The RNA-Protein Interactome of Differentiated Kidney Tubular Epithelial Cells

Significance Statement

RNA-binding proteins (RBPs) are crucial regulators of cellular biology, and recent evidence suggests that regulation of RBPs that modulate both RNA stability and translation may have a profound effect on the proteome. However, little is known about regulation of RBPs upon clinically relevant changes of the cellular microenvironment. The authors used high-throughput approaches to study the cellular RNA‐binding proteome in differentiated tubular epithelial cells exposed to hypoxia. They identified a number of novel RBPs (suggesting that these proteins may be specific RBPs in differentiated tubular epithelial cells), and found quantitative differences in RBP-binding to mRNA associated with hypoxia versus normoxia. These findings demonstrate the regulation of RBPs through environmental stimuli and provide insight into the biology of hypoxia-response signaling in the kidney.

Visual Abstract

Abstract

Background

RNA-binding proteins (RBPs) are fundamental regulators of cellular biology that affect all steps in the generation and processing of RNA molecules. Recent evidence suggests that regulation of RBPs that modulate both RNA stability and translation may have a profound effect on the proteome. However, regulation of RBPs in clinically relevant experimental conditions has not been studied systematically.

Methods

We used RNA interactome capture, a method for the global identification of RBPs to characterize the global RNA‐binding proteome (RBPome) associated with polyA-tailed RNA species in murine ciliated epithelial cells of the inner medullary collecting duct. To study regulation of RBPs in a clinically relevant condition, we analyzed hypoxia-associated changes of the RBPome.

Results

We identified >1000 RBPs that had been previously found using other systems. In addition, we found a number of novel RBPs not identified by previous screens using mouse or human cells, suggesting that these proteins may be specific RBPs in differentiated kidney epithelial cells. We also found quantitative differences in RBP-binding to mRNA that were associated with hypoxia versus normoxia.

Conclusions

These findings demonstrate the regulation of RBPs through environmental stimuli and provide insight into the biology of hypoxia-response signaling in epithelial cells in the kidney. A repository of the RBPome and proteome in kidney tubular epithelial cells, derived from our findings, is freely accessible online, and may contribute to a better understanding of the role of RNA-protein interactions in kidney tubular epithelial cells, including the response of these cells to hypoxia.

RNA metabolism is closely regulated by a group of specialized proteins that are able to directly interact with RNA. RNA-binding proteins (RBPs) influence bound transcripts starting with their biosynthesis and have a significant effect on RNA stability, translation rate, and the velocity of degradation. This effect on the transcriptome consecutively affects the proteome and thereby puts RBPs at the center of regulation for various signaling pathways, metabolism, and cell fate.¹ Whereas in the past RBPs were assumed to primarily contain classic RNA-binding domains (RBDs), recent work has challenged this view.¹ As an example, RNA interactome capture (RIC) approaches in both hepatic and cardiac cells have shown the unexpected potential of cellular metabolic enzymes to have RNA-binding capacity without disposing of classic RBDs.^2,3 Furthermore, a subset of RBPs is capable of binding both RNA and DNA—thereby influencing replication, transcription rate, and the response to DNA damage.⁴ This increasing knowledge of protein-RNA interactions has been greatly facilitated by the development of a number of novel techniques—allowing for the identification of RBPs and their target transcripts—that are mostly based on of UV-induced crosslinking and RNA-protein precipitation coupled with mass spectrometry and RNA sequencing.⁵ In recent years, these techniques have been employed in cells from a number of different tissues¹; however, little is known about the global RNA-binding landscape in kidney cells. Nonetheless, several targeted studies have shown the general potential of RBPs to affect the cellular biology of kidney cells. One of the most extensively studied RBPs, HuR, has been implicated as a key player in the induction of renal fibrosis and inflammation both in vitro and in vivo,⁶ and could be shown to bind Hif1a mRNA thereby increasing both its stability and translation rate.⁷ This finding is of special interest taking into account the low oxygen tension in the renal medulla and the role of hypoxia signaling in both AKI and CKD.^8,9 Seeing that hypoxia leads to a general stop in protein production and a shift toward preferential translation of proteins required for the cellular response to hypoxia, it is likely that RBPs play a major role in the regulation of the hypoxia-associated transcriptome and proteome.^10,11 In order to extend the knowledge toward a more global understanding of the RNA-protein interactome in the kidney, we employed the RIC approach¹²—for the first time—in ciliated cells derived from the inner medullary collecting duct (mIMCD-3) and characterize the modulation of the RBPome by hypoxia.

Methods

Cell Culture and Transfection

Human HEK 293T and murine mIMCD-3 cells were obtained from ATCC and grown in standard media (+10% FBS; DMEM for HEK293T, DMEM-F12 containing 2 mM Glutamax for mIMCD-3) at 37°C, 5% CO₂, and routinely passaged using 0.05% Trypsin. Cells were tested for mycoplasma contamination using a PCR Mycoplasma Test Kit I/C (Venor GeM; Sigma). Transfections were carried out in 60%–80% confluent cells using calcium phosphate¹³ or lipofection (Lipofectamine 2000; Thermo Scientific) according to the manufacturers’ instructions. mIMCD-3 cells were differentiated using serum starvation as described previously.^14,15

Antibodies

Epitope/Name	Manufacturer	ID
FLAG (M2)	Sigma-Aldrich	F1804–1MG
GFP (B2)	SantaCruz BioTech	sc-9996
Hif1a	Cayman Chemical	10006421
Pericentrin	Abcam	AB4448
β-tubulin (E7)	Developmental Studies Hybridoma Bank	E7
Acetylated tubulin	Sigma-Aldrich	T6793
MPDZ	SantaCruz BioTech	sc-136293
KIF13B	Developmental Studies Hybridoma Bank	AFFN-KIF13B-7H5
ARL13B	Proteintech	17711–1-AP

For a list of secondary antibodies see the Supplemental Methods section.

Plasmids and Cloning

All plasmids used in this study were generated using restriction enzyme cloning, modified using QuikChange Mutagenesis (Stratagene), or were purchased from Addgene. For a detailed list of plasmids and primers used in this study see the Supplemental Methods section.

TALEN-Based Transgenesis

Stably integrated transgenic cell lines were generated using TALEN technology by cotransfecting TALEN-encoding plasmids specific for the human AAVS1 locus, as previously described.^16,17 Starting 24 hours after transfection, cell lines were steadily selected with 2 µM Puromycin. All generated transgenic cell lines were genotyped by integration PCR (for details see the Supplemental Methods section) and characterized by western blot and fluorescence microscopy.

mRNA Interactome Capture of Ciliated mIMCD-3 Cells

Oligo (dT) capture and isolation of protein-RNA complexes were performed as previously described.¹⁸ Briefly, for each condition, ten cell culture dishes with 15 cm diameter were used. After reaching a confluency of about 70%, cells were serum starved to induce differentiation and consequently ciliation for a total duration of 30 hours. During starvation, and before hypoxia treatment, the cells were incubated with 200 µM 4-thiouridine (4-SU) for a total of 18 hours to metabolically label RNA. Six hours before harvesting, 50% of the dishes were transferred to a hypoxia cell culture incubator (1% O₂). After crosslinking with UVA (365 nm), cells were harvested and lysed followed by an oligo (dT) pulldown. RNP complexes were treated with RNase/benzonase and submitted to mass spectrometry after a trypsin digest. The details of these procedures are provided in the Supplemental Methods section.

Generation of Cell Lysates for Proteome Analysis

The whole proteome of mIMCD-3 grown under normoxic or hypoxic conditions was measured in three biologic replicates. Approximately 10⁶ ciliated mIMCD-3 cells were washed with PBS, centrifuged (500 × g, 5 minutes, 4°C), snap-frozen, and resuspended in 300 µl 8 M urea. Lysates were sonicated (Ultrasonics Sonifier) and protein concentration was measured using the Pierce BCA Protein Assay Kit. Equal amounts of total protein were reduced with 50 mM DTT and alkylated in the dark using 1:5 total volume of IAA buffer for 1 hour at room temperature. In order to lower the urea concentration to <2 M, each sample was diluted with 50 mM ammonium bicarbonate. Finally, samples were digested at RT overnight with trypsin, stage-tipped, and submitted to mass spectrometry.

Mass Spectrometry

Peptides were separated on a 25-cm long, 75-µm internal diameter PicoFrit analytic column (New Objective) packed with a 1.9 µm ReproSil-Pur 120 C18-AQ media (Dr. Maisch), using EASY-nLC 1000 (ThermoFisher Scientific). The column was maintained at 50°C. Buffers A and B were 0.1% formic acid in water and 0.1% formic acid in acetonitrile, respectively. Peptides were separated on a segmented gradient from 2% to 5% buffer B for 10 minutes, and from 5% to 20% buffer B for 100 minutes at 200 nl/min. Eluting peptides were analyzed on a QExactive Plus mass spectrometer (ThermoFisher Scientific). Peptide precursor mass-to-charge ratio (m/z) measurements (MS1) were carried out at 70,000 resolution in the 300–1800 m/z range. The top ten most intense precursors with charge states from 2 to 7 only were selected for HCD fragmentation using 25% normalized collision energy. The m/z values of the peptide fragments were measured at 17,500 resolution using an AGC target of 2e⁵ and 80-ms maximum injection time and 4.0% underfill ratio. Upon fragmentation, precursors were put on a dynamic exclusion list for 45 seconds.

Immunofluorescence Microscopy

Cells grown on cover slips were washed once with PBS supplemented with Ca²⁺/Mg²⁺ (PBS+), fixed with 4% PFA solution at RT for 10 minutes, and rinsed twice with PBS+. Cells were then blocked in 5% donkey serum and 0.1% Triton X-100 in PBS at RT and incubated in primary antibody at RT for 1 hour. After rinsing the cells three times, secondary antibody was added for 1 hour at RT, after which the stained cells were mounted on glass slides with 10 µl ProLong Gold mounting medium with DAPI to visualize nuclei. Images were acquired using an Axiovert 200M fluorescence microscope (x63/1.2W) equipped with the Axiocam MRm, supporting the ApoTome system and Axiovision 4.8 software or ZEN software (Zeiss). Unless indicated otherwise, all immunofluorescence microscopy was performed in cells kept in normoxia.

Polynucleotide Kinase Assay

HEK293T cells transfected with pcDNA6 expressing triple FLAG-tagged genes of interest were UVC (254 nm) crosslinked, resuspended in lysis buffer (100 mM KCL, 5 mM MgCl₂, 10 mM Tris pH 7.5%, 0.5% NP40, 1 mM DTT, protease inhibitors), and homogenized through a 23G needle on ice. For nuclear proteins, samples were additionally sonicated (Biorupter Pico; ten intervals: 30 seconds on/off). Lysates were cleared by centrifugation (20,000 × g) and treated with 2 U/ml Turbo DNase (Invitrogen) and 40 U/ml RNase I (Ambion) for 15 minutes at 37°C. Immunoprecipitation was performed with anti-FLAG M2 (Sigma) coupled to Protein G Dynabeads (ThermoFisher) for 2 hours at 4°C. Beads were washed five times with washing buffer (500 mM NaCl, 20 mM Tris pH 7.5, 1 mM MgCl₂, 0.05% NP40) and twice with polynucleotide kinase (PNK) buffer (50 mM Tris pH 7.5, 50 mM NaCl, 10 mM MgCl₂, 0.5% NP40). Beads were resuspended in PNK buffer containing 5 mM DTT, 0.2 μCi/μl (γ³²P)ATP (Hartmann-Analytic), and 1 U/μl T4 PNK (ThermoFisher). The labeling was carried out for 20 minutes at 37°C. Beads were washed five times with PNK buffer and boiled with laemmli buffer. The samples were resolved on 4%–12% bis-tris gels (ThermoFisher) and transferred onto Protran Nitrocellulose membrane (Schleicher and Schuell). The blot was exposed to a storage phosphor screen (Amersham) and the signal was detected with a Typhoon scanner (GE Healthcare). IP efficiency was controlled by western blotting with anti-FLAG M2 (Sigma). All PNK assays were performed using cells kept under normoxic conditions.

Data Analysis, Statistics, and Data Sharing

Statistical analysis of both the whole proteome and the RNA interactome raw data were performed with Perseus software,¹⁹ version 1.6.0.7. Default parameters (only identified by site, reverse, and contaminants) were used for filtering the dataset. Raw data and MaxQuant²⁰ (version 1.5.3.8) output were uploaded to the PRIDE repository (http://www.ebi.ac.uk/pride; project accession: PXD010530). More details regarding the proteomics analyses, the consecutive GO term/Pfam/Smart, and KEGG enrichments as well as correlation analyses are provided in the Supplemental Methods section. An interactive online database was created using the “shiny” package in “R” and is provided at http://shiny.cecad.uni-koeln.de:3838/mIMCD_RBPome.

Results

RIC to Identify the Global RBPome in Differentiated Inner Medullary Collecting Duct Cells

To study the RNA interactome in kidney epithelial cells we used the photoactivatable nucleotide 4-SU for crosslinking of RNA-bound proteins.^18,21 To this end, mIMCD-3 cells were grown to 70% confluency followed by starvation-induced differentiation for 12 hours and subsequent 4-SU labeling (Figure 1A). Six of 12 samples (three crosslinked and noncrosslinked, respectively) were exposed to low oxygen tension before crosslinking in order to allow for an analysis of hypoxia-induced changes in the RNA-binding landscape. After 18 hours of incubation with 4-SU, crosslinking was performed in living cells using a UVA pulse and polyA-tailed RNAs were captured using oligo-dT beads (Figure 1A). Visualization of proteins contained in small aliquots of these precipitates on a silver gel confirmed efficient crosslinking (Figure 1B). The clear separation of crosslinked and noncrosslinked samples is depicted in a principal component analysis (Figure 1C). After RNase treatment of the precipitates to remove the non–protein bound RNA fraction, samples were analyzed for protein content using mass spectrometry (Figure 1A). In total, we identified 1302 proteins (Supplemental Table). Because differences in oxygen tension did not lead to major global changes (as indicated in Figure 1C, Supplemental Figure 1), we decided to pool all samples irrespective of the treatment at this point to allow for a comprehensive identification of RBPs. A detailed comparative analysis of hypoxia-induced changes is provided below. Because the comparison of protein abundance in crosslinked over noncrosslinked samples is generally used as an indicator of confidence that a protein identified by RIC is truly an RBP, we classified the proteins according to the following criteria (Supplemental Figure 2A). Class I RBPs are significantly enriched (on the basis of iBAQ intensities) in the crosslinked samples over noncrosslinked samples (t test, FDR<0.1; s0=0.1). Furthermore, all proteins for which significance could not be calculated due to lack of detection in noncrosslinked samples (identification in ≤1 of 6 replicates), despite clear presence after crosslinking (identification in ≥4 of 6 replicates), were attributed to class I. A total of 510 proteins fulfilled these criteria (Supplemental Table). However, proteins that were not significantly enriched may still be RBPs. Consequently, a second class of RBPs—class II—was defined containing all proteins detected in our screen that had previously been identified to show RNA-binding capacity using a recently published compendium, which includes six datasets for Mus musculus and six for Homo sapiens.¹ On the basis of these criteria a total of 1058 proteins fell into these two classes of RBPs (510 class I, 548 class II; see Supplemental Table). These proteins are illustrated in the scatterplot in Figure 1D, depicting the enrichment in crosslinked samples (FDR<0.1 for enrichment over noncrosslinked samples; s0=0.1) as well as the overlap with both the mouse RBPs and the murine orthologs of the human RBPs listed in the compendium.¹ This plot underlines the validity of the dataset because the vast majority of proteins captured had been described to bind to RNA previously (Figure 1D). In order to provide a user-friendly platform to query our data, an online repository was set up that can be interrogated on the basis of user interest (http://shiny.cecad.uni-koeln.de:3838/mIMCD_RBPome).

Figure 1.

RNA-interactome capture reveals hundreds of RNA-associated proteins in differentiated inner medullary collecting duct cells. (A) Culturing scheme depicting the treatment of mIMCD-3 cells and schematic representation of the RIC procedure. mIMCD-3 cells were grown to 70% confluency and starved by serum deprivation to induce differentiation. After 12 hours, the cell culture medium was supplemented with 4-SU. 4-SU labeling and starvation were continued for 12 hours before 20 of 40 dishes were transferred to hypoxic conditions (1% O₂, blue) for 6 hours, whereas the other 20 dishes were kept under normoxic conditions (21% O₂, pink). After this treatment ten dishes of each condition were UVA crosslinked (UVA+), whereas the other ten dishes remained noncrosslinked (UVA−). Aliquots of the collected cell lysates were used for proteome analysis and the rest were submitted to RIC. RIC: after crosslinking and cell lysis, polyadenylated transcripts were captured with oligo-dT beads. The captured RNA-protein complexes were RNase treated and putative RBPs were identified using quantitative label-free proteomics. (B) Analysis of protein enrichment after oligo (dT) capture and RNase treatment by silver staining. The enrichment of a complex protein pattern is only detectable in the lanes with UVA-irradiated samples (+CL) and is absent in the lanes with nonirradiated samples (−CL). Lanes with samples after hypoxic or normoxic treatment are indicated by N or H, respectively. M, molecular weight marker. (C) RNA-bound proteome (RBPome) data of mIMCD-3 cell samples. Principal component analysis of RIC (on the basis of iBAQ intensities) for crosslinked (+CL/red) and noncrosslinked (−CL/black) samples showing a clear separation of the two groups. Open circles represent samples with hypoxia treatment; filled circles represent samples kept under normoxic conditions. (D) Scatter plot of the t test comparison of protein abundance in the crosslinked and noncrosslinked samples. The x axis is the mean log₂ difference in abundance of crosslinked versus noncrosslinked samples; presented on the y axis are the corresponding −log₁₀ P values. The cutoff for significance is an FDR<0.1. RBPs significantly enriched according to our cutoff are represented by large circles. RBPs not reaching significance are represented by small circles. Red, mouse RBPs and mouse orthologs of human RBPs previously described (summarized in Hentze et al.¹ [2018]); black, novel RBPs; gray, others. FC, fold change; Hyp, hypoxia; Norm, normoxia; vs, versus.

Characterization of the Proteins Contained in the RBPome of mIMCD-3 Cells

On the basis of these findings, we went on to perform a detailed comparison of the proteins identified in our RIC with the compendium¹—including both their list of mouse proteins as well as the mouse orthologs of the human RBPs. There was a large overlap with the vast majority of the 1058 class I/class II RBPs (Figure 2A). Interestingly, 25 of the proteins had not been identified by previous screens using mouse or human cells, suggesting that these proteins may be specific RBPs in differentiated kidney epithelial cells (Figure 2A and 1D, see also “RBPome” tab in the online repository). To further characterize the enrichment of RBPs in our dataset compared with the cellular proteome of mIMCD-3 cells, we performed quantitative label-free proteomics of whole-cell lysates. These studies identified 6033 individual proteins that were later used as a background for GO-term analyses. For quantification of hypoxia-associated changes in abundance this list was reduced to proteins identified in at least two of three replicates of either condition (normoxia or hypoxia), resulting in 4883 quantifiable proteins (see “Total proteome” tab in the online repository and Supplemental Table). The comparison of the mIMCD-3 proteome with the list of known mouse RBPs listed in the compendium¹ revealed that about 28% of these proteins were RNA-associated (Figure 2B). Performing the same comparison for the class I RBPs or both class I and II RBPs from our RIC experiment showed, as expected, a much higher proportion of known RNA binders (approximately 91%; Figure 2B). This is not a general capacity of nucleic-acid binding but specific to RNA, which is illustrated by the overlap with proteins contained in the gene ontology (molecular function) terms “RNA-binding” and “DNA-binding” (Figure 2C). Even though the number of proteins overlapping with the term “RNA-binding” was much larger, a significant number of proteins were also assumed to bind to DNA and could be part of the group of “dual binders” that have recently gained increasing attention (Figure 2C).²² Using the whole-cell proteome (described in Figure 2B) as a reference, we asked which known protein domains were enriched among the identified RBPs. Both PFAM and SMART enrichment analyses clearly showed nearly all of the most highly enriched domains to be classic RBDs, with helicase-associated domains on the one hand and the classic RBDs RRM and KH on the other hand being the most prominent terms (Figure 2D). Gene ontology as well as pathways analyses—as before, using the proteome as a reference—of the 510 class I RBPs revealed that top terms in all three GO categories again demonstrated a highly significant overrepresentation of RNA-associated processes (Figure 2E, Supplemental Figure 2, B and C).

Figure 2.

RNA-interactome capture in mIMCD-3 cells clearly enriches for proteins with RNA-binding capacity. (A) Venn diagram depicting the number of RBPs identified in the mIMCD-3 RBPome (white) and the overlap with previously identified mouse (light gray) and human (dark gray) RBPs. Of 1058 RBPs identified in mIMCD-3 cells, 1033 were previously described in mouse (964) or human (858) samples (Hentze et al.¹ [2018]). Twenty-five RBPs were not previously identified in mouse or human. (B) The fraction of known mouse RBPs¹ contained in the mIMCD-3 proteome in comparison with the mIMCD-3 RBPome. Of 6033 proteins measured in the mIMCD-3 proteome, 1695 have been previously identified as mouse RBPs (28.1%). For the 510 class I RBPs, 464 have been previously observed (91%) and, among the combined class I and class II RBPs, 964 of 1058 (91.1%) have been previously identified. We calculate a 3.2-fold enrichment of class I and class I plus class II RBPs in the RBPome over the proteome. Gray, proteins previously identified as RBPs; white, not known to be RBPs (other). (C) Comparison of mIMCD-3 RBPome (gray) and mIMCD-3 proteome (white) for GO terms: “RNA-binding” and “DNA-binding.” (D) Pfam and Smart protein domains with significant enrichment in the mIMCD-3 RBPome. Statistical significance was determined with the Fisher exact test (Perseus software; FDR<0.05). (E) mIMCD-3 RBPome gene ontology enrichment analysis. Here, the ten most significantly overrepresented (black) and underrepresented (gray) GO terms for “molecular function” and “biological process (slim)” are depicted. Statistical significance was determined using the Fisher exact test (Perseus software; FDR<0.05). BP slim, biological process; GO, gene ontology; MF, molecular function.

Proteins Identified for the First Time as RNA Binders in Differentiated mIMCD-3 Cells

As described above, 25 class I RBPs identified in our RIC have not been reported in previous screens of the mammalian RBPome (Figure 2A). Figure 3A provides a list of these novel and potentially context-specific RBPs (“novel RBPs”). As this exclusive detection in mIMCD-3 cells may not only be a consequence of context-specific RNA-binding capacity but could also be explained by absence of protein expression in the cell lines used in previous screens, we queried these proteins using a freely accessible proteome database (“ProteomicsDB”).²³ This analysis clearly showed that all of our novel RBPs (apart from Oasl2 and Gm9242 for which no data are available in “ProteomicsDB”) are expressed in numerous cell lines, including at least one cell line used in previous RIC screens,¹ (Supplemental Figure 3A). We decided to confirm RNA binding for three proteins from our screen (Figure 3B). Specifically, Gadd45gip1 is one of the entirely novel RBPs (Figure 3A), whereas Hic2 and Mfap1a/b had been identified in previous screens but had not been confirmed as individual RNA binders so far. Firstly, we visualized the subcellular localization of these proteins. Mfap1a/b and Hic2 were predominantly nuclear with only a slight cytoplasmic signal, whereas Gadd45gip1 showed both a (peri)nuclear and cytoplasmic localization (Figure 3C). In order to validate their RNA-binding potential we employed PNK assays, revealing a clear-cut radioactive band at the size of the respective protein with a strong enrichment in the crosslinked samples (Figure 3D). These results did not only confirm Hic2 and Mfap1a/b for the first time on an individual protein level to be RNA binders but also showed that our list of entirely novel mammalian RBPs contains true RNA binders as indicated by the result for Gadd45gip1.

Figure 3.

RNA-interactome capture identifies novel RNA binders in mIMCD-3 cells. (A) Table of novel, mIMCD-3–specific RBPs, previously not identified as mouse or human mRNA-interacting proteins. Depicted are the gene names, protein names according to Uniprot and MGI, and the selection criteria. The top 19 proteins (#) were significant in the performed t test (Perseus software). The bottom six proteins (*) were measured at least four times in the crosslinked samples (+CL) and not more than once in the noncrosslinked samples (−CL). (B) List of proteins selected for biochemical confirmation of RNA-binding capacity. The table contains information on gene name, protein name, presence in previous RIC studies as summarized for mouse (Mm) and human (Hs) datasets in the Hentze compendium, classification in the mIMCD-3 RBPome (class), and t test significance. (C) Cellular localization pattern of MFAP1, GADD45GIP1, and HIC2. MFAP1: HEK293T cells expressing an integrated, single copy of the human MFAP1 CDS fused to eGFP, using the TALEN approach, were subjected to fluorescent imaging. GADD45GIP1 and HIC2: HEK293T cells transiently expressing the human CDS of GADD45GIP1 or HIC2 fused to triple FLAG were subjected to immunofluorescent imaging. DAPI was used as a nuclear counterstain. Scale bar, 20 µm. (D) Biochemical validation of Mfap1a/b, Hic2, and Gadd45Gip1 as RBPs. Briefly, the human CDS of MFAP1, HIC2, and GADD45GIP1 were cloned into the 3xFLAG-pcDNA6 and transiently expressed in HEK293T cell. FLAG-tagged proteins were immunoprecipitated from crosslinked (+) and noncrosslinked (−) samples and the associated RNA was labeled by T4 PNK with 32P. The protein-RNA complexes were separated on PAA-gels and blotted onto nitrocellulose membranes. PNK-assay: autoradiograph of the membrane containing the indicated protein with the associated RNA labeled with 32P. Western blot: visualization of FLAG-tagged protein by western blotting with the anti-FLAG antibody. Hs, homo sapiens; Mm, mus musculus; n.d., not detected.

Cilia-Associated RBPs

Primary cilia are a hallmark of differentiated renal tubular epithelial cells. To demonstrate ciliation upon serum starvation–induced differentiation we stained for cilia markers. As shown in Figure 4A and Supplemental Figure 4A, our mIMCD-3 cells showed a consistently high degree of ciliation after 30 hours of serum starvation. We next asked whether RBPs identified in our experiment may also be cilia-associated proteins. To address this question, the RBP compendium¹ as well as our dataset of RBPs was analyzed for overlaps with both cilia-associated protein complexes as characterized by the “SYSCILIA” consortium²⁴ and the ciliary membrane-associated proteome that we had determined using the “APEX” technology.²⁵ Both RBP datasets contained a subset of cilia-associated proteins (Figure 4B, Supplemental Table). The scatterplot in Figure 4C shows cilia-associated proteins identified in the two screens plotted against the enrichment by crosslinking. Regarding the 25 RBPs identified for the first time in mIMCD-3 cells (Figure 3A)—of which none overlapped with the SYSCILIA and APEX datasets—we performed a more detailed literature search and identified five additional proteins (Kif13b, Mpdz, Nme1/2, Wasl, Cdk1) to have a known ciliary function (Figure 4D).^26–30 Because ciliary targeting may be context-specific and none of these proteins had been examined in mIMCD-3 cells before, we confirmed ciliary localization in this cell type for Mpdz and Kif13b by immunofluorescence as a proof of principle (Supplemental Figure 3B).

Figure 4.

Cilia-associated proteins show RNA-binding capacity. (A) Immunofluorescence imaging of ciliated mIMCD-3 cells. The cells were serum starved for 30 hours to induce ciliogenesis, fixed, and stained with antibodies specific for actylated tubulin (green) and pericentrin (magenta). DAPI (blue) was used as a nuclear counterstain. Scale bar, 20 µm. (B) Comparison of known mouse RBPs¹ and mIMCD-3 RBPs with cilia-associated proteins as described by Boldt et al.²¹ (2016) (SYSCILIA) and Kohli et al.²² (2017) (APEX). The compendium of known mouse RBPs (n=1914) contains 93 ciliary proteins as determined by APEX (red) and 20 ciliary proteins characterized by the SYSCILIA consortium (blue). The mIMCD-3 RBPome (n=1058) shares 62 and ten ciliary proteins with the APEX and the SYSCILIA datasets, respectively. (C) Identity of ciliary proteins in the mIMCD-3 RBPome. Depicted in red are proteins overlapping with the APEX dataset²² and/or the SYSCILIA dataset.²¹ For proteins significantly enriched in the mIMCD-3 RBPome the protein name is indicated. Gray, mIMCD-3 proteins without correspondence to APEX and SYSCILIA datasets. For details of the scatter plot refer to Figure 1D. (D) Scatter plot illustrating mIMCD-3 RBPs (never identified in mammalian RIC experiments before) associated with a known ciliary function. A PubMed-based literature search was performed for the 25 mIMCD-3 RBPs (black) shown in Figure 3A. Protein names are indicated for RBPs associated with the search term “cilia.” For details regarding the scatter plot refer to Figure 1D. +CL, crosslinked; −CL, not crosslinked; FC, fold change; vs, versus.

Modulation of the RBPome and Proteome by Hypoxia

Because hypoxia plays a critical role in renal (patho)physiology and is known to affect mRNA stability and translation, we addressed the question of how hypoxia might affect the interaction between RBPs and their target mRNAs. As indicated above, 50% of the cells were exposed to defined levels of hypoxia before crosslinking (Figure 1A). To obtain a view on dynamic changes of RBP–target RNA binding, rather than hypoxia-associated changes in abundance of the respective RBPs, we used a limited duration of hypoxia (6 hours). This treatment did not affect ciliation (Supplemental Figure 4A) but led to a pronounced stabilization of Hif1a and its enrichment in the nucleus (Supplemental Figure 4, B and C), indicating activation of the hypoxia transcriptional program. In order to check what effect this treatment had on global protein expression, we performed quantitative label-free proteomics (Figure 5A, see also “Hypoxia versus Normoxia” tab in the online repository). No proteins were downregulated and the six upregulated proteins reaching significance (FDR<0.1, s0=0.1) had all been previously reported as hypoxia-signaling associated.^{10,11,31–34} Additionally, by calculating the average fold change of the total number of proteins in the whole dataset (FC=0.015, n=4883) and the average fold change of all Hif1a targets (FC=0.45, n=94) depicted in Figure 5A, we confirmed a trend toward a positive regulation of all Hif1a targets in the hypoxia-treated samples. Moreover, our dataset revealed that 46 proteins were exclusively detected under hypoxic conditions but were never identified in normoxia, including Hif1a (Supplemental Table). Seventeen proteins were never measured in hypoxia, including the RNase Drosha that is known to be downregulated in hypoxic conditions³⁵ (Supplemental Table). Taken together, as intended, 6 hours of hypoxia only had a limited effect on global protein expression with few proteins being regulated significantly (Figure 5A, see also “Hypoxia versus Normoxia” tab in the online repository and Supplemental Figure 4D). We then went on to compare the RBPome of cells grown under different oxygen tension. To this end, we performed separate t tests for the normoxia and hypoxia datasets (3 +CL versus 3 −CL; FDR 0.1, S0=0.1) and compared these with the t test of the whole dataset (6 +CL versus 6 −CL) shown in Figure 1D (Figure 5B). This revealed six proteins (Nme1/2, Anxa2, Rplp2, Sdad1, Sod1, Tmem33) that exclusively reached the FDR threshold of class I RBPs in hypoxic cells, and this was also the case for six proteins (Dhx57, Mcat, Ppp1r10, Rps4x, Secisbp2l, Tcp1) in normoxic cells (Figure 5B, see also “Hypoxia versus Normoxia” tab in the online repository and Supplemental Table). To allow for a more quantitative analysis we next correlated the enrichment reached by crosslinking in normoxic and hypoxic cells and found a set of proteins to be more strongly enriched in one of the conditions (normoxia 41 proteins, hypoxia 41 proteins; Supplemental Table, see also “Hypoxia versus Normoxia” tab in the online repository). Using this approach, seven of the 12 proteins (Figure 5B) identified to be specific to one of the two conditions were also found to be outside of the 95% prediction interval, increasing the confidence in these candidates (indicated in black, Figure 5C). Importantly, none of the 12 proteins was altered significantly between the proteomes (Figure 5D), clearly pointing toward hypoxia-associated differential binding. Figure 5E provides more information on the seven proteins identified by both approaches (i.e., derived from the overlap of Figure 5B and C). All putatively hypoxia-associated RBPs in our screen had previously been associated with oxygen metabolism and signaling in the literature (Supplemental Table). Two of them are among the novel RBPs identified in our study, Nme1/2 and Tmem33, both of which reach significance only in hypoxic samples (Figure 5E).

Figure 5.

The RBPome is modulated by oxygen tension. (A) Volcano plot illustrating differentially abundant proteins between the proteomes of hypoxia-treated and normoxic mIMCD-3 cells. The −log₁₀ P value is plotted against the log₂ fold change (hypoxia versus normoxia). Significantly regulated proteins are above the cutoff line and are indicated by name (Perseus software, t test, FDR<0.1, s0=0.1). In total, 4883 proteins were plotted. Red, known Hif1a targets^30,31,34; blue, Kdm3a, manually curated from the literature as an Hif1a target.³⁰ Average fold change of total number of proteins (n=4883) 0.015; average fold change of Hif1a targets (n=94) 0.45. (B) Venn diagram depicting the comparison of RBPomes of mIMCD-3 cells grown in hypoxic and normoxic conditions. Of the 289 (normoxia) and 206 (hypoxia) RBPs reaching the threshold of class I RBPs, six and six proteins are exclusively associated with normoxia or hypoxia, respectively. Gray, normoxia-associated RBPs (class I RBPs reaching statistical significance when comparing three +CL versus three −CL samples in normoxia); blue, hypoxia-associated RBPs (class I RBPs reaching statistical significance when comparing three +CL versus three −CL samples in hypoxia); red, six +CL versus six −CL samples (455 class I RBPs reaching statistical significance in the comparison of the total dataset comprising six +CL and six −CL samples). (C) Scatter plot showing the correlation of hypoxia log₂ fold changes (+CL versus −CL) on the x axis versus the log₂ fold change values of normoxia (+CL versus −CL) on the y axis. The linear regression was calculated with R (black line, lm () method, formula: y=0.3170+0.7069*x). RBPs beyond the calculated 95% prediction interval (outside the gray lines) show a significantly different FC in the two conditions, sometimes going in opposite directions (from positive to negative), suggesting a regulation of the binding to target RNAs in normoxia or hypoxia. Red, class I and II RBPs (with names for proteins above and below the calculated prediction interval); big circles, significant in t test (FDR<0.1); small circles, NS in t test (FDR≥0.1); black, seven RBPs below and above the calculated prediction interval matching the 12 differentially bound RBPs in Figure 5B; gray, others. (D) Volcano plot illustrating the abundance of differentially bound RBPs (Figure 5B) between the proteomes of hypoxia-treated and normoxic mIMCD-3 cells. Red, differentially bound RBPs significant in normoxia; blue, differentially bound RBPs significant in hypoxia. Two proteins (Dhx57and Secisbp2l) were not quantified in the proteome. For details regarding the volcano plot refer to Figure 5A. (E) List of high-confidence RBPs associated with either hypoxia or normoxia. This table shows the seven of 12 differentially bound RBPs that were measured in hypoxia- or normoxia-treated cells, respectively, and in addition were beyond the calculated 95% prediction interval. Given are gene and protein names, significance in t test (+CL versus −CL, in hypoxia or normoxia), information on the presence in previous RIC studies as summarized for mouse (Mm) and human (Hs) datasets in the Hentze compendium,¹ classification in the mIMCD-3 RBPome (class), and classification as to whether the protein is a novel RBP. +CL, crosslinked; −CL, not crosslinked; FC, fold change; vs, versus.

Discussion

Here, we report the first global analysis of RBPs in ciliated epithelial kidney cells, paving the road toward kidney research gaining an insight into the rapidly evolving field of RNA-protein interactions in cellular and organ biology. Our approach (ciliated kidney cells in hypoxia and normoxia) did not only yield an atlas of 1058 RBPs but also allowed for the identification of 25 novel RBPs that were not detected in other systems before. The fact that these novel RBPs are expressed in other cultured cell lines that had been used for RIC screens previously emphasizes the importance of studying such interactions in various cell types and conditions. However, it has to be noted that, apart from context-specific changes in RNA-binding capacity, technical differences (e.g., as to MS/MS) can contribute to such a finding.

Although a few studies have already pointed toward a crucial role of specific RBPs regarding kidney physiology,⁶ the role of this class of proteins in ciliary biology is entirely unclear. It is easy to believe that RBPs—which affect most biologic processes—may also affect cilia formation, signaling, and breakdown through their profound effect on cellular RNA biology. However, it is even more intriguing that datasets like the one presented here allow for the speculation—because a number of the RBPs detected are actually proteins that have been shown to localize to primary cilia—that the cilium might be a novel site of action for RBPs. It is important to note that most of these proteins do not necessarily only localize in cilia and also have other cellular functions. Consequently, such a hypothesis will require further studies comparing ciliated to nonciliated cells and addressing the potential mode of action of putative ciliary RBPs.

It will now be extremely interesting to move these efforts from a characterization of the RBPome toward function in kidney (patho)physiology. Here, in order to provide some first clues, we focused on the potential effect of hypoxia taking into account that the renal medulla is a site of limited oxygen tension and that lack of oxygen is one of the key factors in both AKI and CKD.³⁶ A very interesting example of the potential role of RNA-protein interaction in the response to hypoxia is the targeting of specific mRNAs to polysomes for translation via the 5′ cap binding of the HIF2α-RBM4-eIF4E2 complex.^37,38 Furthermore, key aspects regarding hypoxia are cellular metabolic adaptations (e.g., activation of glycolysis) and recent work has suggested that a large number of metabolic enzymes possess RNA-binding capacity.^2,3 Interestingly, a significant number of enzymes involved primarily in glucose handling in the kidney were identified as RBPs either by our RIC screen or previously published data³ (Supplemental Figure 5A). Importantly, two of the six proteins depicted in Supplemental Figure 5A—Aldolase a and Enolase 1—are below the 95% prediction interval, suggesting significantly increased RNA binding in hypoxic conditions (Supplemental Figure 5, B and C). Future experiments identifying the RNA molecules bound by these enzymes and their effect on enzyme activity will help understanding the effect of this additional layer of regulation regarding hypoxia-induced alterations of cellular metabolism.

In summary, our study provides the first global view on RBPs in ciliated kidney epithelial cells and will—by providing a visualization of large-scale datasets on both the whole-cell proteome and the RBPome of mIMCD-3 cells—serve as a resource for future targeted analyses of specific RBPs and the interaction with their targets in kidney (patho)physiology. In order to allow for optimal usability of this resource we provide these data as an online tool that can be freely accessed and queried (http://shiny.cecad.uni-koeln.de:3838/mIMCD_RBPome).

Disclosures

None.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2018090914/-/DCSupplemental.

Table of contents—Supplemental Tables and Figures.

Supplemental Figure 1. Heatmap of 4 replicates of RNA interactome capture experiments in mIMCD-3 cells.

Supplemental Figure 2. Characterization of the mIMCD-3 RNA interactome.

Supplemental Figure 3. Expression and ciliary localization of novel RBPs.

Supplemental Figure 4. Hypoxia signaling–associated changes.

Supplemental Figure 5. RNA-binding proteins in hypoxia-induced metabolic changes.

Supplemental Table. mIMCD-3 RNA interactome and proteome.