Unexpected Activities in Regulating Ciliation Contribute to Off-target Effects of Targeted Drugs

For several decades, development of cancer therapies has emphasized creating agents that selectively target proteins thought to contribute to cell transformation and tumor growth. This work has been based on information gleaned from the systematic analysis of cell signaling pathways, coupled with the analysis of common mutational or transcriptional targets in cancer genomes, which have collectively identified networks of cancer drivers and their key effectors. Efforts to optimize therapeutic response to targeted inhibitors have been made more challenging by the recognition that signaling responses to drugs used in tumors in vivo sometimes differ significantly from responses to drugs used in culture of tumor cells in vitro. While numerous factors contribute to these differences, one important element is the fact that tumors are complex mixtures of multiple cell types, while cells in vitro typically are grown as monocultures (1). The importance of heterocellular signaling between cancer cells and other cells in the tumor microenvironment is now well recognized. Signals that support tumor growth emanate from cells including cancer cells, stromal fibroblasts, endothelial cells, and infiltrating immune cells, and are transmitted by mechanisms involving both secretion of soluble factors and reconditioning of the extracellular matrix (2). A much less explored topic is how specific cancer drugs may indirectly condition tumor survival and cell-intrinsic signaling by modulating heterocellular signal transmission.

The primary cilium provides a spatially concentrated platform for receiving extracellular cues and inducing intracellular responses for signaling pathways downstream of ligands including Sonic Hedgehog (SHH; ref. 3), WNT (4), Notch (5), and PDGFRα (6), and polycystins (7). As receptors for these ligands localize in sum or in part to the ciliary membrane, activity of these pathways depends, in large part, on the presence or absence of a primary cilium on the cell surface. Some pathogenic conditions, including many “ciliopathies,” are associated with dysfunction or loss of the cilium: among these, one of the most studied has been autosomal dominant polycystic kidney disease (ADPKD), which arises from defects in the cilia-localized polycystins PKD1 and PKD2 and affects as many as 1 in 400 individuals (8, 9). Beyond these inherited syndromes, over the past decade, cilia have emerged as playing multiple important roles in cancer pathogenesis (10). Cilia are retained in a few tumor types, such as medulloblastomas and basal cell carcinomas, which are often dependent on SHH signaling (11). In other tumor types, cilia are lost in the cancer cells, but retained in cells in the tumor microenvironment. For example, in pancreatic ductal adenocarcinoma (PDAC), cancer cells secrete high levels of SHH (12, 13), but downregulate cilia, avoiding autocrine response (14). However, SHH stimulates desmoplasia in the ciliated pancreatic stellate cells in the adjacent stroma (15), inducing the transcription of genes that support formation of a dense, altered extracellular matrix that contributes to the poor response rate of patients to DNA-damaging agents and other drugs (12, 16–19). In addition, SHH causes stromal cells to secrete IGF1 and GAS6, which bind to IGF1R and AXL/TYRO3 receptors on PDAC cells to activate IRS-1, AKT (pT308/pS473), and other prosurvival effectors (14). The consequence of this reciprocal cross-talk between tumor and stromal cells is the creation of a highly tumorigenic microenvironment that supports tumor growth and survival, but also restrains tumor cell metastasis (20, 21).

These observations raise the interesting possibility that if some or many targeted cancer drugs affect cell ciliation per se, they may have unexpected effects in regulating signaling responses that depend on the ciliary signaling platform. In fact, interactions of cancer drugs with ciliation are not hypothetical, as work by us and others investigating ciliation controls have shown that a number of proteins targeted for drug development regulate ciliary assembly and disassembly. Some examples of oncogenic proteins with ciliary functions include Aurora-A (AURKA) kinase (22, 23) and PLK1 (24), each of which has had targeted inhibitors enter the clinic. In this work, we have screened a library of kinase inhibitors, a particularly common source of targeted cancer therapeutics, to gain insight into potential action of these agents in affecting ciliary dynamics, and cilia-dependent signaling. Analysis of screen hits and exploration of their protein targets for the first time identified several kinases, including the innate immune pathway effector IRAK4 as regulators of ciliation, and demonstrated that inhibition of some of these proteins affects intraciliary signaling responses following treatment with SHH. These results have implications for understanding the activity of these kinase inhibitors in cancer treatment, in ciliopathies such as ADPKD, and potentially other conditions.

Human hTERT1-RPE1, HEK293T, CFPAC-1, and murine NIH3T3 cells were obtained from the ATCC and validated by short-tandem repeat profiling. Cell line authentication versus defined control specimens was performed by short tandem repeat (STR) analysis performed by IDEXX Bioresearch (Westbrook, ME). Cell lines, once plated, were cultured no longer than 3 weeks. The patient-derived PNX0010 clear cell renal cell carcinoma (ccRCC) cell line was described previously (25). hTERT-RPE1 cells were cultured in DMEM/F-12 medium (Life Technologies) supplemented with 10% FBS plus 1× penicillin/streptomycin, 1× Glutamax-I (Gibco/Thermo Fisher Scientific), and 0.01 mg/mL hygromycin B (Life Technologies). HEK293T cells and NIH3T3 cells were cultured in DMEM (Life Technologies) supplemented with 10% FBS plus 1× penicillin/streptomycin, 1× Glutamax-I (Gibco/Thermo Fisher Scientific). CFPAC-1 cells were cultured in DMEM (Life Technologies) supplemented with 10% FBS, 1× penicillin/streptomycin, 1× Glutamax-I, 1× sodium pyruvate, and a cocktail of nonessential amino acids (Gibco/Thermo Fisher Scientific). PNX0010 cells were cultured in RPMI1640 (Life Technologies) supplemented with 10% FBS, 1× penicillin/streptomycin, 1× Glutamax-I, 1× sodium pyruvate, and a cocktail of nonessential amino acids (Gibco/Thermo Fisher Scientific).

The hTERT1-RPE1-Arl13bGFP cell line was generated by infection of hTERT-RPE1 cells with a virus containing mammalian L13-Arl13bGFP construct, a gift from Tamara Caspary (The Emory University School of Medicine, Atlanta, GA; Addgene plasmid # 40879, Addgene). The human IRAK4 ORF containing a C-terminal HA-tag was amplified by PCR using the pDONR223_IRAK4_WT vector as a template (a gift from Jesse Boehm, Broad Institute, Cambridge, MA; William Hahn, Dana-Farber Cancer Institute, Boston, MA; and David Root, Broad Institute, Cambridge, MA; Addgene plasmid # 82286) and cloned into pLV-CMV-puro lentiviral vector. To clone human IRAK4 ORF containing a C-terminal HA-tag into the pLV-CMV-puro lentiviral vector, the primers (forward: 5″ – ATC TTT TCT AGA GCC ACC ATG AAC AAA CCC ATA CCC ATA ACA CC – 3″; reverse: 5″ – ATC TTT CTC GAG TTA AGC GTA ATC TGG AAC ATC GTA TGG GTA AGA AGA AGC TGT CAT CTC TTG – 3″) using Ex-Taq Polymerase (Takara Bio USA).

For lentivirus infection, lentivirus was packaged by cotransfection of L13-Arl13bGFP construct with the packaging and envelope plasmids (psPAX2 and pMD2.G) by calcium phosphate transfection into HEK293T cells. Medium was collected after 48 and 72 hours, combined and filtered through a 0.45-μm filter, then applied to hTERT-RPE1 cells. After 72 hours postinfection, hTERT-RPE1 cells were selected with 400 μg/mL zeocin for approximately 2 weeks and additionally selected by limiting dilution assay to generate stable clones expressing Arl13b-GFP.

Shh-N–conditioned medium was collected from HEK-293T cells stably expressing pcDNA3.1 Shh-N (plasmid # 37680, a gift from Philip Beachy, Stanford University, Stanford, CA, to Addgene), with control medium collected from parental HEK-293T cells, as described in (26). To produce Shh-N293T, cells were plated at 70% confluency and transfected with a pcDNA3.1 Shh-N plasmid using Mirus TransIT-LT1 Transfection Reagent (Mirus Bio LLC). At 24 hours posttransfection, the medium was replaced with serum-free Opti-MEM; after an additional 48 and 72 hours Shh-N–conditioned medium was collected, pooled, filtered through a 0.45-μm filter, and stored at −80°C. The concentration of Shh-N in thawed medium used for experiments was measured by ELISA (RAB0431, Sigma-Aldrich Corporation) before use, and was consistently used in a range of 1.5–2 ng/mL. PDGF-AA was obtained from R&D Systems.

The chemical library used for screening includes 178 compounds (Supplementary Table S1), profiled in (27). For compound screening, hTERT-RPE1 Arl13b-GFP were plated at high density (2 × 10⁴ cells) in 96-well plates. All cells were starved for 48 hours in serum-free Opti-MEM medium to induce maximal ciliation. To identify compounds causing ciliary loss, ciliated cells were incubated with compounds in serum-free medium for 2 hours. To identify compounds blocking ciliary loss, after ciliated cells were preincubated with the drugs for 1 hour and then stimulated with 20% FBS for another 2–2.5 hours. For screening, all drugs were assessed at concentrations 2.5 and 0.1 μmol/L, in assays performed in two independent biological replications. Imaging was done twice for each arm: first, prior to addition of compounds, to visualize baseline ciliation, and second, after incubation with compounds with or without serum, for times noted. Cilia was visualized initially for screening at 20× magnification using an ImageXpress micro automated microscope equipped with MetaXpress Software (Molecular Devices). Images of hTERT-RPE1-Arl13b cells before and after drug treatment were analyzed as described in detail by Zhang and colleagues (28). Briefly, acquired images were analyzed with MetaXpress Software (Molecular Devices, www.moleculardevices.com) using the Transfluor module. To identify cilia, the “vesicle” settings were used. Settings criteria for cilia-automated counting are usually determined on the basis of the vehicle “no serum” and vehicle “plus serum” controls, to establish the intensity threshold. Normally, a drug is considered as a “hit” based on approximately 20% change of mean value compared with “no serum control,” based on the performance of two biological repeats performed on separate days, with each well imaged before and after drug addition, and nine separate visual fields analyzed within each well. Statistical significance was evaluated using one-way ANOVA to establish P values.

Ciliation phenotypes of positive hits from the screen were confirmed using confocal microscopy (SP8 Confocal System, Leica Microsystems), using the Arl13b-GFP/hTERT1-RPE1 screening line, and further analyzed using the hTERT1-RPE1 parental line or NIH3T3 cells, using immunofluorescence. These cells were plated at 70% confluence on coverslips coated with collagen (StemCell Technologies) and incubated with Opti-MEM medium without serum for 48 hours. Ciliary stabilization or disassembly was induced by adding medium containing 20% FBS, and ciliation status was observed after treatment with drugs, with or without serum.

Compounds used in validation assays included vehicle (0.01% DMSO; Sigma-Aldrich Corporation), ganetespib (Synta Pharmaceuticals), alisertib (MedchemExpress), sunitinib (LC Laboratories), IRAK1/4 inhibitor (Cayman Chemical), and erlotinib (LC Laboratories).

For RNAi experiments to identify relevant protein drug targets affecting ciliary disassembly, hTERT-RPE1 cells were starved in Opti-MEM serum-free medium for 34 hours, then transfected with SmartPool or individual siRNAs (Dharmacon; Supplementary Table S2) targeting selected genes, using Luciferase GL2 Duplex (D-001100-01-20, Dharmacon) as a negative control. They were then starved for an additional 48 hours (cumulative, 82 hours), after which fresh medium ± 20% serum was added for 2 hours (cumulative, 84 hours). Imaging was performed at the 82 and 84 hour time points. To identify siRNAs affecting ciliary assembly, hTERT-RPE1 cells were first transfected with siRNAs as above. At 24 hours posttransfection, the medium was replaced to serum-free Opti-MEM for another 48 hours, after which fresh medium ± 20% serum was added for 2 hours. The cells were then fixed and processed for immunofluorescence.

For experiments with IRAK4 overexpression, hTERT-RPE1 cells were plated at 70% confluency, transfected with pLV-CMV-puro-IRAK4-HA or pLV-CMV-puro-empty constructs using Mirus TransIT-LT1 Transfection Reagent (Mirus Bio LLC) in serum-free Opti-MEM, and starved for 48 hours. The cells then were fixed and processed for immunofluorescence analysis.

For experiments with inflammatory factors, hTERT-RPE1 cells were plated at 70% confluency, starved in serum-free Opti-MEM for 48 hours, after which fresh medium ± 20% serum, LPS (L2630, Sigma-Aldrich Corporation) or IL1α (200-LA, R&D Systems) was added for 2 hours, then cells were fixed and processed for immunofluorescence analysis.

NIH3T3 fibroblasts were plated at 70% confluency and starved for 24 hours in Opti-MEM supplemented with 0.25% FBS. To investigate whether the drugs affect translocation of Smoothened to the primary cilium, the cells were pretreated with drugs for 1 hour and stimulated with Shh-N–conditioned medium for another hour. The cells then were fixed and processed for immunofluorescence analysis. To evaluate the effect of drugs on PDGFRα signaling, serum-starved NIH3T3 cells were treated with either vehicle or selected drugs for 2 hours or 30 minutes, then stimulated with 50 ng/mL PDGF-AA for 10 minutes, and lysed for Western blot analysis.

tRNA was isolated using a Quick-RNA Kit (Zymo Research) and tested for quality on a Bioanalyzer (Agilent Technologies). RNA concentrations were determined with a NanoDrop Spectrophotometer (Thermo Fisher Scientific). RNA was reverse transcribed using Moloney Murine Leukemia Virus Reverse Transcriptase (Thermo Fisher Scientific) and a mixture of anchored Oligo-dT and Random Decamers (Integrated DNA Technologies). Two reverse-transcription reactions were performed for each sample using either 200 or 50 ng of input RNA in a final volume of 50 μL. Aliquots (2 μL) of the cDNA were used to measure the expression levels of the genes with the primers listed in Supplementary Table S3, and Power SYBR Green or Taqman Universal Master Mix (Thermo Fisher Scientific), on a QuantStudio6 Detection System (Thermo Fisher Scientific). Cycling conditions were 95°C, 15 minutes, followed by 40 (two-step) cycles (95°C, 15 seconds; 60°C, 60 seconds). Ct (cycle threshold) values were converted to quantities (in arbitrary units) using a standard curve (four points, 4-fold dilutions) established with a calibrator sample. Values were normalized to those for mRNA encoding the ribosomal protein 36B4.

To assess the effect of the compounds on AURKA activity, recombinant human GST-tagged Aurora A (A1983, Sigma-Aldrich Corporation) was preincubated with the drugs at a final concentration of 1 μmol/L for 20 minutes on ice. As a substrate for AURKA, recombinant Histone H3 (HH3) was used (SRP0177, Sigma-Aldrich Corporation). The mix containing AURKA, drugs, and HH3 was then incubated with [γ-32P]ATP for 30 minutes at 30°C and processed for SDS–PAGE, followed by autoradiography and Western blotting. Protein loading was quantified on the basis of staining with SimplyBlue SafeStain (LC6060, Thermo Fisher Scientific), and use of antibodies against AURKA (610938, BD Biosciences).

The evaluation of cystic index was assessed as described by Smithline and colleagues (29). Briefly, a grid was applied to the hematoxylin and eosin (H&E) images of kidneys and the index was calculated as a percentage of grid intersections that crossed cysts.

Primary antibodies used in immunofluorescence experiments recognized acetylated α-tubulin (sc-23950) and Smoothened (Smo; sc-166685; Santa Cruz Biotechnology); γ-tubulin (T5192, T5326; Sigma-Aldrich Corporation); γ-tubulin (ab191114), phosphorylated pT²⁸⁸-AURKA (ab83968; Abcam); pericentrin (IHC-00264; Bethyl Laboratories); ARL13B (17711-1-AP; Proteintech Group), GLI3 (AF3690; R&D Systems); and NF-κB p65 (L8F6; 6956; Cell Signaling Technology). Secondary antibodies labeled with Alexa Fluor 488, Alexa Fluor 568, Alexa Fluor 647, and mounting medium Prolong Gold with DAPI to stain DNA were obtained from Life Technologies.

Western blots were performed using antibodies to the following proteins: to phosphorylated pY⁷²⁰-PDGFRα (ab134068), and β-actin (ab20272), from Abcam; to phosphorylated p^S217/S221MEK1/2 (9121), PDGFRα (3174), and MEK1/2 (4694), from Cell Signaling Technology; to IRAK4 (MA5-15883) from Thermo Fisher Scientific, to vinculin (V9131) and β-tubulin (T4026) from Sigma-Aldrich Corporation, to hemagglutinin (HA; sc-7392) from Santa Cruz Biotechnology Inc.

To reprobe blots with multiple antibodies, the membranes were stripped using Restore Plus Western Blot Stripping Buffer (#46430, Thermo Fisher Scientific) for 15 minutes, then washed 5 times, 15 minutes each, in PBS/0.1% Tween-20, blocked in SuperBlock (PBS) Blocking Buffer (#37515, Thermo Fisher Scientific) for 1 hour, and used for staining with antibodies.

For immunofluorescence analysis, cells were fixed with 4% paraformaldehyde for 7 minutes and then in cold methanol for 5 minutes, permeabilized with 1% TritonX-100 in PBS, blocked in PBS with 3% (v/v) BSA, and incubated with antibodies using standard protocols. In some experiments cells were fixed with cold methanol, followed by permeabilization with 1% TritonX-100 in PBS. Samples were imaged at room temperature using an SP8 confocal system equipped with an oil-immersion 363 objective with numerical aperture 1.4 (Leica Microsystems) and LASAF (Leica Application Suite Advanced Fluorescence) software. Images were analyzed using the open-source ImageJ Software (National Institutes of Health, Bethesda, MD).

Quantification of pT²⁸⁸-AURKA immunofluorescence signal in basal bodies was obtained from 16-bit confocal images, as described by Arqués and colleagues (30) using ImageJ software. pT²⁸⁸-AURKA intensity in the “vehicle no serum” group was considered 100%, and the signal intensity after treatment with the other compounds was normalized to this group.

For live-cell imaging, hTERT-RPE1-Arl13b cells (2.5 × 10⁵) were plated in a 35-mm glass bottom dish (P35G-1.5-14-C, MatTek Corporation) and starved for 72 hours in serum-free Opti-MEM medium. Prior to imaging, medium was replaced to fresh Opti-MEM–containing drugs or serum as specified in Results. For compounds stabilizing cilia, the cells were pretreated with compound alone for 1 hour, and then the serum was applied for two hours. Images were obtained every 4 minutes over a 3-hour incubation at 37°C using an SP8 Confocal System (Leica Microsystems) and LASAF (Leica Application Suite Advanced Fluorescence) software.

For cell-cycle analysis, starved hTERT-RPE1 and NIH3T3 cells were trypsinized and collected by centrifugation after 2.5, 8, or 16 hours of drug treatment (with alisertib, ganetespib used at 1 μmol/L, sunitinib, IRAK1/4 inhibitor, and erlotinib used at 2.5 μmol/L), washed with PBS, then fixed in 70% ethanol overnight. The samples were then washed 3× in PBS and incubated with BD propidium iodide PI/RNase Staining Buffer (550825, BD Biosciences) and analyzed by BD FACScan Flow Cytometer (BD Biosciences).

The Institutional Animal Care and Use Committee of Fox Chase Cancer Center approved all experiments involving mice. Conditional Pkd1fl/fl;Cre/Esr1⁺ mice (in which tamoxifen induction of the Esr1 promoter expresses Cre-flox, resulting in inactivation of the Pkd1 gene in vivo), have been described previously (31). Mice were injected intraperitoneally with tamoxifen (250 mg/kg body weight, formulated in corn oil) on P35 and P36 for cyst induction by Pkd1 deletion. Sunitinib Malate (LC Laboratories) was formulated in sterile 0.15 mol/L NaCl with 2% DMSO solution (vehicle) at 20 mg/kg final concentration and administered orally twice daily, using a 5-day on/2-day off schedule. Pkd1^−/− (n = 23) and Pkd1 wt (n = 21) mice were treated sunitinib (Pkd1^−/−, n = 12; Pkd1 wt, n = 10) or vehicle (Pkd1^−/−, n = 11; Pkd1 wt, n = 11) for 10 weeks beginning at 4 months of age, and cyst growth was monitored by MRI. Mice were euthanized at 6.5 months of age to collect kidneys and other organs for analysis. MRI analysis, pathologic assessment of cyst formation, and immunofluorescence assessment of cilia in renal cells was performed as described in detail in (29, 32).

For the analysis of ciliation in normal kidneys, 6-week-old male C.B17/Icr-scid (SCID) mice were used. To analyze ciliation in renal cell carcinoma patient-derived xenografts (RCC PDX), 2 × 10⁶ of PNX0010 cells were injected subcutaneously in the flank region of 6-week-old male SCID mice. After 10 days, animals were randomly assigned to the control or experimental groups (n = 3 mice/group). Mice were treated with vehicle or sunitinib malate at 20 mg/kg final concentration twice daily by oral gavage, using a 5-day on/2-day off schedule, and were euthanized 2 or 16 hours after the final dose.

Organs from euthanized mice were collected, fixed in 10% phosphate-buffered formaldehyde (formalin), embedded in paraffin, sectioned at 5-μm intervals, and processed for H&E staining or immunofluorescence-based IHC as described in (32). For cilia staining, kidney sections were incubated overnight with primary mAbs against acetylated α-tubulin (sc-23950; Santa Cruz Biotechnology), with further 1-hour incubation with secondary antibodies labeled with Alexa Fluor 488. Sections were mounted with Prolong Gold with DAPI to visualize nuclei (Life Technologies). Cilia length was quantified on the basis of the acetylated α-tubulin staining using open-source ImageJ Software (National Institutes of Health, Bethesda, MD).

For statistical analysis, we used Mann–Whitney test for pairwise comparisons, and one-way ANOVA (Tukey tests) for 2 or more group comparisons. Analysis was performed by GraphPad Prism 6 (GraphPad Software).

To investigate the effect of different preclinical and clinical compounds on regulation of the ciliary dynamics, we assessed the activity of a panel of 178 kinase inhibitors of well-defined target specificity (Supplementary Table S1; ref. 27). As a screening platform, we stably expressed cilia-localized Arl13b-GFP (33) into hTERT-RPE1 retinal pigmented epithelial cells to visualize cilia and used two complementary screens to assess drug activity in inducing ciliary resorption in the absence of serum, or in promoting cilia stabilization in the presence of serum (Fig. 1A and B). On the basis of previously defined activities, as controls we used the Aurora-A kinase (AURKA) inhibitor alisertib as an agent to block ciliary disassembly (22, 34), and the HSP90 inhibitor ganetespib as an agent to induce ciliary disassembly (32). Comparison of ciliation pre- and posttreatment using high content microscopy identified 6 kinase inhibitors that caused significant resorption of cilia within 2.5 hours of drug addition, and 5 kinase inhibitors that blocked serum-induced resorption of cilia at the same time point (Supplementary Table S4). Compound phenotypes were confirmed using drug treatment of the parental hTERT1-RPE1 cells and immunofluorescence visualization using antibodies to ARL13B and the basal body marker pericentrin (Fig. 1C and D).

We explored the defined protein targets of the inhibitors (Table 1). Compounds that induced cilium disassembly had targets that included PDGF and VEGF receptor family members (sunitinib, VEGFR inhibitors I and II, SU11652), GSK3β (GSK3β inhibitor XI), and CHK1 (SB218078). In contrast, compounds that blocked ciliary disassembly have been defined as inhibitors of IL1 receptor-associated (IRAK) kinases (IRAK1/4 inhibitor), EGFR (Erlotinib, PD174265), MEK kinase (MEK 1/2 inhibitor), and a dual inhibitor of DNA-PK/PRKDC and PI3K/PIK3CB (DNA-PK inhibitor III). Among these, erlotinib is known to ameliorate cystogenesis in mouse models of ADPKD, a pathologic condition dependent on defective signaling from intact cilia (35), and EGF signaling was recently defined as regulating ciliogenesis (36). Importantly, none of these compounds had ever been described as a regulator of ciliary dynamics, even though some are in common clinical use. For some (e.g., the IRAK1/4 inhibitor), the drug and its targets have never been linked to control of ciliation.

To identify the specific proteins responsible for compound effects on cilia, we used siRNA to individually deplete the protein target(s) of each kinase inhibitor (Fig. 1E; Supplementary Fig. S1A and S1B; Table 1). For this, Arl13b-GFP/hTERT1-RPE1 cells were grown for 34 hours in serum-free medium, transfected with siRNAs, and incubated another 48 hours in serum-free medium, then treated in medium with or without serum for 2 hours: ciliation was assessed at the 82 and 84 hour time points. Among the defined inhibition targets of the drugs, knockdown of LRRK2, PDGFRα, PDGFRβ, GSK3α, and GSK3β significantly reduced ciliation of cells maintained in serum-free conditions; among these siRNAs, depletion of PDGFRα and PDGFRβ caused cells to be nonresponsive to serum (i.e., no further reduction in ciliation was observed). Conversely, siRNAs targeting MERTK, CHEK1/CHK1, IRAK4, EGFR, PRKDC, PIK3CB, and MAP2K2/MEK2 in each case rendered cells resistant to serum-induced ciliary disassembly. Other potential targets of the drugs did not affect ciliation, indicating specificity. Finally, to distinguish effects of specific protein targets on disassembly and initial assembly of cilia, we transfected cells with siRNA in complete media and after 24-hour post transfection replaced the medium with Opti-MEM serum-free to initiate ciliogenesis. In this case, transfection of PDGFRa and GSK3A reduced initial ciliation rates, but depletion of LRRK2, PDGFRB, and GSK3B did not, indicating a specific effect on disassembly (Fig. 1F).

We focused further detailed analysis on sunitinib and erlotinib, as drugs in common clinical use, and on the IRAK1/4 inhibitor, as identifying a new mechanism for ciliary regulation. For each of these compounds, we first established phenotypes were dose dependent (Fig. 2A). Ciliary phenotypes were also confirmed using NIH3T3 cells as an independent cell model (Fig. 2B). As an additional test of specificity, sunitinib and sorafenib have distinct chemical structures, but partially overlapping specificities for kinase inhibition (37). Direct evaluation of sorafenib showed that, like sunitinib, it induced loss of ciliation in a dose-dependent manner, in Arl13b-GFP/hTERT1-RPE1 cells (Supplementary Fig. S2A). Activation of AURKA at the ciliary basal body has been defined as essential for cilium disassembly (38). Strikingly, in spite of the absence of known links between the drugs tested and the control of AURKA activity, drugs that induced ciliary disassembly stimulated the appearance of activated (T²⁸⁸-phosphorylated) AURKA at the basal body within 2 hours of drug addition; conversely, compounds that blocked disassembly impaired the ability of serum to induce T²⁸⁸-phAURKA (Fig. 2C and D; Supplementary Fig. S2B).

To rule out the possibility that the drugs might directly regulate AURKA kinase activity, we performed an in vitro kinase assay, using purified recombinant active AURKA preincubated with each compound at 1 μmol/L final concentration, to assess its ability to phosphorylate a histone H3 substrate. None of the compounds affected AURKA kinase activity (Supplementary Fig. S3A and S3B). These results implied the drugs indirectly regulated AURKA through intercepting or simulating upstream activation signals.

We next used live-cell imaging to compare the effects of these drugs on the dynamics of ciliary disassembly, benchmarking to alisertib and ganetespib as controls (Fig. 2E and F). Serum treatment typically begins to cause ciliary disassembly at 100 minutes after administration, with most cilia resorbing between 150–180 minutes. Significant ciliary disassembly induced by sunitinib was first observed within 20 minutes after drug addition, and most cilia resorbed between 120–150 minutes. We did not observe any signs of cilia decapitation (39) with sunitinib, but rather rapid loss of Arl13b staining across the entire length of the cilia (Supplementary Fig. S4A). Ganetespib had slower kinetics, with resorption first becoming noticeable at approximately 75 minutes after treatment; in this case, decapitation was observed as an early step in the resorption process (Supplementary Fig. S4B). Both erlotinib and IRAK1/4 inhibitor pretreatment caused complete resistance to serum-induced ciliary disassembly for 3 hours. We also found that, compounds stabilizing cilia (alisertib, IRAK1/4 inhibitor, and erlotinib) had very limited effect on length in hTERT-RPE1 cells, and did not affect cilia length in the NIH3T3 cell line (Supplementary Fig. S5A and S5B).

We next evaluated epistatic relationships between inhibitors and promoters of ciliary disassembly by timed addition of these compounds. One hour of pretreatment with alisertib, IRAK1/4 inhibitor, or erlotinib blocked the ability of sunitinib and ganetespib to disrupt ciliary disassembly (Fig. 2G–I). Notably, pretreatment with alisertib, IRAK1/4, or erlotinib each reduced the ability of sunitinib and ganetespib to induce T²⁸⁸-phAURKA (Fig. 2J). Finally, loss of ciliation is typically associated with cell-cycle progression out of early G₁ or G₀ (40, 41). Using FACS analysis, we confirmed that while serum treatment caused a modest shift out of G₁ phase, no consistent pattern was observed with drug treatment of hTERT1-RPE1 cells or NIH3T3 cells within the 2.5-hour time course of our ciliation assay that could explain the observed ciliary phenotypes (Supplementary Fig. S6A and S6B). At later time points (up to 16 hours), treatment of hTERT-RPE1 cells with the compounds demonstrated a significant accumulation of cells in G₂–M cells for ganetespib and alisertib, consistent with previously studies (Supplementary Fig. S6C; refs. 42, 43). A common feature of cellular transformation is loss of cilia (10); for compounds causing increased ciliation, we also investigated whether they possessed similar activities in transformed cancer cells, as well as immortalized but nontransformed models. Treatment of CFPAC-1 pancreatic cell lines and PNX0010 patient-derived xenograft (PDX)-derived renal cell cancer (RCC) cell line with 2.5 μmol/L IRAK1/4 inhibitor or erlotinib for 2 or 4 days showed that while erlotinib significantly increased ciliation in CFPAC-1 but not PNX0010 cells, IRAK1/4 inhibitor did not increase ciliation in either model (Supplementary Fig. S7A and S7B).

The SHH pathway signals through the primary cilium (44, 45), with this signaling essential for multiple biological processes, including induction of desmoplasia among fibroblasts in pancreatic and other cancers (14). Typically, in the absence of SHH ligand, its receptor, PTCH1, resides near the base of the cilia, preventing SMO from entering the cilia. Binding of SHH to PTCH1 causes it to leave the cilia, allowing ciliary entry by SMO, where SMO subsequently activates the GLI signaling pathway at the ciliary tip (Fig. 3A). To assess drug effects on these processes, NIH3T3 cells were treated with vehicle or drugs for 1 hour, then exposed to serum-free medium, with or without Shh-N for 1 hour. At this time point, although some drugs induced ciliary resorption as reported above, at least 30% of the cell population remained ciliated under all conditions, allowing assessment of Smo and Gli3 localization in cilia. In the context of vehicle, Shh-N efficiently recruited Smo to cilia (Fig. 3B and C). In contrast, pretreatment of starved and ciliated NIH3T3 cells with sunitinib or alisertib, markedly reduced the ability of Shh-N to induce ciliary recruitment of Smo. A more moderate but statistically significant effect was also observed with ganetespib and erlotinib treatment (Fig. 3B and C). As a control, we confirmed that treatment of NIH3T3 cells with drugs in the absence of Shh-N did not cause recruitment of Smo to cilia.

As further analysis of this pathway, we examined translocation of Gli3 from the ciliary tip through the main ciliary body upon stimulation with Shh-N (Fig. 3D and E). In agreement with results for Smo, alisertib and sunitinib strongly, and ganetespib and erlotinib more modestly, reduced Gli3 movement from the ciliary tip. Interestingly, although the IRAK1/4 inhibitor compound had no discernible effect on Smo migration into cilia (Fig. 3B and C), it partially reduced Gli3 migration to the ciliary tip (Fig. 3D and E).

PDGF-AA–binding to its cilia-localized receptor PDGFRα causes receptor autophosphorylation (pY⁷²⁰PDGFRα) and phosphorylation of MEK1/2 (p^S217/S221MEK1/2) at the basal body (46). Starved NIH3T3 cells were treated with vehicle or drugs in serum-free medium for 30 minutes or 2 hours, then exposed to PDGF-AA for 10 minutes. Sunitinib reduced phosphorylation of PDGFRα and MEK1/2 at both time points, compatible with its known direct inhibitory activity against PDGFRα. In contrast, ganetespib significantly reduced phosphorylation of PDGFRα and MEK1/2 after 2 hours, but not after 30 minutes of prior treatment, compatible with an effect secondary to loss of ciliation (Fig. 3F and G). Comparison of ganetespib- and sunitinib-induced reduction of phosphorylation and total expression of PDGFRα and MEK1/2 confirmed the drug effects reflected inhibition of activation, rather than destabilization of the proteins (Supplementary Fig. S8). In contrast, alisertib, IRAK1/4 inhibitor, and erlotinib had no significant effect.

In ADPKD, reduction of ciliation in renal cells limits cystogenesis, reducing kidney volume (32, 47); conversely, in cancer, loss of cilia has been associated with tumorigenesis (10). We used two models to explore the effect of sunitinib on ciliation in vivo. In one approach, first, to assess whether sunitinib affected renal cell ciliation or ciliation of tumor cells in vivo, control SCID mice and mice bearing PNX0010 clear cell RCC xenografts were treated daily for 2 weeks with 20 mg/kg sunitinib, and euthanized either 2 hours or 16 hours after last administration of the drug. At both time points, cilia on renal epithelia and xenograft tumors were markedly reduced in both number and length relative to vehicle-treated controls (Fig. 4A–D). Second, as a model for ADPKD we used Pkd1fl/fl;Cre/Esr1⁺ mice, injecting tamoxifen intraperitoneally at postnatal days P35–P36 to cause targeted inactivation of the Pkd1 gene (31, 34, 48). After MRI confirmed initial formation of cysts at 4 months of age, we treated mice with 20 mg/kg sunitinib or vehicle for 10 weeks, from 4 months to 6.5 months of age. MRI imaging after 5 and 10 weeks after the start of treatment, coupled with histopathologic analysis at the experimental endpoint, showed sunitinib markedly reduced renal cystogenesis (Fig. 4E and F), and also significantly reduced kidney volume (Fig. 4G) and cystic index (Fig. 4H). Histopathologic analysis (Fig. 4I and J) confirmed sunitinib treatment reduced both the number and length of cilia in renal tissues, indicating effectiveness of sunitinib on cilia disassembly in vivo was durable over 10 weeks of exposure to the drug.

IRAK4 has not been previously implicated in regulation of ciliation, being better studied as a component of the innate immune signaling pathway (Fig. 5A). However, as some studies have suggested a link between inflammatory signaling and cilia (49–51), particularly in the context of ADPKD, we analyzed its function in greater depth. We first demonstrated that four independent siRNAs or an siRNA pool depleting IRAK4 in each case prevented serum-induced ciliary disassembly (Fig. 5B and C). Conversely, overexpression of IRAK4 in cells reduced ciliation by 60% (Fig. 5D and E). We then systematically performed siRNA depletion of components of the innate immune pathway involving IRAK4. Among these, depletion of IRAK1, MYD88, ILR1, ILR2, PELI2, PELI3, or NF-κB1 did not significantly affect ciliogenesis or disassembly (Fig. 5F). In contrast, depletion of TLR2, and of PELI1 and NF-κB2 strongly inhibited ciliary disassembly (Fig. 5F). The kinase TAK1 has been reported to act downstream of IRAK1/4 in transducing inflammatory signals; however, treatment of ciliated cells with a small-molecule inhibitor of TAK1 (NG25) did not reduce ciliation (Fig. 5G). Examining upstream activators of the IRAK4 pathway, we stimulated ciliated cells for 2 hours with IL1α or LPS, in the absence of serum (Fig. 5H and I). Translocation of NF-κB p65 to the nucleus confirmed effective stimulation of the pathway by each compound (Fig. 5I); in each case, exposure to IL1α or LPS, significantly decreased ciliation (Fig. 5H), implicating multiple elements of the innate immune signaling pathway in control of ciliary dynamics.

The results of this work provide several key insights into the regulation of ciliogenesis by targeted anticancer agents. First, they demonstrate that a number of drugs currently in use or in development for use in cancer or other pathologic conditions have previously unreported activity in regulating cell ciliation. These included commonly used clinical agents such as sunitinib, sorafenib, and erlotinib, among others. Second, investigation of the bioactive targets of these drugs identified a number of genes known to influence ciliation (PDGFRA/B, GSK3A/B, and EGFR) and genes not previously linked to control of ciliation. Among these novel regulators, depletion of LRRK2 caused rapid ciliary loss, while depletion of MERTK, CHEK1, IRAK4, PRKDC, PIK3CB, and MAP2K2 reduced or eliminated the ability of cells to resorb cilia in response to physiologic stimuli. Third, extended analysis of selected drugs confirmed their physiologic action in control of ciliation in vivo, in primary renal cells, RCC xenografts, and ADPKD cysts. Fourth, we demonstrated the relevance of drug action in regulating proximal signaling responses to SHH and PDGF-AA. Fifth, extended analysis of the IRAK4 signaling pathway based on the known interactions of this protein in innate immune response for the first time identified a subset of canonical pathway members that are also required for regulation of ciliation; this may be pertinent to activity of this pathway as a modulator of tumor–stromal signaling.

In all cases, genes and drugs influencing ciliation regulated AURKA activation at the basal body, with induction of disassembly in each case associated with and dependent on AURKA activation, and blockade of disassembly accompanied by failure to activate AURKA at the basal body. This adds the genes identified in this study to a growing number of genes shown to influence AURKA activation in ciliary disassembly (38). Epistasis experiments demonstrate that the IRAK1/4 inhibitor and erlotinib act downstream of drugs that induce ciliary disassembly and are almost as effective as alisertib in blocking ciliary disassembly, suggesting activation of IRAK4 and EGFR are critical component of this process. However, some experiments clearly distinguish the activity of drugs blocking disassembly; for instance, alisertib blocks translocation of SMO into cilia in response to stimulation of cells with Shh-N, whereas IRAK1/4 inhibitor does not. We previously noted defects in ciliary localization of the adenylyl cyclase AC6 in alisertib-treated cells (34); together with the current results, these findings suggest a role for AURKA in regulating ciliary import. It is also interesting that although sunitinib induces resorption, and alisertib blocks resorption, both limit recruitment of Smo and Gli3 following Shh-N treatment; and that IRAK1/4 inhibitor does not affect initial recruitment of Smo, but does reduce Gli3 recruitment. These differences likely reflect roles of the drug targets (e.g., GSK3α/β, AURKA, IRAK4) in phosphorylating specific proteins involved at different stages of the translocation process.

There is growing evidence that the primary cilium is important for various aspects of immunologic response and inflammation (49–51). IRAK4 plays an important role in signal transduction from IL1 receptor (IL1R), IL18 receptor (IL18R), and Toll-like receptors (TLR). The data in this study specifically nominated activation of transmembrane receptors of inflammatory cytokines, and requirement for the kinesin KIF11 and Pellino-1 (PELI1) in disassembly, while excluding a critical requirement for other pathway components such as TAK1. PELI1 is an E3 ubiquitin ligase that binds directly to IRAKs and has been shown to control NF-κB activation and other signaling outputs in response to TLR activation (52, 53). The kinesin KIF11 (also known as Eg-5 and kinesin-5) is essential for mitotic spindle function, and important in formation of neuronal projections; it has been identified as a direct interactor for IRAK4 in high-throughput screening (54). To date the relationship between IRAK4 and KIF11 has not been investigated mechanistically and evidence for a KIF11 function in cilia is limited. However, a recent study of inherited KIF11-associated retinopathies has provided some support for the idea the protein localizes to cilia, and that these diseases represent a class of ciliopathy (55). These relationships are worth further evaluation in the context of ADPKD; like cancer, a disease in which inflammation plays an important role in disease pathogenesis, among many other signaling commonalities (49, 50, 56).

Among the other inhibitor targets identified in this screen, mutations affecting MERTK have also been associated with inherited retinopathies, and defects in retinal phagocytosis (57). Furthermore, expression of MERTK is affected in retinal cells of patients with ciliopathies, although no role for MERTK has as yet been defined in maturation of the retinal pigment epithelium in these patients (58). However, no direct role for MERTK in regulation of ciliation has previously been defined. Activation of CHEK1/CHK1 has been reported to be a consequence of loss of the Glis2/Nphp7 gene, which is associated with progressive kidney atrophy and reverses cystogenesis arising from Kif mutation (59). CHK1 was also described as interacting physically and functionally with NEK8 to regulate DNA damage response in the renal ciliopathy associated with loss of NEK8 (60). To our knowledge, the data in this study are the first to suggest a role for CHK1 in promoting ciliary disassembly. LRRK2, a kinase frequently mutated in cases of autosomal dominant Parkinson's disease, was first identified as a regulator of ciliation based on a proteomic study that identified its phosphorylation of a group of Rab GTPases, including those required for ciliogenesis. A direct role for LRRK2 in this disease was validated using an LRRK2-R14441G knock-in mouse model, which simulated lesions identified in Parkinson's disease (61). Our data use an orthogonal approach to confirm a role for LRRK2 in ciliation.

The results from this study may help explain discordances between the behavior of cancer drugs observed in cell monoculture in vitro, versus in the setting of heterogeneous tissues in vivo. A characteristic of solid tumors is the juxtaposition of cancer cells that are typically unciliated, stromal cells that are typically ciliated, plus immune, endothelial, and additional cells that have various characteristic ciliated states (10, 62). Among the stromal populations, oscillation between ciliated and nonciliated states is an essential component of cell cycle, and there is a long-standing debate as to whether changes in ciliation are passengers or drivers of cell-cycle transit (9). Intriguingly, data in this study indicate removal of cilia from cells quiescent in G₁ is associated with some signs of cycling. Paracrine communication between heterogeneous tumor cell populations supports signaling cascades important for viability and progression of solid tumors. In this study, >6% of assessed kinase inhibitors affected ciliation and cilia-associated signaling of untransformed fibroblasts and epithelial cells. Two drugs, sunitinib and erlotinib, affected ciliation of cancer cells. These results indicate that changes in the balance of ciliation and the function of ciliary signaling systems may arise from previously undescribed activities of targeted cancer therapies in cancer, stroma, and potentially immune cells.

Given that many of the drugs used target signaling proteins important for growth in both cancer and stromal cells, a current limitation of interpreting the in vivo significance of the findings is the difficulty in separating drug effects related to direct activity in cancer cells from indirect activity mediated via interruption or intensification of paracrine signaling circuits. Further investigation is needed to dissect tumor growth in the context of differentially ciliated stroma, ideally incorporating the use of genetic models developed to study ciliary defects. Such studies may identify novel routes for controlling tumor growth, specifically through disruption of paracrine feedback signaling. Nevertheless, the work provides evidence sufficient to suggest a new paradigm for apparent off-target activity of drugs in vivo; through control of the physical platform for heterocellular ciliary signaling.

No potential conflicts of interest were disclosed.

Conception and design: A.A. Kiseleva, V.A. Korobeynikov, E.A. Golemis

Development of methodology: V.A. Korobeynikov, P. Makhov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.A. Kiseleva, V.A. Korobeynikov, A.S. Nikonova, P. Zhang, M.B. Einarson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.A. Kiseleva, V.A. Korobeynikov, A.S. Nikonova, P. Zhang, J.R. Peterson, E.A. Golemis

Writing, review, and/or revision of the manuscript: A.A. Kiseleva, V.A. Korobeynikov, H. Liu, J.R. Peterson, E.A. Golemis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I.G. Serebriiskii, H. Liu

Study supervision: I.G. Serebriiskii, H. Liu, J.R. Peterson, E.A. Golemis

Other (performed experiments): A.Y. Deneka

The authors thank Emmanuelle Nicolas for her help in the production of qRT-PCR data. Figures 1A, 3A, and 5A were created by a modification of Servier Medical Art templates, licensed under a Creative Commons Attribution 3.0 Unported License (https://smart.servier.com). The PNX0010 patient-derived ccRCC cell line was a kind gift of Vladimir Khazak, PhD (NexusPharma, Inc.). This work was supported by the NIH (grant number R01 DK108195 and PA DOH CURE Award Number 4100077072; to E.A. Golemis), by a William J. Avery Postdoctoral Fellowship from Fox Chase Cancer Center (to A.Y. Deneka), by the Russian Government Program for Competitive Growth of Kazan Federal University (to A.A. Kiseleva and A.Y. Deneka), by NCI R50 CA 211479 (to M.B. Einarson), by NIH GM083025 (to J.R. Peterson), and by NIH Core Grant CA006927 (to Fox Chase Cancer Center).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.