An internal deletion of ADAR rescued by MAVS deficiency leads to a minute phenotype

Abstract The RNA-editing protein ADAR is essential for early development in the mouse. Genetic evidence suggests that A to I editing marks endogenous RNAs as ‘self’. Today, different Adar knockout alleles have been generated that show a common phenotype of apoptosis, liver disintegration, elevated immune response and lethality at E12.5. All the Adar knockout alleles can be rescued by a concomitant deletion of the innate immunity genes Mavs or Ifih1 (MDA5), albeit to different extents. This suggests multiple functions of ADAR. We analyze AdarΔ7-9 mice that show a unique growth defect phenotype when rescued by Mavs. We show that AdarΔ7-9 can form a truncated, unstable, editing deficient protein that is mislocalized. Histological and hematologic analysis of these mice indicate multiple tissue- and hematopoietic defects. Gene expression profiling shows dysregulation of Rps3a1 and Rps3a3 in rescued AdarΔ7-9. Consistently, a distortion in 40S and 60S ribosome ratios is observed in liver cells. This dysregulation is also seen in AdarΔ2-13; Mavs−/− but not in AdarE861A/E861A; Ifih1−/− mice, suggesting editing-independent functions of ADAR in regulating expression levels of Rps3a1 and Rps3a3. In conclusion, our study demonstrates the importance of ADAR in post-natal development which cannot be compensated by ADARB1.


INTRODUCTION
Adenosine to inosine deamination (A-to-I editing) in RNA is mediated by adenosine deaminases acting on RNA (ADARs) (1). ADARs bind dsRNAs and catalyze the deamination of adenosine to produce inosine (2,3). Formation of inosine in coding regions can change the genetic information of RNAs since inosine is primarily read as guanosine by the protein translation machinery (4,5).
In mammals, two active ADARs are known today, termed ADAR1 (Adar) and ADAR2 (Adarb1) (13). In mammals, Adarb1 (ADAR2) is mostly expressed in the nervous system and the gastrointestinal tract (14). In contrast, Adar (also known as Adar1) is ubiquitously expressed in two isoforms--an interferon inducible 150 kDa version (15) which is predominately cytoplasmic and a shorter constitutively expressed p110 kDa isoform which is nuclear (16). Both isoforms share common features like three dsRNA binding domains (dsRBDs), a nuclear localization signal (NLS) and a deaminase domain (17). However, ADAR1 p150, exclusively, has two Z-DNA binding domains and a nuclear export signal (NES) at its N-terminus while the shorter isoform has only one Z-DNA binding domain (ZBD) (18). The ZBDs and dsRBDs may confer additional functions that go beyond RNA-editing, such as RNAand DNA-binding. Indeed, editing-independent functions of ADAR have been reported in microRNA biogenesis (10,19), neural induction (20), splicing (21) and translation (22).
Editing-independent functions of ADAR1 are also evident from the phenotypes of the different Adar alleles. Four different Adar knockout mice are reported in the literature depending on the number of exons deleted: Mouse Adar is composed of 15 exons and different number of exons are deleted in the different transgenic mouse lines. Adar Δ2-13 deletes exons 2-13, Adar Δ7-9 deletes exon 7-9, the isoform specific p150 −/− allele selectively deletes exon 1 therefore only allowing for expression of the short p110 isoform (23)(24)(25). Finally, an enzymatically inactive allele Adar E861A/E861A has recently been created that expresses catalytically dead, but RNA-binding-competent ADAR (26). Interestingly, mice with a deletion in Adar die around day 12.5 in utero (23)(24)(25), except the catalytically-dead point

Histology
Spleens, kidneys, intestines and hearts were isolated from littermates 15 days post-partum, fixed overnight in 4% paraformaldehyde, dehydrated, embedded in paraffin and 4 m sections were taken. Hematoxylin and eosin (H&E) staining was carried out following standard protocols. Microscopic analysis and imaging were performed using an Olympus BX61VS slide scanner and OlyVIA 2.9 (Olympus) software.

Western blotting and antibodies
For western blotting HeK293 cells or MEFs were grown in DMEM/FBS/Pen-Strep. To induce the production of ADAR1, cells were transfected with FLAG-tagged ADAR1p150 fused to T2A-GFP in pcDNA 3.1.

RNA-Seq library preparation
Liver was harvested from P15 old mice and homogenized in TriFast™ (Peqlab, Erlangen, Germany) by cutting into small pieces and passing through a syringe with a 18-G needle. Bone marrow cells were isolated by flushing the femur and tibia of the P15 mice with PBS. RNA was isolated following the manufacturer's instructions for TriFast™, treated with DNaseI (New England Biolabs, Ipswich, MA, USA) and then purified by Phenol:Chloroform extraction. Ribosomal RNAs were removed using the Ribo-Zero rRNA removal kit (Illumina, San Diego, CA, USA) and cDNA libraries were subsequently generated using the NEBNext ® Ultra™ Directional RNA Library Prep Kit for Illumina ® (New England Biolabs, Ipswich, MA, USA) and sequenced in a paired-end mode with 125 bp read length on a HiSeq2500 (Illumina, San Diego, CA, USA) machine at a read depth of 31 million reads per sample.

Analysis of RNA-Seq data
Sequence reads were quality trimmed and adaptor clipped using trimmomatic (v0.33) (34). Quality was monitored before and after using FastQC. For quantification of transcript abundance (TPM), a reference transcriptome was obtained from ENSEMBL Biomart for mus musculus (GRCm38.p4) (35) and used together with salmon (v0.9.1) (36) applying sequence-specific and fragment GC bias correction. Resulting read count estimates were used with bioconductor packages tximport (37) and DESeq2 (38) to call differentially expressed genes. Coverage plots for Rps3a1 and Rps3a3 were produced by mapping the trimmed reads against a reference transcriptome containing the mature mRNAs of Rps3a1 and Rps3a3 including 1 kb of flanking regions (since no UTR annotation is available for Rps3a3) using the tool segemehl (v0.2) (39) requiring an mapping accuracy of 0.95. Coverage across the gene body was deduced with tool genomeCoverageBed (40).

Availability of RNA-Seq data
The GEO Accession number for RNA-sequencing data generated for this project and the publicly available data used in this project, are given below: Moreover, ENCODE RNA-Seq data (polyA and total RNA) from wildtype mice at 10 weeks of age and at P0 stage was also analyzed.
The rescued mice showed phenotypes of apparently different penetrance. Analysis of sequencing data from three different tissues from these mice indicated that they were indeed devoid of ADAR1 activity (Supplementary Table S1). While most mice lived until ∼15 days after birth, a few mice survived for several months (the longest living mouse stayed alive until >18 months after birth and was then sacrificed for histological analysis-see below). Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− pups are phenotypically similar, both are developmentally retarded, significantly smaller than their wild-type counterparts and generally immobile (Supplementary Figure S2A). Survival curves revealed that 69.24% of Adar Δ7-9 ; Mavs −/− and 75% of Adar Δ7-9 ; Ifih1 −/− die by day 30 after birth ( Figure 1A). The remaining pups survive beyond 30 days after birth. Out of these, two Adar Δ7-9 ; Mavs −/− lived up to six months and one survived at least till 18 months, while no Adar Δ7-9 ; Ifih1 −/− lived till 6 months of birth. Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− showed deviation from the expected Mendelian ratios (Supplementary Figure S2B). However, it is unclear if the distortion in ratio is due to incomplete rescue of the phenotype resulting in embryonic lethality, or if the rescued mice were cannibalized immediately after birth.
To test whether the partially rescued mice showed any expression of interferon stimulated genes (ISGs) we tested ISG expression in various tissues of Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− mice. Heart, spleen and liver tissues from these rescued mice showed a mild immune response compared to the respective Mavs −/− and Ifih1 −/− littermates ( Figure 1B and C). Consistent with their comparable survival curves, ISG expression was similar in both Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− ( Figure 1B and C). This demonstrated that Mavs −/− and Ifih1 −/− not only rescues the lethality of Adar Δ7-9 mice to the same extent but also alleviates the immune response associated with Adar Δ7-9 to a similar level. In summary, we show that the phenotype of Adar 7-9 ; Mavs −/− mice can be less severe than previously reported and is less severe than that of Adar Δ2-13 ; Mavs −/− which reportedly die immediately after birth but is more severe than Adar E861A/E861A ; Ifih1 −/− that are fully rescued (26,28,29).

Adar Δ7-9 forms a truncated, editing-deficient, mislocalized protein
Since the phenotype of both Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− was less severe than that of Adar Δ2-13 ; Mavs −/− but more severe than the fully rescued Adar E861A/E861A ; Ifih1 −/− we wondered whether Adar Δ7-9 could form a truncated protein. In-silico analysis predicted that deletion of Adar exons 7-9 is in-frame (Supplementary Figure S3). Deletion of exons 7-9 disrupts part of the deaminase domain and the third RNA binding domain (dsRBD3). dsRBD3 embeds a nuclear localization signal (NLS) in human ADAR1 while mouse ADAR1 also harbors a C-terminal NLS (47,48,50). The other two RNA binding domains and the ZBDs remain intact in the Adar Δ7-9 allele (Figure 2A). Indeed, using primers spanning exons 6 and 10 of Adar cDNA, we were able to detect a truncated Adar Δ7-9 transcript in primary MEFs ( Figure 2B). Sanger sequencing verified the truncation to be in-frame. To determine whether the ADAR 7-9 could produce a protein we cloned the ADAR1 isoforms p110 and p150 and their truncated 7-9 counterparts  by western blotting. (F) Detection of ADAR1 in wt and ADAR1 7-9 -derived MEFs after IFN induction using a polyclonal antibody directed ADAR1 p150. (G) Detection of ADAR1 in wt and ADAR1 7-9 -derived MEFs after IFN induction using a monoclonal pan-anti ADAR1 antibody. Arrowhead marks size of predicted p150 7-9 protein and asterisk depicts the size of the p110 7-9 protein. (H) Detection of ADAR1 in wt and ADAR1 7-9 brain lysates. tubulin marks detection of the lower part of the blot with an anti-tubulin antibody. (I) Quantification of ADAR p110 in three independent blots of cell lysates indicates ∼20% expression of p110 7-9 protein (accession no. AAK16102.1). The cloned cDNAs were N-terminally FLAG-tagged and C-terminally fused with a self-cleaving 2A peptide to eGFP (49). Overexpression of these variants confirmed the nuclear localization of ADAR1 p110 and cytoplasmic localization of ADAR1 p150. p110 7-9 still accumulated in the nucleus but showed a less prominent nucleolar localization, suggesting that human and mouse ADAR1 proteins may differ with respect to their nuclear localization signals. In fact, it was shown previously that deletion of a region surrounding the third dsRBD interferes with nucleolar association of mouse ADAR1 while a C-terminal motif was mapped to be crucial for nuclear localization of mouse ADAR1 (50). The localization of p150 7-9 resembled the cytoplasmic localization of full-length p150 ( Figure 2C). Western blot analysis revealed that all four protein versions migrated with the expected patterns, confirming that p110 7-9 and p150 7-9 are expressed when transiently transfected, with no detectable degradation products ( Figure 2D). Interestingly, normalization to the co-translated eGFP indicated that the truncated variants are less stable in cellulo, especially p110 7-9 is more unstable compared to its full-length counterpart p110 ( Figure 2D).
Next, we tested whether full length ADAR1 and ADAR1 7-9 could be detected in MEFs or organs of wt or Adar 7-9 mice. We therefore tested commercial monoclonal anti Adar antibody 15.8.6 (Santa Cruz) on HeK293 cells that were transfected with the above mentioned FLAGtagged ADAR1 p110, p150, p110 7-9 and p150 7-9 constructs. Indeed, the 15.8.6 antibody could detect all four recombinantly expressed versions of the protein ( Figure 2E).
Subsequently, we tested whether full length and truncated p150 and p110 versions of ADAR1 could be detected in two different isolates of MEFs derived from wild-type and mutant Adar 7-9 mice. This was done using an antibody against the p150 version of ADAR1 (Synaptic Systems) ( Figure 2F) or by using a pan-ADAR1 antibody (15.8.6 / Santa Cruz) ( Figure 2G). To induce expression of ADAR1, cells were treated over night with IFN␣. Indeed, all four versions of the protein could be detected expressed from the endogenous locus (Figures 2F, G).
To test, whether the ADAR1 7-9 could also be detected in mice, we tested several organs for the expression of fulllength and ADAR1 [7][8][9] . To this end, ADAR1 7-9 could be most convincingly detected in brains ( Figure 2H, Supplementary Figure S4). Thus, ADAR1 7-9 can give rise to a truncated protein. It should be noted, however, that expression of the ADAR1 7-9 derived proteins was much weaker (about 40%) than that of full-length protein, both in cells and organs ( Figure 2I).

Adar Δ7-9 ; Mavs −/− show defects in multiple organs
Since Adar Δ7-9 ; Mavs −/− and Adar Δ7-9 ; Ifih1 −/− did not show any phenotypic differences and Adar Δ7-9 ; Irf3 −/− could not survive beyond embryonic day E13.5, we focused on Adar Δ7-9 ; Mavs −/− mice for further analysis. Adar Δ7-9 ; Mavs −/− mice were about 3.5-fold smaller than their control littermates ( Figure 3A). Accordingly, kidney and spleen of these animals displayed a reduction in weight. To understand whether these differences were simply due to re-duced overall body size or whether an organ is specifically affected, organ weights were normalized to body weights of the respective animals. This revealed, that the spleen is specifically affected in these animals, whereas the kidney weight is in a comparable ratio to control animals ( Figure  3B). Adar Δ7-9 ; Mav −/− mice showed a strong reduction in the cellularity of the spleen and the bone marrow by 23and 5-fold, respectively ( Figure 3C). The different isoforms of ADAR1 were shown to control development of organs such as kidney, spleen, lymph nodes and intestine, independent of inhibition of interferon-production (29). Histological examination of selected organs at day 14 postpartum revealed pleiotropic effects. In line with a previous study, cellularity of the spleen was generally low (Figure 3D). Similarly, intestinal development and homeostasis were also disturbed throughout the intestine, with strongest effects observed in the small intestine (29). Goblet cells in Adar Δ7-9 ; Mavs −/− mice were preferentially located at the luminal part of the epithelium, whereas they were distributed along the whole villi and crypts of the small intestine and the colon in wild type mice, respectively. Additionally, mild signs of villar fission/fusion and inflammation were detected ( Figure 3D). We could not detect a profound change in the organization of the kidneys in these animals ( Figure 3D). The residual inflammatory signature in the heart of Adar Δ7-9 ; Mavs −/− mice ( Figure 1B) is also not reflected on the histological level, as this tissue appeared comparable to the Adar +/+ ; Mavs −/− littermates (Supplementary Figure S5). Interestingly, the penetrance of the histological phenotype seems variable. We have also performed histological examination of the kidneys and spleen of an 18 month-old Adar Δ7-9 ; Mavs −/− long-time survivor and its heterozygous Adar Δ7-9/+ ; Mavs −/− sibling. However, in this long-time survivor the spleen and liver appeared morphologically normal, indicating that the penetrance of the Adar Δ7-9 ; Mavs −/− phenotype is variable (Supplementary Figure S6).
A previous study showed that Adar −/− ; Mavs −/− animals have specifically reduced splenic B cell numbers controlled by the p150 isoform (29). This study and the reduced cellularity of the bone marrow and the spleen (Figure 3C) prompted us to analyze hematopoietic organs in more detail. Flow cytometric analysis of bone marrow and spleen revealed a specific loss of the mature B cell population (B220 + CD19 + ) in Adar Δ7-9 ; Mavs −/− animals compared to their control littermates ( Figure 4A). Mature B cell numbers were reduced 24-and 50-fold in the bone marrow and the spleen, respectively ( Figure 4B). Intriguingly, the number of immature B cells (B220 + CD19 − ) was not changed (Figure 4B), indicating a specific loss of mature cells either due to e.g. apoptosis or problems of proper homing. The bone marrow is also the primary site of differentiation of circulating myeloid cells, like inflammatory monocytes and neutrophils (51). Contributions of these two cell types to the reduction of bone marrow cellularity were determined by flow cytometry. The fraction of neutrophils (CD11b + Ly6G + Ly6C lo ) was strongly diminished in Adar Δ7-9 ; Mavs −/− animals and the fraction of inflammatory monocytes (CD11b + Ly6G − Ly6C hi ) was increased accordingly within the CD11b + cell population ( Figure  4C). Intriguingly, normalization to total cell numbers re-Nucleic Acids Research, 2020, Vol. 48, No. 6 3293 vealed a profound loss of neutrophils and inflammatory monocytes ( Figure 4D). Overall, the analysis of these two hematopoietic organs indicates that maintenance of proper B cell and myeloid cell numbers is specifically regulated by ADAR1, independent of inflammation. Interestingly, mice expressing the editing-incompetent Adar E861A/E861A variant rescued by concomitant deletion of MDA5 show unaltered hematopoiesis (27), suggesting editing-independent functions of the full-length ADAR1 protein in hematopoiesis.

Adar Δ7-9 ; Mavs −/− show increased apoptosis of B cells and neutrophils
Development of B cells was shown to be dependent on ADAR1 in a conditional mouse ablating the ADAR1 protein in these cells. ADAR1 ablation in pre-B cells induces interferon production and apoptosis upon maturation (52). Neutrophil development is strongly dependent on apop-totic mechanisms to control the massive daily generation of these cells (53). The strong reduction of absolute cell numbers of B cells and neutrophils prompted us to analyze apoptotic cell death in bone marrow and spleen of Adar Δ7-9 ; Mavs −/− animals. Annexin-V staining revealed higher rates of early and late apoptotic neutrophils (subgated for CD11b + Ly6G + Ly6C lo ) and B cells (subgated for B220 + CD19 + ) in the bone marrow ( Figure 5A) and the spleen ( Figure 5B) of Adar Δ7-9 ; Mavs −/− animals compared to their littermate controls.
The higher rate of apoptosis in these two cell populations in Adar Δ7-9 ; Mavs −/− compared to wildtype mice explains the reduction in cellularity of the bone marrow and the spleen. Thus far, initiation of apoptosis in the context of ADAR1 loss was explained by excessive production of interferons, which is rescued upon concomitant deletion of MDA5 (26). The Adar Δ7-9 ; Mavs −/− animals show only slight upregulation of interferon production, therefore,  we conclude that apoptosis is induced by another pathway. Additionally, to the RLR pathway activation, dsR-NAs can lead to autophosphorylation of RNA-dependent protein kinase (PKR) and consequently phosphorylation of eukaryotic initiation factor 2␣ (eIF-2␣) without activation of IRF3. This PKR mediated dsRNA recognition was shown to induce global translation attenuation and apoptosis (43,54). During IFN response, ADAR1 also blocks translation attenuation by preventing hyperactivation of PKR (54).
Rps3a1 and the pseudogene Rps3a3 are dysregulated in Adar Δ7-9 ; Mavs −/− mice Since the histological analysis of P15 Adar Δ7-9 ; Mavs −/− showed variable defects in different tissues like the small intestine, colon, spleen or bone marrow ( Figure 3D, 4A and B) we wished to identify tissue-specific gene-expression changes underlying these defects. We performed RNA-Seq experiments on bone marrow and liver from P15 Adar 7-9 ; Mavs −/− and used Adar +/+ ; Mavs −/− littermates as controls. In the liver of Adar Δ7-9 ; Mavs −/− individuals, a total of 109 genes showed significant differential expression, out of which 66 genes were down-regulated. The down-regulated genes were enriched in liver-intrinsic pathways like oxidation-reduction processes, lipid and fatty acid metabolism and the oxygenase P450 pathway ( Figure 6A and B, supplementary data set). Genes up-regulated in the liver of Adar Δ7-9 ; Mavs −/− mice showed no enrichment for any particular pathway. Differentially expressed genes in the liver also contained genes involved in regulating body size and growth. For instance, insulin growth factor acid labile subunit (Igfals) is down-regulated in liver and reported to be involved in regulating body size and growth. Igfals knockout mice, show a ∼20% reduction in body size and growth beyond 20 days of birth (55). Interestingly, Lepr (leptin receptor) is overexpressed in the liver of Adar Δ7-9 ; Mavs −/− mice. This gene also regulates body weight and pituitary functions. Mutations in Lepr are associated with an obesity phenotype in mice and show elevated plasma levels of leptin, glucose, insulin and corticosterone (56). Another study reported a specific neuronal Lepr knockout causes morbid obesity (57). Additionally, we found neurotrophic receptor kinase 2 (Ntrk2), overexpressed in Adar Δ7-9 ; Mavs −/− mice. This gene is linked to hepatic hematopoiesis and innervation of the liver (58). Out of 109 differentially expressed genes in the liver, 33 (16 upregulated and 17 downregulated in Adar Δ7-9 ; Mavs −/− ) were represented in the interferome database (59). However, these upregulated ISGs showed a log 2 -fold change < 4, indicating a residual, but only mild immune response.
In bone marrow, a dramatic down-regulation of genes was observed. Out of 151 differentially expressed genes only nine were up-regulated in the bone marrow of Adar Δ7-9 ; Mavs −/− mice ( Figure 6A and B, supplementary data set). The down-regulated genes were enriched for immune pathways, B-cell signaling, leukocyte activation and somatic hypermutation of immunoglobulin genes pathways ( Figure  6A and B, supplementary data set). While the downregulation of immune pathways may seem in contrast to the observed upregulation of the same pathway in the liver, the downregulation is most likely caused by the above described reduction of neutrophils and monocytes in the bone marrow of Adar Δ7-9 ; Mavs −/− mice ( Figure 4A-D). Ras-GRP1 was found to be down-regulated in bone-marrow which regulates the development of B1a cells carrying autoantigen receptors (60). This gene was also down-regulated in the cortex of these mice. We also analyzed RNA-Seq data from the cortex of Adar Δ7-9 ; Mavs −/− mice (Kapoor, Licht, Jantsch, unpublished). Here we identified a larger number of 628 differentially expressed genes. This high number can most likely be explained by the larger sequencing depth. Here, 306 genes were up-regulated while 322 genes were down-regulated ( Figure 6A and B, supplementary data set). Both differentially up-and downregulated genes were enriched for processes that are specific to this tissue, like axon-and neuron development (Supplementary Figure S8). Thus, rescued Adar Δ7-9 mice show tissue-specific changes in gene-expression that correlate with tissue-specific defects observed via histological and cytometry analysis.
Interestingly, two genes were commonly deregulated in all the three tissues examined: RPS3a1, a protein-coding gene was down-regulated while its pseudogene RPS3a3 was found upregulated ( Figure 6C). Rps3a1 encodes a ribosomal protein that is a component of the 40S subunit and has three processed pseudogenes dispersed in the mouse genome, Rps3a3 is one of them. Rps3a3 was the most significantly upregulated gene in bone marrow and liver of Adar Δ7-9 ; Mavs −/− mice while it was the third most significantly upregulated gene in cortex ( Figure 6A). Since this ribosomal protein was dysregulated in all examined tissues, we wondered whether translation was altered in Adar Δ7-9 ; Mavs −/− mice.
Ribosomal 40S:60S ratios are distorted in Adar Δ7-9 ; Mavs −/− mice To determine the translation profile in Adar +/+ ; Mavs −/− and Adar Δ7-9 ; Mavs −/− mice, we performed polysome profiling on cell lysates of livers isolated from P15 mice. The polysome profile of Adar Δ7-9 ; Mavs −/− livers showed a decrease in 80S monosomes at the expense of an increase of 60S and to some extend 40S particles ( Figure 7A). This is consistent with a previous report showing that knock-down of Rps3a1 in HeLa cells causes accumulation of free 60S particles resulting from a strong alteration in 40S subunit production due to a failure in the processing of 18S rRNA (61).
Rps3a3 is an intron-less processed pseudogene, likely derived from the parental gene Rps3a1. Rps3a3, shows nine mismatches in its nucleotide sequence compared to the coding region of Rps3a1 which would result in two amino acid changes at the protein level (Supplementary Figure S7A). Also, Rps3a3 has an annotated promoter. This led us to investigate whether Rps3a3 could form a protein product. To achieve this, we cloned Rps3a3 with an N-terminal FLAGtag and C-terminally fused to eGFP with a self-cleaving 2A peptide. Transfection of this construct in HEK293T, showed the presence of GFP indicating that Flag-RPS3a3 was translated (Supplementary Figure S7B). Expression of full-length Flag-RPS3a3 was confirmed by western blotting (Supplementary Figure S7B). Since the putative protein product of RPS3a3 is similar to RPS3A1 differing in only two amino acids, we checked the effect of over-expression of Flag-RPS3a3 by polysome profiling. We overexpressed Flag-RPS3a3 in HEK293T since RPS3a3 is not conserved in humans. Over-expression of RPS3a3 resulted in an accumulation of free 40S and 60S particles and a decrease in 80S particles ( Figure 7B). Western blot analysis on the fractions of polysome profiling confirmed Flag-RPS3A3, to be present in the ribosomal fractions ( Figure 7C). Fractions corresponding to 60S and 40S were verified by probing for Rps19 (a SSU protein) and Rpl11(a LSU protein). Mock transfected cells did not show any bands after probing with Flag antibody (Supplementary Figure S9). The effect of over-expression of RPS3a3 ( Figure 7B) and downregulation of RPS3a1 correlates with the polysome profile of Adar Δ7-9 ; Mavs −/− mice where we observed both 60S and 40S to be accumulating compared to Mavs −/− mice. Thus, a de-regulated RPS3a1 and RPS3a3 might affect the ribosomal subunits of Adar Δ7-9 ; Mavs −/− mice.
We could not detect expression of Rps3a3 in the tissues of Mavs −/− littermates. However, RNA-Seq analysis of wild-type mice at stages E8.5, E9.5 and E10.5 showed high expression of the pseudogene Rps3a3, in various microregions involved in cranio-facial development (41). Highest expression was seen in mandibular arch at these early embryonic stages of wild-type mice, with levels as high as 90 transcripts per million (TPM) ( Figure 7D). Expression of Rps3a3 and Rps3a1 was highly correlated in these tissues ( Figure 7D). Analysis of the ENCODE RNA-Seq data from wildtype mice at 10 weeks of age and at P0 indicated no expression of Rps3a3 with TPM close to zero. All these results confirm that Rps3a3 is not expressed at post-natal stages in wildtype mice in the tissues studied and is only expressed in the micro-regions of facial structures at embryonic stages E8.5, E9.5 and E10.5. While in Adar Δ7-9 ; Mavs −/− pups, expression of the pseudogene is de-regulated and shows a high expression in liver, bone marrow and cortex.

Dysregulation of Rps3a1 and Rps3a3 is not dependent on editing
Rps3a1 is located on chromosome 3 within 10MB of Adar. We therefore examined if the genetic variability arising from the 129/SV-derived congenic interval surrounding the Adar Δ7-9 knockout site could account for the different Rps3a1 expression. Analysis of E11.5 whole-embryo RNA-Seq data of another Adar Δ2- 13 and Adar Δ2-13 ; Mavs −/− strain that was maintained in an C57Bl6N background (28), showed that Rps3a1 and Rps3a3 are also dysregulated in both these Adar allele bearing strains, irrespective of the background (Figure 8). This supports the idea that dysregulation of Rps3a3 and Rps3a1 is most likely caused by Adar-deficiency and is not dependent on the mouse strains investigated. Interestingly, the two Rps3a isoforms are not dysregulated in the brain lysates of 12 week old mice expressing full-length but catalytically dead Adar E861A/E861A ; Ifih1 −/− mice (27). This strongly suggests that dysregulation of Rps3a1 and Rps3a3 is not caused by a lack of editing but rather by other activities of ADAR1 such as RNAbinding ( Figure 8). This notion is also supported by the finding that no editing events could be detected in RNAs encoding Rps3a1 or Rps3a3 in cortex, liver or bone-marrow of P15 pups {Licht, 2019 #8438}.

DISCUSSION
All existing deletion-and point mutation alleles of Adar show a comparable embryonic lethal phenotype that is accompanied by liver disintegration and heightened immune response. However, as the different alleles of Adar produce various fragments of the protein, they may also affect so far unstudied functions of Adar. Indeed, attempts to rescue the various Adar alleles by either deleting Mavs or Ifih1 leads to different phenotypes, depending on the Adar allele used (24,(27)(28)(29). Since, the point mutation Adar E861A/E861A ; Ifih1 −/− shows a complete rescue of lethality and the complete deletion Adar Δ2-13 ; Mavs −/− survives only 24 hours post-partum, it appears that editing-independent functions of ADAR1 are important for normal growth and development in mice. This notion is also consistent with the find-ing that deletion of the interferon-inducible long variant of Adar p150 cannot be rescued by expression of the constitutively expressed version Adar p110 (24). Known editingindependent functions of ADAR1 most likely result from the RNA-binding activity of ADARs that may regulate RNA stability, microRNA biogenesis, splicing and translation (5,7,8,20,21). However, the molecular mechanisms causing post-natal lethality and miniature size of Adar Δ2-13 ; Mavs −/− mice remain unknown.
The two deletion alleles Adar Δ2- 13 and Adar Δ7-9 were always considered functionally identical and have indeed been used interchangeably. Both alleles show embryonic lethality at E12.5. However, our study shows that Adar Δ7-9 pups rescued by deletion of Mavs live significantly longer than Adar Δ2-13 ; Mavs −/− pups. This discrepancy can most likely be explained by the fact that the Adar Δ7-9 allele can form a truncated, editing-deficient protein. Interestingly, deletion of exons 7-9 in mouse Adar removes the region where the nuclear localization signal is located in human Adar. Our study shows that a mouse protein corresponding to truncated p110 Δ7-9 remains primarily nuclear. This suggests that human and mouse Adar may differ in their NLSs. In fact, a critical amino acid for NLS function (R801) is unique to the human but not the mouse protein (48). As expected, p150 Δ7-9 shows cytoplasmic localization.
Interestingly, the Adar Δ7-9 allele had been analyzed previously (29). In this case, a different Mavs allele was used for the rescue of the Adar Δ7-9 allele. In the mentioned study, deletion of Mavs only resulted in a 1 to 2 day survival post partum. In contrast, the mice analyzed by us survive significantly longer. At present, we cannot address whether this is a result of the different Mavs alleles or a result of different strain backgrounds. The mice used in this study had a ∼75% Sv129 and 25% Bl6 background. It is therefore possible that the different backgrounds used contribute to the observed different phenotypes. We can almost certainly exclude differences due to animal housing conditions as mice analyzed in this study were kept in different mouse houses in different facilities over several years and showed comparable litter sizes and survival rates. Moreover, we find sporadic long-term survivors in our litters. These long-term survivors catch up in growth and look normally, at least by histological examination. Unfortunately, the long-term survivors occur at very low frequency, precluding their systematic investigation. In any case, the occurrence of long-term survivors indicates that the penetrance of the Adar Δ7-9 phenotype in a Mavs background may be variable.
The presence of the residual Adar Δ7-9 protein may have different effects. Possibly the remaining dsRBDs of Adar Δ7-9 may shield dsRNAs thereby preventing them from excessive activation of MAVS. A similar function has been postulated for the catalytic dead version, ADAR E861A . This mutant can be fully rescued by a deletion of Ifih1 indicating that RNA-binding of ADAR1 may have critical functions (26). Also the Z-DNA binding domains may interfere with other RNA-processing steps such as PKR activation, miRNA processing, or transposon activation, to name a few. All these functions have been previously reported for ADARs as editing-independent functions (9,10,62,63). In fact, PKR has been shown to be critically involved in the sensing of dsRNAs, particularly in the absence of ADARs (62,64,65). Consequently, it will be interesting to determine whether a concomitant deletion of PKR may further alleviate the phenotype of the Adar Δ7-9 ; Mavs −/− mice.
The phenotype of Adar Δ7-9 ; Mavs −/− mice may also be caused by defects in the production of ribosomal subunits due to dysregulation of Rps3a1 and Rps3a3. This, in turn, may affect translation of mRNAs. As both, Adar Δ2-13 ; Mavs −/− and Adar Δ7-9 ; Mavs −/− mice show a deregulated expression of Rps3a1 and its pseudogene Rps3a3 in all tissues examined at embryonic as well as at post-natal stages, it may be a possible cause for the early lethality observed in both mouse strains. At present, we cannot judge whether the different extents of rescue observed between the two deletion alleles is also due to different expression levels of Rps3a1 and its pseudogenes. Still, to our knowledge this is the first report that demonstrates that ADAR can affect ribosomal subunit levels.
The fact that Rps3a1 and Rps3a3 de-regulation was observed in both Adar Δ7-9 and Adar Δ2-13 mice but not in the Adar E861A/E861A allele, argues for an editing-independent regulation of these ribosomal genes. We cannot exclude the possibility that upregulation of Rps3a3 is the result of down-regulated Rps3a1, or vice-versa. Most interestingly, we could also show that the annotated pseudogene Rps3a3 has indeed the potential to form a protein product that can get incorporated in ribosomal subunits ( Figure 7C). Whether this would also be the situation in vivo remains open. We have also tested whether editing of miRNAs might affect Rps3a1 expression. However, so far, no miRNAs that target the mRNAs encoding this small ribosomal subunit protein have been identified (66).
RPS3a1 has extra-ribosomal functions and has been shown to be involved in cell cycle regulation, cell apoptosis, immunity and signal transduction (67)(68)(69). Enhanced Rps3a1 expression primes cells for apoptosis and suppression of such enhanced expression leads to execution of apoptosis (69). Knock-down of Rps3a1 triggers death of undifferentiated HL-60 cells but not of retinoid-induced differentiated cells. This is consistent with our observation, Adar Δ7-9 ; Mavs −/− show drastically reduced mature B-cells along with massive apoptosis while the numbers of immature B-cells do not change compared to Mavs −/− mice. Most interestingly, Drosophila and zebra fish mutants with knock-down of Rps3a1 show defects in body size and hematopoiesis as seen in Adar Δ7-9 ; Mavs −/− mice (70)(71)(72). Given these results, it is interesting to speculate if defects in apoptosis, hematopoiesis and body size observed in rescued Adar Δ7-9 and Adar Δ2-13 are due to down regulated Rps3a1.

DATA AVAILABILITY
RNA-Seq data generated in this study has been deposited in the European Nucleotide Artiche under project numbers PRJEB31568 and PRJEB31565.