A High-Content Screening Assay for Small Molecules That Stabilize Mutant Triose Phosphate Isomerase (TPI) as Treatments for TPI Deficiency

Abstract

Triose phosphate isomerase deficiency (TPI Df) is an untreatable, childhood-onset glycolytic enzymopathy. Patients typically present with frequent infections, anemia, and muscle weakness that quickly progresses with severe neuromusclar dysfunction requiring aided mobility and often respiratory support. Life expectancy after diagnosis is typically ~5 years. There are several described pathogenic mutations that encode functional proteins; however, these proteins, which include the protein resulting from the “common” TPI^E105D mutation, are unstable due to active degradation by protein quality control (PQC) pathways. Previous work has shown that elevating mutant TPI levels by genetic or pharmacological intervention can ameliorate symptoms of TPI Df in fruit flies. To identify compounds that increase levels of mutant TPI, we have developed a human embryonic kidney (HEK) stable knock-in model expressing the common TPI Df protein fused with green fluorescent protein (HEK TPI^E105D-GFP). To directly address the need for lead TPI Df therapeutics, these cells were developed into an optical drug discovery platform that was implemented for high-throughput screening (HTS) and validated in 3-day variability tests, meeting HTS standards. We initially used this assay to screen the 446-member National Institutes of Health (NIH) Clinical Collection and validated two of the hits in dose–response, by limited structure–activity relationship studies with a small number of analogs, and in an orthogonal, non-optical assay in patient fibroblasts. The data form the basis for a large-scale phenotypic screening effort to discover compounds that stabilize TPI as treatments for this devastating childhood disease.

Introduction

Triose phosphate isomerase deficiency (TPI Df) is a rare degenerative disorder that typically affects young children. The disease belongs to a family of diseases called glycolytic enzymopathies, although it is unique from other members in the family because it is characterized by severe neuromuscular impairment and often progressive neurologic symptoms. ¹ The typical age of diagnosis for TPI Df is between 1 and 5 years of age, and the most common initial symptom is hemolytic anemia because erythrocytes cannot compensate for defective glycolysis through oxidative metabolism. As the disease progresses, patients experience failure in tissues and organs with high energy requirements, including increased susceptibility to infection, neurological dysfunction, and progressive neuromuscular degeneration. ² The end stage of this disease constitutes severe paralysis and the inability to breathe without assistance from a respirator. TPI Df leads to a drastically reduced lifespan, with death typically occurring within 5 years of the initial diagnosis. At this time, there are no medical treatments for patients with TPI Df, aside from supportive care, antioxidants, and possibly dietary changes in the form of the ketogenic diet. ³

TPI is a glycolytic enzyme known to interconvert dihydroxyacetone phosphate (DHAP) and glyceraldehyde-3-phosphate (G3P), which allows for increased energy yield from glucose metabolism during glycolysis. The enzyme is known for being a perfect enzyme, meaning that the kinetics of the enzyme are limited only by the speed of diffusion of the substrate. ⁴ TPI Df is typically caused by specific missense mutations in the TPI1 gene. The most common TPI Df mutation is the TPI^E105D (aka E104D in legacy publications); patients with this mutation can be homozygous TPI^E105D/E105D or compound heterozygous, typically with TPI^E105D and another pathogenic missense mutation resulting in disease. ⁵ These genetic mutations primarily affect the stability of mutant TPI protein. ⁶ While the mutant protein dimer remains catalytically active and functional, it has been shown for several pathogenic mutations that the mutant protein is targeted for degradation through the protein quality control (PQC) pathway, leading to very low levels of TPI within the cell. ⁷ Specifically, both heat shock protein 70 (Hsp70) and heat shock protein 90 (Hsp90) have been demonstrated to play a role in mutant TPI degradation in an animal model of TPI Df. ⁸ In the absence of an established rodent model of TPI Df, which is currently under development in our laboratory, the animal model most extensively studied has been the fruit fly, Drosophila. The TPI gene is highly conserved in Drosophila, its active site and enzymatic mechanism are identical to those of mammals, and glycolysis is highly conserved across organisms, functioning identically in humans, mice, and flies. Drosophila expressing mutant TPI exhibit phenotypes similar to those observed in humans with TPI Df, such as shortened lifespan, locomotor dysfunction, and neuropathology.^7–9

Seminal studies in Drosophila have documented that genetic [by RNA interference (RNAi)] and pharmacological interventions (by inhibitors of the proteasome and Hsp70 and Hsp90) result in elevated TPI levels and reverse the detrimental effects of mutant TPI. ⁸ Hsp70 and Hsp90 are, however, ubiquitously expressed and interact with a wide variety of client proteins and co-chaperones to fulfill their cellular roles. ¹⁰ Although numerous drugs modulate Hsp70, Hsp90, and the proteasome, these drugs are all exceptionally toxic, and currently their clinical significance centers on their intrinsic toxicity and ability to sensitize cells to chemotherapies. We have recently used mutant TPI in a powerful Drosophila genetic screen to elucidate all (or at least most) of the PQC components that regulate TPI turnover. ¹¹ We performed a genome-wide RNAi screen of all known and predicted components of the ubiquitin–proteasome system (UPS) and PQC pathways and numerous unknown proteins with domains similar to those commonly found in UPS and PQC components. Importantly, the screen identified ~27 proteins that are critical to TPI stability (including many that were not previously known to be involved in its turnover), novel proteins with no known or associated PQC function for any substrate, including some that could contribute to protein folding or ubiquitination, or are likely to be involved in co-translational PQC. ¹¹ The majority of these potential targets are known to be much less promiscuous than Hsp70 and Hsp90, which are known to modulate >75% of the proteome, suggesting that discovery of less toxic agents that stabilize mutant TPI should be feasible. ¹¹

In an effort to develop effective and nontoxic treatments for TPI Df, we developed a high-throughput phenotypic screening assay (which captures all mechanisms that govern TPI stability) to identify novel pharmacological regulators of mutant TPI degradation. We developed a modified HEK-293T cell line that expresses mutant TPI^E105D protein fused to a green fluorescent protein reporter (TPI^E105D-GFP). The assay is a phenotypic screen enabling the unbiased assessment of novel compounds as modulators of TPI stability based on a cellular fluorescent readout, which would include compounds that slow the degradation of mutant TPI in human cells. The assay was initially tested by administration of PQC inhibitors that targeted known regulators of mutant TPI and were thus predicted to increase mutant TPI levels, resulting in fluorescence intensity changes in HEK cells stably expressing TPI^E105D-GFP. Once the assay was validated, the National Institutes of Health (NIH) Clinical Collection of compounds was screened in the mutant TPI^E105D-GFP HEK cells to investigate any potential pharmacological regulation of mutant TPI levels, focusing on drugs that have a history of use in human clinical trials. Some of the hits from this pilot screen were further investigated using human patient cell lines and showed increases of endogenous mutant TPI. The assay performed to accepted HTS standards in multiday variability studies and is ready to be used in large-scale HTS. Insights gained from this high-throughput screen may allow for the development of new treatment options for patients with TPI Df.

Materials and Methods

Generating and Culturing HEK293 TPI^E105D-GFP Cells

HEK293 cells expressing a mutant (E105D) TPI-GFP fusion protein were created using the Flp-In System (cat. nos. K6010-01 and K6010-02, Invitrogen, Waltham, MA) following the manufacturer’s protocol. Briefly, a host cell line expressing the transfected Flp-In target site vector, pFRT/lacZeo, was obtained following zeocin-resistant selection. TPI^E105D-GFP fusion construct was cloned into the pcDNA 5/FRT expression vector. Co-transfection of 0.1 µg of the pcDNA 5/FRT construct (390 ng/µL) and 0.9 µg of the Flp recombinase expression vector (pOG44, 127 ng/µL) into the Flp-In host cell line was performed with 3.75 µL Lipofectamine 2000 reagent (Invitrogen). A hygromycin dose–response was performed to determine the minimum concentration of hygromycin B required to kill untransfected host cells. Clones of cells were selected that exhibited hygromycin B resistance. Expression of the TPI^E105D-GFP fusion protein was verified via Western blot. The cells were cultured using standard methods (37 °C, 5% CO₂) in normal media [Dulbecco’s modified Eagle medium (DMEM) with 10% serum, 100 U penicillin/100 µg streptomycin/ml; Lonza, Basel, Switzerland]. To document subcellular localization of mutant GFP-tagged TPI, nuclei were stained with Hoechst 33342 and imaged on a PerkinElmer Opera Phenix confocal high-content reader (PerkinElmer, Waltham, MA) in the ultraviolet (UV) and GFP channels, respectively. A z-series of images (63× water immersion objective, 90 planes, 0.3 µm z-steps) was acquired in both channels, and the color combined stack was subjected to three-dimensional (3D) reconstruction in MetaXpress (Molecular Devices, San Jose, CA). A single slice of the reconstructed image was chosen to illustrate the phenotype of mutant TPI-GFP expression. In addition, a 3D rendering was created in Harmony 5.0 (PerkinElmer).

NIH Clinical Collection

The NIH Clinical Collection is a plated array of 446 small molecules that encompasses US Food and Drug Administration (FDA)-approved drugs and agents that have a history of use in human clinical trials. The collection was assembled by the NIH through the Molecular Libraries Roadmap Initiative as part of its mission to enable the use of compound screens in biomedical research and is maintained as assay-ready daughter plates at the University of Pittsburgh Drug Discovery Institute (UPDDI). The library is maintained under temperature- and humidity-controlled conditions as recommended.^12,13 A full list of compounds in the library is presented in Supplemental Table 1 .

Additional Compounds

A full list of compounds used for the prescreen and for follow-up studies is presented in Supplemental Table 2 .

Assay Development and Prescreen with Selected PQC Inhibitors

HEK293 TPI^E105D-GFP cells were plated (50,000 cells/well in collagen-coated 96-well thin-bottom microplates; cat. no. 655956, Greiner Bio-One, Monroe, NC) and allowed to attach overnight. Cells were treated with four-point, twofold gradients of 5× treatment solutions of test agents (six replicates per condition). The final concentration of vehicle (DMSO) was 0.2%. After 48 h in culture, cells were preincubated with propidium iodide (PI) for 30 min and imaged live in the GFP and Texas Red channels on an ImageXpress Ultra high-content reader (Molecular Devices) using a 20× objective. Images were analyzed with the multiwavelength cell-scoring application using GFP to define and enumerate cells. The integrated GFP intensity per cell was used to quantify mutant TPI expression, the number of cells was used as a proxy for growth inhibition and/or cell loss, and the percentage of cells that exceeded a threshold for PI based on vehicle-treated controls was used to measure acute toxicity (necrosis).

Flow Cytometry

HEK293 TPI^E105D-GFP cells were plated (50,000 cells/well) in the wells of a 96-well plate (100 µl) and treated with 25 µl of a 5× treatment solution of luminespib for 48 h. Cells were detached by adding 25 µl 17.5 mM ethylenediaminetetraacetic acid (EDTA), pH 6.15, for 2 h ¹⁴ directly to growth medium (final concentration: 2.5 mM). Plates were subsequently stained with 25 µl PI [5 µg/ml in phosphate-buffered saline (PBS)], dislodged by pipetting, and analyzed by flow cytometry. Samples were acquired on a BD Fortessa SORP cytometer (BD Biosciences, San Jose, CA). Dislodged (resuspended) cells were run on the BD High Throughput Sampler (HTS) with the following settings to ensure even sampling: sample flow rate: 1.5 µl/s; sample volume: 80 µl; mix volume: 40 µl (40% total volume); mix speed: 200 µl/s; number of mixes: 4; and wash volume: 400 µl. An attempt was made to collect at least 25,000 events per sample. Forward scatter (FSC), side scatter (SSC), GFP (488_515_30), and PI (561_582_15) detectors were optimized on untreated viable HEK293T cells. Samples were analyzed for expression of GFP and viability (PI-negative events). Several population patterns were observed, and correlation with FSC/SSC revealed the following states: viable (live), early-apoptotic, and late-apoptotic cells.

Pilot Screen Using the NIH Clinical Collection

HEK293 TPI^E105D-GFP cells were plated (10,000 cells/30 µl/well) in collagen-coated 384-well thin-bottom microplates (cat. no. 781956, Greiner Bio-One) and allowed to attach overnight. Assay-ready daughter plates of the NIH Clinical Collection containing 2 µL of a 1 mM DMSO stock were reconstituted in DMEM (60 µL) to yield a 30 µM intermediate treatment concentration. DMEM containing 3% DMSO, 60 µl was added to columns 1 and 24, and 60 µL of 600 nM luminespib was added to columns 2 and 23, to serve as negative and positive controls, respectively. Aliquots of 15 µl were transferred directly to cells in 384-well plates using an Agilent Bravo liquid-handling robot (final concentration in assay, 10 µM). For the 1 µM condition, an aliquot of the 30 µM treatment solution was diluted 1:10 and controls added as above. Aliquots of 15 µL were transferred to cells to yield a 1 µM final concentration. Plates were centrifuged for 30 s at 200 ×g and incubated at 37 °C. After 24 h, plates were imaged live in the GFP channel and returned to the incubator; after 48 h, cells were stained with PI, imaged live in the GFP and PI channels on the ImageXpress Ultra high-content reader (four fields, 20×, capturing approximately 1000 cells), and analyzed as described above. Some dose–response studies were done on a PerkinElmer Phenix Plus, capturing approximately 4000 cells in four fields with a 20× air objective.

Culturing Patient Fibroblasts

Patient fibroblasts were obtained from a male TPI Df patient via skin punch using the Stanford University Institutional Review Board (IRB; Registration 5135, e-protocol 28342, Dr. G. Enns). The cells were de-identified, are known only as FB104, and are homozygous for the common E105D mutation (aka E104D in legacy publications). Fibroblasts were cultured using standard methods (37 °C, 5% CO₂) in complete media [DMEM with 10% serum, 100 U penicillin/100 µg streptomycin/ml (Lonza), 2 mM L-glutamine (Gibco), and supplemental non-essential amino acids (Gibco, Carlsbad, CA)].

Western Blotting

FB104 patient fibroblasts (E105D/E105D) treated with luminespib (200 nM), resveratrol (100 µM), itavastatin (5 µM), or DMSO (0.1%; Sigma Aldrich, St. Louis, MO) were trypsinized (0.05% for 5 min), pelleted, resuspended in radioimmunoprecipitation assay (RIPA) buffer with protease inhibitors [phenylmethylsulfonyl fluoride (PMSF; 100 µM), leupeptin (1 µg/µL), and pepstatin A (0.5 µg/µL)], and pulse sonicated. Protein concentrations were determined using a bicinchoninic acid (BCA) assay (Pierce, Waltham, MA). Immunoblotting was performed on whole-cell protein lysates following the addition of an equal volume of 2× sodium dodecyl sulfate–polyacrylamide gel electro-phoresis (SDS-PAGE) sample buffer (4% SDS, 4% β-mercaptoethanol, 130 mM Tris HCl pH 6.8, and 20% glycerol). Proteins were resolved by SDS-PAGE (12%) and transferred onto 0.45 µm polyvinylidene fluoride (PVDF) membrane. The blots were blocked in Odyssey Blocking Buffer (LI-COR, Lincoln, NE) and incubated with anti-TPI (1:5000; rabbit polyclonal FL-249; cat. no. sc-30145, Santa Cruz Biotechnology, Santa Cruz, CA) or anti–beta tubulin (1:1000; mouse polyclonal E7-C; Developmental Studies Hybridoma Bank, Iowa City, IA) diluted in Odyssey Blocking Buffer. Following washes in phosphate-buffered saline with Tween 20 (PBST), the blots were incubated with anti-mouse-IR800 (Fisher Scientific, Waltham, MA) and anti-rabbit-IR680 (Molecular Probes, Eugene, OR) both diluted to 1:20,000 in 0.1% Tween 20 blocking buffer. Blots were washed in PBST and developed using the Odyssey Infrared Imaging System (LI-COR). Quantification of the scanned images was performed digitally using the Image Studio Version 5.2 software (LI-COR). TPI levels were normalized to beta tubulin loading control, and differences in TPI expression were evaluated by a two-tailed Student t test.

Three-Day Variability

Two full 384-well microplates of HEK293 TPI^E105D-GFP cells (10,000 cells/well) were treated with vehicle (1% DMSO) or luminespib (200 nM) on 3 consecutive days using a PerkinElmer MDT (Modular Dispense Technology) or Agilent Bravo liquid handler (Agilent, Santa Clara, CA). After 48 h of incubation, cells were imaged live in the GFP channel as described above and analyzed for TPI-GFP expression using the MetaXpress multiwavelength scoring application. Assay performance was assessed between plates and between days by calculating signal-to-background (S/B) ratios, coefficients of variation (CVs), and Z-factors, as described. ¹⁵

Results

The “common” TPI^E105D mutation is well characterized biochemically and structurally, ¹⁶ and it results in a ~50% reduction in steady-state protein levels in patient cells. ¹⁷ We developed a HEK TPI^E105D-GFP stable knock-in cell line using the Thermo Fisher Flip-In system (Thermo Fisher, Waltham, MA). We previously demonstrated that in compound heterozygotes, the degradation of each mutant protein indeed occurs independently; ¹⁸ thus, mutant GFP-tagged TPI degradation will not be significantly affected by wild-type TPI that keeps the cells healthy. HEK TPI^E105D-GFP cells provide a convenient optical assay platform of steady-state mutant human TPI protein in the context of a human cell. Figure 1 documents stable and uniform expression of mutant GFP-tagged TPI that is consistent with a ubiquitous nuclear and cytosolic localization, suggesting it could serve as a discovery assay for compounds that increase levels of mutant TPI.

Figure 1.

Creation of knock-in cell line. (A,B) Confocal microscopy of stable human embryonic kidney (HEK) cells expressing the common triose phosphate isomerase deficiency (TPI Df) protein fused with green fluorescent protein (HEK293 TPI^E105D-GFP) documents predominantly cytosolic localization of mutant TPI. Images are three-dimensional (3D) reconstructions of a z-series acquired on an Opera Phenix High-Content Screening reader (PerkinElmer) using a 63× water immersion objective. Green, Green fluorescent protein; red, nuclei. (A) and (B) are 3D reconstructions in MetaXpress (Molecular Devices); image in (B) is a blow-up of selected cells in (A). Scale bars: 70 µm and 20 µm, respectively. (C) A 3D rendering in Harmony 5.0 (PerkinElmer) of a z-series of images acquired with a 63X water immersion objective. (D) A representative Western blot of TPI in a sample of the stable Flip-In (Invitrogen) HEK293 cell line. Beta tubulin was used as a loading control. Sizes of molecular-weight markers are in kDa.

Assay Development

We first tested a panel of inhibitors directed at proteins known to target mutant TPI for degradation in Drosophila and examined their effect on HEK TPI^E105D-GFP cells via confocal microscopy. Consistent with Drosophila results, bortezomib, cycloheximide, and luminespib all showed a significant increase in TPI^E105D-GFP fluorescence, demonstrating that TPI^E105D turnover in human cells is similarly regulated as TPI^{sugarkill (sgk)} protein in Drosophila, as well as confirming the utility of the reporter system ( Fig. 2A ). As expected, all agents in this class significantly inhibited cell growth (data not shown) and displayed some acute toxicity as measured by PI staining ( Fig. 2A ). Of the seven agents tested, luminespib produced the highest-fold increase in GFP with the lowest toxicity (Fig. 2A and 2B). Due to the toxicity, we wanted to ensure the increase in GFP was the result of more TPI^E105D-GFP in the live cells and not an artifact of the dying cells. Fluorescence micrographs indicated that PI-positive cells expressed no GFP ( Fig. 2C ). To confirm these results and to eliminate the possibility that morphological changes could contribute to the enhanced GFP levels, we performed flow cytometry, which measures total protein content independent of cell morphology. Flow cytometry confirmed the increase in TPI^E105D-GFP and revealed that the detachment method caused cells to enter early apoptosis ( Fig. 2D , DMSO scatterplot). Luminespib accentuated apoptosis, as evidenced by a shift of cells to late apoptosis ( Fig. 2D , luminespib scatterplot); however, the GFP increase observed was exclusively from the healthy cells, further validating the use of this human TPI^E105D-GFP reporter ( Fig. 2D , histograms). We also performed a pilot test in 384-well plates using a meaningful number of replicates (n = 144) in 384-well plates with luminespib as a positive control, and found that the assay could likely meet universally accepted HTS standards, demonstrating the viability of this screening model and the utility of luminespib as a positive control ( Fig. 2E ).

Figure 2.

Luminespib increases triose phosphate isomerase–green fluorescent protein (TPI-GFP) expression that is quantifiable by high-content analysis. (A) Prescreen of seven putative protein-stabilizing compounds in HEK293 TPI^E105D-GFP identifies luminespib as a strong inducer of TPI-GFP with low cytotoxicity. The x-axis shows the percentage of propidium iodide (PI)-positive cells; the y-axis shows total TPI-GFP intensity per cell; lines represent the mean of vehicle-treated cells. (B) Luminespib shows dose-dependent increases in TPI-GFP with loss of cell attachment but little cell death. Data are mean ± standard deviation (SD) from five independent biological repeats, each performed in quadruplicate. (C,D) Luminespib increases TPI-GFP expression and causes apoptosis in HEK293 cells, but high levels of TPI-GFP do not originate from dying cells. (C) Fluorescence micrographs of TPI-GFP (green) and PI (magenta) from vehicle- and luminespib-treated cells indicate that dead cells do not express TPI-GFP. [Luminespib] = 99 nM. Bar = 30 µm. (D) Flow cytometry of nonpermeabilized but ethylenediaminetetraacetic acid (EDTA)-detached cells stained with PI. Luminespib causes apoptosis, but apoptotic cells are GFP low, consistent with images in (C). Histograms show GFP distribution in the viable cells. Data are from a single well on a 96-well plate; insert shows three technical replicates with mean and SD. Scatterplot is from 30,000 cells treated with (left) DMSO or (right) 99 nM luminespib. (E) Miniaturization and preliminary high-throughput screening (HTS) performance in 384-well plates suggest the assay can meet universally accepted HTS criteria [Z’ > 0.5; coefficient of variation (CV) < 10%; MIN, DMSO; MAX, 100 nM luminespib; n = 144].

Compound Pilot Screen

We conducted a pilot library screen using the NIH Clinical Collection in the HEK TPI^E105D-GFP cell line. The NIH Clinical Collection maintained at the UPDDI contains 446 FDA-approved drugs and small molecules that have a history of use in human clinical trials. The collections were assembled through the Molecular Libraries and Imaging Initiative as part of its mission to enable the use of compound screens in biomedical research. To extract as much information from this library as possible, we screened the library in duplicate at two concentrations (10 and 1 µM) and at two time points (24 and 48 h). Cells were plated in collagen-coated 384-well plates and treated the next day with library compounds in duplicate using the Agilent Bravo pipetting robot. DMSO (1%) and luminespib (200 nM) served as, respectively, negative (MIN) and positive (MAX) controls (n = 32 each). Cells were incubated at 37 °C, scanned on the ImageXpress Ultra high-content reader after 24 h, and returned to the incubator for an additional 24 h. Thirty minutes before the 48 h scan, cells were stained with PI and imaged in the GFP and Texas Red channels to assess TPI^E105D-GFP levels, loss of cell attachment, and acute toxicity (necrosis). Plates were analyzed with the multiwavelength cell-scoring application in MetaXpress; TPI levels were calculated on a per cell basis. Cell densities and TPI^E105D-GFP expression were then normalized to DMSO control.

The performance of the HEK TPI^E105D-GFP cellular assay under conditions of compound screening was excellent ( Fig. 3 ). Total cellular mutant TPI (GFP) was used as the primary screen parameter with luminespib (200 nM) as a positive control. Intraplate S/B values and Z-factors were 1.64 ± 0.11 and 0.30 ± 0.22 on day 1, and 2.16 ± 0.98 and 0.51 ± 0.06 on day 2, respectively. This suggested a 48 h incubation was necessary for optimal assay performance. There was good agreement between replicates (r² = 0.88; data not shown). The 10 µM/48 h condition identified 31 compounds that elevated TPI^E105D-GFP by more than 3 SD greater than the mean of the negative controls (i.e., a pseudo z-score of >3; Fig. 3a , y-axis). Among these were 11 antineoplastic agents ( Fig. 3b , heatmap), which is not surprising because the screen is using transformed cells and the NIH Clinical Collection contains a disproportionately large number of anticancer agents. The majority of these agents could be eliminated by measurements of cell loss ( Fig. 3a , x-axis) and/or PI staining (data not shown), and were excluded from further analysis. The remaining 20 hits contained several clusters, namely, statins (2 compounds), hormones (3), antifungals (2), natural products (3), vasoactives (4), and five singletons with diverse mechanisms of action. Of these, 16 (80%) repeated at the 10 µM condition at the earlier time point, and five repeated in every condition (itavastatin, cerivastatin, artesunate, carvedilol, and milrinone). Compounds that did not repeat were mostly singletons. Although vasoactive agents in general showed high repeat rates, from a translational perspective, they would likely not be useful for the treatment of an early childhood metabolic disease. After these considerations, two classes of agents looked most interesting, statins and natural products, and we followed up on these by investigating resveratrol and itavastatin.

Figure 3.

Pilot screen of the National Institutes of Health (NIH) Clinical Collection. The NIH Clinical Collection was screened in human embryonic kidney (HEK) cells expressing the common triose phosphate isomerase deficiency (TPI Df) protein fused with green fluorescent protein (HEK293 TPI-GFP^E105D cells) at two concentrations and at two different time points. (Left) Scatterplot shows total TPI-GFP intensity per cell, normalized to vehicle control, versus cell loss at the 10 µM, 48 h condition. Data points represent the averages of duplicate plates. Horizontal and vertical lines represent mean + 3 standard deviations (SD) of TPI-GFP expression in DMSO-treated cells and 50% reduction in cell density, respectively. Marked compounds were chosen based on those cutoffs. (Right) The heatmap compares activity of hits against the same concentration at 24 h (column 2), and a lower concentration (1 µM) for both time points (columns 3 and 4). For clarity, only compounds that were active at the 10 µM, 48 h condition are shown; a comprehensive heatmap of all positive compounds that were active at any condition can be found in Supplemental Table 2 . Red, Active; green, inactive.

Hit Validation

Using fresh samples repurchased from commercial suppliers, both agents showed concentration-dependent increases in TPI^E105D-GFP ( Fig. 4a ). A preliminary structure–activity relationship (SAR) study with four statins (lovastatin and mevastatin as examples for natural products, and itavastatin and atorvastatin as examples for synthetics) showed minor differences in potency, with itavastatin being the most potent (EC₅₀ = 1.63 µM; see table in Fig. 4 ). We also tested a synthetic polyphenol (hexylresorcinol) and found that it was much less potent than resveratrol (Fig. 4a), indicating that the observed activity of resveratrol is not a result of nonspecific exposure to phenolic compounds.

Figure 4.

Hit confirmation in triose phosphate isomerase deficiency (TPI Df) patient fibroblasts. (A) Ten-point, twofold dose–response in 384-well plates showing differences in EC₅₀ and magnitude of response for two novel inducers of mutant TPI identified from the National Institutes of Health (NIH) Clinical Collection (itavastatin and resveratrol) and selected structural analogs. Table shows EC₅₀ values in µM as mean ± standard deviation (SD) from (n) independent biological repeats. Data in graphs are the mean ± standard error of the mean (SEM) of quadruplicate wells in a 384-well plate. (B) Representative Western blot of mutant TPI protein levels in [E105D/E105D] patient fibroblasts treated with luminespib (200 nM), resveratrol (100 µM), and itavastatin (5 µM). (B, right) Quantification of mutant TPI levels by densitometry. Data show mean ± SD of five independent biological replicates (closed circles). t-test with Welch correction p-values shown compared with DMSO vehicle control.

Based on these results, we performed Western blots using fibroblasts from a TPI^E105D homozygous patient and confirmed elevated mutant TPI levels by resveratrol and itavastatin using an orthogonal non-imaging method ( Fig. 4b ). An important finding was that in the patient cells, the increase in mutant TPI was small but similar for the three agents. In contrast, in HEK cells, luminespib and resveratrol caused an apparently higher maximal magnitude of response compared with the four statins, which was a result of dramatic changes in cell morphology at the higher doses that did not occur as much with statins (data not shown). This further underscores the need for non-imaging-based secondary assays to rule out artifacts due to changes in cellular morphology (i.e., flow cytometry), and the need to implement and optimize microplate-based assays in patient fibroblasts to validate hits, because those cells did not undergo substantial morphological changes (data not shown). Importantly, the small but significant increase of mutant TPI levels in patient cells is enough to anticipate clinical activity (see Discussion).

Multiday Variability Assessments

Based on the assay development and validation results above, we performed a formal 3-day variability study ( Fig. 5 ), as is standard for HTS implementation of high-content screening assays at the UPDDI (e.g., Refs. ^19–22). Two full 384-well microplates of HEK TPI^E105D-GFP cells were plated and treated on 3 consecutive days with 1% DMSO (MIN) or 200 nM luminespib (MAX) using a multidrop bulk liquid dispenser and a PerkinElmer Janus MDT pipetting robot. After a 48 h incubation, cells were imaged on an ImageXpress Ultra high-content reader and analyzed for TPI-GFP expression. Mean, standard deviation (SD), coefficients of variance (CVs), signal-to-background (S/B) ratios, and Z-factors ²³ were calculated for each plate, for each day of experiments, and across days. Scatterplots illustrate day-to-day performance; tables show calculated intraplate and interplate variability statistics. The assay met accepted HTS criteria for intraplate and interday variability on all 3 days (CV < 10%). The mean Z-factor among all 3 days was 0.51 ± 0.03, conforming to accepted HTS standards and documenting HTS readiness.

Figure 5.

Three-day variability. Two full 384-well microplates were plated and treated on 3 consecutive days with 1% DMSO (MIN) or 200 nM luminespib using equipment to be used in high-throughput screening. SD, Standard deviation; CV, coefficient of variance; S/B ratio, signal-to-background ratio. Scatterplots illustrate day-to-day performance; tables show calculated intraplate and interplate variability statistics.

Discussion

Decreased stability of mutant TPI, resulting in lower steady-state levels of TPI protein, is the underlying cause for a rare but devastating childhood disease termed TPI Df. TPI Df causes hematological, muscular, and neurological symptoms that manifest themselves as anemia within months of birth; progressively worsening neuromuscular symptoms, typically before age 2; and neurodegenerative disease. TPI Df patients lead a miserable life and usually die within 5 years of diagnosis. Neither symptomatic nor curative treatments exist.

The structural basis for decreased protein stability is not known; however, a recent genetic screen for genes that regulate TPI stability has suggested that lower stability of mutant TPI is governed by a variety of proteins, including known PQC regulators, but also cotranslational modulators and a variety of proteins with unknown functions. ¹¹ Therefore, there is no basis for developing targeted therapies. Thus, we reasoned that a phenotypic assay for TPI expression would be most appropriate to discover novel TPI Df therapeutics because such an assay would capture all mechanisms that lead to stabilization of mutant TPI.

To this end, we developed an operationally straightforward cellular assay that quantifies mutant TPI protein. We created a stable reporter cell line expressing a GFP-tagged mutant TPI fusion protein using a site-directed attB/attP insertion method (Thermo Fisher Flip-In System). The system allows for generation of stable reporter lines expressing mutant proteins while preserving endogenous protein expression, which is required for essential proteins such as TPI to keep cells alive. The stable line was characterized by Western blot and immunofluoresence, and the cells are viable and healthy. The cells allow levels of mutant TPI to be directly measured by quantitative fluorescence microscopy in living or fixed cells.

We implemented the HEK293 cell line assay for HTS. Assay development documented that the line responded to PQC inhibitors with increased levels of mutant TPI. Luminespib, an Hsp90 inhibitor, was chosen as a positive control. Counterassays were developed to eliminate toxicity and morphology artifacts, and preliminary HTS performance parameters in a miniaturized format suggested the assay could be implemented for HTS. A pilot screen of the NIH Clinical Collection identified several classes of agents that increased expression of mutant TPI, two of which were confirmed in fibroblasts from a homozyzgous TPI^E105D TPI-Df patient. In particular, resveratrol and itavastatin had properties that might make them candidates for repurposing. Resveratrol as a nutritional supplement would represent an almost immediately applicable treatment for current TPI Df patients. A review of the clinical literature revealed that resveratrol had beneficial effects on skeletal muscle, ²⁴ possibly by attenuating oxidative stress in myoblasts. ²⁵ This could be highly relevant, because muscle wasting is a hallmark of TPI Df and oxidative stress has been observed in the Drosophila model of TPI Df. ²⁶ Statins have been associated with modulation of heat shock proteins, ²⁷ although the clinical picture is heterogeneous. In the right context, however, it is conceivable that statins could mediate protein stability through heat shock protein modulation. Importantly, while the doses of statins that are effective in cardiovascular disease are usually far lower than those observed in the laboratory for other biological activities, itavastatin has been shown to reach peak plasma levels in humans that correlate with concentrations needed to increase TPI^E105D-GFP in HEK cells (0.55 µM). ²⁸

How much of an increase in TPI is needed to have a therapeutic benefit or prevent TPI Df altogether? This is not known, and there is not a great deal of evidence that directly addresses this important question. There are limited patient data from asymptomatic controls, but a published case included a heterozygote mother with a reduced TPI of 584 UI/g hemoglobin (Hb; ~25–30% of the adult reference range of 1407–2133) who was completely asymptomatic. ⁵ In contrast, her child had TPI levels of 289 UI/g Hb (~14–20% of the adult reference range) with severe early-onset TPI Df. ⁵ These data suggest that fully rescuing TPI to wild-type levels is not necessary to have an extremely significant therapeutic benefit and that modest increases would be of therapeutic value even if they did not fully prevent all disease symptoms. Future studies will be needed to verify whether these compounds work similarly in compound-heterozygous TPI Df patient cells and to test efficacy in vivo.

Based on these results, we performed a rigorous 3-day variability assessment that documented HTS readiness, and we are now positioned to conduct a large-scale screen with the HEK TPI^E105D-GFP cell line. Positives that emerge from this effort can be tested in a variety of patient fibroblasts that represent the whole spectrum of disease and in a well-validated in vivo Drosophila model.^5,7 These secondary assays will include measurements of mutant TPI half-life by pulse-chase experiments and activity using an enzymatic assay. ¹⁷ Final validation will occur in a mouse model of TPI-Df that is currently being developed and, while not having been fully validated yet, shows hallmarks of the human disease (data not shown). Collectively, our data document that the HEK TPI^E105D-GFP assay meets HTS performance criteria and can identify low-toxicity, bona fide TPI inducers and stabilizers with different mechanisms of action. The critical path for large-scale HTS will encompass complementary orthogonal assays in TPI-Df patient cells, biochemical characterization of protein stability and activity, and in vivo testing in Drosophila and in a mouse TPI Df animal model that we are currently developing. We are hopeful that this effort will result in the first therapeutics for this devastating childhood disease.

Supplemental Material

sj-pdf-1-jbx-10.1177_24725552211018198 – Supplemental material for A High-Content Screening Assay for Small Molecules That Stabilize Mutant Triose Phosphate Isomerase (TPI) as Treatments for TPI Deficiency

Supplemental material, sj-pdf-1-jbx-10.1177_24725552211018198 for A High-Content Screening Assay for Small Molecules That Stabilize Mutant Triose Phosphate Isomerase (TPI) as Treatments for TPI Deficiency by Andreas Vogt, Samantha L. Eicher, Tracey D. Myers, Stacy L. Hrizo, Laura L. Vollmer, E. Michael Meyer and Michael J. Palladino in SLAS Discovery