Key words
assay validation - screening - antibacterial drug discovery - ChemGPS-NP - natural
products - impedance
Abbreviations
ARS:
aminoacyl-t-RNA synthetase
CV:
coefficient of variation
DLS:
dynamic light scattering
ECIS:
electric cell-substrate impedance sensing
EthD-1:
ethidium homodimer-1
Fab:
fatty acid biosynthesis
FP:
fluorescence polarization
FRET:
fluorescence resonance energy transfer
GFP:
green fluorescent protein
HCS:
high-content screening
HTS:
high-throughput screening
LUO:
laboratory unit operation
MTS:
medium-throughput screening
MTT:
3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide
NP:
natural products
PC:
photonic crystal
S/B:
signal-to-background ratio
S/N:
signal-to-noise ratio
SPR:
surface plasmon resonance
SW:
signal window
T3SS:
type 3 secretion system
TEM:
transmission electron microscopy
uHTS:
ultrahigh-throughput screening
Z′:
screening window coefficient
Introduction
Biomolecular screening involves the exploration of libraries of pure compounds or
extracts for their effects on relevant targets, either actual biomolecules or biological
events, which leads to the identification of actives (“hits”) that can eventually
be further developed for preclinical testing (“leads”) or used as chemical tools.
Based on the throughput, different terms have been coined (see below), but these definitions
have been relative and the use has varied among authors. HTS generally implies that
between 10 000 and 100 000 compounds (or samples) per day are screened, while MTS
refers to campaigns involving around 1000 to 10 000 compounds per day [1]. Screening a lower amount of a sample is then regarded as “low-throughput screening”,
while the other extreme, uHTS, implies that more than 100 000 data points are generated
per day [1], [2]. Regardless of the throughput, three basic and interconnected elements have been
reasonably argued as critical for the success of screening campaigns: 1) targets,
2) screening methods, and 3) chemical libraries.
The selection of a single target has typically been driven by the disease relevance
and the chemical tractability. While the one-target approach has allowed for the exploration
of very large compound collections in a fast and cost-effective manner, it has also
been a reductionist by ignoring the biochemical networks inherently existing in biological
systems. Because of that, recent views have shifted the target selection process towards
finding a set of multiple targets associated with the desired clinical effects [3]. That relates to the second aspect – the selection of screening methods. The more
general choice has been between biochemical and cell-based assays. With the resurgence
of polypharmacology and multiple-target approaches, cell-based methods, in particular
phenotypic and pathway assays, have become more prevalent when performing screenings.
For the first screens, also known as “primary” screens, using either biochemical or
cell-based assays, compounds are tested only once and the follow-up strategies are
decided based upon these results. This implies that the assay has to be shown to perform
robustly. In general, any level of automation of in vitro assays increases throughput, decreases human errors and need for labor, and improves
safety issues. The third critical element is the selection of compounds to be screened.
Very large synthetic chemical collections of millions of compounds have now been made
available, but the positive impact of using natural compound collections is undeniable,
especially in certain therapeutic areas.
Natural compounds have provided milestone drugs, i.e., taxol, statins, and cyclosporine,
with very successful roles in modern pharmacotherapy. For instance, antibacterial
drugs of natural origin (natural compounds or their semisynthetic modifications) account
for more than 65 % of all antibacterial drugs approved over the last 30 years [4]. Another example is provided by anticancer drugs, in which a staggering 50 % of
all existing medications are natural products or molecules derived therefrom [5]. Besides their roles as lead compounds, natural products provide starting scaffolds
for structural optimization via combinatorial chemistry or serve as chemical tools
for the validation of new molecular targets [6]. They are also a more successful choice when screening against protein-protein targets,
because they tend to be larger and, in many cases, more complex than synthetic libraries
[7].
When aimed specifically at the exploration of natural products and the development
of screening assays that encounter specific challenges. The screening of natural origin
samples can involve crude extracts (10–1000 or more compounds), semi-purified mixtures
(5–10 compounds), or structure-elucidated purified single compounds. This structural
complexity of the extracts has historically been the most daunting factor influencing
the work with natural products. From the assay perspective, one challenge is that
screening methods are run under experimental conditions that jeopardize the proper
identification of actives within the complex natural mixtures, thus rendering more
false negatives, as some active components can be present at very low concentrations
[8]. Another problem is that several natural compounds are known to aggregate in biochemical
buffers, causing nonspecific inhibition of unrelated enzymes. For instance, unexpected
behavior of natural retinoids in a filter-binding assay to bovine and reindeer β-lactoglobulin was shown to be explained by detecting aggregate formation [9]. They can also cause interferences with optical detection methods, as they can display
intrinsic fluorescence or cause light scattering, among other things.
Therefore, in this review, several such methodological challenges will be discussed
and concrete approaches will be recommended to circumvent them. To exemplify how these
methodological challenges can be encountered in a relevant area for natural product
researchers, we selected the discovery of antibacterial drugs. This review does not
pretend to be comprehensive, but it does aim to serve as a first-hand tool for natural
product researchers to recognize problems during screening campaigns and adopt specific
resolution strategies.
Development and Validation of Screening Assays
Development and Validation of Screening Assays
The general workflow
The development and implementation of appropriated assays are indispensable steps
prior to the performance of chemical screens. These steps are performed within the
early drug discovery phase commonly refereed as “Target-to-Lead”. The general workflow
of the “Target-to-Lead” phase is schematically represented in [Fig. 1], while specific issues that need to be considered during the discovery of natural
origin drugs are summarized in [Table 1]. Most of the issues that will be discussed in this review are applicable across
all variants of performed screens (from high to low throughput). However, an intentional
emphasis has been put into considerations that are applicable to MTS as this corresponds
with the most frequent type of screens currently performed in academic environments
worldwide. Indeed, with the growing engagement of academic centers in the performance
of MTS (and, in many cases, also of HTS), the term “drug screening” has also become
less used and the more general “chemical screening” has been adopted instead, as this
is not biased towards a therapeutic goal and it better reflects the wider interests
of the academic community [10].
Fig. 1 Workflow describing the general steps of the HTS process (red boxes) within the overall
process of discovering new drug candidates. (Color figure available online only.)
Table 1 Key aspects of the HTS assay optimization, validation, and implementation process
specific to the early drug discovery of natural products.
|
|
General considerations
|
Some key issues NP
|
Discussed here
|
|
HTS assay development
|
Assay configuration
|
|
–
|
|
Assay conditions
|
|
–
|
|
Automation
|
|
Section 3.1
|
|
Miniaturization
|
|
Section 3.1
|
|
HTS statistical quality
|
|
Section 2.2
|
|
HTS assay implementation
|
Identification and elimination of interferences
|
-
Common interferences due to colloidal aggregation of natural compounds need to be
addressed.
-
Methods need to be chosen to exclude optically interfering natural compounds.
-
Chemically reactive NPs can cause false positives or negatives.
|
Section 3.2
|
|
Lead refinement
|
Selection in hit-to-lead process
|
|
–
|
The “Target-to-Lead” phase ([Fig. 1]) starts with the selection of a target, which is greatly influenced by the relevance
to the disease in question and the chemical tractability. At this stage, striking
a balance between the novelty and the validity of the chosen target is an important
goal since this will ultimately determine the type of drug that may be discovered
– a new type of drug or a drug similar to an existing one. Target novelty and validity
generally have an inverse relationship. Validated targets with known disease relevance
generally lack novelty, while on the other hand novel targets typically require further
validation studies (i.e., cell-based and animal trials) to ensure that screens provide
meaningful results and lead to the discovery of clinically valuable compounds [7]. Using drug discovery strategies with both types of targets have pros and cons,
and concrete examples of them will be given in section 4 of this review when discussing
the discovery of natural antibacterial compounds.
Upon selecting a target, an assay method needs to be chosen so that the activity of
the test compounds against the target is measured either in the absence or presence
of cells. The development process ideally aims at providing an assay that combines
simplicity with good sensitivity, reproducibility, and accuracy, as well as amenability to automation and reasonable price
[11], [12]. In terms of simplicity, the goal is to develop assays that do not have many steps
(not more than 5–10) and only necessitate simple operations, with a preference towards
assays that only involve additions, otherwise known as homogeneous. The sensitivity of the assay must be sufficient to permit the identification of low potency or low
efficacy compounds. The accuracy refers to the ability of the assay to provide trustable results as assessed via control
compounds with known pharmacological effects on the measured targets, while the reproducibility of the assay relates to the stability of the response that is offered by the assay,
to ensure that similarly trustable results are obtained when the assay is performed in different wells, plates, days,
and by different human operators or using different types of laboratory equipment
[13]. The reproducibility of the assay is also tightly connected to the assay reagents,
both chemical and biological. These need to be stable and no significant changes in
assay components should be detected, especially in cases of non-commercially available
reagents that may be prepared in-house in different batches, such as recombinant proteins
or cell suspensions.
Assuming that the target is chemically tractable, it is fairly safe to say that most
assays can be engineered and conditions can be optimized so that compounds modulating
the target are found in the tested collections. Examples of the conditions needed
to be optimized for both biochemical and cell-based assays are presented in [Fig. 2]. To follow the optimization process and test the effects that these different conditions
have on the quality of the assay, statistical analysis are performed, which will be
discussed in the next chapter.
Fig. 2 Conditions that typically need to be optimized during the HTS assay development process.
(Color figure available online only.)
When it comes to academic screening centers or groups, they have been particularly
successful in the development and adoption of a cell-based, phenotypic screening (recently
reviewed by [10]) as a strategy to cover diseases that have been neglected by pharmaceutical companies
and access a wider spectrum of targets [14]. The development of high-quality cell-based assays combined with the selection of
well-designed academic libraries (not necessarily larger ones) has translated into
the discovery of highly active molecules that maintain a very high activity in vivo as well (i.e., [15]).
Academic institutions have also contributed with innovative studies incorporating
HCS, not only during the primary screening phase but also for the determination of
the mode of actions of actives (i.e., [16]). The complex data processing, which was the major drawback for the use of HCS,
is now being overcome by the scientific community with the project Open Microscopy
Environment that has provided entirely open solutions, such as OMERO [17]. OMERO allows for the open analysis of complex multidimensional image data as well
as the storage and handling of scientific image repositories. Such tools have allowed
HCS data to be exchanged between multiple platforms and be free of dependency from
closed and often expensive commercial solutions [10]. Furthermore, another interesting feature of academic screening has been the implementation
of whole organism screens employing nematodes or zebra fish, which in some cases have
been applied in conjunction with HCS [18]. More examples of this type of screen will be further discussed in section 4.
Statistical analysis
Assay performance and assay sensitivity have been suggested by authorities in the
field as the two crucial aspects that should be monitored during the development of
screening assays [11], [19]. In general, an assay is thought to perform well if it provides an adequate distinction
between maximum and minimum signals, which, in other words, indicates that the assay
can effectively differentiate between the non-actives that largely populate chemical
collections and the less frequently existing actives. Also, a good reproducibility
needs to be demonstrated. For this purpose, certain statistical parameters characterizing
the assay performance are calculated. In large pharmaceutical organizations, they
are an essential part of the routine operations performed in HTS laboratories, and
a considerable amount of literature has been dedicated to them. However, in academic
groups, especially in those dedicated to the MTS of natural products, they are less
known and their application is consequently much less reported. In [Table 2], a summary of the relevant parameters needed to assess the performance of screening
assays is presented.
Table 2 Statistical parameters needed to characterize the performance of HTS assays.
|
Parameter
|
Equation*
|
Target value
|
Comments
|
Reference
|
|
* In all equations µmax, µmin and SDmax, SDmin refer to the average and standard deviations, respectively, of the maximum (max) and minimum (min) signals in the screening assay
|
|
Screening window coefficient
|
|
0.5–1.0
|
Takes into account the means as well as the dispersion of the assay signals. Z′ is
measured only with controls; Z is measured in the presence of library compounds.
|
[20]
|
|
Signal to noise
|
|
The higher, the better
|
Calculated using only control compounds; measures how robustly maximum and minimum
signals can be differentiated
|
[21]
|
|
Signal to background
|
|
> 2
|
Calculated using only control compounds: ratio of the maximum and minimum signal means.
If the CVs of the signals are under 10 %, the S/B can be lower.
|
[20]
|
|
Coefficient of variation of the signals
|
|
< 15 %
|
Indicates the variation of the signals and can be calculated to compare signals between
plates and days.
|
–
|
|
Coefficient of variation of the assay
|
|
< 20 %
|
The calculation can be applied when SDmin is lower or equal to SDmax.
|
[22]
|
|
Signal window
|
|
> 2
|
Indicates the significant signal between maximum and minimum signals, which is used
to characterize the dynamic signal ratio of the assay.
|
[20]
|
Since first being published in 1999 by Zhang et al. [20], Z′ has become one (if not the most) useful tool to estimate assay performance.
Based upon the assumption that the signals are binomially distributed (99 % of their
data points fall within the ± 3* standard deviation limits), Z′ estimates if a statistically
significant separation between the extreme assay responses (maximum and minimum) exists
in the assay. When the Z′ value is over 0.5 (or 0.4 in the case of cell-based assays),
the assay can be considered a good performing screening method. Conversely, Z′ is
close or equals to 0 when the extreme signals of the assays overlap, which is an indication
of a poor quality assay. As seen in [Table 2], the calculation is very simple, and it is independent of the assay format or the
technology being used. In fact, Z′ is a dimensionless parameter that can be used to
compare between runs performed in different conditions. For calculating Z′, we first
recommend performing normality tests on the control samples with any reliable statistical
software package to ensure that the prerequisite of a binomial distribution is fulfilled.
Then, a second step would focus on preparing and running assay control plates under
different assay conditions ([Fig. 2]), that is, plates containing only control samples (maximum and minimum signals),
to identify and establish the assay performance and variability across different wells,
plates, and days. At this stage, reagent stability should also be studied to exclude
other possible sources of errors during the assay implementation.
The same data generated during this initial optimization run permits the calculation
of S/B and S/N ratios. S/B as a stand-alone parameter has a limited value since it
does not take into account the signals variability. On the other hand, the S/N ratio,
when calculated based on the modified equation published by Bollini et al. [21], does provide a good assessment of both signal window and signal variability. In
any case, both parameters are nowadays typically used in conjunction with Z′.
The Z′ equation can also be rewritten to make the relationship with the S/B ratio
and the signal variations expressed as CVs more evident, as correctly indicated by
[7]:
According to this equation, when the variations of the signals measured by the CV
% are low, the S/B ratio can also be low and the quality of the assay will still be
kept high. Thus, although in principle it is indicated that the S/B ratio is recommended
to be higher than 2 ([Table 2]), an assay with low variability (for instance with a CV of 5 % for both signals)
will result in an assay with a Z′ of 0.55, even if the S/B is only 2. The measurement
of these parameters is crucial to identify the best experimental conditions, but also
once the assay is developed, they also allow for tracking any unexpected change in
performance during the assay implementation stage. Thus, during screening campaigns,
Z′, along with the other statistical parameters, should be calculated for every screened
plate. It has been demonstrated that although Z′ and SW measure the same properties
of the assay, that is, assay signal differences and variability, Z′ is still a better
choice [22]. In our opinion, a more systematic utilization of these parameters in natural product
screens will be a highly beneficial practice for the field.
Examples of Specific Problems Encountered During the Development and Implementation
of Screening Assays of Natural Products
Examples of Specific Problems Encountered During the Development and Implementation
of Screening Assays of Natural Products
Optimization of conditions related to the assay protocol
When optimizing and automating an assay it is essential to define key aspects of the
different assay protocols, i.e., the steps and processes that are to be performed.
This helps in identifying the potential bottlenecks prior to initiating the automation
trials, as well as other potential limiting steps [23], [24].
Fully automated systems are capable of performing full assay procedures unattended,
from the test compound management through the sample preparation and sample analysis
to the data processing. Usually, they utilize a centralized robotic arm, which integrates
compound libraries, several liquid handling stations, analytical devices, plate incubators,
and stackers to the system. These systems are best suited for high compound numbers
assayed with fairly simple protocols. On the other hand, semiautomated systems comprising
a single liquid handling workstation that may be used in combination with a liquid
dispensing unit are more flexible for manual interruptions and reprogramming, and
therefore are usually used in executing sections of more challenging assay protocols.
An example of a semiautomated protocol in comparison with a manual protocol is presented
in [Fig. 3] and further detailed in [Fig. 4]. These semiautomated approaches are generally more accessible to academic researchers
as they may not require massive infrastructure investments and are therefore less
costly to implement.
Fig. 3 General view of a typical protocol flow, exemplified with a manual vs. semiautomated
comparison of an antimicrobial assay.
Fig. 4 Detailed view of the protocol flow (presented earlier in [Fig. 3]) of a manual (A) vs. semiautomated (B) antimicrobial assay, highlighting the concepts of the steps and processes. The processes
are only indicated in the semiautomated assay, within thicker boxes, while steps are
shown within thinner boxes. (Color figure available online only.)
Assay protocols (in semi- or fully automated systems) taking into account LUO thinking
helps to provide an understanding of the relationship between the engineering theory
and performance of actual experimental laboratory operations [25]. A specific sequence of unit operation is called a “process” ([Fig. 4]) and may include one or several individual operations referred to as steps (e.g.,
the addition of cell suspension, removal of culture medium, incubation, washing or
staining steps). This practice improves the project planning, quality, and integrity
of assay protocols, and assures valid interpretation of results from data analysis.
Incubation steps do not typically change between manual or automated assays, and generally
the only precaution would be to ensure that there is enough space in the incubators
to accommodate the larger amount of plates that may be generated in an automated assay.
On the contrary, the liquid handling is probably the most demanding aspect that needs
to be addressed, as it highly affects the quality of the results obtained, and therefore
the liquid handling parameters have to be adjusted to acquire acceptable precision
without interfering with the assay system. Some examples are presented in [Table 3].
Table 3 Some key liquid handling parameters to be taken into account when performing an assay
and troubleshooting strategies, exemplified with the semiautomated assay detailed
in [Fig. 4 B] (the numbering of processes and steps refers also to the LUO shown in [Fig. 4 B]).
|
Process
|
Step
|
Potential problem
|
Critical parameter*
|
Solution
|
|
* Parameter clarification: SH – shaking, TR – tip refreshing, DS – dispensing speed,
AS – dispensing speed, TH – tip height
|
|
1
|
1
|
Suspension uniformity
|
SH
|
|
|
Contamination
|
TR
|
|
|
Mechanical stress on cells
|
DS
|
|
|
2
|
4
|
Cellular detachment during media removal
|
AS, TH
|
|
|
5
|
Cross-contamination within the plate
|
TR
|
|
|
6
|
Stains outside the reaction wells
|
TH
|
-
Dispensing height has to be carefully optimized.
-
Tip refreshing steps may be added.
-
Potentially overstained wells are tracked.
|
Critical steps concern general tolerance of the targets to mechanical stress caused
by pipetting, the dispense speed of all components (e.g., too high of a dispense speed
could cause the fluid forces to interfere with fixation of cells or proteins), removal
(i.e., aspiration) of the supernatant after fixation, and contamination hazards. Contaminations
should be carefully detected and minimized by choosing the most appropriate tips and
liquid handling settings for each reagent [26]. On the other hand, the mixing efficiency is dependent on the correct combination
of liquid volume and method used. When working with cells, it must be remembered that
stirring too vigorously may cause disruption and impair functionality. Evaporation
of liquids may become a problem during prolonged incubations (if the plates cannot
be sealed). Cellular studies are normally run at + 37 °C, which promotes evaporation.
The assay volume determines the plate format and the choice of liquid handling device
[27].
By automating the assays, the number of variables that can be controlled increases
significantly to detail levels that are not possible to achieve by even the most skilled
researchers. Specific variables, such as the distance from the bottom of the well
at which the pipette tip is to be positioned in every dispensed step, the dispensing
or aspiration speed, or the blowout technique, would be impossible to control manually
[27]. Fine-tuning and controlling these variables may in some cases not necessarily translate
into an assay that performs better, but it does generally result into an assay with
better reproducibility from plate-to-plate and day-to-day, as exemplified in [28].
Detection and elimination of interferences during HTS implementation
Interferences due to colloidal aggregation: Colloidal aggregation occurs via self-association of organic molecules in aqueous
buffer solutions and around 95 % of hits in screening campaigns have been attributed
to unspecific, aggregation-based inhibitors [29]. The widespread recognition of colloidal aggregation as a main cause for the occurrence
of false positives is greatly attributed to the work of Shoichet and coworkers (see
references below [29], [30], [31], [32], [33], [34]). They established that many nonspecific inhibitors or “heavy hitters” self-associate
in biochemical buffers, forming spherical particles of various diameters (300–1000 nm)
that are detectable by DLS and TEM.
Criteria for compounds to be deemed aggregators have been proposed to include time-dependence
inhibition of targets, quick inhibition reversal (within seconds) upon the addition
of detergents, inhibition being strongly dependent on experimental parameters (e.g.,
pH, enzyme concentration, protein concentration, and ionic strength), and the occurrence
of steep concentration-response relationships [30], [31]. Because of this, increased Hill coefficients are generally thought to be reliable
predictors of aggregation-based inhibition [29]. Aggregators have been postulated to directly interact with target proteins causing
partial protein denaturation [32] or to sequester proteins leading to a reduced accessibility of the substrate [33]. However, the exact molecular mechanisms taking place are still being investigated.
At a typical screening concentration of 5 µM, about 1–2 % of compound libraries with
”drug-like” properties have been estimated to behave as aggregators, and at 30 µM
that percentage has been shown to increase to 19 % [34]. In both scenarios, the prevalence of aggregators is relevant when considering the
typical hit ratios (< 1 %) of screening campaigns.
Natural products were within the aggregate formers that were first reported by Shoichetʼs
laboratory around 10 years ago. McGovern and Shoichet [31] analyzed 15 nonspecific kinase inhibitors and showed that eight of them were aggregate
formers, from which three molecules were of natural origin ([Fig. 5]). They were the very well-known flavonoid quercetin, indirubin, which is present
in Indigofera tinctoria and claimed to be an antitumor [35], [36], and rottlerin, another phenolic compound naturally existing in Mallotus phillipinensis (“Kamala” tree) and reportedly active as an opener of potassium channels (BKCa++)
[37]. Subsequent investigations confirmed these results and established the promiscuous
enzymatic inhibition profiles of these natural compounds [38], [39].
Fig. 5 Three of the first reported aggregated-induced promiscuous inhibitors of natural
origin. A quercetin, B indirubin, and C rottlerin. (Color figure available online only.)
Recently, a systematic study of the occurrence of colloidal aggregation among purified
natural molecules was performed [40]. These authors screened a small but representative subset of natural phenolic compounds
(117) and found that the proportion of aggregating compounds was around 12 % when
they were tested at a concentration of 10 µM. They showed that flavonoids were more
aggregation prone than other phenolic compound classes such as coumarins and organic
acids. In fact, all of the studied flavonoids (23) formed DLS-detectable aggregates
in at least one of four different tested conditions. The occurrence of aggregates,
however, did not automatically translate into unspecific inhibition and only two flavonoids
(quercetin and rhamnetin) were identified as promiscuous [40]. The study, however, gave the foundations for another equally plausible scenario,
that aggregation could also lead to false negatives by reducing, for instance, the
concentration of the compounds available in the solution. Based on all these findings,
a necessity has surged of acknowledging aggregate formation as a likely source of
either false or negative hits when screening natural compound collections.
A first step to exclude false positives after performing a primary screening campaign
is to check if any of the identified natural hits have been previously flagged as
aggregators. For flavonoids, we recommend checking on the list compiled in [40] (available at: http://www.mdpi.com/1420–3049/17/9/10774), while for other types
of natural compounds, the list of aggregators that is maintained by Shoichetʼs laboratory
could be consulted (available at: http://shoichetlab.compbio.ucsf.edu/take-away.php).
In biochemical assays, a simple experimental way to preclude interfering aggregators
is by retesting their inhibitory effects in the presence of non-ionic detergents.
If the inhibition is significantly attenuated by small amounts of non-ionic detergent,
the compound is likely to act via aggregation. Detergents (0.01–0.1 %) have been proposed
to disrupt aggregate formation as well as dissociate the protein-aggregate interaction.
Feng et al. [29] performed a large detergent-based campaign with more than 70 k molecules and concluded
that inclusion of 0.01 % Triton X-100 effectively reverses the promiscuous inhibition
caused by more than the 95 % of the aggregators. This strategy can be optimized to
perform well in either lower throughput (i.e., cuvettes or 6-well plates) or higher
throughput formats (i.e., 96-, 384-, or 1536-well plates). Other detergents such as
Tween-20, CHAPS12, saponin 10, and digitonin have been shown to be applicable as well
[41]. Experiments can be run separately in the presence and absence of the detergent.
A precautionary note is that when using the detergents, they should be preferably
added to the buffer before any other component. The introduction of detergents in
many different assay formats has been proven possible without compromising the assay
quality [41]. In assays that cannot tolerate non-ionic detergents, for instance, in cell-based
assays, a suggested possibility has been to use 1 mg/ml of bovine serum albumin (BSA)
instead [30], but this molecule can sometimes sequester non-promiscuous hits [34] or cause other types of interferences [42].
Although the detergent sensitivity concept is widely applied for excluding unspecific
inhibitors, other methods have been implemented that allow a direct and noninvasive
quantification of the formed aggregates. Among them, high-throughput screening assays
based on SPR using Biacore technology [39] and PC biosensor microplates [43] have been reported. From these studies, it has also become clear that some of the
interactions between aggregates and protein targets are spontaneously reversible and
this can add additional complexity to the process of flagging and removing aggregators
from natural libraries.
Other features of the functional behavior of hits can similarly help in distinguishing
the “false” (or promiscuous) from the “true” ones. For instance, if an inhibitor is
found to display a competitive kinetic mechanism, this compound can be regarded as
less likely to be an aggregator [34]. This consideration is based on the structural similarities that competitive inhibitors
typically share with the substrate and their ability to recognize specifically the
catalytic sites, which differs from the unspecific nature of the aggregates-induced
inhibition. On the other hand, the preservation of inhibitory activity after spinning
compounds for several minutes in a centrifuge also indicates that aggregates are not
being formed.
Apart from experimental approaches, attempts have been made to characterize the physical-chemical
properties of the aggregators on the hopes of applying in silico calculations to predict the aggregation potential. Two features have been suggested
to potentially distinguish between aggregators and non-aggregators: clog P and aqueous
solubility. Based on these features, from of a set of 111 compounds, a valid distinction
in more than 80 % of the cases was done, with aggregators exhibiting higher clog P
and lower aqueous solubility [44], [45]. A better prediction (correct in over 90 % of cases) has further been achieved by
a more complex recursive partitioning model [44].
We have, for instance, mapped the chemical space occupied by aggregators and non-aggregators
using ChemGPS-NP, a freely available chemography tool [46], [47], applicable to natural compounds ([Fig. 6], unpublished results). The used aggregators and non-aggregators (96 compounds) have
been obtained from the publicly available repositoire of Shoichetʼs laboratory, mentioned
earlier. Regions of the chemical space, as defined by a combination of descriptors
characterizing molecular size (PC1), aromaticity (PC2), and lipophilicity (PC3), have
been seen to overlap between aggregators and non-aggregators. As pointed out by other
authors [44], the selection of proper descriptors and development of simple models that could
accurately predict the aggregating behavior of compounds is a challenging task. We
believe this is an area in which more research needs to be performed in order to facilitate
follow-up studies during the reconfirmation stage of a large number of primary hits.
Fig. 6 A Workflow of the in silico process used for mapping of the chemical space of aggregators and non-aggregators.
B 3D representation of the chemical space of aggregators (blue dots) and non-aggregators
(red dots), using the Principal Component Analysis (PCA)-based chemical space navigation
tool ChemGPS-NP (unpublished results). As schematically represented in A, the analysis uses 2D descriptors (35) describing the physical-chemical properties
of the compounds that are calculated from SMILES. Salts, hydration information, as
well as counter-ions are excluded from SMILES. For analysis of the chemical space,
the first four dimensions (PC1–PC4) are plotted using the software Grapher 2.1 (MacOS
X, US).
Optical interferences: Crude extracts have a maximized chemical diversity and do not require any purification
steps, but in order for the activity to be detected, they often need to be screened
at higher concentrations due to the low concentrations of their active components.
A major drawback of screening concentrated crude or semi-purified extracts is that
color interference, autofluorescence, or light scattering by particulated samples
(as those present in lab dust) can occur, which generates false positives and negatives.
Similarly, colored and/or autofluorescent pure compounds can cause artifactual results.
Indeed, many natural compounds are rigid and planar and possess multiple conjugated
aromatic moieties, which increase the probability of endogenous fluorescence [48]. This is the case in widely distributed natural molecules such as coumarins, anthraquinone
derivatives (for instance, hypericin, present in the alcoholic extract of Hypericum perforatum) and pigments such as carothenes, chlorophyll, or chlorophyll breakdown products
such as phaeophorbide A [49]. Additionally, the aging of samples can result in the formation of degradation products,
which can be strongly light absorbing compounds even in the visible range (400–700 nm)
[50].
In these cases, the compound spectral properties cause interferences with the light
detection step of the screening assay, and they are mostly predominant in assays that
are run in absorbance and fluorescence (FI, FP, and FRET) modes [51]. Such interferences are manifested by a typical increase in the background signal
of the assay but also by the participation in unwanted FRET with the assay fluorophore
[48]. Given that a vast majority of screening assays is nowadays run with absorbance-
or fluorescence-based technologies, these issues cannot be ignored. Moreover, the
increased use of homogeneous assays also accentuates these problems as the test samples
remain in the wells during the entire duration of the assays.
Because these interferences are technology dependent, suggested solutions typically
involve changes in the protocols or ultimately in the detection methods. The simplest
strategy for dealing with minor optical interferences is to include one step in which
the absorbance or fluorescence of the interfering molecules is measured in the absence
of any other reaction component, which is then subtracted from the signal detected
in the real biochemical or cell-based assay. However, in many cases, the compound
fluorescence can be higher than that of the fluorophore, even at relatively low concentrations
(10 µM) [52], and this strategy is, thus, not sufficient. The problem is additionally aggravated
when higher concentrations of extracts or pure compounds (> 10 µM) are tested in cell-based
assays, since different cell types can display endogenous fluorescence in various
conditions. This solution is also not operational for unwanted FRET artifacts, which
are more difficult to identify and correct. In the literature, a common bypass of
this issue has been to exclude optically interfering compounds from follow-up studies,
but in our opinion this strategy hampers the identification of natural scaffolds that
could have otherwise held promise as starting points for lead refinement strategies
[48]. A better solution is offered by simple mathematical procedures that can be applied
to correct for both increases and decreases from the baseline caused by interferences
from test compounds, as described in [50]. Another contribution focuses on different strategies that can be applied to tackle
this type of interference, specifically in FRET-based assays [53]. However, even if data can be corrected with these procedures, the data would eventually
need to be rejected in cases where the interferences result in more than a 2-fold
change in the signal [50].
A second-tier strategy that we can recommend is the utilization of fluorescence assays
that rely instead on red-shifted dyes or longer wavelength tracer fluorophores to
avoid spectral overlap with organic compounds that absorb in the ultraviolet region
or other autofluorescent molecules (such as coumarins). Several dye classes have been
developed in recent years with absorption maxima beyond 520 nm, extending to nearly
800 nm, from which it is possible to select for nearly all types of assay applications.
The suitability of this approach for natural product screening has been documented.
Red-shifted fluorogenic substrates have been shown to reduce interferences during
the screening for protease inhibitors from natural extracts, from prokaryote, fungal,
and plant sources, as well as pure natural compounds [54]. Also, an FP assay using red-shifted dyes has been developed to screen for kinase
inhibitory activity resulting in significantly less interferences from constituents
of microbial extracts when compared to a fluorescein-based competitive FP assay or
a [33P]ATP Flashplate assay [55]. However, this strategy is also not exempt of limitations, as it may not prevent
the interferences from red- and far-red emitting pigments such as chlorophyll and
other naturally occurring porphyrins.
A third strategy for overcoming these limitations is the implementation of methods
that entirely preclude the use of chemical labels, otherwise known as label-free.
Particularly in cell-based assays, label-free methods rely on impedance-based measurements
to detect changes in the electric properties or the passage of ions through the cells.
These changes can be brought about by cellular changes in the attachment to electrodes
located in dishes, for instance, in 96-well microplates or by the activity of different
receptor types (i.e., GPCRs, tyrosine kinase) [56], [57]. Beneficial aspects of these methods include that they can be applied without any
restriction in colored and autofluorescent samples, they are noninvasive and they
offer continuous readouts and a simultaneous view of short- and long-term cellular
events with very minimal labor involved. Moreover, because cells are not stained,
fixed, or altered in any way at any point, samples can then be interrogated for the
presence of metabolites or for other responses using chemical labels. Such an additional
interrogation allows for obtaining a multicomponent response from a single culture
of cells, which diminishes biological variability. Currently available label-free
technologies, for instance, Electric Cell-Substrate Impedance Sensing (ECIS, from
Applied Biophysics), Epic System (from Corning), or xCELLigence (from ACEA Biosciences),
were originally accepted with a very slow pace in the drug discovery scenario, but
over the last ten years they have increasingly attracted interest as their throughput
and robustness have increased.
Until now, label-free methods have been shown to be excellent tools for tracking cytotoxic
effects, also in the case of natural samples. Investigations have been performed on
the cytotoxicity of a large collection of extracts from Bangladeshi traditional medicinal
plants against pancreatic and breast cancer cells [58], [59], [60]. These studies have followed a tiered screening approach in which the first screen
is performed in all samples using a label-free PC biosensor, followed by two other
follow-up assays using conventional labels (MTT proliferation and caspase 3-induction).
Their approach ensures that none of the tested extracts (more than 55 in two of the
studies) would be excluded due to interfering optical signals during the first screen.
The PC biosensors are located in 96-well microplates and they produce a highly localized
shift in reflected wavelength at the site of cellular attachment that is coupled to
an image detection system that scans the biosensor surface and has sufficient resolution
to monitor the attachment/detachment of individual cells. Image analysis can be used
to study the cellular population of cells in the wells, which can be readily translated
to a simple cell count. Using the label-free biosensor assay, researchers have been
able to rapidly differentiate and classify the effects on cancer cells of several
plant extracts with a previously unknown function [58], [59].
In a recent work, Kling et al. 2013 [61] used the ECIS method to screen for the neuroprotective activity of 19 phenolic compounds
such as flavonoids, flavonoid metabolites, phenolic acids, and their methyl esters
(including several colored compounds) after induction of oxidative stress with tert-butyl hydroperoxide, and compared their output with a conventional cytotoxicity assay,
based on an endpoint measurement with the MTT probe. This work documented the benefits
of studying neuroprotection mechanisms via the recording of continuous cellular responses
with ECIS. The described method was also particularly advantageous for dealing with
compounds like quercetin or kaempferol, which have been shown to interfere with the
performance of redox probes, such as MTT [62], [63], [64].
In our laboratory, studies have been performed on the cytotoxicity of several natural
extracts, and one of them (coded NP1, unpublished results) has offered a challenge,
as this plant extract interferes with many commonly used viability assays. For instance,
NP1 reduces resazurin in the absence of cells, likely due to redox-active constituents,
and it also increases the signal of calcein and ethidium homodimer 1 (EthD-1; components
of the commercial LIVE/DEAD viability/cytotoxicity kit), which to a certain extent
could be explained on the basis of the extractʼs autofluorescence. Thus, NP1 can cause
an overestimation of the cell viability, leading to false negative results.
These problems are circumvented when the label-free ECIS assay is applied for the
cytotoxicity measurement of crude extracts. In [Fig. 7], the impedance curves recorded with ECIS showed an overall concentration- and time-dependent
cytotoxicity of NP1 (unpublished results). In this assay, increases in the impedance
are associated to cellular attachment while decreases are caused by toxic insults
that can result in the loss of the cellular integrity of the cellular monolayer formed
on the ECIS electrodes, as exemplified by the addition of NP1 (200 µg/ml, [Fig. 7]).
Fig. 7 Impedance changes recorded by ECIS in GT1–7 cells treated with a crude plant extract
(coded NP1) at different concentrations (50, 100, 200 µg/ml). ECIS instrumentation
model Z (Applied Biophysics) was used. 8W1E electrodes were pretreated with 10 mM of cysteine
as recommended by the manufacturers. Electrodes were filled with 400 µl of DMEM, and
impedance recorded at 16 kHz every 60 seconds for 1 hour. Cells (400 µl, 4 × 105 cells/ml) were then added and impedance continued to be measured. Twenty-four hours
after adding the cells, 40 µl of the medium was replaced by 40 µl of the NP1 or medium
in the untreated control wells.
Specific advantages of using ECIS in this type of study are readily noticeable. The
first one is related to the continuous nature of the cellular responses that are measured.
For example, after adding NP1 (100 µg/ml, [Fig. 7]), a sudden decrease of about one-fourth of the impedance values compared to untreated
cells is detected, but within a few hours, a recovery of the cells is recorded and
after 40 hours, less than a 10 % decrease of the impedance values is detected. This
indicates that early toxic events triggered by NP1 are temporarily buffered by the
cells, which would have been overlooked in a conventional cytotoxicity assay performed
only at a single time point (for instance, at 48 hours). Another advantage of the
ECIS studies is the possibility of recording delayed cytotoxicity events. In [Fig. 7], it can be seen that after 48 hours, the cells can tolerate NP1 (100 µg/ml) with
less than a 25 % decrease in impedance values compared to untreated cells. However,
as time proceeds, the impedance steadily drops and a clear toxicity is seen after
72 hours with an over 50 % decrease in impedance values. This delayed toxicity would
have also been undetected had the measurements been conducted only using acute cytotoxicity
assays.
Label-free methods, on the other hand, are not exempt of disadvantages. Among them
are the difficult interpretations of the results and the lack of full understanding
of the biological significance of some measured signals [65], [66]. These two elements seem to relate to the biophysical nature of the measured responses
that are mostly associated with complex physical responses and are not always fully
elucidated from a biochemical perspective. Also, the utilization of these methods
by the academic community has been hampered by their high consumable costs. However,
upon identifying natural compounds or extracts with optically interfering compounds,
these methods could offer a new alternative for further investigation of some biological
activities. Based on this, it is our view that more dedicated research in label-free
bioactivity screens of natural products will benefit drug discovery.
Interferences due to the meniscus effect: Optical signals can also be attenuated by reasons that are not associated to the
optical properties of the compounds, but instead by the surfactant-like physical properties.
Although these interferences are not as frequent as the ones described above, some
natural samples containing surfactants such as saponins can cause a deepening of the
meniscus of the liquid in microtiter well plates, which results in a decrease of the
path length in the liquid column that can reduce the amount of absorbed light being
measured through the liquid column. This interference typically results in a lower
signal when running absorbance-, fluorescence-, and luminescence-based assays in top-reading
instruments. Methods for correcting them have been described recently [50].
Antibacterial Screening of Natural Products As a Case Study
Antibacterial Screening of Natural Products As a Case Study
Having discussed general aspects of the development of screening assays and implementation
stages, we will now focus on an area in which natural products offer a successful
case study: antibacterial drug discovery. Despite the immense impact that antibacterial
drugs have had on overall life expectancy and quality, bacterial infections remain
a persistent health burden both in industrialized countries and the third world [67]. Considering the aspects of both the bacterium and the host, the two obvious challenges
in treating bacterial infections are the selection of resistant mutants and the increasing
number or opportunistic infections among immunocompromised persons [68], [69].
Antibacterial agents have been the focus of natural product research for several decades,
and continue to be among the most widely screened biological activities among natural
product scientists. One evident reason for this is the success of natural products
as antibacterial drugs and drug leads. According to a recent report by Cragg and Newman
[4], 75 % of all new small molecule antibacterial agents approved by the FDA between
1981 and 2010 were of natural origin. However, antibacterial drug discovery in general
is currently a matter of major concern due to low success rates, and several pioneers
and professionals in the field have pointed out the need for reevaluating the screening
strategies to reconcile our research practices with the demands set by resistance,
persistence, and opportunistic infections [70], [71], [72], [73]. The wide variety of antibacterial assays reflects the extensive efforts to identify
new drug molecules within this therapeutic. The rest of this section aims to give
a short overview of the different assays available, highlighting some of the points
discussed in the previous sections.
Phenotypic assays for bacterial growth inhibition
The classical antibacterial screening strategy has relied on exposing bacterial cultures
grown on agar plates to test samples and measuring the inhibition zones (areas with
no bacterial growth) around the samples. This methodology was used during the discovery
of virtually all antibiotics during the 1960s, many of which still form the basis
of our antibacterial drug arsenal [74]. It is still widely used for low-throughput bioactivity studies and has been recently
applied, among other studies, to the evaluation of plant extracts selected for ethnobotanical
use in different areas [75], [76]. The popularity of the this assay for screening the activities of plant extracts
most likely reflects tradition rather than rational selection of an assay method least
prone to interference with the test material. As recently pointed out by Gertsch (2011),
the scientific knowledge needed for choosing the biologically most relevant means
for studying the biological activities of complex plant-based mixtures, in which synergism,
for instance, could take place, is currently missing to a large extent [77]. The ethnobotanical studies typically involve some tens of samples in maximum, reflecting
the limitations that the method has for screening purposes. Since the assay endpoint
relies on a visually measurable inhibition zone, the assay requires large amounts
of reagents and laboratory space, and suffers from a slow and labor-oriented readout
because of the need for manual determination of the zones of inhibition. Nevertheless,
the assay can be miniaturized and automated by using image-based screening platforms
capable of quantifying the bacteria-free zones on a much smaller scale [78], [79].
To meet the needs set by increasing sample numbers, various microtiter plate-based
assay formats have been established, relying on the measurement of bacterial biomass
or metabolic activity of bacteria growing as a suspension [80]. Considering time and reagent consumption per sample, such assays are much better
suited for large sample collections or projects with bioassay-guided isolation of
the active ingredients within extracts, as shown by the recent identification of novel
broad-spectrum antibacterial agents among natural products using a pH-dependent metabolic
dye [81] or a study on Pseudomonas aeruginosa using optical density measurements [82]. Integrating such methods with other technologies, such as microfluidics and fluorescence-aided
cell sorting, has also allowed for the development of sophisticated platforms of bacterial
viability evaluation [83].
In antibacterial assays, potential sources of interference are dominated by the colored
or fluorescent nature of the tested samples, since all classical viability probes
rely on photometric or fluorometric readouts within the spectral area that is also
covered by natural products and other small molecules, as discussed earlier. One practical
means for overcoming this type of interference is by the baseline measurement of each
sample prior to the bacterial growth phase, allowing for the elimination of potentially
increased background values in each sample separately. Another potential source of
compound-borne interference is the ability of some redox-active compounds to directly
convert viability dyes whose conversion to a colored or fluorescent form relies on
cellular oxo-reductive metabolism. If such a feature is suspected in the antibacterial
assay, we recommend performing counter-screenings in the absence of living cells to
detect any direct interactions between the dye and the samples to be tested.
From a biological point of view, the major difference between the above-mentioned
microtiter plate-based assays and the classical agar culture assays is the state of
bacterial community. The microtiter plate-based assays rely on liquid cultures of
planktonic bacteria, in which the cell population is maintained in a logarithmic growth
phase. In contrast, bacteria grown on agar plates grow attached to the solid medium
surface, which, as discussed below, is considered to represent a biologically more
relevant growth state of most pathogenic bacteria.
More focused or hypothesis-driven phenotypic screening assays have been set up by
using reporter gene assays with fluorescent or luminescent bacterial strains. Depending
on the specific setup, the reporter systems are used to identify inhibitors of, e.g.,
bacterial transcription [84] or translation [85]. Technically speaking, the time span of these assays are typically much shorter
than in classical growth inhibition assays, primarily for two reasons. First, changes
in transcription, translation, or other similar events can be detected within a few
hours when the monitored event is the direct target of the inhibitor in question.
Second, the use of fluorescent and luminescent reporter enzymes typically yields readouts
with a good SW, providing the basis for a good assay performance even with short exposure
times. The prerequisite for achieving biologically meaningful screening assays by
this approach is the validation of the reporter gene insertion stability within the
bacterial genome and characterization of the growth kinetics of the recombinant strain.
Similar to pure growth inhibition assays, autofluorescent natural products may interfere
with fluorescent readouts, while luminescent reporter genes do not suffer from such
interferences. However, since the light generated by luciferases originates from an
enzymatic reaction, any inhibitor, activator, or stabilizer of the reporter enzyme
can suppress the amount of light to be detected and may thus be identified as a false
positive in any screen utilizing the reporter [86]. The proportion of firefly luciferase inhibitors in an unbiased chemical library
has been estimated to be approximately 3 % [87], and screening data on luciferase inhibitors has also been made publicly available
(PubChem AID 411). To control false positives or negatives due to reporter enzyme-directed
effects, consulting such publicly available data, testing the activity of hit compounds
on purified luciferase, and performing a hit confirmation are our recommended approaches.
Also, it is worth to keep in mind that luciferase-inhibiting natural compounds may
still exhibit genuine bioactivities not related to this property. For instance, the
extensively studied polyphenol resveratrol has been reported to inhibit firefly luciferase
[88] and should therefore be treated with special caution in luciferase-based assays,
but certainly not all biological activities reported for resveratrol can be attributed
to luciferase-related artifacts. Therefore, we do not consider excluding compounds
like resveratrol from natural product libraries necessary or even beneficial.
With respect to intracellular bacteria, image-based HCS assays provide a more efficient
and informative way to alleviate the laborious assay protocols often involving the
readout of the results under a microscope. Simultaneous detection of bacterial replication
centers and host cell components have been described for different bacteria [88]. While most bioactivity screens utilizing the manually determined intracellular
bacteria load have involved less than one hundred samples (e.g., [89], [90]), the screening of 57 000 compounds with an HCS assay using GFP-expressing Mycobacterium tuberculosis and RAW macrophages as host cells illustrates the possibility to potentially increase
the throughput associated with HCS platforms [91].
The improved information content achieved by HCS applications in an antibacterial
assay context refers typically to either information on bacterial antigen localization,
detection of specific stages in the bacterial life cycle, or simultaneous detection
of host cell viability. While the two former cases may provide significant benefits
in a hypothesis-driven or targeted antibacterial screening, the latter aims to simply
exclude toxic compounds without the need for additional counter-screens. Even though
the concentration-dependent effects of different antibiotics on bacterial cell morphology
are relatively well-known, increasing primary screen information content by achieving
mechanistic data from image-based screening platforms is limited by the resolution
of fluorescent microscopes due to at least one order of magnitude smaller bacterial
cell size compared to mammalian cells. However, the recent work by Peach et al. [92] described the development of a software platform capable of predicting mechanisms
of action of antibacterial ingredients within marine natural product extracts using
Vibrio cholerae cell morphological phenotypes and a training set of antibiotics with known mechanisms
of action.
The nematode worm Caenorhabditis elegans has become an established model for studying bacterial pathogenicity in a whole organism
and, more recently, also gained popularity as an in vivo screening model for antibacterial activity [93]. The ability of C. elegans to feed on pathogenic bacteria and the death of the organisms upon the infection
has formed the basis for straightforward assays detecting worm viability using live/death
dyes. With its size of 1 mm, C. elegans can be dispensed with liquid handling instruments. Academic screens using a C. elegans rescue assay with tens of thousands of compounds and natural product extracts have
been carried out on Enterococcus, Pseudomonas, and Vibrio species [94], [95]. The power of C. elegans rescue assays in antibacterial screening is obvious for two reasons: i) it detects
not only inhibitors of bacterial replication but also compounds suppressing in vivo virulence of the target bacterium, and ii) it can exclude compounds showing acute
toxicity on the host organism or pharmacokinetic properties limiting the penetration
to target tissues [93]. While HCS and other phenotypic platforms are, in some cases, limited by the sample
numbers they can be adopted to, many academic screening campaigns involve small or
medium size collections of compounds or extracts that can easily be tested in assays
with a high information content and, thus, we encourage the academic natural product
research community to take full advantage of this aspect.
Target-based approaches for identifying inhibitors of bacterial growth
Screening of classical antibiotic targets: Clinically approved antibacterial drugs target only a few bacteria-specific structures,
such as ribosomes and enzymes essential for cell wall or folate biosynthesis. Target-based
screening assays, both for ligand-binding and functional enzyme assays have been described
for all these targets and can be readily used for screening ([Table 4]). A primary advantage of these targets is that their essential nature for bacterial
growth is well validated and the proteins in question have been extensively characterized
by biochemical and structural biology means. Many of these proteins and assay reagents
are also readily commercially available (for representative examples, see references
in [Table 4]).
Table 4 Some examples of assays for antibacterial targets with existing antibiotics that
are in clinical use.
|
Target
|
Assay format
|
Detection mode
|
Application
|
Reference
|
|
Penicillin-binding protein
|
Competitive binding assay
|
Fluorescence Polarization; labeled penicillin as competing ligand
|
Screening of pooled small molecules
|
[136]
|
|
Ribosome
|
Binding assay
|
Fluorescence quenching of a fluorophore-labeled ribosome
|
Screening of a small set of soil microbe extracts
|
[137]
|
|
Ribosome
|
Competitive binding assay
|
Fluorescence, labeled neomycin
|
Characterization of aminoglycoside binding
|
[138]
|
|
Ribosome
|
Competitive binding assay
|
FRET; coumarin conjugated aminoglycoside and ribosome
|
Characterization of aminoglycoside libraries
|
[139], [140]
|
|
Topoiso-merase
|
Enzyme inhibition assay
|
Fluorescence Intensity; fluorophore-
labeled oligonucleotide as substrate
|
Screening of small molecule and NP extract libraries
|
[96]
|
|
Dihydrofolate reductase
|
Enzyme inhibition assay
|
Fluorescence Intensity; NADH levels determined with resazurin
|
Screening of small molecules/synthetic and natural
|
[141]
|
|
Dihydropteroate synthase
|
Enzyme inhibition assay
|
Radiometric; substrate and product separation by TLC
|
Screening of pyrimidine libraries
|
[142]
|
Many of these targets have been exhaustively tracked by chemical inhibitor screens,
and scaffolds of known ligands have been widely diversified by medicinal chemists,
reflecting the fact that the vast majority of newly approved antibiotics during the
past decades are derivatives or analogues of previously known drugs [71], [73]. However, previous and recent work on these targets illustrate that the target-based
approach can also be used for successful identification of inhibitors or modulators
that are structurally unrelated to the known antibiotics. Examples in this respect
include the identification of several non-beta lactam structured ligands of Neisseria
gonorrhoeae, a penicillin-binding protein, and anziaic acid isolated from Hypotrachyna sp. as an inhibitor of topoisomerase 2 (an enzyme also known as DNA gyrase) [96].
Despite the massive amount of work put into the known targets, only a limited number
of drug molecules suitable for clinical use are available for certain targets. For
example, mupirocin, a monoxycarbolic acid derivative originally isolated from Pseudomonas fluorescens, remains the only clinically approved inhibitor of aminoacyl-tRNA synthetase (ARS;
a family of enzymes involved in bacterial tRNA biosynthesis), but its use has been
limited to topical applications such as wound infections due to its properties not
being suitable for systemic use [97]. Trimethoprim, on the other hand, remains the only approved antibacterial drug targeting
dihydrofolate reductase [98]. Identifying new molecules chemically unrelated to the previously used drugs can
be considered one valuable approach to broaden our antibacterial drug arsenal.
However, the usefulness of mupirocin and trimethoprim, as well as most other antibiotics
in clinical use, is limited by the emergence of resistant bacterial strains, in part
due to the accumulation of low-affinity variants of the target protein. Screening
of the wild-type enzymes and bacterial strains is therefore not a sufficient means
for identifying clinically useful compounds for further development. Instead, including
the low-affinity mutants of the protein is recommended to overcome this problem [70]. Other resistance mechanisms, such as compound inactivating enzymes or efflux pumps,
can also be included within the screening platforms, which may be of particular benefit
when screening plant-derived material. While high-potency growth inhibitors are typically
not found among pure compounds isolated from plants, some classical examples of synergistic
combinations of compounds found within plant extracts are known [99]. Assay methods and advances in identifying inhibitors of both beta-lactamase enzymes
and bacterial efflux pumps have been previously reviewed [100], [101] and are, thus, not covered here in more detail.
Screening of novel targets: It is generally thought that finding more potent inhibitors of known targets will
not solve the resistance problem in the long run, and huge efforts have been made
in order to find new targets for antibacterial discovery. Genomics analyses of pathogenic
bacterial species have indicated that there are 100–200 conserved bacterial genes
with no close homologues in eukaryotes [71]. The pharmaceutical industry has invested significant amounts of time and resources
to validate the essentiality of these gene products for bacterial replication, to
develop screening assays on them, and to conduct large scale screening campaigns on
tens of different targets [71], [72]. As a result of these efforts, only a small number of lead compounds have been taken
to preclinical studies, and the majority of them have never been taken to clinical
trials. Many of the screening assays developed within the process have been published
and thus they can be adopted for the screening of chemical collections harboring compounds
with more antibiotic-like properties. For example, bacterial ribosome biosynthesis
can be targeted by screening inhibitors against a bacterial GTPase in an assay using
the isolated bacterial enzymes with generally applicable screening methodologies for
GTP level detection [102]. As the target proteins are of bacterial origin, their expression and production
in quantities required for HTS or MTS is generally not the rate-limiting factor for
screening, and generic protocols for detecting the enzymatic activity of question
are often readily available.
As mentioned, success on developing new drug candidates by applying target-based approaches
has been limited. A major limitation of the target-based approach is that it gives
no information on the compoundʼs ability to penetrate bacterial membranes and thus
reach the target site. Particularly with gram-negative bacteria, penetration through
the bacterial membranes poses a significant challenge with regards to the compounds
physicochemical characteristics and is thus critical for the biological activity first
hand [70]. One means for taking this into account in target-based screening is by designing
whole cell assays for the targets of interest by comparative screening of wild-type
and target-depleted strains. By silencing the target of interest by RNAi or other
means, the silenced and wild-type strains of the bacterium are expected to have different
susceptibilities towards a small molecule modulator of the target. This approach has
been successfully used, for example, in the discovery of new and previously known
fatty acid biosynthesis (Fab) inhibitors among natural product extracts targeting
S. aureus FabF7FabH [103], [104]. However, the widely studied Fab as an antibacterial target has revealed one additional
challenge in target-based approaches: the essentiality of the target may not be directly
interpretable based on its conserved nature, since gram-negative bacteria have later
been shown to be resistant to Fab inhibitors [105].
Generally speaking, screening with biochemical assays is prone to false positives
due to aggregating behavior of some small molecules, as discussed earlier. In fact
beta-lactamases are among the most widely studied enzymatic targets in this respect,
and according to the data from Shoichetʼs laboratory, hit lists from screening campaigns
with this target are dominated with promiscuous inhibitors, with an occurrence reaching
97 % of all screening hits [106]. Similarly, Newton et al. [107] reported a screening assay for the synthesis of mycothiol, an essential Mycobacterium tuberculosis, in which 65 of the screening hits from a collection of 2024 compounds were found
to be promiscuous, nonspecific inhibitors or to interfere with the photometric readout.
Assaying bacterial virulence factors
An alternative antibacterial strategy aims at the identification of small molecules
inhibiting the activities of virulence factors. One approach, in this respect, is
based on the phenotypic assays for virulence factor gene regulation [108], [109], while others have addressed the question via the specific virulence mechanisms
[110], [111]. One widely conserved virulence factor especially among gram-negative bacteria is
the type 3 secretion system (T3SS), a syringe-like protein complex responsible for
exosis of bacterial products. Screening assays for the discovery of T3SS inhibitors
have been described on several bacteria, generally applying fluorescent of luminescent
T3SS substrates detectable from extracellular space samples [110], [111]. Another virulence factor targeted by recent screening campaigns is bacterial motility,
for which a screening assay utilizing a miniaturized version of the classical soft
agar method in combination with viability staining to distinguish growth inhibitors
from motility inhibitors has been described [112]. In addition, bacteria-specific exoenzymes have inspired the development of target-based
screening assays based on the detection of the cleavage products [113].
Screening for anti-biofilm compounds
According to an estimate by the U. S. National Institutes of Health, over 65 % of
bacterial infections are nowadays recognized to be caused by biofilms [114]. The main challenge posed by biofilms is their increased tolerance to chemotherapy
and the host immune responses, which stems from a variety of factors that include
the interbacterial communication networks, the production of the extracellular matrix,
and the presence of persisters [114], [115]. Nonbacteriocidal or bacteriostatic approaches for screening anti-biofilm compounds
have involved, for instance, targeting signaling pathways [116], [117], [118]. Interest in biofilm research has highly increased during the past few years, and
efforts towards producing standardized data have expanded involving both assay development
and database integration [28], [119], [120], [121], [122].
Plants have classically not been considered as sources of potent antibacterial compounds
[70]. Yet, although no efficient inhibitors of bacterial growth have been discovered
from plants, several plant-derived compounds are known which target bacterial populations
by other means. Extracts from garlic and Elmleaf blackberry (Rubus ulnifolius), as well as flavonoids isolated from different plants, have shown anti-biofilm activity
[123], [124], [125], [126], [127], and in the case of garlic extract, the anti-biofilm activity has also been confirmed
in a mouse model [128]. In another study, the anti-biofilm activity of a natural compound originating from
garlic was traced to metabolites putatively affecting interbacterial communication
and occurred at concentrations that do not affect planktonic bacteria growth [129].
In a screening-compatible manner, a typical procedure to measure biofilm modulating
effects is by detecting the biomass of a biofilm grown on the bottom and walls of
the wells in a microtiter plate by crystal violet staining. Additionally, metabolic
activity of the bacterial biofilms can be determined with viability dyes such as resazurin,
and the extracellular matrix can be quantified by a specific dye. Validated screening
platforms based on such methods have been described and successfully used for identification
of organic small molecules preventing biofilm formation and/or destructing preexisting
biofilms (i.e., [130], [131]). Attempts to miniaturize the crystal violet assay in the 96-well format have been
successful and, for instance, a semiautomated protocol for crystal violet staining
has been discussed earlier as an example here ([Fig. 4]) [28]. To increase throughput, other alternative methods such as an attachment assay described
for a Pseudomonas aeruginosa luciferase expressing strain have been developed [131]. In addition, an HCS assay for anti-biofilm studies has been described, based on
GFP-expressing bacterial or a combination of fluorescent viability dyes [132], [133]. When compared to crystal violet staining, the HCS assay was stated to be significantly
more sensitive in detecting surface attachment and other early events in the biofilm
life cycle, or simultaneous quantification of non-biofilm forms of the bacteria [132]. Achieving data on biofilm architecture in a high-throughput format may yield screening
hits with characteristics different from those identified with conventional methods,
but, to the best of our knowledge, reports on medium- or large-scale anti-biofilm
screens using HCS platforms have not been published thus far.
When it comes to selecting the proper measurement endpoint for anti-biofilm screens,
several authors have noticed the importance of combining biomass and viability with
matrix measurement methods (i.e., [119], [134]. Various lines of evidences have shown that compounds that are regarded as effective
when they inhibit biomass and/or biofilm viability could, in some cases, promote overproduction
(or maintenance) of the biofilm matrix, which thus facilitates biofilm colonization
in the long term [135]. Thus, including matrix detection assays could significantly enhance the understanding
of the biofilms responses towards anti-biofilm agents and give a better assessment
of their genuine clinical relevance. Also, in our opinion, an essential issue that
needs to be acknowledged in antibacterial screening is the fact that neither biofilms
nor suspended bacteria exist as an isolate lifestyle. Bacteria dynamically switches
between them upon changes in host or environmental conditions, and thus the knowledge
of the effects that test compounds may have in one or the other state is essential
as well.
Conclusions
The screening of natural products is undoubtedly complicated by their chemical complexity,
and even the purified natural compounds are known to be structurally unique and challenging
from a bioactivity perspective. However, proper assay validation and implementation
can help overcoming these difficulties and lead to the discovery of meaningful natural
lead compounds.
Over the last few years, academic screening has become increasingly engaged in chemical
screenings, and has come to provide mature and innovative contributions in a way that
has reshaped the field, traditionally dominated by the pharmaceutical companies. With
the opening of new academic centers worldwide, as well as the launching of large open
initiatives (such as EU-OPENSCREEN), a vast array of infrastructures and compound
collections have (and will continue to) become available for an even wider community.
Thus, it is our goal that some of the strategies discussed here offer methodological
guidelines for natural product researchers, as well as encourage others to embrace
new efforts in the rewarding path of discovering new drugs from natural sources.
Acknowledgements
The authors thank the Academy of Finland for financial support (WoodyFilm project,
decision 264 064; ArtFilm project, decision 272 266) as well as the Drug Discovery
and Chemical Biology (DDCB) network of Biocenter Finland. The authors also thank Malena
Skogman, Ph.D. and Daniela Karlsson, Ph.D. for kindly contributing some of the drawings
and images utilized in [Figs. 1], [3], and [6].
