In healthy brain tissue, editing levels for the recoding sites within the glutamate receptor are highly uniform across individuals [10,12,24]. As a first step, we tested whether the same is true for other editing sites occurring within coding sequences expressed in other tissues.

First, we investigated 32 skin samples. Although mice ADAR1 or ADAR2 knockout die in-utero or shortly after birth, RNA editing is implicated as relevant to the skin in humans by the observation that mutations in ADAR1 lead to Dyschromatosis symmetrica hereditaria [27], a pigmenting genodermatosis with an autosomal dominant inheritance reported predominantly in Japanese and Chinese individuals [28]. It is interesting to note that the effect of these mutations on ADAR1 function or editing pattern has not been found.

We looked at A-to-I editing levels in three recently discovered, mouse-conserved, targets within coding sequences: FLNA, CYFIP2 and BLCAP [29,30]. Filamin A (FLNA) displays an A→I editing site in its transcript (chromosome X:153,233,144, edited by both ADAR1 and ADAR2 [31] ), resulting in a Q→R substitution at amino acid 2341 in the human protein. The CYFIP2 (cytoplasmic FMR1 interacting protein 2) transcript encodes a protein of 1253 amino acids, and undergoes A→I editing (by ADAR2 [31]) at chromosome 5:156,669,386, resulting in K→E substitution at amino acid 320. BLCAP (bladder cancer associated protein), is highly conserved among species, having 91% and 100% identity at the DNA (coding region) and protein levels, and is recoded by editing. Here, we tested one editing site in the nucleotide encoding the second codon of BLCAP, located at chromosome 20:35,580,986, resulting in a Y→C substitution. Sequenome analysis for RNA editing [32] has shown an average editing level of 8, 9 and 14% for FLNA, CYFIP2 and BLCAP, respectively, in the skin samples. Remarkably, in all three genes, we observed a comparable level of editing for the 32 individuals, as mirrored by the low standard deviations - in all three sites the standard-deviation of the editing levels across individuals is about 1/3 of their averaged editing level (standard deviations were 2.7, 3 and 4.2 for FLNA, CYFIP2 and BLCAP, respectively, where editing levels are measured on a 0-100 scale as usual; see Figure 1A). These finding suggested that the protein diversity derived from editing is tightly regulated in these targets. Editing levels of FLNA and CYFIP2 in skin were lower than the values reported for normal brain tissue, while in BLCAP, the level was comparable to the one in normal brain tissue, oral cavity and lung [13]. This observation is consistent with previous reports suggesting that BLCAP is edited (almost) only by ADAR1, while the other two sites are edited by both ADARs or by ADAR2 alone [31].

Next, we tested the variability in editing levels of non-recoding sites. In particular, we focused on editing targets within the primate-specific Alu repeats which constitute the vast majority of editing sites in the human genome. In this category we looked at a single editing site in 3 targets selected from the RNA editing database [18]: FYN (last intron, chr6:112,094,677; (hg18)), CARD11 (the 14th intron, chr7:2,942,082), and KLRD1 (3' UTR, chr12:10,359,728). Using Sequenom analysis of the above 32 skin samples, we found low average editing levels in these sites, 19, 2, and 3% (with standard deviations of 11, 3 and 3) for FYN, KLRD1 and CARD11, respectively (Figure 1B, C). In order to quantify the degree of consistency among individuals, we compared the individual editing levels of the 6 different sites in the FLNA, CYFIP2, BLCAP, KLRD1, CARD11 and FYN genes. A statistically significant difference (p < 0.05 by Mann-Whitney test) between editing levels of the different sites was observed for 13 out of the 15 pairs of editing sites tested (see supplementary Table 1). We therefore conclude that the variation of results across individuals is significantly lower than the difference in editing efficiency between different sites (F-ratio 41.8 by ANOVA analysis; p-value 1.1E-28).

Most editing sites in the human transcriptome occur in clusters where a number of nearby sites undergo editing. Therefore, the question arises whether editing regulation occurs at the cluster level or at the site level. That is, whether regulation is able to distinct editing sites residing in the same highly-edited region. To answer this question, we direct sequenced an Alu repeat within the last intron of the FYN gene for 32 different human skin samples and analyzed 7 different editing sites in this region (see Additional file 1). Twenty two of the samples resulted in high-quality sequence data. Distinct editing levels were found for the 7 sites (Figure 2). Mann-Whitney analysis showed the editing levels of different sites to be distinct: 20 out of 21 comparisons resulted in a significant (p < 0.05, supplementary Table 2) difference. ANOVA analysis resulted in F-ratio 68.25 (p = 4.9E-40), demonstrating that the difference in editing efficiency among various sites in the same Alu repeat is an order of magnitude larger than variability in editing efficiency of specific sites across individuals.

It is established that editing levels are particularly high in some tissues, including the brain [33]. This can be attributed to the elevated level of ADAR2 in the brain. Here we aim at comparing the editing levels in a number of non-brain tissues. For this purpose we analyzed editing data [18] for clusters of editing sites in the 3'UTR Alu sequences of 3 different genes: MDM4, NRIP3 and THOC5 (see Additional file 1) in the following tissues: lung, kidney, prostate, uterus, liver and glioma tumor (not including normal brain tissue). We found low variability among different tissues and a clear distinction between different editing sites of the 3 genes. That is, given the six values of editing levels in the six non-brain tissues, for two different sites within the same Alu repeat, the tissue-to-tissue variance is low enough to allow one to tell (using a standard statistical test, e.g. t-test or Mann-Whitney) that the two sets of six measurements describe two sites differing in their editing efficiency. The tissue-to-tissue similarity is high enough to allow pair-wise distinction between different sites in a cluster in 51/91, 148/300 and 50/105 of the pairs, in MDM4, NRIP3 and THOC5, respectively (p < 0.05; Mann-Whitney test. supplementary tables 3-5). ANOVA analysis yielded F-ratios 12.71, 18.20 and 5.29 (p-values 7.5E-16, 1.4E-26 and 4.7E-07) for MDM4, NRIP3 and THOC5, respectively, again showing that the tissue-to-tissue variability is an order of magnitude lower than the site-to-site variability within the same cluster. The standard-deviation to mean ratio is 0.34, 0.49 and 0.62 for MDM4, NRIP3 and THOC5 (averaged over all sites in the same gene).

Regulated editing events might be recruited for functional processes. As an example, we studied the Alu editing in the NARF gene. In this gene, insertion of Alu retrotransposon pair into an intron led to editing of the Alu repeats, which in turn created a novel primate-specific alternatively spliced exon [34]. We looked at the tissue-to-tissue consistency of the editing levels in 5 different sites within this Alu using the data published by Lev-Maor et al [34]. In concordance with our previous data, the editing levels of NARF were highly consistent among different human tissues including transformed and cancerous human cell lines (Hela, 293T, MCF7, SKOV3 and MDA, Figure 3): statistically significant difference (p < 0.05) was observed in 9/10 comparisons (supplementary table 6). ANOVA analysis yields an F-ratio 182.6 (p = 7.39E-33), demonstrating once more the individual to individual scatter to be much lower than editing-efficiency variability.

The study was approved by the Institutional Helsinki Committee at Sheba Medical Center, Tel Hashomer, Israel and informed consent was properly obtained by all participants. Human skin tissues were frozen by liquid nitrogen after their removal at surgery and kept at -70°C until further use. Thirty two skin samples were tested; 20 inflammatory skin lesions with the following clinical and pathological diagnoses: Atopic/nummular Dermatitis n = 7, Drug eruption n = 2, Psoriasis n = 1, Allergic contact dermatitis n = 1 and Cutaneus T-cell Lymphoma (Mycosis Fungoides) n = 9. 12 Normal skin samples were collected at the Chaim Sheba Medical Center.

Total RNA was isolated using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Random-primed cDNA synthesis was done on 2 μg of total RNA using M-MLV reverse transcriptase (Invitrogen) according to the manufacturer's instructions. For analyzing editing levels we used Sequenom (San Diego, CA) MassARRAY Compact Analyzer and MassARRAY Assay Design 2 software, as described before [32]. The primer sequences and reaction conditions are available as supplementary data (Additional file 1).

In order to compare editing levels of different sites in a highly edited region within an Alu repeat, we direct sequenced the 3' UTR of FYN transcripts. PCR reaction for cDNA products was carried out using 100 ng (1 λ) of cDNA obtained as described above, at an annealing temperature of 60°C. The PCR product was separated by 1.5% agarose gel electrophoresis, and extracted using QiaQuick gel extraction kit (QIAGEN) and according to the manufacturer's instructions. cDNA sequencing was carried out in genetic analyzer 3100 (Applied Biosystems/Hitachi; Foster City, CA), and according to the manufacturer's sequencing protocol. High-quality sequences were analyzed. Sequencing results were read using Sequencher 4.2 software (1991-2004 Gene Codes Corp., Ann Arbor, MI); editing quantification was carried out using DS gene 1.5 software (Accelerys Inc. Discovery Studio 2003; San Diego, CA).

Mar. 2006 (hg18) assembly