Identification of RNA-binding proteins in macrophages by interactome capture

Pathogen components, such as lipopolysaccharides of Gram-negative bacteria that activate Toll-like receptor 4, induce mitogen activated protein kinases and NFκB through different downstream pathways to stimulate pro- and anti-inflammatory cytokine expression. Importantly, post-transcriptional control of the expression of Toll-like receptor 4 downstream signaling molecules contributes to the tight regulation of inflammatory cytokine synthesis in macrophages. Emerging evidence highlights the role of RNA-binding proteins (RBPs) in the post-transcriptional control of the innate immune response. To systematically identify macrophage RBPs and their response to LPS stimulation, we employed RNA interactome capture in LPS-induced and untreated murine RAW 264.7 macrophages. This combines RBP-crosslinking to RNA, cell lysis, oligo(dT) capture of polyadenylated RNAs and mass spectrometry analysis of associated proteins. Our data revealed 402 proteins of the macrophage RNA interactome including 91 previously not annotated as RBPs. A comparison with published RNA interactomes classified 32 RBPs uniquely identified in RAW 264.7 macrophages. Of these, 19 proteins are linked to biochemical activities not directly related to RNA. From this group, we validated the HSP90 cochaperone P23 that was demonstrated to exhibit cytosolic prostaglandin E2 synthase 3 (PTGES3) activity, and the hematopoietic cell-specific LYN substrate 1 (HCLS1 or HS1), a hematopoietic cell-specific adapter molecule, as novel macrophage RBPs. Our study expands the mammalian RBP repertoire, and identifies macrophage RBPs that respond to LPS. These RBPs are prime candidates for the post-transcriptional regulation and execution of LPS-induced signaling pathways and the innate immune response. Macrophage RBP data have been deposited to ProteomeXchange with identifier PXD002890.

Following activation of inflammation-related genes (5,6), post-transcriptional checkpoints are critical for the precise immune response modulation (7)(8)(9). Information about post-transcriptional mechanisms that regulate protein synthesis downstream of TLR4 to adjust the range and extent of the immune reaction is still fragmentary.
RNA-binding proteins (RBPs) that coordinate mRNA turnover and mRNA translation contribute to rapid and purposeful immune cell responses. Specific RBPs that interact with AU-rich elements (AREs) in mRNA 3' untranslated regions (3'UTR) (ARE-BPs) have been shown to directly regulate cytokine mRNA translation and/or stability. AREs were first discovered in the short-lived human and mouse tumor necrosis factor (TNF) mRNAs (10). Besides TNF, ARE-BPmediated regulation also controls the synthesis of other pro-and anti-inflammatory factors, such as interleukins and inducible nitric oxide synthase (7,11,12). Several ARE-BPs have been identified (7,12): target mRNA translation is inhibited by T-cell-restricted intracellular antigen 1-related protein, CUG-repeat binding protein 2 and Fragile-X-related protein (13)(14)(15), while target mRNA decay is initiated by tristetraprolin (TTP), butyrate response factor 1 and 2 and KH-type splicing regulatory protein (16)(17)(18) through degradation factor recruitment. In contrast, Y-box binding protein 1 and Hu-antigen R (HUR) stabilize ARE-containing mRNAs (19)(20)(21), the latter has also been shown to regulate mRNA translation (22)(23)(24). Furthermore, AUF-1 (heterogeneous ribonucleoprotein D, HNRNP D) either inhibits or promotes target mRNA decay (25,26). However, RBPs not only directly regulate stability and/or translation of cytokine mRNAs. Recently we could show that the synthesis of TLR4 downstream transforming growth factor-β-activated kinase 1 (TAK1), an essential signaling molecule in accurate cytokine expression control, is controlled by HNRNP K in murine macrophages (27). In vitro RNAbinding assays revealed that the HNRNP K homology domain 3 of HNRNP K interacts specifically with a U/CCCC (n) motif in the TAK1 mRNA 3'UTR. HNRNP K depletion in macrophages did not affect TAK1 mRNA synthesis, but increased its translation. The resulting elevated TAK1 protein level changed the macrophage LPS response to an earlier and extended P38 phosphorylation, enhancing cytokine mRNA synthesis. This suggests that LPS-induced TLR4 activation abrogates TAK1 mRNA translational repression by HNRNP K and the newly synthesized kinase TAK1 boosts the macrophage inflammatory response (27).
To systematically identify regulatory RBPs that modulate the LPS-induced macrophage response, we employed RNA interactome capture (28) combining UV-induced protein-RNA crosslinking in LPS-activated and untreated RAW 264.7 macrophages with oligo(dT) capture of polyadenylated RNAs and bound RBPs after cell lysis, and subsequent identification of eluted proteins by mass spectrometry. Our analysis identified 402 RBPs in macrophages, referred to here as macrophage RNA interactome, including 91 proteins not previously annotated as RBPs.
An comparison of the macrophage RNA interactome with the RNA interactomes of HeLa cells (29), HEK293 cells (30) and murine embryonic stems (ES) cells (31) identified 32 RAW 264.7 cell-specific RBPs. Of that group, 19 proteins that lacked RNA-related functional annotations were classified as novel macrophage RBPs. From these RAW 264.7 cell-specific RBPs we selected two candidates: P23, which acts as a heat shock protein 90 (HSP90) co-chaperone (32) and was shown to possess cytosolic prostaglandin E2 synthase 3 (PTGES3) activity (33); and the hematopoietic cell-specific LYN substrate 1 (HCLS1, HS1) Figure 1A). Enrichment and isolation of RBPs bound to polyadenylated RNA was performed as described (28,29

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)
In each of two replicate experiments, three samples were compared: (1) LPS treatment (+LPS), UV-crosslinking (+CL), (2) no LPS (-LPS) (+CL) and (3) controls (ctrl.) (+LPS, -LPS), both non-crosslinked (no CL). Controls were combined because they contain only few proteins and the background was usually fairly constant. Protein samples were processed as described before (28,29). Briefly, proteins were digested, followed by a stable-isotope labeling step via reductive methylation (using differential labels producing 'light', 'intermediate' and 'heavy' peptides in the respective samples) ( Figure 1A). Samples were combined, and peptides were fractionated by iso-electric focusing. The twelve fractions generated were analyzed by liquid chromatography (LC) coupled to an OrbitrapVelosPro mass spectrometer (Thermo Fisher Scientific) using a Proxeon nanospray source. Reverse phase chromatography was performed with a nanoAcquity

MS data processing
Raw files were processed with MaxQuant version 1. Razor peptides, which represent non-unique peptides assigned to the protein group with the most other peptides, following Occam's razor principle (39) and unique peptides were quantified only as unmodified peptides. Cysteine carbamidomethylation and methionine oxidation were set as fixed and variable modification, respectively. The minimum peptide length was set to 6 amino acids, the enzyme specificity was set to trypsin/P, the maximum allowed miss-cleavage was set to 2, and the false discovery rate (FDR) was set to 0.01 for both peptide and protein identifications. Re-quantification and match between runs were also performed unless stated otherwise. The protein identification was reported as a "protein group" if no unique peptide sequence to a single database entry was identified. Statistical analysis was performed using the Hochberg's method. The following criteria were applied: EASE score ≤0.1, at least three proteins could be assigned to one term and corrected p-value ≤0.05.

Experimental Design and Statistical Rationale
In each of the two replicate experiments, three samples were compared: 1] LPS treatment (+LPS), UV-crosslinking (+CL), 2] no LPS treatment (-LPS), UV-crosslinking (+CL) and 3] controls (+LPS, -LPS), both non-crosslinked (no CL). The controls (ctrl.) of the two experiments were combined because they contain only few proteins, and the background was usually fairly constant.
We have digested each of these three protein samples, followed by a stable-isotope labelling step producing 'light', 'intermediate' and 'heavy' peptides in the respective samples ( Figure 1A).
Statistical analysis was performed with the Limma package in R/Bioconductor (41, 42) applying adjustment of p-values for multiple testing with Benjamini and Hochberg's method.

RNA preparation and quantitative real-time PCR
For analysis of precipitated RNA covalently bound proteins were removed by proteinase K.
Samples from input and eluate were pre-incubated with 5×proteinase K buffer (50 mM Facility, Australian National University). Reads were mapped to the mouse transcriptome and all non-rRNA reads were assigned to categories.

In vitro transcription
RNA for spike in controls and for extraction control was transcribed with T7 MEGAscript® Kit (Thermo Fisher Scientific).

Immunoblot analysis
Western blot assays were performed as described previously (51) and analyzed on a LAS-4000 system (GE Healthcare Life Sciences).

Immunofluorescence and Fluorescence in situ hybridization (FISH)
Immunofluorescence staining was essentially performed as described in (48) with specific antibodies against HCLS1 and P23 and FISH with an oligo(dT) probe as in (52).

Immunoprecipitation
Crosslinked RAW 264.7 cells were collected by centrifugation for 5 min at 500×g and 4°C and lysed in 1 volume IP buffer (46) by passing ten times each through a 20G and subsequently a 26G needle. Supernatant representing cytoplasmic extract was stored at -80°C. Anti-P23 or anti-HCLS1 antibodies were incubated with 40 µl Protein G sepharose overnight at 4°C. Antibody coupled beads were incubated 2 h with 1 mg cytoplasmic extract derived from RAW 264.7 cells (untreated or 2 h LPS-induction) in IP buffer. Luciferase control IP was performed as described (27). Beads were washed twice in IP buffer and either boiled in 2×SDS sample buffer for Western blot analysis or resuspended in Trizol for RNA isolation.
His-P23 and peptide variants as well as His-HCLS1 were expressed as described for His-HNRNP K (54).

Pulldown of His-tagged proteins
40 pmol His-P23 and peptide variants, His-HCLS1 or His-PRMT1 were immobilized on 30 µl Ni-NTA agarose in IP buffer and subsequently incubated with 50 µg of total RNA isolated from untreated or LPS-induced RAW 264.7 cells. Co-precipitated RNA was isolated using Trizol with firefly luciferase RNA as extraction control.

Comprehensive analysis of RNA-bound RBPs in differentially treated macrophages
Stability and translation of several cytokine mRNAs is controlled by ARE-BPs (7,(10)(11)(12). In addition, HNRNP K was identified as a specific regulator of TAK1 mRNA translation in LPSinduced macrophages (27). We aimed to systematically analyze RBPs that modulate the macrophage LPS response. To characterize the differential RNA-binding properties of RBPs in untreated and LPS-induced RAW 264.7 macrophages we applied RNA interactome capture, which had already successfully been employed to determine the RBP repertoire in other contexts were not detected in the eluates ( Figure 1C, lanes 3-5 and 8-10). RNA isolation from the input and elution from the oligo(dT)-beads yielded comparable total RNA amounts for the samples from untreated and LPS-induced RAW 264.7 cells, indicating no major changes in the overall mRNA pool, bound to oligo(dT) ( Figure 1D). QPCR analysis further demonstrated that specific RNA amounts eluted from oligo(dT) were comparable independent of macrophage LPS treatment ( Figure 1E). The yield of exogenously added polyadenylated Renilla luciferase mRNA (Renilla poly(A) + ) was higher than that of non-adenylated firefly luciferase mRNA (firefly poly(A) -) ( Figure 1E, left and middle panel). This was supported by the relative abundance of the two spikein controls in the elution fractions as quantified by RNA-Seq ( Figure 1F). In addition to the exogenously added mRNAs we investigated binding of an endogenous mRNA encoding NDUFV1 that is expressed constitutively in untreated and LPS-induced macrophages (27). 40% of NDUFV1 mRNA was recovered in the elution of the crosslinked samples compared to input ( Figure 1E, right panel). The analysis of the eluted non-ribosomal RNA population by limited high-throughput sequencing revealed that mRNAs represent the major fraction of non-ribosomal RNA, independent of macrophage LPS-induction or UV-crosslinking prior to oligo(dT) capture ( Figure 1G).
These data indicate that the UV-induced crosslinking between RNAs and RBPs was effectively established, while abundant proteins that lack RNA-binding activity could not be detected.

Classification of RBPs in RAW 264.7 cells
To characterize the protein pool that is bound to polyadenylated RNA, RBPs from untreated and  Figure 3B).

RBPs specifically identified in murine macrophages
To identify cell-specific RBPs we compared the murine macrophage RNA interactome with previously published RNA interactomes of murine ES cells (31) and two human cell lines, HeLa (29) and HEK 293 (30) ( Figure 4A). This analysis classified 69 proteins only identified in murine cells ( Figure 4A and B), of which 32 were exclusively detected in macrophages ( Figure 4A and C). This group of 32 RBPs only identified in macrophages includes 19 proteins ( Figure 4C) with activities so far not directly related to RNA (Table 1). In

P23 and HCLS1 bind directly to RNA
From the 19 macrophage-specific proteins that were designated as novel RBPs (Table 1)  Immunofluorescence analysis revealed that both proteins are localized in the cytoplasm, at least at steady state ( Figure 5B). Therefore, we applied cytoplasmic extracts to address the interaction of endogenous P23 and HCLS1 with polyadenylated RNA in untreated macrophages and after LPS-induction. Quantification revealed that both proteins were immunoprecipitated at comparable levels independent of LPS treatment ( Figure 5C and D, upper panel).
Specific co-precipitation of polyadenylated RNA was verified for both proteins, HCLS1 ( Figure 5C, middle panel) and P23 ( Figure 5D, middle panel) in a dot blot-based Biotin-Streptavidin assay, which utilizes biotinylated oligo(dT) hybridization to immobilized poly(A) + RNA. Interestingly, a smaller proportion of poly(A) + RNA was co-precipitated with P23 from cytoplasmic extracts of LPS-induced macrophages ( Figure 5D, middle panel, lanes 2 and 5). This was not due to RNA preparation differences as proven by equally detectable exogenously added firefly luciferase mRNA, shown in RT-PCR analysis ( Figure 5D, lower panel).
To test whether the poly(A) + RNA population binds directly to P23 and HCLS1, we purified macrophage mRNA and constructed expression vectors for both polypeptides. For initial in vitro interaction studies recombinant His-tagged proteins were purified from E. coli ( Figure 6A). His-P23 and His-HCLS1 immobilized on Ni-NTA agarose precipitated specifically poly(A) + RNA from total RNA isolated from untreated and LPS-induced RAW 264.7 cells, whereas His-PRMT1 did not ( Figure 6B). Interestingly, P23 precipitated more poly(A) + RNA isolated from untreated macrophages than from LPS-induced macrophages ( Figure 6B, C), consistent with the co-immunoprecipitation of poly(A) + RNA with endogenous P23 ( Figure 5D). Empty Ni-NTA beads (-protein in the dot blot) ( Figure 6C), which did not show poly(A) + RNA binding were used for normalization ( Figure 6B). Differences in co-precipitation were not due to varying extraction efficiencies shown by RT-PCR analysis with primers for firefly luciferase extraction control ( Figure 6D).
To validate the interaction of HCLS1 and P23 with poly(A) + RNA observed in the immunoprecipitation assay ( Figure 5C and D) and with recombinant protein ( Figure 6B These results suggest that the two proteins P23 and HCLS1, which were so far not classified as related to nucleic acid binding, exhibit RNA-binding activity. Panther protein class annotation (56, 57) of these newly identified RBPs revealed an overrepresentation of cytoskeletal proteins ( Figure 4C, Table 1 During apoptosis P23 is cleaved by CASPASE-3 and 8 (86). Interestingly, we discovered that LPS-induction of RAW 264.7 macrophages leads to miR-155-mediated CASPASE-3 downregulation and decelerated apoptosis, thereby sustaining macrophage activity (87). This is in agreement with a role of P23 in preventing ER-stress-induced apoptosis (88,89).