Mutations in ADAR1 cause Aicardi-Goutières syndrome associated with a type I interferon signature

Abstract

Adenosine deaminases acting on RNA (ADARs) catalyze the hydrolytic deamination of adenosine to inosine in double-stranded RNA (dsRNA) and thereby potentially alter the information content and structure of cellular RNAs. Notably, although the overwhelming majority of such editing events occur in transcripts derived from Alu repeat elements, the biological function of non-coding RNA editing remains uncertain. Here, we show that mutations in ADAR1 (also known as ADAR) cause the autoimmune disorder Aicardi-Goutières syndrome (AGS). As in Adar1-null mice, the human disease state is associated with upregulation of interferon-stimulated genes, indicating a possible role for ADAR1 as a suppressor of type I interferon signaling. Considering recent insights derived from the study of other AGS-related proteins, we speculate that ADAR1 may limit the cytoplasmic accumulation of the dsRNA generated from genomic repetitive elements.

Aicardi-Goutières syndrome (MIM 225750) is a genetically determined inflammatory disorder particularly affecting the brain and skin. In its most characteristic form, AGS is a clinical mimic of in utero–acquired infection¹ and, like congenital infection, is associated with increased production of the antiviral cytokine interferon α (IFN-α)². AGS can result from mutations in any one of the genes encoding the DNA exonuclease TREX1 (AGS1)³, the three non-allelic components of the RNase H2 endonuclease complex (RNASEH2B, AGS2; RNASEH2C, AGS3; and RNASEH2A, AGS4)⁴ and the deoxynucleoside triphosphate triphosphohydrolase SAMHD1 (AGS5)^5,6. Although AGS is most typically inherited as an autosomal recessive trait⁷, rare examples of disease due to de novo dominant mutations in TREX1 have been reported^8-10.

Studies of the function of TREX1 have delineated a cell-intrinsic mechanism for the initiation of an autoimmune response by interferon (IFN)-stimulatory nucleic acid^11,12, begging the question of the source of nucleic acid inducing the type I IFN–mediated immune disturbance in AGS. In this regard, it has been shown that TREX1 can metabolize reverse-transcribed DNA and that single-stranded DNA derived from endogenous retroelements accumulates in TREX1-deficient cells¹¹. On a related note, TREX1 (ref. 13), SAMHD1 (refs. 14–16) and RNase H2 (ref. 17) have been implicated in the metabolism of the (exogenous) retrovirus HIV-1. Perhaps most notably, a recent study showed rescue of the lethal inflammatory TREX1-null mouse phenotype by a combination of reverse transcriptase inhibitors (antiretroviral therapy as used to treat HIV-1)¹⁸, suggesting that the accumulation of cytosolic DNA in TREX1-null cells can be ameliorated by inhibiting endogenous retroelement cycling.

To define other genes relevant to the AGS phenotype, we undertook whole-exome sequencing in four individuals with a clinical diagnosis of AGS, all of whom screened negative for mutations in TREX1, RNASEH2A, RNASEH2B, RNASEH2C and SAMHD1. Using in-solution hybridization followed by massively parallel sequencing, we derived over 2 Gb of mapped sequence for each subject, such that an average of 56-fold coverage was achieved across the exome for all the samples (Supplementary Table 1). We performed an analysis of the called nonsynonymous, splice-site, substitution and coding insertion and/or deletion exome variants under a model of a rare autosomal recessive disorder. Visual inspection of the generated data identified two affected individuals, AGS81_P1 and AGS219, who each had two nonsynonymous coding alterations in ADAR1, a gene we had already highlighted as a candidate for AGS in view of its known role as a suppressor of type I IFN signaling^19,20. Sanger sequencing confirmed the variants in these individuals, as well as in two further affected siblings from family AGS81. In light of these data, we proceeded to sequence ADAR1 in other individuals lacking mutations in TREX1, RNASEH2B, RNASEH2C, RNASEH2A and SAMHD1 (AGS1–AGS5, respectively) from our AGS cohort.

In total, 12 affected individuals from 8 families harbored biallelic ADAR1 variants (Fig. 1, Table 1 and Supplementary Table 2), which were considered likely pathogenic on the basis of species conservation (Supplementary Figs. 1 and 2) and the output of pathogenicity prediction packages (Supplementary Table 3). In these families, all parents tested were heterozygous for one putative mutation. Two further unrelated affected individuals, AGS150 and AGS474, harbored a single mutation encoding p.Gly1007Arg that was not present in either parent. Genotyping of microsatellite markers was consistent with stated paternity, indicating that this variant had arisen de novo in both cases (Supplementary Table 4). Of the nine distinct ADAR1 mutations we identified, the c.577C>G (p.Pro193Ala) transversion was seen in the compound heterozygous state in five families of European ancestry. This same variant was observed in 41 subjects (32 of 4,350 European-Americans and 9 of 2,203 African-Americans) annotated on the Exome Variant Server database, whereas none of the other ADAR1 variants present in our AGS cohort were found in more than 12,000 control alleles.

Figure 1.

Schematic of the human ADAR1 gene. (a) ADAR1 spans 26,191 bp of genomic sequence on chromosome 1q21.3 (154,554,533–154,580,724). Neighboring genes are also shown. Cen, centromeric; tel, telomeric. (b) Position of identified mutations within the genomic sequence of the ADAR1 long isoform (p150). The number of alleles with each mutation is shown in parentheses. *, the mutation encoding p.Gly1007Arg identified as a single heterozygous de novo change in two families; †, the same mutation identified in identical twins (therefore counted once). Numbers given above the gene indicate the relevant exons (only exons with mutations are numbered). The shorter isoform (p110) of ADAR1 starts at c.886 of the p150 isoform. (c) Position of identified variants within the ADAR1 p150 1,226-amino-acid protein. Numbers above the protein are the amino-acid count at the exon boundaries. The shorter isoform starts at amino acid 296 of p150, giving rise to a 931-amino-acid protein. (d) Schematic of the position of protein domains and their amino-acid boundaries in the p150 isoform of ADAR1. Note that the p110 isoform does not include the Zα DNA/RNA-binding domain and nuclear export signal.

Table 1.

Ancestry, pedigree structure, consanguinity status and sequence alterations in ADAR1 mutation–positive families in an AGS cohort

Family	Ancestry	Individuals tested	Consanguinity	Nucleotide alteration	Exon	Amino-acid alteration
AGS81	Norwegian	3A, M, F	−	c.[577C>G]+[2675G>A]	2, 9	p.[Pro193Ala]+[Arg892His]
AGS93	Italian	1A, M, F	−	c.[577C>G]+[2608G>A]	2, 8	p.[Pro193Ala]+[Ala870Thr]
AGS107	Pakistani	2A, M, F	+	c.3337G>C (hom)	14	p.Asp1113His
AGS150	Brazilian	1A, M, F	−	c.3019G>A (het, de novo)	11	p.Gly1007Arg
AGS219	Pakistani	1A	+	c.3335A>T (hom)	14	p.Tyr1112Phe
AGS228	Indian	1A, M, F	−	c.2997G>T (hom)	11	p.Lys999Asn
AGS251	White British	1A, M	−	c.[577C>G]+[2615T>C]	2, 8	p.[Pro193Ala]+[Ile872Thr]
AGS327	Italian	1A	−	c.[577C>G]+[1076_ 1080del]	2, 2	p.[Pro193Ala]+[Lys359Argfs*14]
AGS430	Spanish	2Aa, M, F	−	c.[577C>G]+[2675G>A]	2, 9	p.[Pro193Ala]+[Arg892His]
AGS474	European-American	1A, M, F	−	c.3019G>A (het, de novo)	11	p.Gly1007Arg

A, affected individual; M, mother; F, father; het, heterozygous; hom, homozygous.

^aIdentical twins.

ADARs catalyze the hydrolytic deamination of adenosine to inosine in dsRNA²¹. Four ADARs have been described in mammals (ADAR1, ADAR2, ADAR3 and TENR), although only ADAR1 and ADAR2 are known to have catalytic activity. ADAR1 is encoded by a single-copy gene that maps to human chromosome 1q21. Two main isoforms of ADAR1 are present in mammalian cells: a truncated ADAR1 protein (p110; nucleotide, NM_001025107.2; protein, NP_001020278.1) is constitutively expressed, whereas a full-length form of ADAR1 (p150; nucleotide, NM_001111.4; protein, NP_001102.2) is IFN inducible²². Both isoforms have been shown to shuttle between the nucleus and the cytoplasm. ADAR1 is a modular protein with a C-terminal deaminase catalytic domain, three centrally located dsRNA-binding domains (dsRBDs) and one or two N-terminal Z-DNA–binding domains; compared with p110, the p150 isoform of human ADAR1 possesses an additional 295 N-terminal amino acids containing a nuclear export signal and an extra Z-DNA/Z-RNA–binding domain (designated Zα) (Fig. 1)²³.

Of the eight amino-acid substitutions identified in our AGS cohort, seven involve residues situated in the catalytic domain of ADAR1 (Fig. 2). Five of these seven catalytic domain residues (Arg892, Lys999, Gly1007, Tyr1112 and Asp1113) lie along the surface of the protein that interacts with dsRNA (Fig. 2a), and the two others (Ala870 and Ile872) lie internal to the domain structure and are predicted to destabilize the protein (Fig. 2b,c). In contrast, Pro193 is positioned within the Z-DNA/Z-RNA–binding domain. In the wild-type protein, Pro193 makes direct contact with the nucleic acid, and substitution of this residue with alanine removes important atomic interactions between the protein and DNA/RNA (Fig. 2d,e).

Figure 2.

Structural context of ADAR1 protein substitutions. (a) The surface of the ADAR domain (dark pink) with surface substitutions highlighted in bright pink. The active site contains a zinc ion (black) in the center. Arg892His, Lys999Asn, Gly1007Arg, Tyr1112Phe and Asp1113His are all on the same side of the domain as the active site, and all have the potential to alter charge and/or hydrogen bonding characteristics of the surface in the region that is likely to be responsible for RNA binding. (b,c) Models of wild-type (b) and mutant (c) residues at position 870 in the ADAR domain. Interactions between the residue and the surrounding protein structure are indicated by all-atom contact dots (blue). Green dots represent energetically favorable van der Waals interactions, whereas red and pink spikes indicate unfavorable van der Waals overlaps. The Ile872Thr substitution introduces an unsatisfied hydrogen bond donor/acceptor group, which is destabilizing but not easily depicted. (d,e) Interactions of the proline residue at position 193 (Pro193) in the Z-DNA–binding domain (d). Contact dots (blue, green, yellow) indicate favorable interactions between Pro193 and the DNA backbone (white). These interactions are absent in the mutant form (e). (f) Modeling of the deaminase domain of ADAR2 suggests contact with dsRBD2 close to Gly1007, highlighting the possibility for an arginine residue to make functionally important polyvalent interactions.

More than 130 different ADAR1 mutations (Supplementary Table 5) have already been documented in individuals with dyschromatosis symmetrica hereditaria 1 (DSH), an autosomal dominant disorder characterized by the childhood onset of hypo- and hyperpigmented macules on the face and dorsal aspects of the extremities^24,25. The frequent observation of stop and frameshift ADAR1 variants in individuals with DSH indicates haploinsufficiency as the likely molecular pathology. However, ADAR1 missense variants, spread throughout the gene, are also commonly seen in association with the DSH phenotype. All except one of the ADAR1 mutations recorded in our AGS cohort were missense variants, and protein blotting of lymphoblastoid cells from affected individuals showed normal levels of both the constitutive and IFN-inducible isoforms of ADAR1 (Fig. 3). We predict that the proteins containing the amino-acid alterations seen in individuals with AGS act as hypomorphs and that, as in the Adar1-null mouse, complete loss of ADAR1 protein activity is embryonic lethal¹⁹.

Figure 3.

Protein blot of lymphoblastoid cell lines (LCLs). Protein blot analysis of ADAR1 expression in Epstein-Barr virus (EBV)-transformed LCLs from one unrelated control and two affected individuals (AGS93, AGS107). Whole-cell lysates were derived from 1 × 10⁷ cells per sample, and 10 μg of total protein was loaded per lane. To test antibody specificity and confirm IFN induction of the p150 isoform of ADAR1, unstimulated cells were compared to IFN-stimulated cells. The antibody to ADAR1 recognizes both the p110 and p150 isoforms. Immunoblotting of tubulin (50–55 kDa) was used as a loading control.

Of the nine discrete ADAR1 mutations observed in our AGS cases, only the allele encoding p.Gly1007Arg has been reported previously^26,27. Uniquely, this mutation was described in two individuals with DSH also demonstrating neurodegeneration with dystonia and intracranial calcification (Supplementary Table 6). Using a known ADAR1 editing substrate, miR376-a2, we found that, of six ADAR1 p110 mutants (Ala870Thr, Ile872Thr, Arg892His, Lys999Asn, Gly1007Arg and Asp1113His) expressed following transfection of plasmid constructs into HEK293 cells, only the Gly1007Arg variant showed a significant effect on editing, with levels of editing equivalent to those seen with inactive protein (Fig. 4a). Modeling to position the deaminase domain active site at the target adenosine along the dsRNA substrate (Fig. 2f) predicts that the deaminase domain also contacts dsRBD2 close to Gly1007. The proximity of Gly1007 to the RNA backbone, and the possibility for an arginine residue to make polyvalent interactions there, suggests a mechanism whereby Arg1007 might confer a dominant-negative effect: by binding more tightly to RNA, the mutant protein could act as a competitive inhibitor of wild-type protein, while being itself catalytically inactive. In keeping with this model, a plasmid expressing Gly1007Arg ADAR1 showed stronger inhibition of wild-type ADAR1 than equivalent amounts of a plasmid expressing catalytically inactive ADAR1 (Fig. 4b). Although these observations might explain the seemingly unique nature of the Gly1007Arg variant, how the other mutations that were tested in our system cause an aberrant phenotype remains unclear at this time. Possible explanations might relate, for example, to non-editing functions of ADAR1 (ref. 28) or to editing substrate²⁹ or cell type specificity¹⁹.

Figure 4.

Site-specific and competition editing assays. (a) HEK293 cells were co-transfected with 500 ng of a plasmid expressing miR376-a2 and 500 ng of a plasmid expressing wild-type (WT) ADAR1 or ADAR1 mutants. Background editing in HEK293 cells with 500 ng of substrate plasmid is approximately 20%. As previously observed, only one monomer in an ADAR1 dimer is required to be enzymatically active, such that addition of inactive protein initially increases editing⁴⁰. Editing activity is expressed as a proportion of WT ADAR1 editing activity, which is 1 (y axis). Error bars, s.e.m. ***P = 0.0005. (b) Competition between ADAR1 and inactive ADAR1 mutants. HEK293 cells were co-transfected with 200 ng of a plasmid expressing miR376-a2 and 200 ng of a plasmid expressing ADAR1 p110 in the presence of increasing amounts of a plasmid expressing a catalytically inactive form of ADAR1 (Glu912Ala)²⁸ or Gly1007Arg. Error bars, s.e.m. **P = 0.0026.

The recurrent p.Pro193Ala alteration lies in the Zα DNA/RNA-binding domain, thus implicating the IFN-inducible p150 isoform of ADAR1 in the AGS phenotype. Mice lacking ADAR1 die by around embryonic day 12.5 owing to defective hematopoiesis and wide-spread apoptosis, which are associated with global upregulation of IFN-stimulated genes (ISGs), indicating that ADAR1 acts as a suppressor of type I IFN signaling¹⁹. In light of these observations, using whole blood from 8 ADAR1 mutation–positive individuals, we performed quantitative RT-PCR to analyze the mRNA levels of 15 ISGs. Compared to nine controls, all tested individuals with mutations in ADAR1, including AGS150 and AGS474 harboring a de novo heterozygous allele encoding p.Gly1007Arg, showed a consistent pattern of ISG upregulation (Supplementary Fig. 3). We then undertook an analysis of a subset of the 6 most highly expressed ISGs in 10 ADAR1 mutation–positive AGS cases, 6 sets of parents with heterozygous mutations in ADAR1 and 18 ADAR1 mutation–positive individuals with DSH (Fig. 5 and Supplementary Table 7). For the six ISGs assayed, expression was variably higher in AGS heterozygous parents and DSH cases versus controls, whereas individuals with a clinical diagnosis of AGS (due either to biallelic mutations in ADAR1 or a heterozygous mutation resulting in a p.Gly1007Arg amino-acid substitution) had even higher levels of expression.

Figure 5.

Quantitative RT-PCR of a panel of six ISGs in whole blood measured in individuals with AGS, their parents and individuals with DSH. Scatter plots showing log₁₀-transformed RQ values for a panel of 6 ISGs measured in whole blood from 10 AGS cases with mutations in ADAR1, 6 sets of parents heterozygous for mutations in ADAR1, 18 individuals with ADAR1 mutation–positive DSH and 20 healthy controls. All genes were significantly upregulated in AGS cases (P < 0.001) compared to controls. RQ is equal to 2^−ΔΔC_T, with −ΔΔC_T ± s.d., that is the normalized fold change relative to a calibrator. ***P ≤ 0.001; **P ≤ 0.01; *P < 0.05.

The skin lesions typical of DSH have not been described in AGS (they are distinct from AGS-related chilblains)³⁰, and none of the ADAR1 mutation–positive AGS cases included in our study had obvious features of DSH. Of note, DSH has only very rarely been reported outside of Japan and China. Moreover, even within known families segregating the DSH phenotype, a marked variability in expression is well recognized³¹. Although our findings do not necessarily implicate dysregulation of type I IFN in the DSH phenotype, they suggest that missense and null heterozygous mutations in ADAR1 are consistent with both DSH (depending on, for example, ancestry) and with carrier or, in the case of the p.Gly1007Arg alteration, affected status for AGS. Thus, we believe that the lack of DSH skin features in our AGS cases and their heterozygous parents most likely relates to non-penetrance and/or variable expressivity due to ancestry or other genetic or non-genetic factors, and we predict that two individuals with DSH would have a 1 in 4 risk of having a child with AGS.

Because inosine is recognized as guanosine by the translation and splicing machineries, editing of adenosine to inosine can alter the protein-coding information of messenger RNA and the structural stability of dsRNA. The first identified ADAR-edited substrates were in codons, and ADARs were presumed to function primarily in proteome diversification. However, although codon editing is clearly important, it represents only a small fraction of editing events in the transcriptome, with editing sites in RNA derived from non-coding regions, most particularly Alu elements, being vastly more prevalent^32-35. The biological function of such repetitive element editing is uncertain³⁶. Recent studies have highlighted an antagonistic interaction between ADARs and the RNA interference machinery, suggesting that a pool of common RNA substrates is capable of engaging both pathways³⁷ and specifically implicating the binding properties of the p150 isoform of ADAR1, beyond its editing activity^28,38, in this relationship. How, then, does loss of ADAR1 activity lead to upregulation of type I IFN signaling? Theoretically, wild-type ADAR1 might edit specific currently undefined transcripts that are important in IFN regulation. Alternatively, and perhaps more likely, lack of editing in ADAR1-deficient cells may lead to an increase in immunoreactive dsRNA³⁷ and/or result in failure to generate inosine:uracil (IU)-dsRNA with an intrinsic function in suppressing IFN induction³⁹. Considering insights derived from the study of TREX1, SAMHD1 and RNase H2, we speculate that ADAR1 has a role in the metabolism of retroelements and that a major contribution of ADARs to evolutionary fitness may be to regulate the accumulation of dsRNA generated from basally transcribed repetitive sequences within the genome.

URLs. UCSC Human Genome Browser, http://genome.ucsc.edu/; Ensembl, http://www.ensembl.org/; dbSNP, http://www.ncbi.nlm.nih.gov/projects/SNP/; Exome Variant Server, National Heart, Lung, and Blood Institute (NHLBI) Exome Sequencing Project (ESP) (accessed 21 March 2012), http://snp.gs.washington.edu/EVS/; PolyPhen, http://genetics.bwh.harvard.edu/pph/; SIFT, http://sift.jcvi.org/www/SIFT_enst_submit.html; Align GVGD, http://agvgd.iarc.fr/index.php; Clustal Omega, http://www.ebi.ac.uk/Tools/msa/clustalo/; Protein Data Bank (PDB), http://www.pdb.org/; Alamut, http://www.interactive-biosoftware.com/.

ONLINE METHODS

Affected individuals and families

All affected individuals included in this study had a clinical diagnosis of AGS that was based on the presence of early-onset encephalopathy (at <18 months of age), negative investigations for common prenatal infections, intracranial calcification with or without white matter changes in a typical distribution, elevated levels of IFN-α with or without pterins in the cerebrospinal fluid and/or chilblains. Clinical information and samples were obtained with informed consent. The study was approved by the Leeds (East) Research Ethics Committee (reference 10/H1307/132).

Exome sequencing

Genomic DNA was extracted from lymphocytes from affected individuals and parents by standard techniques. For whole-exome analysis, targeted enrichment and sequencing were performed on 3 μg of DNA extracted from the peripheral blood of four individuals (AGS81_P1, AGS125, AGS163 and AGS219). Enrichment was undertaken using the 38 Mb SureSelect Human All Exon kit (Agilent) following the manufacturer’s protocol, and samples were paired-end sequenced on an Illumina HiSeq 2000. Sequence data were mapped using BWA (Burrows-Wheeler Aligner) against the hg18 (NCBI Build 36) human genome as a reference. Variants were called using SOAPsnp and SOAPindel (from the Short Oligonucleotide Analysis Package) with medium stringency and were then filtered for those with greater than 5× coverage.

Sanger sequencing

Primers were designed to amplify the coding exons of ADAR1 (Supplementary Table 8). Purified PCR amplification products were sequenced using BigDye terminator chemistry and an ABI 3130 DNA sequencer. Mutations were annotated on the basis of the reference cDNA sequence NM_001111.4, with nucleotide numbering beginning from the first A in the initiating ATG codon.

Gene expression analysis

The expression of 15 genes known to be interferon stimulated was assessed in whole blood. Total RNA was extracted from whole blood using the PAXgene RNA isolation kit (PreAnalytix). RNA concentration was assessed using a spectrophotometer (FLUOstar Omega, Labtech). Quantitative RT-PCR analysis was performed using TaqMan Universal PCR Master Mix (Applied Biosystems) and cDNA derived from 40 ng of total RNA. The relative abundance of target transcripts, measured using TaqMan probes for Ly6E (Hs00158942_m1), MX1 (Hs00895598_m1), USP18 (Hs00276441_m1), RSAD2 (Hs01057264_m1), OAS1 (Hs00973637_m1), IFI44L (Hs00199115_m1), IFI27 (Hs01086370_m1), ISG15 (Hs00192713_m1), IFIT1 (Hs00356631_g1), IFI44 (Hs00197427_m1), IFI6 (Hs00242571_m1), SIGLEC1 (Hs00988063_m1), IFIT3 (Hs00155468_m1), IRF7 (Hs00185375_m1) and STAT1 (Hs01013989_m1), was normalized to the expression level of HPRT1 (Hs03929096_g1) and 18s (Hs999999001_s1) and assessed with Applied Biosystems StepOne Software v2.1. Statistical significance between groups was determined by t tests using DataAssist v2.0 (Applied Biosystems). Data from affected individuals are expressed relative to the average of nine normal controls. A subset of the six most highly expressed ISGs (RSAD2, IFI44L, IFI27, ISG15, IFIT1 and SIGLEC1) were measured in 20 controls, 10 ADAR1 mutation–positive cases, 6 sets of ADAR1 heterozygous parents and 18 individuals with ADAR1 mutation–positive DSH. RNA from individuals with DSH was extracted using the QIAamp RNA Blood Mini kit (Qiagen), and cDNA was derived as above. Statistical significance between groups was determined by Kruskall Wallis tests using GraphPad Prism 5. The median fold change of the 6 ISGs compared to the median of the 20 healthy controls was used to create a score for each affected individual, similarly to previously described methods^41,42.

Interferon stimulation

EBV-transformed lymphoblastoid cells from affected individuals and controls were counted using a Bright-Line Hemacytometer (Sigma). We then transferred 10 ml of cells at a concentration of 1 × 10⁶ cells/ml to each of two T25 flasks per cell line. One flask for each cell line was stimulated with 1,000 international units (IU)/ml of human IFN-α (human Intron A, Shering-Plough) for 24 h.

Protein analysis

Whole-cell lysates were prepared from lymphoblastoid cells (1 × 10⁷ cells per sample) using 10 mM EDTA–RIPA buffer containing protease inhibitors (Roche). For protein blot analysis, 10 μg of total protein was loaded onto 8% SDS-PAGE gels, and electrophoresis was performed using the Mini-PROTEAN system (Bio-Rad Laboratories). Following wet-blotting transfer of the proteins onto PVDF membrane (Amersham), non-specific antibody binding was blocked using 5% normal goat serum (Vector Laboratories) and 2% BSA (Sigma) in 0.1% TBS-Tween (Sigma) overnight at 4 °C with gentle agitation. Rabbit primary antibody to ADAR (Sigma, Prestige Antibodies, HPA003890) was incubated with the membranes for 2 h at room temperature using a dilution of 1:800. As a loading control, membranes were incubated with a 1:4,000 dilution of rabbit primary antibody to α/β-tubulin (Cell Signaling Technology, 2148). Horseradish peroxidase (HRP)-labeled goat secondary antibody to rabbit (1:10,000; Cell Signaling Technology, 7074) was incubated with the membranes for 1 h at room temperature to detect both ADAR1 and tubulin antibodies. Signal was detected using a 1:10 dilution of Enhanced Chemiluminescence reagents (Lumigen).

Microsatellite genotyping

To exclude non-paternity, informative polymorphic microsatellite markers on chromosomes 3 (D3S3640 and D3S3560), 11 (D11S913 and D11S1889) and 20 (D20S847, D20S896 and D20S843) were genotyped using DNA from AGS150, AGS475, their parents and an unrelated control sample. DNA samples were amplified by standard PCR (primer sequences available upon request). Each amplicon was mixed with Hi-Di Formamide (Applied Biosystems) and 500 ROX Size Standard (Applied Biosystems) and run on the Genetic Analyzer 3010 capillary electrophoresis system. Results were analyzed with GeneMapper v4.1 software (Applied Biosystems).

Editing assays

Transient transfections were performed with slight adaptations to a previously described protocol²⁸ in 24-well plates seeded the day before transfection with 0.14 × 10⁶ cells using Lipofectamine 2000 and OptiMEM I reduced serum medium. Cells were transfected with 500 ng of plasmid expressing a known ADAR1 editing substrate, miR376-a2, together with 500 ng of plasmid expressing wild-type ADAR1 p110 protein or the mutants Ala870Thr, Ile872Thr, Arg892His, Lys999Asn, Gly1007Arg and Asp1113His, for which coding sequences had been subcloned into the pcDNA3.1 expression vector. Reduced serum medium was replaced after 6 h with DMEM, and cells were harvested after 48 h. Total RNA was extracted with TRIzol reagent (Invitrogen), purified and treated with TURBO DNase (Ambion) and RNasin Plus RNase Inhibitor (Promega). Reverse transcription was carried out using SuperScript II and random hexamers, and cDNA was amplified by PCR with Platinum Taq DNA polymerase and specific primers for pri-miR376-a2 (Supplementary Table 8). The 300-bp PCR product was purified with exonuclease in combination with shrimp alkaline phosphatase (SAP) before sequencing. RNA editing of the pri-miR376-a2 transcript was identified as an adenosine-to-guanine change within the cDNA sequence, and the editing level was expressed as a percentage. Peak heights were measures for edited (G) and unedited (A), the editing ratio percentage was calculated by (G / (A + G)) × 100. Results are the average of two separate experiments. Statistical significance between groups was determined by two-tailed t tests using Excel.

Protein modeling

The ADAR1 substitutions p.Ala870Thr, p.Ile872Thr, p.Arg892His, p.Lys999Asn, p.Gly1007Arg, p.Tyr1112Phe and p.Asp1113His all fall within the adenosine deaminase domain. No crystal structure is available for this domain in human ADAR1; thus, a comparative model was constructed using the deaminase domain of human ADAR2 (Protein Data Bank (PDB) 1zy7)⁴³. The sequences of ADAR1 and ADAR2 were aligned using ClustalW⁴⁴, and 25 models were built using Modeller⁴⁵. The model with the lowest discrete optimized protein energy (DOPE) score was selected (representing the most probable model judged by the fit to the Modeller statistical potential). Residues 975–996 could not be modeled accurately, as they have no equivalent in ADAR2; for this reason, these residues were not analyzed further. Models of mutations were built using KiNG⁴⁶, hydrogen atoms were added with Reduce⁴⁷, and all-atom contacts were calculated with Probe⁴⁸. In each case, all low-energy rotamers⁴⁹ were considered, and the rotamer with the best Probe score was used. We assumed a similar binding mode for ADAR2 to that described for the tRNA deaminase–tRNA anticodon complex⁵⁰, allowing us to orientate the deaminase domain of ADAR2 on dsRNA and position it relative to the ADAR2-dsRBD2 complex⁵¹. The crystal structure of the ADAR1 Z-DNA–binding domain⁵² (PDB 1qbj) was used to analyze the likely structural effect of the p.Pro193Ala substitution. As the side chain of alanine has no degrees of freedom from rotatable dihedral angles, the position of the side chain is determined by the conformation of the protein backbone.