Translating in vitro CFTR rescue into small molecule correctors for cystic fibrosis using the Library of Integrated Network-based Cellular Signatures drug discovery platform.

Cystic fibrosis (CF) is a lethal autosomal recessive disease caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene. The common ΔF508-CFTR mutation results in protein misfolding and proteasomal degradation. If ΔF508-CFTR trafficks to the cell surface, its anion channel function may be partially restored. Several in vitro strategies can partially correct ΔF508-CFTR trafficking and function, including low-temperature, small molecules, overexpression of miR-138, or knockdown of SIN3A. The challenge remains to translate such interventions into therapies and to understand their mechanisms. One approach for connecting such interventions to small molecule therapies that has previously succeeded for CF and other diseases is via mRNA expression profiling and iterative searches of small molecules with similar expression signatures. Here, we query the Library of Integrated Network-based Cellular Signatures using transcriptomic signatures from previously generated CF expression data, including RNAi- and low temperature-based rescue signatures. This LINCS in silico screen prioritized 135 small molecules that mimicked our rescue interventions based on their genomewide transcriptional perturbations. Functional screens of these small molecules identified eight compounds that partially restored ΔF508-CFTR function, as assessed by cAMP-activated chloride conductance. Of these, XL147 rescued ΔF508-CFTR function in primary CF airway epithelia, while also showing cooperativity when administered with C18. Improved CF corrector therapies are needed and this integrative drug prioritization approach offers a novel method to both identify small molecules that may rescue ΔF508-CFTR function and identify gene networks underlying such rescue.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Although several CFTR modulators have been developed, a substantial treatment gap remains and many patients with cystic fibrosis (CF) would benefit from improved corrector or other targeted therapies. Therefore, there is a need for additional modulator therapies to maximize pulmonary health in people with CF. Several in vitro strategies to correct ΔF508‐CFTR have been identified, including overexpression of miR‐138, knockdown of SIN3A, and low temperature.

WHAT QUESTION DID THIS STUDY ADDRESS?

By querying genomic signatures associated with rescue of CFTR in Library of Integrated Network‐based Cellular Signatures, we aimed to identify small molecules to restore function to the ΔF508‐CFTR protein and elucidate the mechanisms of efficacious molecules.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

We identified eight small molecules that restored partial function to ΔF508‐CFTR in a cell culture model, including XL147, which was shown to rescue ΔF508‐CFTR in cells from four donors. Post hoc analysis indicated that ribosomal stalk proteins may be involved in XL147‐mediated CFTR rescue and subsequent experiments identified several genes, including RPL32, that contribute to CFTR rescue when knocked down.

HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

This study supports a transcriptomic signature approach to identifying and repurposing drugs for the treatment of CF lung disease, while also providing lead compounds and genetic targets for the development of CF therapies.

INTRODUCTION

Cystic fibrosis (CF) is a lethal autosomal recessive disease caused by mutations in the CF transmembrane conductance regulator (CFTR) gene. The most common CFTR mutation, termed ΔF508, causes protein misfolding, resulting in proteasomal degradation. If ΔF508‐CFTR is allowed to traffic to the cell membrane, anion channel function may be partially restored, although the residency and open‐state probability of the mutant protein are reduced. , As ~90% of patients with CF have at least one ΔF508 allele, there is substantial interest from academic and industry laboratories in identifying interventions that might restore function to this misprocessed protein. To this effect, we previously reported that transfection with an miR‐138 mimic or knockdown of SIN3A in polarized primary cultures of CF airway epithelia increased ΔF508‐CFTR mRNA and protein levels, promoted trafficking of the mutant protein to the membrane, and partially restored cAMP‐stimulated chloride conductance. Additionally, Denning and colleagues reported that ΔF508‐CFTR processing reverts toward that of wild‐type CFTR during low temperature incubation.

However, a challenge remains to translate such interventions into therapies and to understand their mechanisms. One approach for connecting such interventions to small molecule therapies that have previously succeeded for CF and other diseases is through the use of mRNA expression profiling and iterative searches for small molecules with similar expression signatures. The Connectivity Map (CMAP) is a comprehensive catalog of gene expression profiles from cultured human cells treated with bioactive small molecules, with pattern‐matching algorithms to mine these data. We previously used a transcriptomics‐based drug discovery approach by using CMAP (build 02) to identify drugs that mimicked the miR‐138 overexpression and SIN3A knockdown treatments. Signatures generated in Calu‐3 epithelia treated with a miR‐138 mimic or SIN3A Dicer‐Substrate Short Interfering RNAs (DsiRNAs) matched signatures of 27 small molecules. We reported the identification of four small molecules that partially restored ΔF508‐CFTR function in primary CF airway epithelia. Of these, pyridostigmine showed cooperativity with corrector compound 18 (C18) in improving ΔF508‐CFTR function, highlighting the utility of a transcriptomic signature approach in drug discovery. CMAP has also been used to successfully investigate multiple cancers, inflammatory bowel disease, and human skeletal muscle atrophy. , , , These studies provide support for the use of connectivity mapping to link gene expression, disease, and small molecule therapeutics.

Build 02 of CMAP, which we used in our initial connectivity mapping study, housed ~7000 gene expression profiles representing over 1300 compounds tested in three cell lines at multiple doses and durations. More recently, the Library of Integrated Network‐based Cellular Signatures (LINCS) was introduced, which now contains over 3,000,000 gene expression signatures collected from the treatment of ~33,000 small molecules and ~9200 genetic perturbagens (shRNA, cDNA, and CRISPR) on over 200 cell types. To produce such high‐scale data, the CMAP team developed the L1000 platform, a rapid high‐throughput gene expression profiling technology, which computationally converts raw fluorescence into gene expression signatures that can be queried using the LINCS software. In this study, we further expanded and improved our transcriptomics‐based drug discovery pipeline by including low temperature treatment signatures in our query and using the more robust LINCS drug discovery platform.

METHODS

Generation of gene sets and queries of the LINCS database

Differential gene expression analysis was performed using the Partek Genomics Suite Gene Expression (St. Louis, MO) workflow to identify the most significant up‐ and downregulated genes. A p value cutoff less than 0.02 and fold change greater than 1.2 gave roughly 200 genes in each category, which were uploaded on to the internal LINCS database, at the time not publicly available. Every reference signature in the database was compared with the query signatures and given a score termed the “connectivity score” based on the extent of similarity between the two. Scores range from +1 (indicating higher similarity) to 0 (no similarity) and to −1 (meaning that the two signatures are opposite of each other). Candidate drugs with a connectivity score nearest to +1 were selected. LINCS also used stringent controls, as well as biological and technical replicates, in all experiments. These included: (1) correlation coefficient across replicates, (2) signal strength measured as the ratio of the top 50 and bottom 50 expressed genes, and (3) correlation of rank, determined by examining the percent correlation of differentially expressed genes between the replicates (Tables [Link], [Link], [Link]). Candidate drugs with a favorable score in all three factors, a positive enrichment score (drugs that most closely mimicked our interventions), and the highest relative frequency (prioritizes the most important hits by controlling for variability in sampling size) were prioritized for further screening. The top 75 compounds for each query were included in the initial list before duplicates and unavailable compounds were removed, resulting in a final list of 107 small molecules.

Small molecules

Small molecules were obtained from the Broad Institute DOS Library (Cambridge, MA, USA), Cayman Chemical (Ann Arbor, MI, USA), or Sigma‐Aldrich (St. Louis, MO, USA). For detailed information on small molecules, see Table S4.

Cultured cells

The CF bronchial epithelial cell line CFBE41o−, hereafter termed “CFBE,” was originally developed by immortalization of human CF airway cells and later stably transduced with a ΔF508‐CFTR expression cassette. These cells were cultured as previously described. , , Primary human airway epithelia were cultured as previously described.

Functional screen of compound activity by conductance assay

Effects of selected compounds were tested using the robotic Transepithelial Current Clamp TECC‐24 assay (EP Design, Bertem, Belgium) with CFBE cells grown on microporous membranes of Transwell plates (Corning, Corning, NY, USA), representing a planar array of 24 Ussing chambers with a 6 mm diameter insert containing the epithelial monolayer. Cells were pretreated 24 h before experiments with Log‐titrated and 0.2% DMSO‐normalized compounds with and without 3 µM C18 (1, 3, 10 µM final concentrations in 250 µl apical +750 µl basolateral solution, incubated at 37°C). An addback treatment was performed 90 min prior to the start of electrophysiology experiments. The 10 µM C18 and 0.2% DMSO were used as positive and negative controls, respectively. Drugs were tested in duplicates and controls in triplicates. For CFBE cells, baseline sodium conductance was suppressed by the apical addition of 3 µM benzamil prior to transepithelial current analysis. After 20 min of baseline measurements, CFTR‐dependent chloride conductance was stimulated with the apical and basolateral addition of 10 µM forskolin and 1 µM ivacaftor (Vertex Pharmaceuticals, Boston, MA, USA), followed by 20 µM Inh‐172‐induced CFTR inhibition 60 min later. The area under the curve (AUC) of chloride conductance was calculated for the 60‐min interval between additions of 10 µM forskolin and 1 µM ivacaftor (20 min mark) and 20 µM Inh‐172 (80 min) for all small molecules.

Transepithelial chloride current studies

Transepithelial chloride current measurements were made in Ussing chambers ~10 days post‐seeding. Briefly, primary cultures were mounted in the Ussing chamber and transepithelial chloride current was measured under short‐circuit conditions. After measuring baseline current, the transepithelial current (It) response to sequential apical addition of 100 μM amiloride, 100 μM 4,4′‐diisothiocyanoto‐stilbene‐2,2′‐disulfonic acid (DIDS), 10 μM forskolin and 100 μM 3‐isobutyl‐1‐methylxanthine (IBMX), and 100 μM GlyH‐101 was measured. Studies were conducted with a chloride concentration gradient containing 135 mM NaCl, 1.2 mM MgCl2, 1.2 mM CaCl2, 2.4 mM K2PO4, 0.6 mM KH2PO4, 5 mM dextrose, and 5 mM Hepes (pH 7.4) on the basolateral surface, and gluconate substituted for chloride on the apical side.

Statistical analysis

For conductance studies, the AUC of transepithelial conductance measurements between the additions of 10 µM forskolin and 1 µM ivacaftor and 20 µM Inh‐172 were calculated. AUC ratios were computed for small molecule‐treated replicates relative to the mean AUC of control experiments. Experiments with a baseline transepithelial resistance less than 400 G·cm2 were considered toxic and removed from further analyses. One sample, one‐sided t‐tests were performed to evaluate whether small molecule AUC ratios were greater than control AUC ratios, with Benjamini‐Hochberg multiple hypothesis adjustment applied to the results of each small molecule. The AUC thresholds were established based on experience from other small molecules screened at Rosalind Franklin University for rescue of ΔF508‐CFTR. A stringent threshold of AUC greater than 1.25 was used for the initial screen. Data are presented as a mean ± SE of individual data points. Adjusted (adj) p < 0.05 was considered significant. For chloride current studies, the average change in peak transepithelial current (It) was calculated and statistical significance was determined by one sample, one‐sided t‐tests with Benjamini‐Hochberg multiple hypothesis adjustments (adj p < 0.05).

RNA‐sequencing

Total RNA was isolated from CFBE cells using the mirVana miRNA isolation kit (Ambion, Austin, TX, USA). Total RNA was tested on an Agilent Model 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Samples with an RNA integrity number (RIN) greater than 8.0 were selected for further processing. Libraries were prepared using the TruSeq RNA Sample Prep (Illumina, San Diego, CA, USA) and submitted to the Iowa Institute of Human Genetics Genomics Division for deep sequencing. Data were processed using Kallisto and Sleuth. ,

Network analysis

A STRING network consisting of CFTR as the central hub, 250 max interactors in the first shell, and 10 max interactors in the second shell was extracted from string‐db.org. This network had an interaction score of 0.400 and included all interaction sources. Transcripts per million (TPM) values for genes with an average TPM greater than 1 in XL147 and/or DMSO were inputted along with the STRING network into the DeMAND (Detecting Mechanism of Action based on Network Dysregulation) algorithm with default parameters. The code used for the DeMAND analysis is available at https://github.com/matthewdstrub/XL147. A detailed description of the DeMAND algorithm can be found in ref. 22

Oligonucleotide reagents

DsiRNAs were obtained as TriFECTa kits from IDT (Coralville, IA), each containing three pre‐designed DsiRNAs per gene (Table S5). To ascertain the specificity of the oligonucleotides, we harvested RNA from cells transfected with the three pooled oligonucleotides per gene and measured the expression of multiple genes. RNA was harvested from each sample 24 h post‐transfection. First‐strand cDNA was synthesized using SuperScript II (Invitrogen, Waltham, MA, USA), with oligo‐dT and random‐hexamer primers. Primers for each gene were designed and produced by IDT and validated in HEK cells. Quantitative real‐time polymerase chain reaction (RT‐PCR) was performed using the QuantStudio 6 Flex Real‐Time PCR system (Applied Biosystems, Waltham, MA, USA). All experiments were performed in quadruplicate. Following validation of knockdown efficiency, three DsiRNAs per gene were pooled and CFBE cells were reverse‐transfected using Lipofectamine RNAiMAX (Invitrogen) and grown on microporous membranes of Transwell plates 120 h prior to the electrophysiology measurements.

RESULTS

Identification and screening of candidate small molecules using the Library of Integrated Network‐based Cellular Signatures

Microarray signatures generated in CFBE cells treated with miR‐138 mimic, SIN3A DsiRNA, or low temperature (27°C; Gene Expression Omnibus Series GSE142610) were used to generate gene sets, consisting of up‐ and downregulated genes, to query LINCS. Candidate small molecules with signatures closely mimicking those of our queries, strong correlation across replicates, signal strength, and conservation across a minimum of two treatments were prioritized in this study (Tables [Link], [Link], [Link]). We screened 107 small molecules (Table S4) for functional rescue of ΔF508‐CFTR in CFBE cells. CFBE cells were basolaterally and apically treated with compounds of interest for 24 h, with addback performed 90 min prior to electrophysiological measurements. Small molecules were tested at three concentrations (1, 3, and 10 µM) and in the presence or absence of C18, an analog of VX‐809 (Lumacaftor; Vertex Pharmaceuticals). Five molecules showed significant activity (A893, K659, K750, Dorsomorphin, and XL147; Figure 1a) with four also showing synergistic rescue with C18 (all except A893; Figure 1b).

FIGURE 1

Eight small molecules rescue ΔF508‐cystic fibrosis transmembrane conductance regulator (CFTR) function in cystic fibrosis bronchial epithelia (CFBE) cells. Transepithelial conductance (Gt) in response to forskolin and ivacaftor (F&I) and the CFTR inhibitor‐172 (inh‐172) under open circuit conditions was measured in polarized CFBE (ΔF508/ΔF508) cells using the TECC‐24 assay. The area under the curve (AUC) of transepithelial conductance measurements between the additions of F&I and inh‐172 were calculated. The ratio of the AUC of compound‐treated cells compared to the AUC of DMSO‐treated cells are displayed for active small molecules identified by (a) the original LINCS screen and (c) relatives of the original LINCS hits. The ratio of the AUC of cells treated with compounds in the presence of C18 compared to cells treated with only C18 are displayed for active small molecules identified in (b) the original LINCS screen and (d) relatives of the original LINCS hits. Small molecules were administered basolaterally for 24 h and addback of molecules was performed 90 min prior to the electrophysiology assay. Concentrations of small molecules are noted in the keys. DMSO was administered at 0.2% concentration and C18 was administered at 10 µM when used as a control and 3 µM when used in combination with candidate small molecules. n greater than or equal to four for all tested conditions. Error bars indicate standard error of the mean. *Indicates adjusted p value less than 0.05. #Indicates toxicity at noted concentration. Gray dotted line represents the threshold AUC ratio of 1.25. Relationships of molecules to parent compounds are noted under brackets in panels c and d. Note the difference in Y‐axis scales across panels

We next screened an additional 28 small molecules (Table S6) that were chemically related to the five active molecules identified (Figure 1a,b), with the rationale being that the active molecules represent an efficacious structure or residue, or target a specific pathway. Twenty relatives were structural congeners of the active molecules. Because XL147 is a member of the PI3K inhibitor drug class, an additional eight small molecule inhibitors of PI3K were also tested. Treatment with three congeners of K659 significantly increased transepithelial chloride conductance in CFBE cells (Figure 1c), while also showing synergistic rescue in combination with C18 (Figure 1d).

Testing of efficacious compounds in primary human airway epithelia

After screening 135 small molecules in CFBE cells, we identified eight active molecules that significantly rescued ΔF508‐CFTR transepithelial chloride conductance. Next, we selected small molecules with an AUC ratio of 1.25 or greater (i.e., a 25% or greater increase in conductance), which we considered to be the threshold of substantial activity, for testing in primary human airway epithelia. All eight molecules were subsequently tested in well‐differentiated primary human airway epithelia from a single ΔF508/ΔF508‐CFTR donor (Table S7), with six showing significant improvement in ΔF508‐CFTR‐mediated transepithelial chloride conductance in at least one of our experimental regimens (i.e., alone or in combination with C18; Figure 2a,b). Two of the small molecules produced AUCs greater than 1.25; dorsomorphin was efficacious alone (Figure 2c), whereas XL147 showed activity alone and in the presence of C18 (Figure 2c,d). Dorsomorphin and XL147 were further tested in primary airway epithelia from four additional ΔF508/ΔF508‐CFTR donors. XL147 had an average increase of 94.9% in peak chloride current after the addition of forskolin and Ivacaftor compared to DMSO. When in combination with C18, XL147 had a 56.5% increase compared to C18 alone (Figure 3). Dorsomorphin did not significantly improve chloride current compared to controls in these donors.

FIGURE 2

Six small molecules rescue ΔF508‐cystic fibrosis transmembrane conductance regulator (CFTR) function in primary human airway epithelial cells from an individual ΔF508/ΔF508‐CFTR donor. Transepithelial conductance (Gt) in response to forskolin and ivacaftor (F&I) and inhibitor‐172 (inh‐172) under open circuit conditions was measured in well‐differentiated primary ΔF508/ΔF508 human airway epithelial cells using the TECC‐24 assay. The area under the curve (AUC) of transepithelial conductance measurements between the additions of F&I and inh‐172 were calculated. The ratio of the AUC of compound‐treated cells compared to the AUC of control‐treated cells are displayed for small molecules that showed efficacy in the absence (a) or presence (b) of C18. Compounds with an AUC greater than or equal to 1.25 in the absence or presence of C18 are displayed in panels c and d, respectively. Small molecules were administered basolaterally for 24 h and addback of molecules was performed 90 min prior to the electrophysiology assay. Concentrations of small molecules are noted in the keys. DMSO was administered at 0.2% concentration and C18 was administered at 10 µM when used as a control and 3 µM when used in combination with candidate small molecules. n greater than or equal to four technical replicates for all tested conditions. Error bars indicate standard error of the mean. *Indicates adjusted p value less than 0.05. Gray dotted line indicates AUC ratio of 1.25. Note the difference in Y‐axis scales across panels

FIGURE 3

XL147 rescues ΔF508‐CFTR function in primary human airway epithelial cells. Average change in transepithelial current (It) in response to forskolin and IBMX (F&I) and GlyH‐101 under short circuit conditions was measured in well differentiated primary ΔF508/ΔF508 human airway epithelial cultures from four donors. Small molecules were administered basolaterally for 24 hs prior to the electrophysiology measurements in Ussing chambers. XL147 was administered at 3 µM, Dorsomorphin at 10 µM, C18 at 6 µM, and DMSO at 0.2%. Error bars indicate standard error of the mean. *Indicates adjusted p value less than 0.05

RNA‐sequencing analysis of XL147

Following the identification of XL147 as an efficacious modulator of ΔF508‐CFTR in multiple primary human airway epithelia donors, we performed RNA‐sequencing of CFBE cells treated for 24 h with XL147 at its most effective concentration (3 µM). RNA from CFBE cells treated with the negative control (0.2% DMSO) and positive control (6 µM C18 and 5 µM VX‐809) treatment groups were also sequenced. XL147 treatment resulted in 464 significantly upregulated genes and 169 significantly downregulated genes at false‐discovery rate (FDR) less than 0.05. The 50 most upregulated and downregulated genes ranked by FDR are highlighted in Tables S8 and S9, respectively.

RNA‐sequencing analysis of C18‐treated CFBE cells revealed that only seven genes were differentially expressed when compared to DMSO. SLC38A2, SLC7A11, AMIGO2, PHLDA1, CYP1B1, DKK1, and ABCG2 were all significantly upregulated, whereas no genes were downregulated. Likewise, only nine genes were differentially expressed in VX‐809 treated CFBE cells compared to DMSO. HR, PHLDB2, STC2, INHBA, OGT, and TENM3 were significantly upregulated, whereas HIST2H2BE, SUOX, and PLK4 were significantly downregulated. The low number of genes with altered expression in C18 and VX809 strongly suggests that these corrector compounds are unlikely to act primarily through a transcriptional mechanism.

Network analyses of XL147

To elucidate the possible mechanisms of XL147’s effects on ΔF508‐CFTR, our RNA‐sequencing data were run through Gene Set Enrichment Analysis (GSEA). Enriched hallmark terms with FDR q‐value less than 0.25 for XL147 treatment included TGF‐β signaling, apical surface, and PI3K/AKT/mTOR signaling (Table 1). Downregulated terms included large ribosomal subunit and immune response in mucosa. To further identify specific genes involved in XL147‐mediated ΔF508‐CFTR rescue, we used the Detecting Mechanism of Action by Network Dysregulation (DeMAND) algorithm. Briefly, we inputted our XL147 and DMSO RNA‐sequencing profiles, as well as a CFTR‐centric regulatory network from the STRING protein‐protein interaction (PPI) networks database. DeMAND evaluated the dysregulation of each PPI by first using a Gaussian Kernel method to smooth the dysregulation co‐expression scatterplots for any two interacting genes, generating interaction probability densities. Next, the density differences pre‐ and post‐XL147 treatment were evaluated using KL‐divergence and the statistical significance of each KL‐divergence was assessed by gene pair shuffling. Last, the global dysregulation of each individual gene was determined by integrating the p values of all its network interactions. The DeMAND output yielded a rank‐ordered list of potential genes contributing to the mechanism of XL147 (Table S10). Numerous genes related to the large ribosomal subunit appeared in this output. As this GSEA hallmark term was downregulated in XL147 (Table 1), we hypothesized that knockdown of ribosomal‐related proteins could increase ΔF508‐CFTR trafficking to the cell surface and restore partial function.

TABLE 1

GSEA results of XL147 RNA‐sequencing data compared to DMSO

Term	p value	FDR q‐value	Direction in XL147
Oxidative phosphorylation	0.0214	0.0920	Down
Organellar large ribosomal subunit	0	0.0936	Down
Deoxyribonucleotide biosynthetic process	0	0.0972	Down
Angiogenesis	0.0188	0.1846	Up
Epithelial mesenchymal transition	0	0.1853	Up
Hypoxia	0	0.1869	Up
IL2 STAT5 signaling	0.1356	0.1936	Up
Apical surface	0.0980	0.1951	Up
Myc targets, version 2	0.0311	0.2011	Up
PI3K AKT mTOR signaling	0.0198	0.2068	Up
Glycolysis	0.0523	0.2077	Up
IL6 JAK STAT3 signaling	0.0302	0.2156	Up
TNFα signaling via NFκB	0.0775	0.2172	Up
KRAS signaling up	0.0792	0.2217	Up
Interferon alpha response	0.1021	0.2251	Down
Apical junction	0	0.2317	Up
Innate immune response in mucosa	0	0.2367	Down

Abbreviations: FDR, false‐discovery rate; GSEA, Gene Set Enrichment Analysis.

siRNA‐mediated knockdown of ribosomal stalk proteins rescues ΔF508‐CFTR

We used pools of three DsiRNAs per gene to knockdown ribosomal stalk and other ribosome‐related genes in CFBE cells. Interestingly, knockdown of RPL32, RPL5, RPL9, and RPS8 resulted in significant restoration of transmembrane chloride current in ΔF508‐CFTR CFBE cells (Figure 4). Of these, knockdown of RPL32 yielded the greatest rescue, with an increase of 168.9% in peak chloride current after the addition of forskolin and IBMX compared to DMSO. Knockdown of RPL32 also showed an increase of over 50% in peak chloride current when compared to knockdown of SYVN1. As Lukacs and colleagues previously identified RPL12 as a CFTR modifier, we used knockdown of RPL12 as an additional positive control. When compared to RPL12 knockdown, cells treated with the RPL32 DsiRNAs showed an increase in peak chloride current of 54.6%.

FIGURE 4

DsiRNA‐mediated knockdown of ribosomal stalk proteins rescues ΔF508‐CFTR function in CFBE cells. Average change in transepithelial current (It) in response to forskolin and IBMX (F&I) and GlyH‐101 under open circuit conditions was measured in CFBE cells. Three DsiRNAs per gene were pooled and CFBE cells were reverse‐transfected using Lipofectamine RNAiMAX and grown on microporous membranes of Transwell plates 120 h prior to the electrophysiology measurements. Error bars indicate standard error of the mean. *Indicates adjusted p value less than 0.05 when compared to DMSO. Knockdown of RPL32 also resulted in a statistically significant increase in transepithelial current compared to knockdown of positive controls RPL12 and SYVN1. n = 4–5 per candidate target gene. DsiRNAs, Dicer‐Substrate Short Interfering RNAs

DISCUSSION

Since the discovery of the CFTR gene in 1989, great progress has been made in both the development of therapeutics for CF and the understanding of CFTR biogenesis. In 1992, Denning and colleagues discovered that ΔF508‐CFTR is temperature sensitive and provided proof‐of‐principle that mutant CFTR could escape proteasomal degradation and retain partial function. With this knowledge and the advent of high‐throughput screening technologies, several small molecule CFTR modulators have been investigated for therapeutic purposes. The first US Food and Drug Administration (FDA) approved small molecule for the correction of ΔF508‐CFTR was VX‐809, which showed modest efficacy in clinical trials. Recently, the triple combination therapy of elexacaftor‐tezacaftor‐ivacaftor (Trikafta; Vertex Pharmaceuticals) was FDA‐approved and this three drug combination shows greater efficacy than VX‐809. However, despite an average increase of 13.8% in the forced expiratory volume in one second (FEV1) in patients with one ΔF508 allele and a minimal‐function mutation on the second allele, the improvement in lung function does not reach the levels observed in carriers of CFTR mutations. , Furthermore, nearly one‐third of ΔF508 homozygotes experience less than a 5% increase in FEV1 on Trikafta and some individuals do not tolerate the medication. Unfortunately, CF remains progressive and fatal and improved treatments are needed. Likewise, the CFTR interactome remains incompletely understood and the identification of CFTR effectors could lead to improved therapeutics.

We previously reported that knockdown of SIN3A or overexpression of miR‐138 rescues ΔF508‐CFTR to the cell surface and restores partial function. However, translating these discoveries into therapies remains a challenge, as the efficient delivery of siRNAs or microRNA mimics to airway epithelia is difficult. Therefore, we attempted to connect such interventions to small molecule therapies via mRNA expression profiling. Our strategy was to identify small molecule‐induced gene expression responses that were similar to the expression induced by SIN3A knockdown, miR‐138 overexpression, or low temperature incubation. Such genomic signature strategies have previously worked for other diseases, including muscle atrophy, inflammatory bowel disease, and various cancers. , ,

The present work expands upon our previous genomic signature approaches to prioritize and test compounds for rescue of ΔF508‐CFTR trafficking and function. We previously queried genomic signatures for miR‐138 overexpression and SIN3A inhibition in the CMAP and iteratively mined candidate drugs with similar effects on gene expression. We reported that the treatment of human primary airway epithelia with seven of these small molecules improved ΔF508‐CFTR trafficking greater than C18 at two or more doses. Four of the molecules also improved ΔF508‐CFTR maturation in CFBE cells. Following the introduction of LINCS, we integrated gene expression and pathway information from a variety of sources and the subsequent query aided us in identifying an additional three compounds that restored function to CFTR in human primary CF airway epithelial cells homozygous for ΔF508.

Likewise, Pesce et al. (2016) had previously queried gene expression signatures characterizing low temperature in CMAP and found that anti‐inflammatory glucocorticoids increased mutant CFTR function in a bronchial epithelial cell line. However, these glucocorticoids were found to be ineffective in primary bronchial epithelial cells. Our strategy differed from the Pesce et al. analysis in several ways. First, we performed pattern‐matching using LINCS, which contained 3,000,000 expression signatures (compared to 7000 in CMAP) of 33,000 small molecules (1300 in CMAP), and 9200 genetic perturbagens in over 200 cell lines (3 in CMAP). Furthermore, to produce such high‐scale data, the L1000 rapid high‐throughput gene expression profiling platform was developed and incorporated into the LINCS software. We reasoned that this considerable expansion of the LINCS dataset compared to its predecessor CMAP would enrich our chances of finding efficacious small molecules. Interestingly, we did not find that glucocorticoids were positively connected to the inputted low temperature signature (Table S3), resulting this ineffective drug class not being included in our candidate list. Second, we queried not only gene expression signatures characterizing low temperature, but also included signatures of miR‐138 mimic and SIN3A DsiRNA, whose gene expression signatures differed greatly from low temperature, suggesting different mechanisms of rescue. Querying multiple forms of ΔF508‐CFTR rescue also increased the likelihood of identifying correctors. Indeed, XL147 only appeared in the SIN3A DsiRNA output. By only querying low temperature, XL147 would not have been discovered. Last, we performed a more high‐throughput screen, testing 135 small molecules compared to the 20 compounds investigated in the aforementioned study. Interestingly, Hegde et al. (2015) performed a reverse connectivity mapping experiment by extracting and deconvoluting the transcriptional profiles of ΔF508‐CFTR proteostasis regulator drugs housed in CMAP to identify signaling cascades regulating proteostasis. Although we used known sources of CFTR rescue to identify small molecule correctors, their approach used known correctors to identify genes or proteins involved in CFTR proteostasis. We adopted a similar approach following the discovery of XL147, as we then performed RNA‐sequencing on cells treated with XL147 and inputted the resulting transcriptomic data into the DeMAND algorithm to identify novel CFTR effectors.

In the current study, we queried high quality gene expression profiles of SIN3A knockdown, miR‐138 overexpression, and low temperature incubation in a beta version of LINCS and identified eight out of 135 small molecules that partially restored chloride‐mediated current in ΔF508‐CFTR CFBE cells. This screen yielded an unusually high percentage of hits compared to traditional high‐throughput screens and provided confidence in our approach.

We focused on the small molecules that were most efficacious in CFBE cells for subsequent studies in primary epithelial cells. XL147 was found to partially restore CFTR‐dependent chloride current to ΔF508‐CFTR human primary airway epithelial cells from four donors. XL147 is a class 1 PI3K inhibitor that reversibly binds to lipid kinases in an ATP‐competitive manner, thereby inhibiting the production of the secondary messenger PIP3 and preventing activation of the PI3K signaling pathway. As activation of the PI3K pathway is frequently linked to tumorigenesis, XL147 has been investigated as an orally bioavailable cancer drug in several clinical trials for endometrial carcinoma, metastatic breast cancer, and glioblastoma (e.g., NCT01013324, NCT01042925, and NCT01240460; clinicaltrials.gov), although it has not yet been FDA approved. The PI3K/Akt/mTOR signaling pathway has also been targeted for the treatment of CF. Coppinger and colleagues used protein interaction profiling and global bioinformatic analyses to identify upregulation of mTOR activity in CFBE cells. Inhibition of the PI3K/Akt/mTOR pathway with multiple unique inhibitors yielded an increase in CFTR stability and expression. Subsequent analyses determined that the most efficacious inhibitor, MK‐2206, rescued ΔF508‐CFTR by restoring autophagy, potentially through its presumed target BAG3. Additionally, Rotin and colleagues identified three small molecule inhibitors of the PI3K/Akt/mTOR signaling pathway (FPA‐124, PI‐103, and 10‐DEBC) that restored partial short‐circuit current and/or rescued CFBE maturation in ΔF508‐CFTR MDCK and ΔF508‐CFTR 293MSR‐GT cells, respectively. In a related study, Trzcinska‐Daneluti and colleagues discovered that shRNA‐mediated knockdown of FGF receptors (FGFRs) and downstream signaling proteins resulted in the partial rescue of ΔF508‐CFTR maturation and surface stability. FGFRs are known activators of the PI3K/Akt/mTOR pathway. IPMK, a PI3K that acts as a molecular switch for Akt and can stimulate or inhibit PI3K/Akt signaling, was also identified as a suppressor of ΔF508‐CFTR rescue. Interestingly, how PI3K inhibition promotes rescue of ΔF508‐CFTR is not fully known. Akt has been shown to promote cAMP‐mediated trafficking of wild‐type CFTR to the cell surface, although it has also been reported to decrease surface expression of plasma membrane proteins. , , Whether XL147 rescues CFTR‐mediated Cl− current to ΔF508‐CFTR via inhibition of the PI3K signaling pathway requires additional investigation. GSEA of our XL147 RNA‐sequencing gene expression profile indicated an upregulation of PI3K/Akt/mTOR signaling, perhaps suggesting that XL147 is acting through an alternative mechanism. In addition, testing of several other PI3K inhibitors yielded modest results.

To further elucidate the potential mechanism of XL147, we used our RNA‐sequencing data to query the DeMAND algorithm. The output consisted of a rank‐ordered list of genes that could be mechanistically responsible for the activity of XL147. This list was enriched for ribosomal stalk proteins. RNAi knockdown of four candidates partially rescued chloride‐mediated current in ΔF508‐CFTR CFBE cells. Interestingly, Lukacs and colleagues identified a component of the ribosomal stalk, RPL12, as an effector of ΔF508‐CFTR. Silencing of RPL12 attenuated the translational elongation rate and increased the folding efficiency and conformational stability of ΔF508‐CFTR. Additionally, knockdown of RPL12 in combination with VX‐809 restored ΔF508‐CFTR function to ~50% of the wild‐type channel in primary human airway epithelial cells. Lukacs and colleagues also observed that silencing of RPLP0, RPLP1, and RPLP2, all components of the ribosomal stalk, partially rescued ΔF508‐CFTR function. Our GSEA analysis revealed that genes related to the large ribosomal subunit were significantly downregulated in our RNA‐sequencing data of XL147 treated cells compared to DMSO, further supporting the hypothesis that XL147 could be rescuing ΔF508‐CFTR through the silencing of ribosomal stalk proteins. Generally, the roles of ribosomal stalk proteins in disease have been enigmatic and understudied. The relationship between CFTR biogenesis and the ribosomal stalk requires further investigation.

The latest clinical trials for CFTR modulators have consisted of multi‐drug combinations, as it is likely that a single corrector is inefficient to restore ΔF508‐CFTR maturation and function. The redundancy and complexity of quality control mechanisms that detect and degrade ΔF508‐CFTR likely contribute to the need for additional therapies , , , , such as the recently FDA‐approved triple combination of elexacaftor‐tezacaftor‐ivacaftor. Indeed, XL147 also showed cooperativity when administered with C18 and cotreatment of the two small molecules yielded a significant increase in CFTR‐dependent Cl− current compared to C18 alone. C18 is an analog of the FDA‐approved VX809 (Lumacaftor) and affects the biosynthetic processing of ΔF508‐CFTR, while likely also inducing conformational changes that improve the gating defect. , C18 and VX‐809 are not believed to act through a transcriptional mechanism and may act as chemical chaperones for the misfolded protein. Indeed, our RNA‐sequencing data of C18 and VX‐809 treated cells support this hypothesis, as both small molecules induced very few differentially expressed genes when compared to DMSO. Therefore, these molecules are excellent candidates for cotreatment with small molecules that act through a different mechanism, such as the transcriptionally active XL147. The cooperativity between XL147 and C18 supports the utility of modulator cotreatment with transcriptionally active small molecules and CFTR‐interacting drugs.

Although the clinical utility of XL147 as a modulator for CF requires further studies, the discovery of its effect on ΔF508‐CFTR processing and function highlights the genomic signature pattern‐matching strategy. Furthermore, LINCS continues to expand its database and develop new publicly available tools. The BD2K‐LINCS Data Coordination and Integration Center has developed methods to connect cellular phenotypes with molecular signatures. For example, Harmonizome is a biological knowledge engine containing information about genes and proteins from 70 resources, allowing for discovery across diverse sets of omics resources. Additionally, Drug Gene Budger allows for the identification of small molecules that up‐ or downregulate an individual gene of interest. Such tools provide further opportunities for drug discovery in CF and other diseases.

CONFLICT OF INTEREST

The authors declare no conflicts of interest.

AUTHOR CONTRIBUTIONS

M.D.S., S.R., A.S., A.L., R.J.B., and P.B.M. designed research. M.D.S., S.R., D.Y.B., E.A.M., A.S., and A.L. performed research. M.D.S., S.R., D.Y.B., A.A.P., and P.B.M. analyzed data. M.D.S. and P.B.M. wrote the manuscript.

Table S1

Click here for additional data file.

Table S2

Click here for additional data file.

Table S3

Click here for additional data file.

Table S4

Click here for additional data file.

Table S5

Click here for additional data file.

Table S6

Click here for additional data file.

Table S7

Click here for additional data file.

Table S8

Click here for additional data file.

Table S9

Click here for additional data file.

Table S10

Click here for additional data file.