Stem-loop structure preference for site-specific RNA editing by APOBEC3A and APOBEC3G

Cell culture, plasmids and transfection

Cell cultures of primary monocyte-enriched PBMCs, exposure to hypoxia (1% oxygen) and interferon type 1 were performed as previously described (Sharma et al., 2015).

Plasmid constructs for expression of human A3A cDNA, for the generation of C-terminal Myc-DDK-tagged A3A and A3G, pcDNA 3.1(+) vector (used as an empty vector control) were obtained from sources mentioned in Sharma et al. (2015) and Sharma et al. (2016).

The TLA-HEK293T human embryonic kidney cells (293T cells) (Open Biosystems, Lafayette, CO, USA) were transfected with plasmid DNA using the jetPRIME (Polyplus-transfection) reagent as per the manufacturer’s instructions. The transfection efficiency was 60%–80% as assessed by fluorescent microscopy of cells that were transfected with the pLemiR plasmid (Open Biosystems, Lafayette, CO, USA) for expression of a red fluorescent protein. Cells were harvested 2 days following transfection.

Purification of recombinant A3A proteins

The WT A3A was purified as described in Sharma et al. (2015). Briefly, Rosetta 2(DE3)pLysS E. coli (EMD Millipore, Burlington, MA, USA) transformed with a bacterial expression construct for C-terminal His₆-tagged WT A3A was grown in Luria broth at 37 °C. The cells were induced for expression of the recombinant protein with 0.3 mM isopropyl β-D-1-thiogalactopyranoside and cultured overnight at 18 °C. A3A protein was purified from the lysates by affinity chromatography using the Ni-NTA His bind Resin (EMD Millipore).  The concentrated protein was stored in 25 mM Tris (pH 8.0) with 50 mM NaCl, 1 mM DTT, 5% v/v glycerol and 0.02% w/v sodium azide at −80 °C.

Predicting RNA secondary structures

18 nucleotides (with 7 nucleotides flanking on each side of the tetra-loop sequence) of WT SDHB, TMEM109 and APP RNAs were folded using the Mfold nucleic acid folding program (Zuker, 2003). 18 nucleotides of WT PRPSAP2 RNA were folded using both Mfold and RNAfold 2.3.2 (Zuker, 2003; Lorenz et al., 2011). No optional parameters were used. A single structure along with the minimum free energy value for the structure was obtained for the selected RNAs and is represented in Fig. S1.

RNA mutagenesis and RNA editing assays

The DNA templates for generating WT and mutant SDHB (except M1, M6, M7), TMEM109 and APP RNAs were amplified using oligonucleotide primers listed in Table S1. M1, M6 and M7 SDHB RNAs were generated from the 1.1 kb complete SDHB ORF encoding plasmid (RC203182, Origene) following site-directed mutagenesis and XhoI linearization of the plasmid DNA. Sanger sequencing was performed on all DNA templates to confirm the desired mutations, which were then in vitro transcribed to generate RNAs using reagents and methods provided with the MEGAscript or MEGAshortscript T7 Transcription Kit (Life Technologies, Carlsbad, CA, USA). RNAs isolated from the transcription reaction were treated with DNAse I (Thermo Fisher Scientific, Waltham, MA, USA) and their integrity was verified by electrophoresis on an agarose gel.

In vitro RNA-editing assay with purified APOBEC3A contained 1–10 µM APOBEC3A, 50 pg of synthetic RNAs, 10 mM Tris (pH 8.0), 50 mM KCl and 10 µM ZnCl₂. The reactions were incubated for 2 h at 37 °C. RNA was purified from the reactions using TRIzol (Life Technologies, Carlsbad, CA, USA) as per the manufacturer’s instructions and reverse transcribed to generate cDNAs as described previously (Sharma et al., 2015). The 136C>U editing of the WT and certain SDHB RNA mutants (M1, M3–M7) was assessed by allele-specific AS-RT–qPCR as described previously (Sharma et al., 2015; Baysal et al., 2013), whereas RNA editing levels for remainder of the mutant RNAs along with the WT controls were determined by Sanger sequencing, using the primers listed in Table S1, because these mutants could not be amplified by AS-RT-qPCR reverse primers (Table S2, Data S1). We have previously shown (Sharma et al., 2015, Fig. S8) a strong positive correlation (r = 0.94) between SDHB 136C>U RNA editing level measurements obtained by AS-RT-qPCR and Sanger sequencing, although estimates obtained by Sanger sequencing were somewhat higher (slope of correlation = 0.71). Thus, it is possible that true editing levels in SDHB constructs M2, M8, M9 and M10, which were estimated by Sanger sequencing (Data S1), might be actually lower (by approximately 30%).

Since in vitro RNA editing by A3G has not yet been demonstrated, to examine the impact of RNA mutations in the A3G substrate PRPSAP2 on RNA editing, we co-transfected mutated PRPSAP2 expression plasmid with A3G expression plasmid in 293T cells. The mutations were performed by site-directed mutagenesis (New England Biolabs, Ipswich, MA, USA) in the PRPSAP2 expression plasmid (clone ID Ohu59963, RefSeq accession XM_011523960; GenScript). Total RNA was isolated and RT-PCR was performed using a PRPSAP2-specific forward primer and a vector specific reverse primer complementary to the DDK tag sequence (Table S1). These primers specifically amplified the plasmid derived PRPSAP2 transcripts but not the endogenously expressed transcripts, allowing us to directly examine the impact of RNA mutations on A3G-mediated RNA editing.

Estimation of RNA editing levels by Sanger sequencing

Sequencing primers (Integrated DNA Technologies) for the WT and mutant cDNAs generated from RNAs are listed and underlined in Table S1. The PCR products were examined by agarose gel electrophoresis to verify their size and then sequenced on the 3130 xL Genetic Analyzer (Life Technologies, Carlsbad, CA, USA) at the RPCI genomic core facility as described in (Sharma et al., 2016). The major and minor chromatogram peak heights at putative edited nucleotides were quantified with Sequencher 5.0/5.1 software (Gene Codes, Ann Arbor, MI) in order to calculate the editing level for the position (Data S1). Since the software identifies a minor peak only if its height is at least 5% that of the major peak’s, we have considered 0.048 (=5∕(100 + 5)) as the detection threshold (Sharma et al., 2016; Data S1).

Preference for stem-loop structure for site-specific A3A and A3G-mediated RNA editing

Previous studies have shown that A3A-mediated DNA deamination of synthetic oligonucleotides occurs non-specifically at TC dinucleotides (Chen et al., 2006; Shinohara et al., 2012; Sharma et al., 2015; Chan et al., 2015). However, A3A-mediated cellular ssRNA editing is site-specific, and bioinformatic analyses predicted that approximately 70% of the edited Cs in A3A’s RNA substrates are located within secondary structures (Sharma et al., 2015). The most common secondary structure is predicted to be comprised of a CAUC tetra-loop flanked by an average of three palindromic nucleotides (Sharma et al., 2015). Similarly, bioinformatics analyses predicted that ∼98% of the edited Cs in A3G RNA substrates are located within secondary structures; the most common structure comprising of CNCC (N is any nucleotide) flanked by an average of four palindromic nucleotides (Sharma et al., 2016). Separately, while validating edited sites in primary monocytes by Sanger sequence analysis, we observed that a silent A/G single nucleotide polymorphism (SNP) in the A3A substrate C1QA mRNA (rs172378) markedly increased C>U RNA editing three nucleotides upstream of the polymorphism (Fig. 1A). The A>G change in the C1QA mRNA (rs172378) is predicted to increase the stem length and subsequently the stem stability of a putative stem-loop structure, resulting in increased RNA editing. While the CCCCCUCGG(a/a) (expressed SNP variation in lower case) sequence shows 11% and 21% editing in 2 donors, CCCCCUCGG(a/g) increased the average editing to 40% when monocyte-enriched peripheral blood mononuclear cells (MEPs) were exposed to hypoxia/IFN-1 (Fig. 1A). Although the estimated RNA editing level may have been affected by background noise (in donor 2 and 3 hypoxia/IFN-1 samples), visual inspection of the chromatograms clearly show that the heterozygous donor 1 cellular RNAs are edited at a higher level compared to homozygous donor 2 and 3 cellular RNAs.

We thus hypothesized that stem-loop RNAs are preferred substrates for editing by APOBEC3A and -3G proteins. We selected three site-specifically edited A3A mRNA substrates-SDHB (NM_003000: c.136C>U, R46X), APP (NM_001204302: c.1546C>U, R516C), TMEM109 (NM_024092: c.109C>U, R37X) (Sharma et al., 2015; Sharma et al., 2017) and one such A3G substrate, PRPSAP2 (NM_001243941: c.664C>U, R222W) (Sharma et al., 2016) for further analysis. On analysis of 18 nucleotides of RNA sequence containing the target C by the Mfold (Zuker, 2003) or RNAfold (Lorenz et al., 2011) nucleic acid folding prediction programs, secondary structures with ΔG values between −5 to −6 kcal/mol are predicted for SDHB, APP and PRPSAP2 RNAs (Fig. S1). The predicted secondary structure for these RNAs is a tetra-loop with the edited C at the 3′ end of the loop flanked by a stem containing 3–5 base pairs (bp). TMEM109 is predicted to form a hepta-loop with a four bp stem and a ΔG value of −1.7 kcal/mol. To test the importance of stem-loop structures for A3A and A3G-mediated RNA editing, we created various mutations (see methods) in the putative loop and stem regions of A3A substrates SDHB, APP and TMEM109 and the A3G substrate PRPSAP2 (Figs. 1B–1D) and assessed their editing levels. SDHB, APP, TMEM109 RNAs show ∼83%, 24%, 51% site-specific editing in an in vitro system, respectively and PRPSAP2 shows ∼44% RNA editing in a cell based system. The RNA editing levels were analysed by AS-RT-qPCR for the SDHB mutants, except those which did not have a reverse primer compatible for AS-RT-qPCR analysis of RNA editing (see methods; Table S2). The remainder of the SDHB, APP, TMEM109 and PRPSAP2 mutants were analysed by Sanger sequencing (Data S1). In either method for assessing RNA editing levels, WT RNA substrates were used as a positive control. For convenience in data interpretation, RNA editing of WT substrates is set to 100% and that of mutant RNAs is reported as a fraction of that observed with the WT substrates.

We tested the importance of the −1 nucleotide (nt) (immediately 5′ to C) in A3A and A3G substrates. C>U editing sites are most commonly present within a CCAUCG sequence motif in ssRNA A3A substrates (Sharma et al., 2015). Changing the −1 U to C, (UC>CC) in the predicted loop region of the SDHB RNA (Fig. 1B, M1), markedly reduced A3A-mediated RNA editing from the normalized value of 100% to 19%. Unlike most A3A substrates, which prefer U at −1 position, in APP RNA the −1 nt is occupied by C. Interestingly, substituting C at −1 with U in APP RNA (CC>UC), increased editing to 364% (Fig. 1C, M1A). The majority of C>U editing sites in A3G substrates are present within a CNCC [A/G] sequence and therefore prefer C at −1 position (Sharma et al., 2016). Changing −1 nt to G (CC>GC) in the A3G substrate PRPSAP2 RNA loop markedly reduced RNA editing to 15% as compared to WT (Fig. 1D, M1P). These results suggest a preference for U and C at the −1 position in the loop regions of A3A and A3G substrates, respectively.

We next tested the importance of the location of the reactive C within the predicted loop region of the A3A RNA substrate, SDHB. As mentioned above, our computational analysis predicts that the edited C is generally located at the 3′ end of the tetra-loop (Sharma et al., 2015; Sharma et al., 2016; Sharma et al., 2017). Changing the position of edited C one nucleotide upstream within the loop in SDHB RNA, while maintaining U at −1 position, greatly reduced RNA editing to 10% (Fig. 1B, M2). This result suggests that position of the reactive C within the loop is critical for RNA editing.

The majority of known A3A and A3G RNA substrates are predicted to form a tetra-loop structure (Sharma et al., 2015; Sharma et al., 2016; Sharma et al., 2017). To test whether the size of the loop plays a role in RNA editing, we created substitutions that increase or decrease the predicted loop size in SDHB and PRPSAP2 RNAs (Fig. 1B, M3 and M4; Fig. 1D, M2P and M3P). Increasing from a tetra-loop to a penta-loop (Fig. 1B, M3) reduced RNA editing to 10% in the SDHB RNA, and decreasing the size to a tri-loop (Fig. 1B, M4) diminished editing to 60% as compared to the WT SDHB RNA. Changing the size of the loop (penta- or tri-loop) of the PRPSAP2 RNA abolished A3G-mediated RNA editing (Fig. 1D, M2P and M3P). These results suggest that a larger loop is detrimental to both A3A and A3G-mediated RNA editing, whereas reducing the size of the loop to three nucleotides may be tolerated better.

We next tested whether the sequence and/or structure as well as stability of the predicted stem are determinants of RNA editing. We weakened or disrupted the predicted stem by decreasing the number of complementary base pairs in SDHB and PRPSAP2 RNAs (Fig. 1B, M5, M6 and M7; Fig. 1D, M4P and M5P). All of these changes reduced SDHB RNA editing 5–10 fold to 16%, 12% and 10%, respectively (Fig. 1B M5, M6 and M7) and abolished A3G-mediated RNA editing of PRPSAP2 (Fig. 1D, M4P and M5P). Further, altering (inverting/swapping) the sequence of the stem while maintaining base-pairing of the SDHB RNA (Fig. 1B, M8 and M9) also reduced RNA editing levels to 37% and 17% as compared to WT, respectively. These observations suggest that both sequence and stability of the RNA structure are important for optimum RNA editing.

Usually the +1 position (with regard to C) in A3A substrates is occupied by a G base-paired with C or in some cases A, which was substituted with A in M8 (37% editing) and C in M9 (17% editing) (Fig. 1B). Hence, to test the importance of G at +1 position, we created another mutant that retained the first base pair of the predicted stem (G at +1) as WT SDHB, but remainder of the stem sequence and structure was similar to the M9 SDHB mutant (Fig. 1B, M9 and M10). Changing C at +1 position in the M9 mutant to G in the M10 mutant (Fig. 1B) increased A3A-mediated RNA editing from ∼17% to 130%, respectively (Fig. 1B). These results suggest that the structure/stability rather than sequence of the predicted stem, other than G at +1 position determines the level of RNA editing.

To further examine the importance of the stability and structure of the stem, we analyzed the A3G RNA substrate, PRPSAP2 and the A3A substrate TMEM109. Interestingly, weakening the putative stem by substituting two G-C base pairs with A-U base pairs in PRPSAP2 RNA only affected RNA editing slightly (80%) (Fig. 1D, M6P). As mentioned above, disrupting the predicted stem structure abolished RNA editing in PRPSAP2 (Fig. 1D, M5P). However, on swapping the 5′ and 3′ sequence while maintaining the stem complementarity as well as the first C-G base pair, increased RNA editing to 180% as compared to WT PRPSAP2 (Fig. 1D, M7P). Similarly, when we compare the SDHB RNA mutants M6 and M10 (Fig. 1B), restoring the stem stability and structure, while maintaining G at +1 position increased RNA editing from 12% to 130%. These results provide further evidence that stem stability and G at +1 position, rather than nucleotide sequence in the remainder of the predicted stem region determine the level of RNA editing.

As mentioned above, for the A3A substrate TMEM109, the Mfold program predicts a hepta-loop flanked by a four bp stem (Fig. S1). However, if the unpaired adenosine in the hepta-loop region bulges out then we predict WT TMEM109 RNA to form a tetra-loop with C at the 3′ end of the loop, G at +1 position base paired with C and a 5 bp long stem (Fig. 1C). To test the effect of perfect stem complementarity on TMEM109 RNA editing level, we deleted the unpaired adenosine (Fig. 1C, M1T). Unlike for WT TMEM109 (ΔG =  − 1.7 kcal/mol), the Mfold program predicts a ΔG value of −5.2 kcal/mol for TMEM109 M1T structure (Fig. S1), suggesting an increase in secondary structure stability. Deletion of the unpaired adenosine to obtain perfect stem complementarity resulted in an increase in the RNA editing level of TMEM109 from 100% to 122% (Fig. 1C, M1T).

Taken together, our results show that for site-specific RNA editing, A3A and A3G prefer a stem-loop secondary structure, with C at the end of the tetra-loop as well as specific nucleotides at 5′ and 3′ positions immediate to the reactive C , and suggests that the sequence of the predicted stem other than at +1 position is not as important as the stability of base pairing.