|Home | About | Journals | Submit | Contact Us | Français|
Eukaryotic initiation factor 2 (eIF2) B is a guanine nucleotide exchange factor that plays a central role in translation initiation and its control, especially in response to diverse cellular stresses. In addition, inherited mutations in human eIF2B subunits cause a fatal brain disorder commonly called childhood ataxia with central nervous system hypomyelination or leukoencephalopathy with vanishing white matter. In yeast, inhibiting activity of eIF2B up-regulates expression of the transcriptional activator general control nondepressible (GCN) 4. We report here evaluation of high-throughput screening (HTS) using a yeast-based reporter gene assay, in which strains containing either wild-type or a mutant eIF2B were screened in parallel to identify compounds modifying eIF2B-dependent responses. The goals of the HTS were twofold: first, to discover compounds that restore normal function to mutant eIF2B, which may have therapeutic utility for the fatal human disease; and second, to identify compounds that activate a GCN4 response, which might be useful experimental tools. The HTS assay measured cell growth by absorbance, and activation of gene expression via a β-galactosidase reporter gene fusion. Because mutant eIF2B activates GCN4 in the absence of stress inducers, the mutant strain was screened for reduction in GCN4 activation. HTS revealed apparent mutant-selective inhibitors but did not reliably predict selectivity as these hits affected both wild-type and mutant strains equally on dose–response confirmation. Wild-type strain results from the HTS identified two GCN4 response activators, both of which were confirmed to be wild-type selective in dose–response testing, suggesting that these compounds may activate GCN4 by a mechanism that down-regulates eIF2B activity.
The study of cellular responses to environmental stress has provided valuable insight into disease processes and novel approaches for therapeutic intervention. Many such studies have focused on the yeast Saccharomyces cerevisiae because of the similarity of its cellular control mechanisms to those in mammalian cells and the relative ease with which it can be manipulated. The response of yeast to a variety of environmental stresses, including amino acid or purine starvation and oxidative stress,1 is mediated through the transcriptional activator protein general control nondepressible (GCN) 4p (Gcn4p). Gcn4p levels are tightly controlled by modulating the translation of its mRNA2 and the stability of the protein.3 Therefore expression of GCN4 is tightly regulated, a process known as general amino acid control.2,4
In general amino acid control, amino acid starvation results in the accumulation of deacylated tRNAs that bind to and activate the protein kinase Gcn2p. Active Gcn2p phosphorylates the translation initiation factor eukaryotic initiation factor 2 (eIF2) on a conserved serine residue at position 51 of its α-subunit. eIF2 is an essential G protein factor that when bound to GTP recruits initiator methionyl tRNA to the ribosome for each protein synthesis initiation event. Gcn2p-mediated phosphorylation of eIF2 antagonizes this by converting eIF2 into an inhibitor of its own guanine nucleotide exchange factor, eIF2B.5 This reduces protein synthesis initiation rates in cells. However, the translation of GCN4 mRNA is specifically stimulated under these conditions, resulting in the downstream transcriptional activation of numerous genes.6 Many Gcn4p target genes encode amino acid biosynthetic enzymes, including HIS4, that enable cells to overcome the starvation imposed (Fig. 1, left and middle). Gcn2p-mediated activation of GCN4 in response to oxidative stresses appears to be a critical factor allowing yeast cells to avoid damage from diverse oxidants such as hydrogen peroxide and the mutagens cumene hydroperoxide and cadmium.1,7
In mammals mGCN2 similarly responds to amino acid starvation to activate translation of activating transcription factor 4 (ATF4). Both ATF4 and GCN4 possess upstream open reading frames that limit the flow of scanning ribosomes to the authentic AUG start codon under non-stress conditions, but that allow enhanced GCN4/ATF4 translation under stress situations using a mechanism that requires resumed scanning and re-initiation. Recent studies have shown that mGCN2 may have diverse functions as it is important in the brain for sensing dietary amino acid levels and for long-term memory formation.8,9
eIF2B is a multisubunit protein that is required for protein synthesis initiation and its regulation in all eukaryotic cells.5 It is a complex of five nonidentical subunits (α, β, γ, δ, and ε) that are well conserved from yeast to human. eIF2B appears to be a key mediator of gene-expression control in response to environmental stress. In yeast several stress-response pathways, in addition to those involving Gcn2p, converge on eIF2B, and the events triggered downstream of eIF2B may differ depending on the nature of the stress. For example, during nitrogen starvation amino acid breakdown becomes an important nitrogen source. The breakdown products of this catabolism are fusel alcohols—fermentation-derived higher-order alcohols such as butanol or isoamyl alcohol—that function as potent chemical messengers apparently acting directly on eIF2B to inhibit global translation activation while specifically activating GCN4 translation. Thus the addition of exogenous fusel alcohols rapidly increases levels of Gcn4p and slows growth.10 It has been shown that the effect of the fusel alcohols is not mediated through Gcn2p and eIF2 phosphorylation and also that the downstream translational outcome differs from the Gcn2p-mediated response to amino acid starvation.11
Activation of GCN4 may also be induced by specific pharmacologically active agents, giving rise to phenotypic responses similar to those associated with environmental stress. Volatile anesthetics, such as isofluorane or halothane, can reversibly inhibit protein synthesis in both mammalian cells and yeast. These responses are in part caused by eIF2B-dependent mechanisms.12,13 Notably, the immunosuppressive and anticancer agent rapamycin activates GCN4-mediated stress responses through its inhibition of Tor1p and Tor2p, which are the yeast homologs of mammalian target of rapamycin, a mammalian phosphoinositide 3-kinase-related kinase.14,15 Mammalian target of rapamycin is currently the subject of intense study because inhibitors of this key regulator of cell growth may have broad therapeutic potential.16–19 An intriguing recent finding is that GCN4 activation mediates replicative lifespan extension in yeast. Thus, increases in yeast lifespan caused by dietary restriction or treatment with the antibiotic diazaborine—to reduce levels of 60S ribosome subunits—were shown to be dependent on activation of GCN4.20
Translation of GCN4 is also derepressed by specific mutations in eIF2B that give rise to the phenotypic responses such as slow growth normally induced upon environmental stress (Fig. 1). These yeast eIF2B mutants provide valuable tools to study a rare but devastating human disease. Mutations in human eIF2B genes have recently been found to cause the fatal disorder called childhood ataxia with central nervous system hypomyelination (CACH) or vanishing white matter disease (VWM).21–24 Disease symptoms are apparently related to a reduction in the catalytic activity of the guanine nucleotide exchange factor activity of eIF2B. CACH/VWM is inherited as an autosomal recessive trait, and progression and symptoms of the disease are exacerbated by febrile infections or minor head trauma, usually leading to death by the age of 2–5 years.21 Recent evidence implicates an abnormally heightened stress response as the cause of this extreme sensitivity to environmental insults. Up-regulation of the transcriptional activator ATF4 was observed in human embryonic kidney 293 cells containing human eIF2B mutations.25 Furthermore, primary fibroblasts from CACH/VWM patients showed abnormally high levels of ATF4 in response to stress.26 These findings extend to yeast, in which dysfunctional eIF2B has been shown to lead to translational activation of GCN4, the yeast homolog of human ATF4.23
We report here results of a high-throughput screen designed to discover chemical modulators of stress-response pathways in yeast. We focused specifically on eIF2B, because of its central role in activation of GCN4 and the identification of human eIF2B mutants as the cause of CACH/VWM disease. A reporter gene assay was developed in which both cell growth and expression of the GCN4 gene were measured in two yeast strains, wild-type S. cerevisiae and an eIF2Bε-R284H mutant that is equivalent to the clinically relevant human eIF2B5-R299H mutation. Both mutant and wild-type strains contain a chromosomally integrated GCN4-dependent reporter gene fusion to β-galactosidase. The assay was optimized for HTS by adaptation of a commercially available coupled β-galactosidase–firefly luciferase system for use with yeast cells.27
Mutant and wild-type yeast strains were screened in parallel to identify compounds acting specifically on either strain. The parallel screening strategy was designed to eliminate nonspecific compounds directly following HTS and thereby minimize the number of cherry-picks required. A compound that restores normal function to mutant eIF2B should give selective reduction of β-galactosidase activity in the mutant but not the wild-type strain. Such reduction in GCN4 activation in the strain containing mutant eIF2B might indicate a compound capable of restoring normal function to human eIF2B that would have therapeutic utility against CACH/VWM disease. Conversely, a compound that activates GCN4 expression through an eIF2B-mediated mechanism should selectively enhance β-galactosidase activity in the wild-type strain. These compounds might be useful chemical probes to study eIF2B-mediated pathways leading to stress responses. Comparison with known modulators of eIF2B such as the fusel alcohols may lead to further insights into the mechanism of action of these agents. Given the recently discovered roles for Gcn4p in protection against oxidative stress and enhancement of yeast replicative lifespan,7,20 activators of GCN4 may have potential to forestall the damaging effects of aging and age-related diseases.
This assay represents a novel approach to a target lacking useful chemical probes for mechanistic studies and pharmacological intervention. As such this project warranted significant development effort as it furthered the goals of the National Institutes of Health-funded Molecular Libraries Screening Center Network to address targets in underexplored complex areas of biology. We describe here technical challenges encountered in miniaturizing to a robot-compatible 384-well format and the implementation of a parallel screening strategy to discover modulators of stress-response pathways in yeast.
Wild-type yeast cells strain GP4213 (MATα leu2-3 leu2-112 ura3-52::HIS4–lacZ ino1 gcd6Δ gcn2Δ::hisG pAV1265[GCD6 CEN6 LEU2]) and an isogenic gcd6 mutant strain GP4198 (MATα leu2-3 leu2-112 ura3-52::HIS4–lacZ ino1 gcd6Δ gcn2Δ::hisG pAV1744[gcd6-R284H CEN6 LEU2]) were stored at –80°C as single-use frozen stocks, prepared as described below. YPD medium (10 g/L yeast extract, 20 g/L Bacto Peptone, and 20 g/L glucose), Breathe-Easy™ sealing films (catalog number Z380059), and buffer salts were purchased from Sigma-Aldrich (St. Louis, MO). The 384-well polypropylene V-bottom plates (catalog number 781280) were from Greiner Bio-One (Monroe, NC), and 384-well clear plates (catalog 3702) and white plates (catalog number 3652) were from Corning Life Sciences (Lowell, MA). Protease inhibitor tablets (catalog number 1873580) were from Roche Applied Science (Indianapolis, IN). The Beta-Glo® assay system (catalog number E4780) was from Promega (Madison, WI). Heat seals were applied using a Plate Loc heat sealer (Velocity 11, Menlo Park, CA), and polystyrene lids were from Nunc (Rochester, NY). The compounds tested here were supplied by BioFocus DPI (South San Francisco, CA) from the Molecular Libraries Small Molecule Repository.28
Yeast strains were taken from storage in a −80°C freezer and streaked onto a YPD plate, and colonies were grown by incubation of the plate for 3 days at 30°C. Cell stocks for the two yeast strains were prepared by inoculating a single colony into 50 mL of YPD medium in a 250-mL Erlenmeyer flask and growing the culture overnight at 30°C with vigorous shaking (200 rpm). Cell concentration was determined after overnight growth by measurement of absorbance at 595 nm (A 595). The cell culture was centrifuged at 3,000 g for 5 min, and the pelleted cells were resuspended into YPD medium containing 15% (vol/vol) glycerol to give a cell suspension with A 595 = 5. The cells were aliquoted for single use, frozen on dry ice, and stored at −80°C.
Compounds from the Molecular Libraries Small Molecule Repository were shipped on dry ice from BioFocus DPI as frozen 10 mM solutions in dimethyl sulfoxide (DMSO) in heat-sealed 384-well polypropylene plates. Prior to compound plating and shipping, quality control testing at BioFocus DPI confirmed compound solubility in DMSO, the presence of the expected molecular ion in liquid chromatography-mass spectrometry, and >90% purity by evaporative light scattering or ultraviolet absorbance at 214 nm. On arrival, plates were stored in desiccators (<20% relative humidity) at −25°C. After storage for approximately 3 months, plates were allowed to warm to room temperature in desiccators prior to use in HTS assay.
The yeast growth assay was designed to determine turbidity of the cell suspension after overnight growth by measurement of A 595. A serial dilution of yeast cells was used to determine the linear range of absorbance values as a measure of cell density. Yeast cell stock frozen at −80°C was thawed and serially diluted in a clear 384-well plate in YPD medium to give 50 μl per well of dilutions ranging from 1:5 to 1:500. A 595 was measured using an Envision spectrophotometer (PerkinElmer, Boston, MA).
The optimal time to incubate the yeast suspensions in the HTS assay prior to recording growth was selected by determining a growth time course of over 24 h. For each time point a clear polystyrene 384-well plate was filled with 25 μl per well of YPD medium, and wells designated as blanks were filled with an additional 25 μl of medium in place of yeast cells. Thawed wild-type and mutant yeast cells were diluted 1:500 in YPD medium, and 25 μl of one of the diluted stocks was inoculated to the appropriate wells. Plates were sealed with Breathe-Easy membranes, and the yeast cultures were incubated at 30°C for 4, 8, 16, or 24 h. After incubation, the Breathe-Easy membranes were removed, and A 595 was measured.
To determine the effect of DMSO, both wild-type and mutant yeast cells were grown for 16 h in the presence of 0–1.6% (vol/vol) DMSO. The growth assay was as described above, except that DMSO was mixed with medium in the assay plate prior to the addition of yeast cells. Wells without added DMSO were included as controls.
The assay protocol followed here is based on one described previously.27
Absorbance assay to record yeast growth. For mutant yeast strain HTS, V-bottom polypropylene 384-well plates were filled with 25 μl of YPD medium using a Multidrop-384 reagent dispenser (Thermo Scientific, Franklin, MA), and an additional 25 μl of YPD medium was added to column 1. Compounds were added by pintool transfer on an EP3 workstation (PerkinElmer). A 384-pin pintool (V&P Scientific, San Diego, CA) was used to transfer 120 nl of a 10 mM solution in DMSO into the medium-containing V-bottom growth plates, giving a final concentration of 24 μM of each compound in 0.24% DMSO. Wells in columns 1, 2, 23, and 24 were reserved for controls and blanks and contained 0.24% DMSO but no compound. Thawed mutant yeast cells were diluted 1:500 in YPD medium, and 25 μl was inoculated in all columns except 1 and 23 in the V-bottom growth plates using a Multidrop Micro reagent dispenser (Thermo Scientific). Likewise, 25 μl of diluted wild-type yeast cells was inoculated in column 23 of each V-bottom plate. Plates were sealed with Breathe-Easy membranes, and the yeast cultures were incubated for 16 h at 30°C. Following growth the cell suspension was mixed on the EP3 workstation by 10 aspirate–dispense cycles using 30-μl disposable tips, in which 25 μl of suspension was aspirated from 0.1 mm above the well bottom at a rate of 30 μl/s. Placing the tips as close as possible to the well bottom ensured adequate mixing of cells, while avoiding bubble formation. Immediately following mixing, 5 μl of the cell suspension was used for the luminescence assay (see below). Thirty microliters of the portion that remained was transferred to a clear 384-well plate, and A 595 was recorded.
Beta-Glo luminescence assay to record β-galactosidase activity. Twenty microliters per well of Z-buffer (82 mM disodium hydrogen phosphate, 9 mM sodium dihydrogen phosphate, and 0.1% sodium dodecyl sulfate) was added to white 384-well plates using a Multidrop Micro reagent dispenser, followed by 5 μl of cell suspension by tip transfer from V-bottom yeast growth plates (see above). After cells and Z-buffer were mixed and incubated for 20 min at room temperature to permeabilize the cells, Beta-Glo reagent (25 μl) was added using a Multidrop Micro reagent dispenser, and luminescence was measured on an Envision plate reader after a further 90-min incubation at room temperature to allow the luminescence to reach a steady state.
The protocol for wild-type yeast strain HTS was identical to that described above for the mutant strain except that wild-type yeast was added to all columns except 1 and 23 and mutant yeast was added to column 23 only.
This assay was identical to the HTS assay, except that 16 twofold dilutions of each compound were tested to give a final concentration range of 100 μM to 3 nM in 1% DMSO. Compounds were twofold serially diluted in DMSO as follows: hits selected from HTS were resupplied by BioFocus DPI as frozen 10 mM solutions in 23 μl of DMSO. Compounds were arranged 20 per plate in wells A3–A22 of 384-well V-bottom polypropylene plates. For dose–response testing frozen plates were allowed to warm to room temperature in desiccators prior to serial dilution using an EP3 pipetting workstation. DMSO (23 μl) was added to wells A1, A2, A23, and A24, followed by 11 μl per well of DMSO added to the entire plate except row A using a Multidrop Micro reagent dispenser. The compounds were twofold serially diluted by transfer of 11 μl row-by-row from row A to row P using a single row of disposable tips on an EP3 384-tip pipetting head, after which 11 μl was discarded from row P. The resulting dose–response plates contained 16 twofold dilutions (10 mM–305 nM) of each compound, arranged one compound per column in columns 3–22. Plates were heat-sealed and stored at room temperature for 1–2 days in desiccators prior to use.
Data were analyzed using ActivityBase (IDBS, Guildford, UK). Each HTS plate contained compounds (25 μM in 0.24% DMSO) in columns 3–22, controls (wild-type cells for wild-type screen and mutant yeast cells for mutant screen) in columns 2 and 24, reference yeast strain (wild-type cells for mutant screen and mutant cells for wild-type screen) in column 23, and blanks (YPD medium) in column 1. Percentage inhibition of growth or percentage inhibition of luminescence was calculated for each compound from the signal in absorbance units or luminescence units, respectively, and the mean of the plate controls and the mean of the plate blanks using Eq. 1:
The reference yeast strain (column 23) is not used in the calculation. It was used to monitor consistency of cell growth throughout the screen. Percentage inhibition of luminescence was corrected for cell growth using Eq. 2:
Results were retrieved from ActivityBase using the SARgen data retrieval tool, arranged in Excel (Microsoft, Redmond, WA), and deposited into PubChem.29
Dose–response plates contained one compound per column in columns 3–22, controls (wild-type cells for wild-type dose–response plates and mutant yeast cells for mutant dose–response plates) in columns 2 and 24, reference yeast strain (wild-type cells for mutant dose–response plates and mutant cells for wild-type dose–response plates) in column 23, and blanks (YPD medium) in column 1. Each column from 3 to 22 contained 16 twofold dilutions of a single compound, ranging in concentration from 100 μM to 3 nM. IC50 and EC50 values were calculated in an Excel-based ActivityBase template as follows. Percentage activity was calculated for each dilution of each compound from the signal in luminescence units and the means of the plate controls and plate blanks using Eq. 3:
In the case of inhibitors of luminescence, no dose-dependent effect on growth was observed, so luminescence assay dose–response curves were fit, and IC50 values were calculated without correction for cell growth. Conversely, some of the enhancers of luminescence showed a dose-dependent effect on growth, so the luminescence assay percentage activity values at each compound dose were corrected for cell growth as shown in Eq. 4:
Dose–response curves of percentage activity were fit in XLfit (IDBS), using a four-parameter logistic fit (Eq. 205 with maximum percentage activity and minimum percentage activity fixed at 100 and 0, respectively). Results were retrieved from ActivityBase using the SARgen data retrieval tool, arranged in Excel, and deposited into PubChem.30
The goal of the screening described here was twofold: first, to identify compounds that restore the function of mutant eIF2B in yeast cells; and second, to find activators of GCN4 in the presence of functional wild-type eIF2B. A reduction in eIF2B function reduces the overall translation rate and tends to cause a reduced rate of growth. It additionally activates translation of the stress-response gene GCN4, so that the strain expressing the reduced-function mutant eIF2B has elevated Gcn4p protein levels and elevated activation of GCN4 target genes, including HIS4. 23,27 Thus any compound that restores mutant eIF2B activity should selectively reduce GCN4 activation in the mutant yeast strain. Conversely, compounds that activate GCN4 in an eIF2B-dependent manner should give selective enhancement of reporter gene activity in the wild-type strain.
Modifiers of amino acid metabolism are known to influence signaling to eIF2B/GCN4. For example, 3-aminotriazole is an inhibitor of imidazoleglycerol-phosphate dehydratase, encoded in yeast by HIS3, which catalyzes the sixth step in histidine biosynthesis. 3-Aminotriazole and other compounds activate the protein kinase Gcn2p. To avoid identifying known or other activators of Gcn2p, which would be indirect modulators of eIF2B, the strains used here were deleted for GCN2. In addition, amino acid analogs that can be incorporated into proteins are toxic when yeast are grown in minimal medium. GCN4-mediated activation of amino acid biosynthetic enzymes can overcome this toxicity.23 To avoid selection of toxic amino acid analogs, yeast strains were grown in rich medium (YPD) for this analysis.
Compounds were screened in parallel against two yeast strains, wild-type S. cerevisiae and an eIF2Bε-R284H mutant that is equivalent to the clinically relevant human eIF2B5-R299H mutation. Both mutant and wild-type strains contain a chromosomally integrated gene fusion of HIS4 to β-galactosidase, allowing GCN4 activation to be monitored using a coupled β-galactosidase–firefly luciferase readout adapted from commercially available reagents.27 Constitutive down-regulation of eIF2B activity and hence elevated GCN4 activity are often accompanied by slow growth.10,23 In this work, loss of eIF2B function in the mutant strain was observed to cause a modest reduction in overnight yeast growth (Figs. 1 and and2).2). However, this change was not statistically significant in the presence of 0.25% DMSO, the DMSO content selected for HTS (Fig. 3). Compounds that restore the function of eIF2B in the mutant cells therefore were expected to modestly enhance the growth rate of the mutant yeast strain. Conversely, activators of GCN4 in the wild-type strain may slow its growth. Thus an HTS strategy was designed in which compounds were screened for their effects on both overnight yeast growth and β-galactosidase activity. The two sets of results were combined by correcting the β-galactosidase data to take account of each compound’s effect on growth. In the case of the mutant eIF2B strain this correction was used to reveal compounds that enhanced growth to such a degree that inhibition of β-galactosidase expression was masked. These compounds would show little apparent uncorrected inhibition of β-galactosidase expression due to the increased cell number relative to the plate controls. This analysis also allowed us to exclude toxic compounds that showed apparent inhibition of β-galactosidase activity that was in fact due the lower cell number resulting from growth inhibition or cell killing. In the case of compounds activating GCN4 in the wild-type strain, growth correction revealed enhancement of luminescence that would otherwise have been masked by the slowed growth that often accompanies GCN4 activation.
Before embarking upon HTS, it was necessary to ensure that robotic handling would allow accurate and precise measurements of yeast cell growth and reporter activity and to assess whether the strains were compatible with DMSO, the vehicle used to deliver test compounds.
Growth assay. To allow sufficient time to discern test compound effects on yeast growth, the assay was designed to measure cell density after overnight incubation with each compound. In addition to the importance of ensuring that the yeast cells remained in an exponential growth phase, it was also necessary to confirm the linearity of the absorbance readings used to measure cell turbidity. Above a certain threshold, light scattering causes the correlation between cell density and absorbance to deviate from linearity.31 A standard curve of serially diluted yeast stock showed that absorbance values up to approximately 0.4 are directly proportional to cell density (Fig. 2A). Overnight growth curves confirmed that 16-h growth of a 1:1,000 dilution of the frozen yeast stocks gave absorbance values within the linear range of detection (Fig. 2B). Sensitivity of yeast cell growth to DMSO was also assessed as several reports have described effects of DMSO on yeast cells.32 As shown in Fig. 3 there was no significant effect on growth at 0.25% DMSO, but a reduction of almost 25% in the presence of 1% DMSO. Interestingly, growth of the mutant yeast was consistently repressed more strongly than that of wild-type yeast. To prevent inconsistencies and false-positives due to growth suppression, we limited the DMSO concentration in the HTS to 0.25%.
β-Galactosidase assay. Measurement of β-galactosidase activity after overnight yeast growth showed that, as previously observed,27 the mutant eIF2B strain gave at least twofold higher luminescence than the wild-type (Fig. 4). The rate of light output reached a steady state after 90 min at room temperature, allowing for a single end-point read after this time in the HTS. The observed difference between mutant and wild-type luminescence defined the assay signal window in the screen for compounds that restore function to mutant eIF2B; the goal was to identify compounds that reduced the mutant strain β-galactosidase activity towards the lower activity of the wild-type strain. Conversely, an increase in β-galactosidase activity in the wild-type strain was expected to reveal compounds that activate GCN4 in the presence of wild-type eIF2B. Both yeast strains were included in each HTS plate to monitor the magnitude and consistency of the luminescence values. Thus, one control column in mutant HTS plates contained wild-type yeast, and likewise wild-type plates contained a control column of mutant yeast.
To ensure well-to-well consistency of the luminescence readings, a critical parameter that had to be carefully optimized was the mixing and transfer of the cell suspensions from the yeast growth plate to the β-galactosidase assay plate. After overnight growth the yeast cells needed to be resuspended to ensure efficiency of transfer of a small volume of cells for the luminescence assay. Mixing in a flat-bottom clear plate using 30-μl tips on the EP3 liquid handling workstation tended to concentrate the cells to the rims of the well; it was found to be extremely difficult to get uniform mixing in these plates. Therefore, we decided to grow cells in V-bottom polypropylene plates. Efficient mixing necessitated slow aspiration of the cell suspension with each pipette tip a fraction of a millimeter above the well bottom, and this in turn required very precise adjustment of the EP3 head to ensure uniform alignment of the tips across an entire 384-well plate. Using V-bottom plates we were able to achieve robust growth and reproducible mixing, leading to consistent results in the luminescence assay as shown in the pilot HTS plate in Fig. 5.
A total of 27,199 compounds were screened once at a concentration of 24 μM against both the eIF2B mutant and the corresponding wild-type yeast strain for their effects on cell growth and expression of the β-galactosidase reporter. Most of the yeast growth plates gave robust Z’-factors, whereas a substantial number of the luminescence assay plates gave a Z’-factor below the 0.5 value that is often used as an indicator of a robust HTS assay (Fig. 6). Based on the mean coefficients of variance of the plate controls, which varied between 14.6% and 20.4% for the four sets of plates, we selected compounds exceeding a threshold of 50% deviation from the plate control mean for further evaluation. The modest Z’-factors suggested that a substantial number of the putative hits would be false-positives and that weak compounds might be lost in the assay noise. Nevertheless, the combination of the 50% hit threshold and the 24 μM screening concentration was expected to reveal actives with the desired IC50 or EC50 values below 10 μM. Analyses of the actives inhibiting mutant luminescence and those enhancing wild-type luminescence are shown in Tables 1 and and2,2, respectively.
Screening for inhibitors of reporter gene function. Without correction for cell growth, 355 actives inhibited mutant luminescence, a hit rate of 1.35% (Table 1). Assuming a normal distribution of the data, 108 compounds at random (0.4%) would be expected to exceed the 50% inhibition threshold, based on this threshold being 2.7-fold greater than the SD of the HTS data.33 Therefore, even taking into account the expected scatter of the data, the mutant luminescence assay gave a relatively large number of hits. After correction for cell growth, this number increased to 540. Although 75 compounds appeared to be toxic based on their inhibition of yeast growth and were removed from the original list of 355, a further 260 compounds in addition to the original 355 showed enhanced cell growth such that growth correction revealed inhibition of luminescence, so these were included as active. Similar analysis of the wild-type luminescence assay used as the counterscreen revealed 448 compounds that gave >30% inhibition after correction for the effect of each compound on wild-type cell growth. Comparison of the mutant and wild-type lists of actives revealed 146 compounds that inhibited luminescence in both the mutant and wild-type strains; exclusion of these compounds left 394 hits that appeared selective for the mutant strain. Analysis of activation results (Table 2) revealed 420 compounds that gave >50% enhancement of wild-type luminescence prior to growth correction. Growth correction increased this number substantially, but a number of compounds caused severe growth inhibition suggestive of toxicity. Compounds that caused >50% inhibition of growth were judged to be toxic and excluded from further analysis, leaving 985 compounds that gave 50% activation in the wild-type strain after growth correction. Of these, 896 appeared to be selective for the wild-type and were selected for retesting. The 1,290 actives, consisting of 394 inhibitors of mutant luminescence and 896 activators of wild-type luminescence, from the consolidated list were tested in a dose–response assay consisting of 16 twofold dilutions starting at a top concentration of 100 μM. As in the HTS, overnight cell growth and β-galactosidase expression was determined with both mutant and wild-type yeast.
Only 17 out of 394 putative inhibitors of mutant strain luminescence gave an IC50 <100 μM against the mutant strain, a retest rate of 4.3%. Interestingly, 13 of the confirmed actives were compounds from the original 355 mutant luminescence actives prior to correction for cell growth, and only four additional compounds were found in the additional 260 compounds selected from the HTS as inhibitors of mutant luminescence after correction for apparent enhancement of cell growth. Furthermore, none of the additional 260 compounds showed concentration-dependent enhancement of overnight cell growth, suggesting that the elevated cell number values in the HTS did not represent reproducible enhancement of growth. Thus the luminescence HTS results prior to growth correction were in fact more reliable predictors of activity (or lack thereof) on dose–response confirmation than the growth-corrected results. Testing of the putative mutant-selective luminescence inhibitors in dose–response against the wild-type strain revealed that all the confirmed active compounds gave similar IC50 values against both mutant and wild-type cells; thus none of the compounds was selective for the eIF2B mutant. Results are shown in Table 3, and IC50 curves for the most potent compounds are given in Fig. 7. To ensure that compounds acting selectively on mutant eIF2B had not been missed in the hit selection process, two more groups of samples were tested in dose–response: first, compounds that were actives in the mutant luminescence assay before correction for cell growth but were just under the 50% inhibition threshold after correction; and second, compounds that appeared to be potent but nonselective inhibitors in the original HTS. All of the confirmed actives from these selections showed similar IC50 values against the mutant and wild-type strains. Table 3 also includes luciferase enzyme inhibition data reported by the San Diego Center for Chemical Genomics (SDCCG).34 Comparison of our results with the biochemical assay data from the SDCCG shows that the activity of all but five of the confirmed hits in the cell-based screen may be accounted for by inhibition of the luciferase enzyme in the Beta-Glo reagent. The fact that the active compounds are luciferase inhibitors accounts for their equal potency in both wild-type and mutant screens.
Screening for activators of reporter gene function. Out of 896 putative activators of wild-type luminescence, only two showed dose-dependent enhancement of luminescence, a retest rate of 0.2%. Both confirmed actives were compounds from the original 420 actives prior to correction for cell growth; none was confirmed from the additional 476 compounds selected from the HTS as activators of wild-type luminescence after correction for inhibition of cell growth. Both actives showed modest concentration-dependent inhibition of overnight cell growth, up to 19% and 10% inhibition, respectively, at the highest concentration tested (100 μM). Therefore, EC50 values were calculated from a curve fit to the data after correction for cell growth. Both active compounds were confirmed to be selective enhancers of luminescence in the wild-type strain, suggesting that they are in fact activating GCN4 via an eIF2B-mediated mechanism. Results are shown in Table 4, and EC50 curves are given in Fig. 8. Inspection of the luciferase enzyme inhibition data from the SDCCG34 showed that neither of the confirmed enhancers had any significant effect on luciferase enzyme activity.
The screening described here achieved our second goal, to find activators of GCN4 in the presence of functional wild-type eIF2B, but not our first, to identify compounds that restore the function of mutant eIF2B in yeast cells. Overall, HTS and dose–response confirmation successfully identified two compounds that appeared to cause a stress response resulting in GCN4 activation, but none that reduced GCN4 activation by selectively restoring mutant eIF2B function.
Preliminary studies confirmed our expectation of a high inhibition rate for the β-galactosidase reporter assay. The parallel wild-type assay thus served as a counterscreen to aid in the identification of specific inhibitors of the mutant eIF2B yeast strain by eliminating nonspecific hits such as luciferase inhibitors or modulators of the expression or activity of β-galactosidase working by a non–eIF2B-related mechanism. Conversely, the mutant eIF2B assay served as a counterscreen to identify compounds that selectively enhance wild-type luminescence by eliminating nonspecific enhancers of the luminescent readout.
In the search for compounds that selectively restore mutant eIF2B function, no mutant growth enhancement was observed on dose–response testing, and all the compounds inhibiting mutant luminescence were equally potent against wild-type. The fact that we found active compounds with IC50 values ranging from 2 to 100 μM against both mutant and wild-type suggests that the HTS would have identified selective compounds if they were present in the compound collection. Given the complex biology, the chances of finding a compound capable of restoring function to mutant eIF2B may be very low, necessitating the screening of several hundred thousand compounds. A successful outcome from such a large screen requires an efficient process of elimination to focus attention on hits with the desired activity profile. In this regard the HTS reported here was not as efficient as originally expected. A sizeable number of compounds appeared from the HTS to be selective but were found to be nonselective on dose–response testing (Table 3). Even compounds such as 2 and 3 that gave 75–85% inhibition of mutant luminescence and 0–15% inhibition of wild-type luminescence (with or without growth correction) gave almost identical IC50 values on retesting. Furthermore, the retest rate for both inhibition of mutant strain luminescence and enhancement of wild-type luminescence was very low. In both cases <5% of the compounds that exceeded the threshold of 50% deviation from the HTS controls were active in dose–response testing.
The luminescence inhibition results suggest that, despite considerable optimization effort, the assay is not sufficiently robust to allow for reliable predictions to be made from single HTS data points obtained at one compound concentration. Such single-point data are usually sufficient when screening one parameter; true actives can be identified among a sizeable number of false-positives. However, in a complex multiparameter study such as that described here elimination of apparently selective false-positives at the HTS stage may require replicates at a single concentration or testing of multiple concentrations. An alternative is to select a large number of primary hits from a single HTS and postpone correlation of multiple readouts until dose–response testing. Use of the luminescence assay alone (i.e., without data correction based on a parallel growth assay) as the primary HTS might allow for an assay design that avoids the cell transfer step on the EP3 workstation. This step was a significant source of assay variability and required very careful optimization. Adequate mixing of the cells with the Beta-Glo reagent would nevertheless remain critical to obtaining robust data; thus growth of the yeast in round-bottom white plates might be required so that the Beta-Glo reagent could be added directly to the yeast culture. Based on the results of our 27,199-compound HTS, screening of a large compound library (300,000–500,000 compounds) against the wild-type and eIF2B mutant yeast strains in separate HTS campaigns would produce many thousands of hits, requiring extensive cherry-picking and dose–response profiling to identify true hits among false-positive compounds. Therefore, it may be more efficient to obtain duplicate data on each compound in parallel HTS to improve the robustness of the correlation of the parallel datasets than to postpone this correlation until after dose–response confirmation.
In wild-type yeast, the two activating compounds enhanced β-galactosidase luminescence linked to GCN4 expression and slowed growth, similar to the behavior of agents such as fusel alcohols that are known to activate GCN4 expression.10 The fact that the luminescence-enhancing compounds were inactive in the yeast strain containing the reduced-activity mutant eIF2B probably reflects the fact that GCN4 is already activated in these cells. As the protein kinase gene GCN2 was deleted from the screen strain, the activator compounds cannot be functioning by stimulating eIF2α kinase activity. Therefore these compounds must be acting downstream and could be directly targeting eIF2B. Both active compounds appear quite selective for the GCN4-β-galactosidase screen. At the time of writing, compound 26 had been tested in a total of 260 assays reported to PubChem,35 and the only activity confirmed in dose–response testing was an IC50 of 34 μM against the serine protease complement factor C1s.36 From 243 assays37 compound 27 had been reported as an agonist of the dopamine D1 receptor with an EC50 of 4.5 μM 38 and an inhibitor of 15-lipoxygenase with an IC50 of 15.8 μM.39 A search in SciFinder (Chemical Abstracts Service, Columbus, OH) revealed that neither 26 nor 27 had been reported in the literature. Further studies will be required to explore the mechanism of action and utility of these compounds.
Nuzhat Motlekar, Penn Center for Molecular Discovery, Institute for Medicine and Engineering, and Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, Pennsylvania.
Rogerio Alves de Almeida, Faculty of Life Sciences, University of Manchester, Manchester, United Kingdom.
Graham D. Pavitt, Faculty of Life Sciences, University of Manchester, Manchester, United Kingdom.
Scott L. Diamond, Penn Center for Molecular Discovery, Institute for Medicine and Engineering, and Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, Pennsylvania.
Andrew D. Napper, Penn Center for Molecular Discovery, Institute for Medicine and Engineering, and Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, Pennsylvania. Current affiliation: Nemours Center for Childhood Cancer Research, Wilmington, Delaware.
The work reported here was supported by the National Institutes of Health Molecular Libraries Screening Center Network (grant U54HG003915-02) and grant ELA 2005-008C5 from the European Leukodystrophy Association.
No competing financial interests exist.