The RNA-binding protein AATF coordinates rRNA maturation

Apoptosis-antagonizing transcription factor (AATF) is a predominantly nuclear protein essential for both embryonic development and tumor growth. Several studies have shown a role of AATF in the modulation of cellular signal transduction pathways such as p53-, mTOR- and HIF-signaling. However, the exact molecular functions underlying its essential nature to cell proliferation and survival have remained elusive. Interestingly, several lines of evidence point towards a pivotal role of this protein in ribosome biogenesis and the maturation of ribosomal RNA. In this study, we identify AATF in a screen for RNA-binding proteins. Importantly, CLIP-sequencing shows a predominant association with ribosomal RNA precursor molecules. Furthermore, AATF binds to mRNAs encoding for ribosome biogenesis factors as well as snoRNAs. These findings are complemented by an in-depth characterization of the protein interactome of AATF, again containing a large set of proteins known to play a role in rRNA maturation. Consequently, our multilayer analysis of the protein-RNA interactome of AATF reveals this protein to be a central hub in the coordination of ribosome biogenesis.


INTRODUCTION
AATF, also known as CHE-1, was originally identified as an RNA polymerase II interacting protein with anti-apoptotic capacities. Through its interaction with RNAPII, AATF modulates the function of a row of transcription factors including pRB, p65 and STAT3 (1)(2)(3). Taking into account its role in the prevention of apoptosis the involvement of AATF in the DNA-damage response and its impact on p53 function are explicitly interesting. The functional impact of these findings is emphasized by data showing AATF as a pro-tumorigenic factor in several tumor models (4,5). Upon DNA damage AATF gets phosphorylated by checkpoint kinases leading to its interaction with the NF-kB subunit p65 and relocalization to the p53 promoter (3,6,7). However, AATF also modulates the specificity of p53 by shifting its binding preference towards target genes leading to growth arrest over those that mediate apoptosis (5,8). Furthermore, AATF has been shown to play a role in other key pathways involved in tumorigenesis, namely mTOR-and HIF-signaling (9,10). Seeing the impact of this protein on signal transduction in apoptosis and tumor formation, its potential as a therapeutic target in cancer therapy has been discussed extensively (5,11,12). Interestingly, AATF -whilst also having been detected in cyto-and nucleoplasm -primarily appears to localize to the nucleolus with the nucleolar fraction having been associated with c-JUN-mediated apoptosis (13). With nucleoli being the sites of rRNA transcription and maturation, it is important to note that AATF was recently identified by independent RNAi screens for factors involved in ribosomal subunit production (14,15). This connection is extremely interesting due to two considerations. Firstly, cell proliferation and ribosome biogenesis rate are closely intertwined, and increased ribosome formation is linked to tumorigenesis (16,17).
Secondly, blocking ribosome formation induces ribosome biogenesis stress leading to the activation of p53 mediated by inhibition of MDM2. Vice versa, induction of ribosome biogenesis inhibits p53 (16,18,19). However, the mode of action of AATF in ribosome maturation has remained unclear. A first insight came recently from a study by Bammert et al. (20) that could show AATF as part of a nucleolar protein complex (termed ANN complex). Here, AATF, together with NOL10 and NGDN, was essential to efficient generation of the small ribosomal subunit (SSU) -in line with findings from previous screens of ribosome biogenesis factors (14). Yet, the molecular function that AATF fulfils within this complex and the question which one of the three proteins mediates binding to rRNA precursors remained elusive. Work published recently by Piñeiro et al. (21) shows AATF to be amongst 211 RNA polymerase I-dependent RNA-binding proteins and to bind to pre-rRNA molecules.
In line with their findings, we identified AATF to be an RNA-binding protein. Together, their and our data reveal AATF to be the protein mediating RNA-binding in the ANN-complex. We complement this finding with the identification of AATF binding sites in the 45S rRNA precursor and an extensive characterization of its protein and RNA interaction partners placing AATF at a central position in rRNA maturation and ribosome biogenesis.

Molecular cloning and design of small interfering RNAs (siRNA)
For the generation of GFP-tagged AATF transgenes, the AATF wild type sequence, truncated or mutated versions of the protein generated by overlap extension PCR were cloned into the AAV CAGGS GFP plasmid (Addgene #22212). siRNAs targeting the 3' UTR of human AATF were custom designed by Dharmacon (accession number AJ249940.2).

Cell culture, transfection and generation of single-copy transgenic cell lines using TALEN
Human HEK 293T, HCT116 and U2OS cells were grown in standard media at 37°C, 5% CO 2 and routinely passaged using 0.05% Trypsin. Mycoplasma contamination was checked using a commercial kit (Venor GeM, Sigma). Transfections were carried out on 60-80% confluent cells using calcium phosphate as described previously (22) Stably integrated transgenic cell lines were generated using TALEN technology by co-transfecting TALEN encoding plasmids specific for the human AAVS1 locus as previously described (23). 24h after transfection, cell lines were steadily selected with 2 µM Puromycin. All cell lines were genotyped by integration PCR and phenotyped by both immunoblot and immunofluorescence.

Western blot
Cells were washed once with ice-cold PBS, lysed on ice in either iCLIP lysis buffer or modified RIPA buffer for 15 minutes and then sonicated (BioRuptor Pico). Lysates were then run on home-made PAGE gels, transferred onto PVDF membranes and blocked in 5% bovine serum albumin. Primary antibodies were incubated either for 1h at room temperature or overnight at 4°C. Secondary HRP coupled antibodies (Jackson ImmunoResearch,) were incubated for 1h at room temperature. Bands were visualized using home-made ECL, at a Fusion chemoluminescence imaging system (Peqlab) (see Suppl. Methods for a list of primary antibodies used).

Immunofluorescence/Microscopy
U2OS and HTC116 cell lines were grown on coverslips coated with 100 µg/ml Collagen I to 60-70% confluency. Cells were washed once with PBS supplemented with Ca 2+ /Mg 2+ (PBS+), fixed with icecold Methanol at -20°C for 5 minutes and rinsed three times 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 4°C overnight. Secondary antibody (Cy3-conjugated, Jackson ImmunoResearch) was added for one hour at RT, after which the stained cells were mounted on ProLong Gold mounting medium with DAPI (Invitrogen) to visualize nuclei. Images were acquired using a Zeiss epifluorescence microscope (Zeiss Axiovert 200M) as well as Zeiss ZEN software and analyzed using ImageJ/Fiji (24).

eCLIP-seq, Read Processing and Cluster Analysis
eCLIP of AATF in K562 and HepG2 cells using primary anti-AATF antibody (A301-032A lot 001, Bethyl) was performed as previously described (25). In addition to standard read processing and processing of reads to identify unique genomic mapping, reads mapping ribosomal RNA were quantified using a family-aware repeat element mapping pipeline that identifies reads unique to 45S pre-rRNA, 18S rRNA, or 28S rRNA respectively (Van Nostrand, E.L., et al. in preparation). To quantify relative enrichment between IP and input, relative information was calculated as the Kullback-Leibler divergence: , where p i is the fraction of total reads in IP that map to position i, and q i is the fraction of total reads in input for the same position. Regarding the definition of non-rRNA interaction partners we filtered the total dataset (K562 and HePG2 cells, 2 replicates each) for significant peaks over input (log 2 FC ≥ 3 and -log 10 p-value ≥ 5) and collapsed all peaks mapping to one transcript to define a list of targets. For AATF eCLIP data accessibility refer to "Availability" section below. and alkylated (40 mM CAA, 30 min, 20°C in the dark). Proteins were digested at 37°C for 16h using trypsin and lysC (both at a 1:75 enzyme-protein ratio) using standard protocols. Formic acid was added to a final concentration of 1% to stop proteolysis. Samples were loaded onto StageTips and labeled with stable isotopes using on-column dimethyl labeling (27). Eluted peptides of the corresponding light, medium and heavy-labelled channels were mixed and dried down in a vacuum concentrator. Peptides were resuspended in 5% DMSO / 1% formic acid and stored at -20°C prior MS analysis (28,29). For details on MS processing and data analysis of the protein interactome see Suppl.

Coimmunoprecipitation and sample preparation for MS/MS
Methods. TRIzol extraction and concentrated with RNA clean and concentrator columns. An equal amount of immunoprecipitated RNA was reverse transcribed using SuperScript III as described above.

AATF is an RNA-binding protein
We performed an analysis of proteins associated with polyA-tailed RNA using RNA interactome

AATF is associated with ribosomal RNA
In order to determine the identity of the RNA molecules bound to AATF we analyzed enhanced crosslinking and immunoprecipitation (eCLIP) data generated by the ENCODE consortium (Fig. 1C).
On the one hand unexpectedly, with AATF having been identified in a screen for proteins associated with polyA-tailed sequences, the vast majority of transcripts bound are ribosomal (Fig. 1D). On the other hand, this finding is well in line with the subcellular localization of AATF to nucleoli, the site of rRNA generation as well as previously published data (15,20,35) and is technically explained by RNA-RNA hybridization between mRNA and rRNA molecules as set forth recently by Piñeiro et al. (21).
The enrichment of rRNA species is strongly significant when compared to a similar analysis of 223 other publicly available datasets of 150 RNA binding proteins generated by the ENCODE consortium ( Fig. 1D). The overrepresentation is strongest regarding the 45S rRNA precursor molecule ( Fig. 1D/E).

Specific binding of AATF to cleavage sites in ribosomal precursor RNAs
In order to get a better understanding of the actual position of the binding sites in the 45S rRNA precursor, taking into account recently published data on the potential role of AATF in ribosome biogenesis (14,15), we focused on the spacer sequences outside 18S and 28S rRNA (which are essentially very abundant RNA species and may also partly be contaminants) containing the cleavage sites essential to rRNA maturation. Doing so, we discovered a number of highly specific peaks ( Fig.   2A). These peaks are strongly enriched compared to the 150 ENCODE RBPs when looking at the spacer regions whilst partly less specific regarding the 18S and 28S region. When examining the peaks in the spacer regions in more detail, we noticed a close proximity to sites that are cleaved specifically during rRNA maturation (Fig. 2B) (14). More specifically this is the case for cleavage sites involved in small subunit (SSU) processing as depicted for the sites in the 5' external transcribed spacer (5' ETS) -01, A0 and 1 -and the first sites in internal transcribed spacer 1 (ITS1) following the sequence of 18S (15,36). This finding is in line with recently published data showing that loss of AATF or other components of the so-called ANN complex results in reduced cleavage activity at these sites indicating reduced SSU processome activity (14,20). Other non SSU associated sites in ITS1, ITS2 and 3'ETS are still detected but show a much lower signal (e.g. 02, Fig. 2B). In line with a role of AATF in ribosome maturation, siRNA-mediated knockdown of AATF leads to a strong reduction of cellular rRNA (Fig. 3A). This finding was confirmed using qPCR, where 18S rRNA levels were significantly reduced after loss of AATF (Fig. 3B). In order to confirm this to be a specific effect of AATF depletion, we repeated the experiment with a siRNA targeting the 3'UTR of AATF in both wildtype cells (WT) and cells harboring a single-copy transgene of GFP-tagged AATF lacking the 3'UTR.
This transgene rescued the effect of AATF knockdown on global rRNA levels (Fig. 3C). Of note, cell number (data not shown) and tubulin staining did not differ between treatments.

Nucleolar localization of AATF is required for its association with rRNA precursors
AATF has been described as being primarily nuclear with recent reports confirming an accumulation of the protein in nucleoli. Previous work indicated that the C-terminal portion of AATF is required for this specific subcellular localization (13,20). However, the actual nucleolar localization signals (NoLS) had not been identified. Prediction of putative NoLS contained in AATF using the Nucleolar localization sequence Detector (NoD) (37) revealed the presence of two putative NoLS in the Cterminal portion of the protein (Fig. 4A, Suppl. Fig. 1A). In order to visualize the subcellular localization of AATF we generated stable cell lines harboring a single-copy transgene encoding GFP-

AATF binding to other RNA species
As indicated in Fig. 1D AATF binds -even though the largest fraction is rRNA -also to other RNAspecies. Here, mRNA is the most common type (69% protein coding) followed by several types of non-coding RNAs (Fig. 5A, Suppl. Table 1). Seeing the association with both ribosomal RNA and mRNAs we were wondering whether the associated coding transcripts were functionally linked to ribosome biogenesis. Interestingly, GO term and pathway analyses confirm such a link with the most overrepresented terms including "ribosome" for KEGG pathways as well as "rRNA processing", "ribosome", "translational initiation" and "structural constituent of ribosome" regarding gene ontologies ( Fig. 5B). As shown in Fig. 5A several non-coding RNA-species co-precipitate with AATF including snoRNAs and miRNAs. This finding is especially interesting taking into account the important role snoRNAs and associated snoRNPs play in ribosome biogenesis (16,38,39). Interestingly, snoRNAs of the C/D class are overrepresented compared to H/ACA snoRNAs and scaRNAs indicating a potential specific function associated with C/D box snoRNPs (Fig. 5C/D) (40). Since snoRNAs are key players in the site-directed nucleotide modification of ribosomal RNA precursors, which play an important role in their maturation, we set forth to analyze whether loss of AATF may have a global impact on the abundance of these modifications. However, quantification of the most common variants by mass spectrometry did not reveal any significant differences between cells transfected with a siRNA targeting AATF and the respective scrambled controls (Suppl. Fig. 2C). Regarding the fact that miRNAs were also among the transcripts AATF knockdown did not show any impact on overall miRNA abundance (Suppl. Fig. 2A). However, several specific miRNAs showed a highly significant dysregulation after loss of AATF with miR-25 and miR-509 being the most prominent examples (Suppl. Fig. 2B).

The AATF protein interactome confirms a strong link to ribosome biogenesis
Since AATF had been associated with other nucleolar proteins before (20) and recruitment of components of the ribosome biogenesis machinery to the 45S precursor may be one of the key functions of an rRNA-associated RBP, we decided to characterize the protein interactome of AATF by mass spectrometry using immunoprecipitation (AP-MS) of FLAG-tagged AATF expressed from a single-copy transgene. In order to ascertain specificity, we performed an immunoprecipitation of FLAG-tagged GFP and compared the abundance of proteins in both datasets performing a t-test.
Doing so, we identify 165 proteins significantly co-immunoprecipitated with AATF (log 2 FC ≥ 2 andlog 10 pvalue ≥ 1.3). Those can be considered as bona-fide AATF interactors (Suppl. Table 2). Whilst some of these bona-fide interactors had been described before in the literature or were classified as physical interactors of AATF in BioGRID (V3.4.153), the majority of proteins detected had not been described to be associated with AATF before (Fig. 6A/Suppl. Table 2). Interestingly, an analysis of the most overrepresented GO terms and KEGG pathways showed primarily terms associated with the ribosome and its biogenesis (Fig. 6B). A closer look at the proteins behind these terms revealed the interactome to contain a large number of known r-proteins and rRNA processing factors (as recently identified by several screens) (14,16) as well as nucleic acid associated enzymes including helicases known to be involved with snoRNA binding or release such as DHX15 (41) (Fig. 6C). Interestingly, more than 80% of the bona fide interactors have been identified as RBPs in independent screens themselves (Fig. 6D). As the AATF interactome contains such a large number of putative or confirmed RBPs we were wondering whether many of these interactions were actually RNA-dependent, as had been shown previously for other RBPs (42). However, repeating the AP-MS experiment including an on-column RNase/benzonase treatment showed that the majority of the interactors do not depend on RNA but are rather direct protein-protein interactions (Fig. 6E, Suppl.

in the t test performed between AATF IP treated with
RNase versus AATF IP normalized) (Fig. 6E, Suppl. Table 2). In order to obtain a better view of which proteins may at least show a partial dependency on RNA we repeated this analysis using alleviated thresholds (log 2 FC ≤ -1, -log 10 pvalue ≥ 1.0) which still results in only 19 putatively RNA-dependent interactors of AATF (Fig. 6E, Suppl. Table 2). Interestingly, a significant proportion of these 19 proteins (~30%) are classical ribosomal proteins, whilst this is only the case for 21 out of 146 of the other AATF interactors (14%) (data not shown). However, functional analyses (GO-terms and KEGG-pathways) did not reveal any further obvious differences between putatively RNA-dependent andindependent protein interactors. Only two proteins were completely lost after nuclease treatment -RNA-binding protein SLIRP and protein phosphatase PPP1CB (Fig. 6E, Suppl. Table 2). Since inhibition of RNAPI altered the nuclear distribution of AATF (Suppl. Fig. 1D) we went on to ask the question whether the large number of RBPs in its interactome contained other RNAPI-dependent RBPs. Interestingly, overlapping our data with the recent global identification of RNAPI-dependent RBPs by Piñeiro et al. (21) revealed AATF to interact with both RNAPI-dependent (46) andindependent (59) RBPs (Fig. 6F).

A combinatorial approach to the RNA and protein interactome suggest AATF to be a central hub in SSU maturation
Taking into account the published data on a putative role of AATF in SSU maturation (14,15) as well as our data regarding the binding site in 45S rRNA localizing closely to SSU cleavage sites ( Fig. 2A/B) we continued our analysis of the protein and RNA interactome focusing on the key complexes involved in SSU maturation (Fig. 7). Here, AATF shows a striking interaction with protein components of the three UTP complexes, the C/D and H/ACA snoRNPs and the exosome complex (Fig. 7).
Furthermore, AATF does not only bind the proteins but also several mRNAs encoding for key components of these complexes -such as the enzymatic subunits FBL and DKC1 -which may add a regulatory level to the interaction of AATF with SSU processome complexes (Fig. 7). Interestingly, as determined in our AATF eCLIP experiment, U3 snoRNA itself is among the RNA interaction partners of AATF as well (Fig. 7D). Altogether, out of the snoRNAs known to be required for rRNA processing,

DISCUSSION
AATF had previously been associated with ribosome biogenesis due its nucleolar localization and the impact of loss of the protein on the generation of the 40S subunit and rRNA maturation (14,15).
Furthermore, AATF has been shown to be part of a protein complex (ANN complex) required for efficient generation of the SSU (20). However, the actual RBP associating this protein complex to pre-rRNA molecules had not been identified due to a lack of canonical RNA-binding domains. Very recently, work by Piñeiro et al. showed AATF to be an RNAPI-associated RBP itself (21). Their findings are in clear accordance with our results showing that AATF has been identified in a row of RNA-interactome capture screens in different species (32,33,44). A key finding of our study is the identification of the actual RNA-molecules bound by AATF by eCLIP with a clear overrepresentation of ribosomal RNAs. Furthermore, this approach allowed us to localize the binding events in the 45S precursor molecule primarily to the cleavage sites required for SSU maturation strengthening the hypothesized role of AATF in this process. Several lines of published evidence also point towards this direction. Interestingly, Bfr2 -the budding yeast orthologue of AATF -was identified in a screen that analyzed constituents of yeast 90S particles assembled using plasmid-encoded 3'-truncated pre-18S RNAs and was shown to be specifically bound to the 5'ETS (45). Furthermore, Bammert et al. (20) had shown recently that loss of the ANN complex (containing AATF) resulted in 45S processing defects at the cleavage sites that are necessary for SSU generation. Our data provide strong indications that AATF is the actual RBP of this complex. However -just like AATF itself -both NGDN and NOL10 were detected in screens for polyA-tailed RNA-associated proteins and may possess the capacity to bind RNA themselves as well (31,33). As mentioned above, none of the ANN complex constituents -including AATF itself -possesses a canonical RNA binding domain. While Bammert et al. (20) argued that RNA binding of the ANN complex may be mediated by the WD40 domain of NOL10 which had previously been shown to mediate RNA-binding of other proteins (46), the intrinsically disordered, basic C-terminus of AATF, 20% of which is made up of lysine (K) and arginine (R) residues, appears likely to contain the RNA binding domain. This is also in line with recent studies that could emphasize the importance of intrinsically disordered regions (IDR) for RNA-protein interactions (32,47,48) and the results of recent proteome-wide screens for RNA-binding domains (48,49). Interestingly, work by He et al. showed the two NoLS, the deletion of which resulted in a loss of rRNA binding in our hands, were directly associated with RNA (49). Nevertheless, future experiments will be necessary to further dissect the exact molecular requirements of the physical interaction between AATF and RNA.
Our multilayer analysis -involving the global identification of the protein and RNA interactomeprovides further evidence regarding the role of AATF in ribosome maturation. Beyond rRNA precursor molecules themselves AATF binds other RNA species important to the biogenesis of the ribosome.
On the one hand, the mRNAs co-precipitating with AATF primarily encode proteins involved with RNA metabolism and ribosome maturation. Here, it is intriguing to hypothesize that AATF may exert a regulatory function regarding the post-transcriptional regulation of these proteins. On the other hand, snoRNAs are highly enriched in the AATF interactome pointing towards a role of AATF in recruiting snoRNAs to pre-rRNA molecules. The fact that AATF is associated with U3 snoRNA -taking into account that cleavage in the 5'ETS and the ITS1 of 45S pre-rRNA strongly depends on the U3 snoRNP -corroborates this hypothesis. This view is further complemented by the protein interactome containing numerous factors required for ribosome biogenesis. Since the vast majority of protein interactions -apart from SLIRP and protein phosphatase PPP1CB -does not depend on RNA but appears to be mediated by direct protein-protein binding, AATF is likely to be a central hub in the coordination of protein-RNA supercomplexes in rRNA maturation.
How is the molecular function of AATF in ribosome biogenesis linked to the previously described phenotypes regarding cellular proliferation and tumorigenesis? With rRNA availability being a central requirement for cellular survival and cell division, it is not surprising that loss of AATF in a knockout mouse model led to early embryonic lethality (35). As to human disease this protein has been implicated to play an important role in cancer biology, a hypothesis that is partly based upon its ability to inhibit apoptosis. AATF has previously been found to be amplified or overexpressed in both hematological and solid tissue tumors, to correlate with poor prognosis and reduced survival (4)(5)(6)9,10,50) and to mediate its effects on apoptosis by the modulation of p53 abundance and function (4,6). Furthermore, increased ribosome biogenesis is not only employed by tumors to increase their proliferative potential but appears to be a risk factor for cancer onset on its own (51). However, AATF appears to ensure cellular survival not only in the setting of cancer. Since its protective role has also been in the setting of oxidative stress exposure to different cell types including renal tubular cells in a model of acute kidney injury (52)(53)(54). The known link between ribosome stress and p53 activity in the light of our new data allows for the exciting hypothesis that AATF mediates its effects on cell death employing this pathway. Based on the interaction between r-proteins and the E3 ubiquitin ligase MDM2 defects in ribosome maturation activate p53 (18,39,(55)(56)(57). Loss of AATF would thus increase p53 activity linking the roles of AATF in ribosome biogenesis and p53 activity. In the light of an increasing number of studies trying to target ribosome biogenesis in this setting (58) and taking into account that AATF has been shown to sensitize cancer but not normal cells to antineoplastic drugs (5), future research shedding light on these aspects will be highly valuable.

AVAILABILITY
AATF K562 eCLIP data has been deposited at the ENCODE Data Coordination Center      (14) and human RNA helicases (47)