wells using a Multidropꢀ

17,18

excitation sculpting,

using biselective 3 ms shaped 180

degree flip-back pulses. The suppression of the DMSO peak

also resulted in suppression of the low-field methylene dou-

blet (2.4 ppm) of TCEPO (Fig. 2 and Supplementary Data). The

excitation sculpting sequence (Bruker Topspin pulse program

zgesgp) had two pairs of 2.5 ms gradient pulses, whereby

the entire pulse sequence contained a transverse relaxation

2

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

3

84 Reagent Dispenser (Thermo Scientificꢀ). The wells were

activity across the assays it has been tested in against the ex-

pected incidence of activity. The latter is determined from the

larger set of data for all compounds. The pBSF score is the

negative logarithm of a chance and reflects the probability

that the observed tally of activity for the compound occurs

randomly, given the expected incidence of activity. If this

probability is very low, it suggests that the observed activity

pattern is an outlier and goes beyond what is expected.

In practice, such outliers tend to be compounds more col-

loquially known as ‘‘frequent-hitters,’’ anomalous inhibitors

that appear to give rise to inhibition readouts in a larger-

than-expected number of assays. Typical threshold values

for pBSF lie in the 2–3 range, corresponding to a false-

positive chance of 1%–0.1% that the compound is not an

outlier, indicating a large likelihood that the compound is an

anomalous inhibitor.

allowed to react for 60 min at room temperature after addition of

RZ/TCEP solution and were then read on a PHERAstar (BMG

LABTECH) with Ex =560 nm/Em =590 nm.

Horseradish Peroxidase/Phenol Red Based Assay

13

We refer to Johnston et al. for a complete description of

the assay. Their procedure was followed using 1 mM TCEP

(

Sigma Aldrich, #C4706) as the reducing agent. In summary,

compounds (200 nL, 10 concentrations 100 to 0.2 mM in

DMSO) were loaded into clear polystyrene 384-well Greiner

7

81101 plates using an Echo (Labcyte, Inc.). On-board DMSO

and NSC663284 (1.5 mM in DMSO; Sigma Aldrich, #N7537)

control wells were used. TCEP (0.5 M, pH 7.4) was diluted to

1

mM in HBSS (Sigma Aldrich, #H6648). The phenol red-HRP

detection reagent was prepared (400 mg/mL phenol red [Sigma

Aldrich, #P3532] and 25 U/mL HRP [Sigma Aldrich, #P2088]

in HBSS). Twenty microliters of the TCEP solution was added

to all wells and after 15 min of incubation at ambient tem-

perature, 20 mL of the HRP-phenol red solution was added to

all wells. After 5 min of incubation at ambient temperature,

Selection of a set of putative redox-active compounds. Results from

historical high-throughput screens were used to identify

compounds potentially acting through a redox-cycling

mechanism by focusing on a subset of relevant assays. Data

from a set of 21 screens run at AstraZeneca and targeting

various redox-susceptible proteins, including cysteine prote-

ases, metalloproteases, and protein tyrosine phosphatases,

were gathered (Supplementary Table S1), and pBSF scores

corresponding to results in these assays were calculated. Only

compounds with a pBSF higher than or equal to three were

included. These compounds show clear evidence of anoma-

lous hitter behavior and will likely interfere through unknown

mechanisms, including redox activity.

1

0 mL of NaOH (0.5 M) was used to stop the reaction. All ad-

ditions were completed using a Multidrop 384 Reagent Dis-

penser (Thermo Scientific). Each well was measured using a

PHERAstar (BMG LABTECH), with 600 nm absorption.

EC50 curves. For both assays described above, EC50s were de-

termined using 10 concentrations, with a maximum concen-

tration of 100 mM, in half-step dilutions, and in triplicate.

Original compound stock was kept in 10mM DMSO solutions.

Twelve wells of DMSO (-ve control) and 12 wells of NSC663284

As non-redox mechanisms can raise issues in assays more

generically, additional sets of ‘‘background’’ assays were col-

lated and the same approach was applied to obtain pBSF scores

based on these sets. Since all selected redox-susceptible assays,

except one, are Fluorescence Intensity, F o¨ rster Resonance En-

ergy Transfer, or spectrophotometric assays, sets of background

assays for each of these categories were selected, as well as a

generic set of diverse assays to assess general promiscuity of

compounds (Supplementary Data and Supplementary Table S2).

pBSF values were derived for each of these background sets. A

set of putative redox-active compounds was then selected by

retaining those tested at least once and with a pBSF lower than

2, that is, ‘‘clean’’ in each set of background assays. This set

comprises 1,097 compounds and is referred to as S.

(+ve control) were present on each assay plate and used to apply

a plate-based normalization of assay signals in Genedata

ꢁ

Screener 12 (Genedata AG, Basel, Switzerland, 2015). The data

were plotted utilizing Genedata Screener’s ‘‘Smart Fit’’ option to

apply a four-parameter fit. This enabled the software to auto-

matically fix the asymptotes to 0% and 100% if required. A

minimal nHill value of 0.5 and a maximal nHill value of 8 were

permitted. All curves had to exhibit a span in percentage effect

of greater than 50%, but less than 150% to be classed as a valid

fit. Automated outlier masking was enabled.

Data Mining Methodology

Characterization of anomalous inhibitor behavior. We character-

ized both generic ‘‘frequent-hitter’’ behavior using a previously

S consists of compounds found to act as frequent hitters in

redox-susceptible assays, but not in other assays, suggesting

that these are either redox active (e.g., RCCs) or compounds

actually capable of binding to several proteins of one class

(such as protein tyrosine phosphatases inhibitors containing

1

9

described method and potential redox-cycling behavior by

analyzing a subset of data derived from activity data for redox-

susceptible targets. In summary, the method calculates a score

(

pBSF) for a compound by comparing the actual incidence of

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 3

TARNOWSKI ET AL.

phosphate-mimetic groups). The latter were removed in a

further step comprising visual inspection and checks of his-

torical data.

approach, which is generally applicable when the range of the

activity seen in the primary HTS is limited.

HTS Compound Set for Triage

Derivation of fragments enriched for redox behavior. Compounds

in S were fragmented and an enrichment analysis was per-

formed to identify overrepresented and underrepresented

fragments (see Supplementary Data). This led to the manual

selection of 42 representative enriched fragments. Binding

properties of these fragments were assessed subsequently by

investigating existing complexes in the Protein Data Bank

A set of 4,921 compounds was identified based on actives

from a MALT1 HTS campaign. MALT1 is a cysteine protease.

In the course of the HTS workup, these actives were subjected

to a hit triaging cascade to filter out, among others, com-

pounds that appeared false positive as a result of redox-related

interference. A subset of this set was processed using our

NMR-based assay as part of the cascade.

2

0

(

PDB) (see Supplementary Data for details).

The group of compounds associated with an enriched

fragment f was defined as the set of compounds tested in at

least one of the selected redox-susceptible assays and con-

taining f, but no other representative enriched fragment (un-

less latter fragment is a subfragment of f). To determine the

frequent-hitter and redox-cycling tendencies of a given

fragment, we tested for an increased proportion of frequent

hitters among its associated group with regard to the redox

target and background sets of assays. One-tailed binomial

Comparison of HRP and RZ Assays

A set of 4,921 compounds were screened in the HRP and

RZ assays to assess their correlation. The HRP assay followed

1

3

the Johnston et al. protocol utilizing 1 mM TCEP for a 15-

min reaction. This provided linear detection of H O pro-

2

duction up to 100 mM, although there was capacity in the

system for nonlinear detection up to 400 mM. The RZ assay

was optimized to use 5 mM, performed over 60 min, and

capable of linear signal detection for full conversion of the

dye to resorufin. Differences in the relative sensitivity of the

respective assays to detect RCCs resulted in varying and

bespoke criteria needing to be adopted when assigning ac-

tivity. This was exacerbated by the anomalous (non-

sigmoidal) fitting of a number of EC50 curves. The classifi-

cation of actives was determined as follows and provides, in

our opinion, an accurate representation of compound ac-

tivity across the three assays: compounds with a pEC50 of

four or less in the RZ assay were designated as inactive (N),

thereby classifying as active (A) all compounds that gave

rise to bell-shaped curves, and/or had a pEC50 > 4. EC50s

derived using the HRP assays turned out to be too weak to

be used consistently for classification, and therefore the com-

pounds with %effect at the highest concentration (100mM)

>5 · standard deviation of the inhibition values of the null

control wells were labeled as active.

21

tests were applied to frequent-hitter classifications using

pBSF >3. A test was applied on each pair of fragment and set of

2

assays. Each test was followed by Holm correction and

considered positive if P < 0.001. A positive test indicates an

increased incidence of frequent hitters for the group of com-

pounds associated to a fragment in a set of assays.

RESULTS AND DISCUSSION

Redox Assays

Two assays with a suitably high throughput have been re-

ported in the literature for the detection of redox effects like

reducing agent-mediated redox cycling. One uses resazurin

(

RZ) to detect compounds that are able to catalyze its radical-

11

mediated conversion to resorufin (referred to as the RZ assay).

The other employs horseradish peroxidase (HRP) to detect for-

mation of hydrogen peroxide by monitoring the oxidation of

13

phenol red by HRP (referred to as the HRP assay). The latter

assay will detect effects of direct oxidation, as well as effects

resulting from superoxide anion and peroxide formation. Tria-

ging is generally done in two different ways—a worst-case

scenario is checked where a compound is tested at relatively

high concentration for any redox activity or the level of redox

activity (e.g., determined as EC50) is compared to the level of

activity in the primary HTS target. Considering the level of re-

dox activity in the context of the primary target activity is

relevant, but can be ambiguous, for example, as a result of

anomalous redox curve shapes. We focus here on the former

Curves determined for our set of compounds using the RZ

assay were commonly found to exhibit bell-shaped curves.

11

This behavior has previously been noted by Lor et al. In such

cases, the curve-fitting software (Genedata Screener) was

unable to assign a pEC50 value; these compounds were as-

signed as active when %effect at any concentration tested

>5 · standard deviation of the null control wells.

RZ and HRP assays show a reasonable correlation in the

designation of redox activity (Fig. 1). The majority of com-

pounds flagged as active by the HRP assay is also flagged by the

RZ assay (95%); a small fraction of HRP inactives (8%) is flagged

4

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 5

TARNOWSKI ET AL.

as active by the RZ assay. Of the RZ actives, approximately two

thirds are designated active by the HRP assay; only 1% of the RZ

inactives are labeled as active by the HRP assay.

NMR Assay

To characterize the sensitivity and specificity of the RZ and

HRP assays, we developed a novel 1D NMR assay with an ac-

curate, oxidation-dependent readout orthogonal to the read-

outs of the biochemical assays. This assay detects the formation

of the TCEPO, which arises for example, from the formation of

hydrogen peroxide catalyzed by presence of the compound

The RZ assay shows a hit rate of 21% across the compounds

tested, versus 15% in the HRP assay. This may either be a

consequence of the difference in the way these assays detect

redox behavior or be indicative of one assay having a greater

sensitivity for false positives than the other. Because a positive

readout in the HRP assay indicates formation of hydrogen

peroxide, such a result strongly suggests a redox-cycling

mechanism. Therefore, a compound active in the HRP assay

through an H O -mediated mechanism must appear active in

2

3

(Fig. 2a).

The NMR redox assay takes advantage of the pH-dependent

1

changes in the H NMR spectra of TCEP and TCEPO. The assay

1

is designed to be carried out at neutral pH, at which the H

1

spectrum of TCEP is highly broadened, while the TCEPO H

2

the RZ assay (which employs a measure of radical formation).

Conversely, compounds that appear active in the RZ assay

need not always show up in the HRP assay if their mode of

action does not give rise to formation of H O . This seems to be

NMR spectrum presents two well-resolved multiplets (Fig. 2

and Supplementary Data). Samples initially contain only TCEP

and the compound of interest in the aqueous buffer. The assay

requires no protein and little optimization is needed. However,

since the redox mechanism is catalytic, it is important to

minimize any variation in timing between preparation of the

sample and NMR measurements. In our case, the samples were

prepared by a sample preparation robot, which executes a pi-

petting script at a certain time point before start of the NMR

experiment acquisition. After mixing, the sample was delivered

2

supported by the data (Fig. 1). An interesting observation is

that compounds are generally active in the RZ assay at 10- to

1

00-fold lower concentrations than in the HRP assay. This

may reflect differences in assay setup as optimized in this

study, for example, time in the presence of reducing agent and

reducing agent concentration.

23

1

Fig. 2. (a) Conversion of TCEP to TCEPO by hydrogen peroxide. From Tan et al. (b) Part of 1D H NMR spectrum showing the high field

multiplet signal from TCEPO. The TCEPO signal is integrated and normalized relative to a control sample lacking compound. Compound

AZ4266 has no redox activity, whereas A9045, AZ3789, and AZ5571 are redox active and cause an increase in TCEP oxidation compared to

the control. (c) Hydrogen peroxide calibration curve with peroxide concentration on x-axis and TCEPO integral relative to control on y-axis.

6

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

to the NMR spectrometer by a SampleRail sample delivery

system. The automated setup eliminates timing variability be-

tween samples (see Materials and Methods section for details)

and is superior to manual preparation and data acquisition in

precision in timing.

derived from the RZ and HRP assay data. Figure 3a–d con-

tain receiver-operator curves (ROCs) depicting sensitivity

(true-positive rate) and 1-specificity (1-true negative rate)

for the RZ and HRP assays detecting NMR actives, as well as

logistic plots for the same. The ROCs suggest that the clas-

sification of the HRP assay is much in line with the relative

integral from the NMR assay (Fig. 3a), and the logistic re-

gression graph (Fig. 3c) suggests that it flags a subset of

nearly all NMR-active compounds over a relative integral

threshold of approximately 12. The graphs for the RZ assay

(Fig. 3b, d) paint a different picture and suggest that the RZ

assay has a somewhat lesser ability than the HRP assay to

discern true NMR positives from false positives (ROC Fig. 3b),

apparently recognizing a subset of NMR actives across the

lower range of relative integral values (regression plot Fig. 2b).

The tables in Figure 1b suggest that the hit rate of the RZ

assay could be about 100 times as large as that of the HRP

assay (1 HRP active vs. 109 RZ actives that are not active in

NMR), but it should be recognized that the RZ assay detects

additional modes of action.

The risk of false-positive results is low since the assay is

based on direct observation of the TCEPO signal and there is

little opportunity for interference by other components in

the system. A false-positive risk could arise, however, from

redox-active impurities in the compound samples that are

not visible in the NMR spectrum, or from an overlap be-

tween compound and TCEPO multiplet signal, leading to

overestimation of the integral. The presence of overlapping

compound signals can easily be assessed by recording ref-

erence spectra of the compound of interest in the absence of

TCEP.

The assay specifically detects redox-active compounds

acting through the RCC mechanism since TCEP oxidation in

the absence of a protein must proceed by production of hy-

drogen peroxide. The hydrogen peroxide calibration curve in

Figure 2c shows that addition of up to 3 mM hydrogen per-

oxide to a sample results in a relative TCEPO integral, which is

up to 130-fold larger than that in a sample containing no

hydrogen peroxide. Further addition of hydrogen peroxide

does not increase the TCEPO integral, indicating that this is the

maximum relative integral, corresponding to full conversion

of the 3 mM TCEP that was originally present in the sample.

The largest value for the relative TCEPO integral observed for a

compound in our experiments was 24, which suggests that

What we are really interested in is how the HRP and RZ

assays can be used as high-throughput proxy to find those

compounds that the NMR assay will flag as active. Using the

HRP and RZ activity (% readout) at 100 mM as indication of

activity, we construct the ROCs shown in Figure 3e and f, and

logistic plots in Figure 3g and h. The ROCs suggest that neither

assay is very predictive of compound activity in the NMR

assay, and in particular for the RZ assay, false-positive ac-

tivity affects the capability of ranking NMR-active com-

pounds using the %-readout at 100 mM. In the case of the RZ

assay, this results in a high hit rate. In summary, the HRP and

RZ assay seem capable of picking up approximately half of the

compounds that are active in the NMR assay, and the HRP

assay appears to do this with the lowest incidence of false

positives. Further work suggested that the hits picked up by

the HRP and RZ assays, but not by the NMR assay, in part,

arise from redox-mediated effects, as removal of reducing

agent removes activity for some (results not shown). Neither

the RZ nor HRP assay is suited for identifying redox cyclers

exclusively.

1

00 mM of added compound was able to generate around

00 mM hydrogen peroxide within the timeframe from sample

5

mixing to the end of data acquisition.

Triaging of RZ and HRP Hits by NMR

A combined set of 3,266 compounds tested in RZ and HRP

assays was tested in the NMR assay. Compounds were re-

garded as NMR redox active if they increased the amount of

oxidized TCEP by at least 10% compared to the control (a

relative integral fold difference of 1.1 or greater). Both the RZ

and HRP assay were compared to the NMR redox assay

(

Fig. 1b).

The hit rates in the NMR, HRP, and RZ assays within the

Use of NMR Data in the MALT1 Hit Triaging Cascade

The NMR assay was trialed in a hit triaging cascade for a

cysteine protease (MALT1). Before the start of the HTS cam-

paign, NMR was used to evaluate MALT1 inhibitors observed

in a 10,000-compound validation set. Thirty two of the hits

resulting from this run were soluble enough to be tested in

the NMR assay and 13 proved to be redox active. The large

set of 3,266 amount to 3%, 1% and 5% respectively. Sur-

prisingly, neither the RZ nor the HRP assay flags all the NMR

actives—they flag 40 and 50 of 88 NMR-active compounds,

respectively. Using the NMR relative integral signal as an

indication of the degree of true redox-cycling behavior, we

can assess whether this signal predicts the classification

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 7

Fig. 3. Prediction of HRP and RZ assay classification by NMR signal. (a, b) Receiver-operator curves for relative NMR integral predicting the HRP

and RZ activity flags; red line indicates random response. (c, d) Logistic regression plots for same. (e, f) Receiver-operator curves for %effect at

100 mM compound concentration predicting NMR activity, for HRP and RZ (maximum effect at -100%); (g, h) logistic regression plots for same.

8

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

proportion of redox actives among the hits from the validation

set led to a change in the HTS strategy; a high-throughput RZ

assay was included in the screening cascade. This work was

performed before the large-scale comparison of results be-

tween the HRP and RZ assays. As such, at the time, the RZ

assay was deemed the most suitable high-throughput ap-

proach for hit triage in that project. Of the MALT1 hits arising

from the screen, 122 were measured in the NMR redox assay.

Nine out of these 122 hits showed redox activity, that is, the

proportion of false positives had decreased from 40% in the

validation set to 7% in the final HTS output, which is, in part,

explained by a relatively high content of ‘‘bad’’ compounds in

the validation set. This demonstrates the value of screening

validation sets for flagging potential problems, as well as the

benefit of including redox artifact assays in the screening

cascade.

HRP actives, confirming that both assays identify similar

numbers of compounds. The corresponding incidence of re-

dox activity in the set of confirmed actives are 20% (RZ) and

11% (HRP). This makes sense as the confirmation of the target

activity seen in the HTS for these compounds suggests they are

better behaved and more likely to be true actives. Again, the

hit rate of the RZ assay is shown to be higher compared with

the HRP assay. This may be expected as the RZ assay will pick

up multiple modes of action, whereas the HRP assay specifi-

cally detects formation of H O .

2

The remaining question is then around the nature of the

compounds that are flagged as active by the HRP and RZ

assays. Using our frequent-hitter descriptor, we assign all

compounds as well or badly behaved. Latter compounds have

shown activity in many HTS assays, and not necessarily as a

result of redox activity. For the set of unconfirmed com-

pounds, we see an enrichment of bad compounds in the RZ or

HRP actives subset (*35% of the redox actives is classified as

bad, vs. *10% of the compounds that are not redox active,

and this is similar across HRP and RZ data). This suggests that

both the RZ and HRP assays are sensitive to known frequent

hitters. The fraction gives an indication of the false-positive

rate being potentially as bad as 1 in 3 in this case, although it is

interesting to note that this is similar across either HRP or RZ

assay. Looking at the contingent of frequent-hitting com-

pounds within this set, both assays can be seen to designate a

large fraction of these as ‘‘redox-active’’ (HRP, 73%; RZ, 88%).

This suggests a sensitivity of either assay to anomalous

binders, as not all of these compounds will be redox-active.

Using Historical Data to Assess Anomalous Behavior

We derived a general frequent-hitter score (pBSF) to ana-

lyze hits for anomalous behavior other than redox. In short,

the descriptor captures frequent-hitting behavior based on

historical, in-house HTS data by comparing a compound’s

activity across many HTS campaigns to the expected activity

1

9

for an ‘‘average’’ compound. Details are described elsewhere.

The score enables identification of compounds that may affect

the assays through other modes of interference.

Further analysis of the HRP and RZ results suggests that a

strong signal (% effect at 100 mM) in either assay is indicative

of a compound showing frequent-hitter behavior across many

HTS assays. Forty-six percent of HRP actives, 21% of RZ ac-

tives, and 50% of NMR actives were frequent hitters according

to their generic pBSF scores. The high overall hit rate seen in

the RZ assay combined with relatively low frequent-hitter

incidence again suggests that this assay may have a high

false-positive rate.

Analysis of Historical Results

for Redox-Susceptible Targets

To focus the historical data analysis further on modes of

interference that are relevant for redox-sensitive proteins, we

performed a data-mining analysis for a subset of targets from

the following classes: cysteine proteases, metalloproteases,

and protein tyrosine phosphatases. A set of 1,097 compounds

found to be frequent hitters in redox-susceptible assays, but

not in background assays was selected. The fragmentation and

fragment enrichment analysis run on this set of likely redox-

active compounds led to the selection of 42 representative

enriched fragments. We tested for an increased incidence of

frequent hitters for the groups of compounds associated to

each fragment (Table 1; fragments in Fig. 4; see also Sup-

plementary Fig. S1). A ‘‘Y’’ at the intersection of the row

corresponding to group i and the column corresponding to the

set of assays j indicate that an increased proportion of com-

pounds from group i were found to be frequent hitters in set j.

‘‘N’’ indicates that no significant increase was detected.

Comparison of Redox Activity in Subsets

of Actives and Inactives

To compare the false-positive behavior of the HRP and RZ

assays in more detail, we compared incidence of redox activity

across a set of compounds confirmed active in the primary

assay, and a set of compounds that showed no activity in the

primary assay. Data in this case consisted of dose–response

data, and were collated retrospectively across multiple assays

following primary screening. Counts for these subsets can be

found in the Supplementary Data. In the set of unconfirmed

active compounds (i.e., compounds more likely to be false

positives), the incidence of RZ actives is 63%, and 51% for

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 9

Table 1. Results of the Binomial Tests Assessing Increased Proportion of Frequent Hitters in Groups of Compounds

Representing Each Fragment

Fragment

group

Number of

compounds

Redox-susceptible assays

Background assays

CP

MP

PTP

All RS

FI

FRET

Spec

Tech

Diverse

Comments

1

2

3

4

5

6

7

8

9

1772

7252

151385

6769

160085

13665

9706

17548

47

N

Y

N

Y

MP only

Y

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

MP only

Ambiguous

Not Redox

Ambiguous

Not Redox

Ambiguous

Redox

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

N

Y

1

2

3

0

1

2

3

4

5

6

7

8

9

0

1

2

3

4

5

6

7

8

9

0

1

259

86

N

Y

Not Redox

Redox

5844

141

35

N

Y

Not Redox

MP only

N

Y

214

6

N

Y

Redox

169

39

Y

Ambiguous

Redox

Y

N

Y

N

Y

4

N

Y

N

Y

Not Redox

MP only

2504

425

804

275

979

230

218

183

165

2

Y

N

Redox

N

Not Redox

2

6

(continued)

10

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

Table 1. Continued

Fragment

group

Number of

Redox-susceptible assays

Background assays

compounds

CP

MP

PTP

All RS

FI

FRET

Spec

Tech

Diverse

Comments

Redox

3

4

2

3

4

5

6

7

8

9

0

1

2

11

3

Y

N

Y

N

Y

N

Y

N

Y

N

Redox

2

Y

Redox

10053

1553

38

N

Not Redox

N

Y

1971

21

N

419

46

5

The table reports ‘‘Y’’ when the test is positive (increased proportion of frequent hitters) and ‘‘N’’ when the test is negative.

CP, cysteine proteases; MP, metalloproteases; PTP, protein tyrosine phosphatases; All RS, all redox-susceptible targets; FI, fluorescence intensity assays; FRET, F o¨ rster

Resonance Energy Transfer; Spec., spectrophotometry; Tech., technology.

Surprisingly, most fragments whose corresponding group

of compounds exhibited an increased incidence of frequent

hitters for cysteine proteases did not show an increased inci-

dence of frequent hitters for metalloproteases. A study of the

assay protocols revealed that only one out of the nine me-

talloproteases assays used a reducing agent, while eight out of

the nine cysteine proteases assays and all three of the protein

tyrosine phosphatases assays did. In the absence of a reducing

agent, RCCs are not expected to inhibit the redox-susceptible

targets. Thus, the analysis of metalloproteases can be regarded

as a negative control and it highlights that inclusion of a

reducing agent in the assay is required to see redox-cycling

frequent-hitter behavior.

34). These fragments are thus likely to convey redox-cycling

properties and their compounds were kept for further exper-

imental testing. The four remaining groups showed an am-

biguous profile. They exhibited an increased incidence of

frequent hitters both for sets of redox-susceptible assays and

for background sets of assays, suggesting that these could act

through redox cycling or another mechanism. Their corre-

sponding compounds were also kept for experimental testing.

It is important to emphasize that the presence of such frag-

ments alone is not a sufficient condition to make a compound

an RCC, but our analysis indicates that such compounds are

more likely to display redox-cycling behavior.

For each of the 8 + 4 fragment groups selected for further

experimental testing, a few representative compounds were

tested in the HRP, RZ, and NMR redox assays. Selection took

into account three criteria: pBSF value for all susceptible

targets, diversity of structures, and purity. We prioritized

compounds with sample purity over 80%, with diverse struc-

tures, and with a high pBSF in the set of redox-susceptible

assays. All were tested in HRP and RZ assays, and a subset of

compounds was tested in the NMR redox assay. Results can be

found in Table 2.

Four groups showed an increased incidence of frequent

hitters for metalloproteases, but not for other redox-susceptible

assays. Given that eight of the nine metalloprotease assays did

not contain a reducing agent, the corresponding fragments

were not considered likely to give rise to redox-cycling prop-

erties. Twenty-six groups did not show a statistically significant

increase in the incidence of frequent hitters for any set of

redox-susceptible assays.

Eight groups exhibited an increased proportion of frequent

hitters for redox-susceptible targets, but not for any back-

ground set of assays (fragments 9, 11, 16, 18, 28, 32, 33, and

The results from the in silico analysis are mostly confirmed

experimentally. For five out of eight compound groups selected

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 11

Fig. 4. Manually selected fragments representative of the enriched fragments in S. Numbers correspond to those in Table 1. Redox

fragments are those assigned as redox in Table 1. See Supplementary Data section ‘‘Fragment selection’’ and Figure S1 for further details.

12

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

Table 2. Experimental Results in the RZ, HRP, and NMR Redox Assays for the Fragments Identified Using Our In Silico Analysis

Number of compounds tested in

Active compounds in

RZ HRP

NMR

At least At least All

In silico results

Fragment

RZ and HRP

NMR

A

wA mA sA one assay

two

assays

‘

‘Redox’’: Increased incidence of frequent

9

10

5

8

10

3

0

1

0

3

2

1

0

1

0

4

2

1

0

3

1

0

2

8

5

1

0

5

0

2

3

0

1

8

7

0

2

1

0

5

0

hitters in redox susceptible but not

in background

1

2

3

1

6

8

2

3

4

3

5

8

7

1

3

0

3

5

3

10

6

8

10

0

10

8

10

4

10

1

9

4

2

3

2

3

1

‘‘Ambiguous’’ Increased incidence of

10

71

7

3

0

6

3

frequent hitters in redox susceptible

and in background

6

0

4

0

8

1

0

3

1

7

6

7

6

a

‘‘Redox-clean’’

—

71

2

0

2

0

a

Redox-clean compounds refer to compounds tested in all twelve redox-susceptible HTS assays employing reducing agent and inactive in all of them.

A, active; wA, weakly active; mA, medium active; sA, strongly active (activity categories are defined in the Results and Discussion for each assay).

to represent fragments that are suspected to cause redox effects

by in-silico methods, at least one compound was active in all

three redox assays. The ‘‘ambiguous’’ category is still ambigu-

ous experimentally and suggests disagreement between the

NMR, HRP, and RZ assay results. One fragment showed strong

activity in all three assays, while the others only show activity

in the RZ or NMR redox assays.

The remaining groups showed no HRP activity. At least 30%

of compounds from groups 9, 11, 3, 5, and 8 were not active in

any assay. Where possible, compounds selected for experi-

mental testing had high pBSF for redox-susceptible targets.

For groups with a lot of representative compounds such as

groups 3, 5, and 8, this means that compounds with high pBSF

have been chosen. The absence of activity in all redox assays

thus suggests alternative frequent-hitter mechanisms for

compounds harboring these motifs. For groups 3, 5, and 8, a

few compounds may be redox cycling (as confirmed by the

redox assays), but different frequent-hitter mechanisms may

feature here, as suggested by the in silico analysis. For groups

9 and 11, the activity rates in the NMR redox assays are low,

but not zero, and this assay did not seem prone to give false

negatives. We conclude that each of these fragment sets

contain RCCs that are not detected by the HRP assay. It is likely

that these groups have very weak redox-cycling properties

only detectable in the NMR assay. The absence of RZ-active

compounds for cluster nine suggests that the RZ assay is prone

to give false-negative results for this category.

The ‘‘redox-clean’’ compounds included as control did not

show any activity in redox-susceptible HTS assays. Any positive

result in the redox assays can therefore be considered a false

positive. False positives do not appear to be a problem in the HRP

and NMR assays since these assays do not detect any activity in

this set of compounds. The RZ assay flags two compounds from

this set as active, which may be compounds containing impurities

or compounds that are unstable and reactive.

The HRP assay only shows activity for groups 16, 18, 28, 32,

3

3, 34, and 17. Except for group 32, all these groups exhibit

high activity rates and show activity on at least one occasion

in all three redox assays, and these observations suggest

strong redox-cycling properties for these compounds.

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 13

TARNOWSKI ET AL.

The results confirm that the HRP assay is susceptible to false

negatives when the redox-cycling activity is weak, but that it

rarely gives false positives. The NMR and RZ assays both de-

tect more RCCs than the HRP assay, and they both successfully

identify compounds from clusters identified as redox active in

silico, where the HRP assay fails. However, results for the RCCs

still suggest substantial false-negative rates for the RZ assay.

In summary, we have analyzed the capacity of RZ, HRP,

and a novel NMR-based redox assay to flag suspicious com-

pounds. The former two biochemical assays are considered high

throughput, whereas the NMR assay is medium throughput.

Overall hit rates for these assays within a set of HTS actives are

different, 1% for the HRP assay, 3% for the NMR assay, and 5%

for the RZ assay. Surprisingly, neither the RZ nor the HRP assay

flags all of the NMR actives—they identify 40 and 50 of 88

NMR-active compounds, respectively, and may be related to

sensitivity differences arising from differently optimized assay

readouts.

mechanisms other than redox, and that compounds flagged in

the RZ and HRP assays may be undesirable still for reasons

other than redox-cycling behavior. Closer inspection of a set of

‘‘compounds-of-interest’’ (confirmed-active and confirmed-

inactive compounds for the primary target) yields numbers in

the same range, but also suggests that both HRP and RZ assays

behave similarly in terms of flagging frequent hitters. We

conclude that the NMR redox assay offers a novel and reliable

way of identifying RCCs at a medium throughput, with the HRP

assay being a straightforward higher-throughput option with

reasonable recognition and low false-positive rate. The RZ as-

say has a similar ability to recognize compounds active in the

NMR assay as the HRP assay, but its higher overall hit rate

necessitates further deconvolution of output.

DISCLOSURE STATEMENT

All authors were employees of AstraZeneca when working

on this project.

Computational analysis enabled assessment of the abil-

ity of each assay to identify redox-cycling frequent-hitter

compounds (RCCs), and the results suggest that all three

assays are capable of identifying compounds that frequently

hit RCC-susceptible targets in HT screens. This suggests that

even if the primary reason for activity is not pure redox-

cycling behavior, the flagged compounds are generally likely

to misbehave and of lesser interest. Historical HTS data

proved to be a valuable source of information with regard to

the detection of frequent-hitter compounds acting through a

specific mechanism. In addition to identifying previously

identified problematic RCCs, the computational analysis run

on these data generalizes the results and reveals redox-

cycling fragments that identify compound sets enriched in

redox-active behavior. Hits from redox-susceptible screens

that contain such fragments can thus be flagged as suspi-

cious, regardless, when historical data for these compounds

are limited.

REFERENCES

1

2

.

Drews J: Drug discovery: a historical perspective. Science 2000;287:1960–1964.

Wigglesworth MJ, Murray DC, Blackett CJ, Kossenjans M, Nissink JW: Increasing

the delivery of next generation therapeutics from high throughput screening

libraries. Curr Opin Chem Biol 2015;26:104–110.

3. Thorne N, Auld DS, Inglese J: Apparent activity in high-throughput screening:

origins of compound-dependent assay interference. Curr Opin Chem Biol 2010;

14:315–324.

4

5

.

Feng BY, Simeonov A, Jadhav A, et al.: A high-throughput screen for

aggregation-based inhibition in a large compound library. J Med Chem 2007;

50:2385–2390.

Roche O, Schneider P, Zuegge J, et al.: Development of a virtual screening

method for identification of ‘‘frequent hitters’’ in compound libraries. J Med

Chem 2002;45:137–142.

6

.

Baell J, Walters MA: Chemistry: chemical con artists foil drug discovery. Nature

2014;513:481–483.

7

Soares KM, Blackmon N, Shun TY, et al.: Profiling the NIH Small Molecule

Repository for compounds that generate H2O2 by redox cycling in reducing

environments. Assay Drug Dev Technol 2010;8:152–174.

8.

Bova MP, Mattson MN, Vasile S, et al.: The oxidative mechanism of action of

ortho-quinone inhibitors of protein-tyrosine phosphatase alpha is mediated by

hydrogen peroxide. Arch Biochem Biophys 2004;429:30–41.

Using observed behavior in historical HTS data of redox-

sensitive and non-redox-sensitive target assays, we designated

sets of ‘‘bad’’ and ‘‘suspicious’’ RCCs, as well as ‘‘clean’’ com-

pounds. The HRP, NMR, and RZ assays identified 76%, 84%,

and 80% of ‘‘bad’’ RCC frequent hitters, respectively, and 54%,

9. Brisson M, Nguyen T, Wipf P, et al. Redox regulation of Cdc25B by cell-active

quinolinediones. Mol Pharmacol 2005;68:1810–1820.

1

0. Smith GK, Barrett DG, Blackburn K, et al.: Expression, preparation, and high-

throughput screening of caspase-8: discovery of redox-based and steroid diacid

inhibition. Arch Biochem Biophys 2002;399:195–205.

6

8% and 73% of ‘‘suspicious’’ RCC frequent hitters, respec-

1

1. Lor LA, Schneck J, McNulty DE, et al.: A simple assay for detection of small-

molecule redox activity. J Biomol Screen 2007;12:881–890.

2. Lal M, Rao R, Fang XW, Schuchmann HP, von Sonntag C: Radical-induced

oxidation of dithiothreitol in acidic oxygenated aqueous solution: a chain

reaction. J Am Chem Soc 1997;119:5735–5739.

3. Johnston PA, Soares KM, Shinde SN, et al.: Development of a 384-well

colorimetric assay to quantify hydrogen peroxide generated by the redox

cycling of compounds in the presence of reducing agents. Assay Drug Dev

Technol 2008;6:505–518.

tively. They also identified 11%, 20%, and 17% of the ‘‘clean’’

set, respectively, which may be false-positive results, or due to

these compounds not having been tested enough times to be

identified as problematic by their data yet.

1

Analysis of anomalous-binder behavior of compounds in a

wider set of HTS assays suggests that 46%, 50%, and 21% of

HRP, NMR, and RZ actives are frequent hitters that can act by

14

ASSAY and Drug Development Technologies APRIL 2018

ª MARY ANN LIEBERT, INC.

UTILITY OF REDOX ASSAYS

1

4. Sameshima T, Miyahisa I, Yamasaki S, Gotou M, Kobayashi T, Sakamoto J: High-

throughput quantitative intrinsic thiol reactivity evaluation using a fluorescence-

based competitive endpoint assay. SLAS Discov 2017:2472555217704654.

5. Wigle TJ, Swinger KK, Campbell JE, et al.: A high-throughput mass spectrometry

assay coupled with redox activity testing reduces artifacts and false positives in

lysine demethylase screening. J Biomol Screen 2015;20:810–820.

6. Krezel A, Latajka R, Bujacz GD, Bal W: Coordination properties of tris(2-

carboxyethyl)phosphine, a newly introduced thiol reductant, and its oxide. Inorg

Chem 2003;42:1994–2003.

Address correspondence to:

J. Willem M. Nissink, PhD

Oncology IMED

AstraZeneca

Darwin Building

3

10 Cambridge Science Park

Milton Road

7. Hwang TL, Shaka AJ: Water suppression that works—excitation sculpting using

arbitrary wave-forms and pulsed-field gradients. J Magn Reson Ser A 1995;112:

Cambridge CB4 0WG

United Kingdom

2

75–279.

1

8. Dalvit C: Efficient multiple-solvent suppression for the study of the interactions

of organic solvents with biomolecules. J Biomol Nmr 1998;11:437–444.

9. Nissink JWM, Blackburn S: Quantification of frequent-hitter behavior based

on historical high-throughput screening data. Future Med Chem 2014;6:

E-mail: willem.nissink@astrazeneca.com

1

Abbreviations Used

FI ¼ fluorescence intensity

1

113–1126.

2

0. Berman HM, Westbrook J, Feng Z, et al.: The protein data bank. Nucleic Acids

FRET ¼ F o¨ rster resonance energy transfer

HRP ¼ horseradish peroxidase

HTS ¼ high-throughput screening

RCC ¼ redox-cycling compound

ROC ¼ receiver-operator curves

RZ ¼ resazurin

Res 2000;28:235–242.

1. Hollander M, Wolfe, DA, Chicken, E: Nonparametric Statistical Methods. John

Wiley & Sons: New York, 1973, pp. 15–22.

2. Holm S: A simple sequentially rejective multiple test procedure. Scandinavian

journal of statistics 1979;6:65–70.

3. Tan Z, Ihnat PM, Nayak VS, Russell RJ: Quantitative analysis of tris(2-

carboxyethyl)phosphine by anion-exchange chromatography and evaporative

light-scattering detection. J Pharm Biomed Anal 2012;59:167–172.

TCEP ¼ tris(2-carboxyethyl)phosphine

TCEPO ¼ P-oxide of TCEP

ª MARY ANN LIEBERT, INC.

ꢀ

VOL. 16 NO. 3

ꢀ

APRIL 2018 ASSAY and Drug Development Technologies 15

Article Doi

Utility of Resazurin, Horseradish Peroxidase, and NMR Assays to Identify Redox-Related False-Positive Behavior in High-Throughput Screens

DOI: 10.1089/adt.2017.838

Source and publish data:

Authors:

Article abstract of DOI:10.1089/adt.2017.838

Full text of DOI:10.1089/adt.2017.838

Products guided by the article

R&D Labs maybe for 5962-40-3

Relevant to this article

Hot Product