RNA editing plays an important role in the diversification of gene products. There are two different types of enzymatic deamination which lead either to the conversion of adenosine to inosine (A-to-I RNA editing) or of cytidine to uridine (C-to-U RNA editing). While the role of A-to-I RNA editing in health and disease is being investigated by many laboratories, research on C-to-U RNA editing is still at a rudimentary level (Meier et al., 2016).

In this study, we focused on C-to-U RNA editing and the involvement of the apolipoprotein B (ApoB) mRNA editing enzyme catalytic polypeptide-like Apobec-1. Successful cloning of APOBEC-1, the first identified member of the APOBEC gene family of cytidine deaminases (Navaratnam et al., 1993; Teng et al., 1993), was a milestone in research on C-to-U RNA editing. The role of APOBEC1 and auxiliary proteins in pre-mRNA editing of ApoB in liver and small intestine was elucidated in great detail (Backus and Smith, 1992; Schock et al., 1996; Dance et al., 2002; Smith, 2007). However, C-to-U RNA editing may also play a critical role in the central nervous system. In fact, C-to-U RNA editing of gene transcripts coding for the neurotransmitter receptor for glycine (GlyR) was discovered over 10 years ago (Meier et al., 2005). In the case of GlyR, C-to-U RNA editing leads to amino acid substitution of leucine for proline and receptor gain-of-function (Meier et al., 2005; Legendre et al., 2009).

Using bulk material of resected hippocampi, GlyR C-to-U RNA editing was shown to be increased in patients with severe intractable temporal lobe epilepsy (iTLE) (Eichler et al., 2008). Increased C-to-U RNA editing leads to neuronal gain-of-function through presynaptic activity of RNA-edited GlyRs, resulting in facilitation of neurotransmitter release (Bischofberger et al., 2006; Meier et al., 2014). This ultimately and persistently tilts the balance between excitation and inhibition in a bidirectional way (Eichler and Meier, 2008). Depending on the neuron type that is afflicted with increased GlyR C-to-U RNA editing, cognitive dysfunction or persistence of contextual fear memory can thus arise through enhanced function of glutamatergic or parvalbumin type GABAergic neurons, respectively, which has been shown in a corresponding animal model of the disease (Winkelmann et al., 2014; Çaliskan et al., 2016). Thus, methods for neuron type-specific detection of RNA-edited GlyR protein are ultimately required to elucidate the role of this gain-of-function neurotransmitter receptor in iTLE.

We present here novel molecular and chemical tools for detection of C-to-U RNA editing at the single cell level. Furthermore, the results identify an agonist that allows discrimination between RNA-edited and non-RNA-edited GlyR proteins in hippocampal neurons. The combination of these novel molecular and chemical tools enable a proof-of-principle demonstration of the role of Apobec-1 in C-to-U RNA editing of GlyR and expression of the RNA-edited GlyR protein. Moreover, bioinformatic analysis revealed worldwide differences in the allelic distributions of two different APOBEC1 80I- or 80M-coding alleles. Using a new PCR-based restriction fragment length polymorphism (RFLP) approach we furthermore provide a retrospective assessment of the iTLE patients analyzed in 2008 (Eichler et al., 2008) with regard to expression of Apobec-1 80I and 80M variants. The results showed that iTLE patients with 80I expression mostly experienced simple or complex partial seizures, whereas patients with 80M expression exhibited secondarily generalized seizure activity. This establishes the basis for a new mapping of epidemiology and semiology of iTLE with regard to APOBEC1 80M/I alleles and opens up new ways to functionally characterize the Apobec-1 80M and 80I variants in the brain and in iTLE patients. Together, these results advance our understanding of C-to-U RNA editing at the single cell level and establish expression of the Apobec-1 80I and 80M variants as new genetic risk factors of iTLE.

Recent studies identified the intracellular large cytosolic loop between transmembrane domains 3 and 4 (TM3-4 loop) as a highly efficient nuclear targeting domain (Melzer et al., 2010; Förstera et al., 2014). Here, we revisited and corroborated these findings and furthermore show that addition of the transmembrane domain 4 (TM4) efficiently prevents nuclear import of GlyR α1 and α3 TM3-4 loops and hence is a dominantly acting cytosol-retaining domain. Figure 1A shows the molecular design of the constructs used here. We used TM3-4 loops of GlyR α1^ins and α3K splice variants as described (Förstera et al., 2014). These domains were cloned downstream of mCherry. For control purpose, mCherry alone or equipped with SV40 NLS (KKKRK) were used. Furthermore, we included the TM4 in the TM3-4 loop constructs of GlyR α1^ins and α3K. Figure 1B shows representative images of transfected HEK293T cells. Note that the GlyR α1^ins and α3K TM3-4 loops (red) efficiently targeted the nucleus (green), while the TM4 domain efficiently overrode the nuclear targeting activity of the TM3-4 loops and led to almost exclusive cytosolic localization of the mCherry signal (Figure 1C). Quantification of the ratio between nuclear and cytosolic fluorescence intensities and statistical analysis demonstrated the dominant effect of the TM4 over TM3-4 loops (Figure 1, Table 1).

These clear-cut results prompted us to develop a C-to-U RNA editing sensor (Figure 2). As the GlyR α3K TM3-4 with and without TM4 provided most accentuated results regarding nuclear vs. cytosolic targeting, the human Apobec-1 consensus sequence derived from apolipoprotein B was inserted between the C-terminal end of the TM3-4 loop of GlyR α3K and the N-terminus of the GlyR α3 TM4 domain (Figure 2A). However, in contrast to the constructs used in Figure 1, the fluorescent reporter protein was inserted upstream of the C-to-U RNA editing sensor GlyR TM3-4 domain to indicate Apobec-1-dependent RNA editing through translocation of the fluorescence into the nucleus. Editing of the CAA codon to the STOP codon UAA should thus result in nuclear targeting of the reporter fluorescence protein (here EGFP upstream of the sensor domain). In order to test the efficacy of the editing sensor, we used the recently established faithful multicistronic co-expression technology (Tang et al., 2009). Editing sensor was either expressed alone or in combination with human Apobec-1 complementation factor (ACF) and Apobec-1 80I or 80M variants, using the Thosea asigna virus 2A self-processing peptides (Tang et al., 2009; Figure 2B).

We investigated the functionality of these constructs using the C-to-U RNA editing-deficient liver carcinoma cell line HEPG2 (Chen et al., 2010). As Apobec-1 was previously associated with C-to-U RNA editing of apolipoprotein B and GlyR (Backus and Smith, 1992; Meier et al., 2005), we assessed the effect of Apobec-1 80M and 80I on the subcellular localization of the editing sensor. Figures 3A–C show representative images of HepG2 cells expressing EGFP-editing sensor (A, control) or ACF-Apobec-1 80M/I-editing sensor constructs shown in Figures 2B, 3B,C. Quantification using line scans of fluorescence intensities in cytosol and nucleus of the transfected cells (Figure 3D) corroborated the marked first impression of prominent nuclear targeting of the sensor upon 2A-peptide-dependent co-expression of 80M or 80I Apobec-1 variants [mean fluorescence ratios between nucleus and cytosol; editing sensor alone: 0.761 ± 0.114 (mean ± SD), editing sensor with Apobec-1 80M or 80I: 2.031 ± 0.646 (mean ± SD) and 2.324 ± 0.659 (mean ± SD), respectively; Figure 3E].

To pursue these promising results, we next characterized the performance of the editing sensor in rat primary hippocampal neurons (Figure 4). Figures 4A–C shows representative images of primary neurons expressing EGFP-editing sensor (Figure 4A, control) or ACF-Apobec-1 80M/I-editing sensor constructs shown in Figures 2B, 4B,C. As in HEPG2 cells, quantification using line scans of fluorescence intensities in cytosol and nucleus of the transfected neurons (Figure 4D) revealed that nuclear targeting of the editing sensor was increased upon 2A-peptide-dependent co-expression of 80M or 80I Apobec-1 variants [mean fluorescence ratios between nucleus and cytosol; editing sensor alone: 0.763 ± 0.187 (mean ± SD), editing sensor with Apobec-1 80M or 80I: 1.940 ± 0.316 and 2.197 ± 0.407 (mean ± SD), respectively; Figure 4E].

Our pharmacological screens (Schneidereit et al., 2017) suggested NH₄ as agonist of RNA-edited GlyRs, which may allow discrimination of non-edited and edited GlyRs. To address this possibility in detail, we used transfected HEK293T cells and performed whole cell patch clamp analysis (Figure 5). We focused here on the GlyR α2 subunit because primary hippocampal neurons at DIV8-13 predominantly express GlyR α2 (Meier and Grantyn, 2004; Raltschev et al., 2016). However, as both splice variants of GlyR α2 (Figures 5A,B) and the GlyR α3L subunit are expressed in the iTLE hippocampus (Eichler et al., 2008), we included both GlyR α2 splice variants (Figure 5B) and GlyR α3L (Supplementary Figure 1) into our analyses. Thus, we co-expressed EGFP and GlyR α2A-192P/L, α2B-192P/L, or α3L-185P/L in HEK293T cells and recorded the currents in response to 5–50 mM NH₄ and 100 μM glycine (Figure 5, Supplementary Figure 1). Quantification of the current ratios revealed that 10 mM NH₄ is the most suitable concentration for effective discrimination between non-edited and edited α2- and α3-GlyRs (Figure 5B, Supplementary Figure 1). Furthermore, strychnine blocked NH₄-elicited currents, indicating that NH₄ indeed activated GlyRs (for values see Table 2, Supplementary Table 1).

To investigate endogenous expression of RNA-edited GlyR protein, we transfected primary hippocampal neurons with the editing sensor, in which EGFP was substituted by mCherry (above, Figures 2–4), alone or in combination with ACF-Apobec-1 80M/I-EGFP variants (Figure 6A). As control, EGFP was co-transfected with the editing sensor. The co-transfection experimental approach was used here to detect subcellular localization of the Apobec-1 variants and the editing sensor simultaneously in live neurons. Figure 6A shows representative images of live neurons expressing mCherry editing sensor (middle panels) with either EGFP (control, upper row), Apobec-1 80M-EGFP (middle row), or Apobec-1 80I-EGFP (lower row). Analysis of the fraction of neurons with nuclear sensor localization revealed that 25.5% of EGFP-expressing neurons had a nuclear sensor localization (12 out of 47; Figure 6B, upper row). Upon coexpression of Apobec-1 80M, 77.5% (62 out of 80) showed a nuclear sensor localization (Figure 6B, middle row), whereas 98.6% (71 out of 72) of neurons showed nuclear sensor localization upon Apobec-1 80I expression (Figure 6B, lower row). We finally tested neurons with nuclear sensor localization for NH₄ responsiveness (Figures 6C–E). In the EGFP control group, the mean ratio between current amplitudes elicited by 10 mM NH₄ and 100 μM glycine was 0.033 ± 0.009 (n = 16; Figure 6D). Both, Apobec-1 80M and 80I variants increased this ratio to 0.064 ± 0.021 (n = 16) and 0.089 ± 0.032 (n = 10), respectively (Figure 6D). The fraction of NH₄-responsiveness was then calculated according to amplitude ratios between NH₄- and glycine-dependent responses that were above 0.033, which corresponded to the mean ratio we measured in the EGFP control group (Figure 6E). Although slightly increased in both 80M and 80I conditions, the ratios between current amplitudes elicited by 10 mM NH₄ and 100 μM glycine in NH₄-responsive neurons was not significantly different (control: 0.076 ± 0.016, n = 5; 80M: 0.151 ± 0.033, n = 6; 80I: 0.140 ± 0.042, n = 6). However, this revealed that Apobec-1 80I almost doubled the fraction of NH₄-responsive neurons (60%, 6 out of 10), compared to control neurons (EGFP, 31.25%, 5 out of 16 neurons) or neurons with Apobec-1 80M expression (37.5%, 6 out of 16; Figure 6E). Thus, human Apobec-1 80I is more efficient than the 80M variant in generating RNA-edited GlyR protein in primary rat hippocampal neurons.

Two different genomic human APOBEC1 variants are annotated (NCBI Reference Sequences NM_001644.4 and AAA64230.1 coding for 80M and 80I, respectively). Through a bioinformatic analysis of 1,000 Genome Project Phase 3 allele frequencies we investigated the world-wide distribution of genotypes corresponding to APOBEC1 80M- and 80I-coding alleles (Figure 7, Supplementary Table 2; note that the human APOBEC1 reference gene sequence is annotated in antisense direction of the ORF-coding sequence). This analysis revealed that the 80I-coding allele is most frequent world-wide (Figure 7A, “ALL,” G: 66%). The European population leads with 89% of individuals harboring the 80I-coding allele (Figure 7A, “EUR”), followed by the South Asian population (Figure 7A, “SAS,” G: 77%), the American population (Figure 7A, “AMR,” G: 62%), East Asian and African population (Figure 7A, “EAS,” “AFR,” G: 56%, and 50%, respectively). Figures 7B,C reveal the distribution of homo- and heterozygous allele carriers. The broader European population (EUR) featured 496 out of 503 reads (98.6%) which detected at least one allele coding for 80I; while the frequency drops to 75.0% in the African population (AFR); see Supplementary Table 2 for details.

Given the world-wide heterogeneity of 80I- and 80M-coding APOBEC1 allele frequencies and regarding the preference of the Apobec-1 80I variant to generate RNA-edited GlyR protein, we wondered whether iTLE patients analyzed in 2008 with regard to GlyR RNA editing (Eichler et al., 2008) expressed the 80I- or 80M-coding allele. For this purpose, we developed an assay for detecting Apobec-1-coding mRNAs. The 80M-coding sequence gives rise to an NlaIII restriction site CATG and thus allows discrimination of 80M- and 80I-coding APOBEC1 gene transcripts according to different restriction fragment lengths (Figure 8A). As control and restriction fragment lengths reference, cDNA clones encoding Apobec-1 80M or 80I were processed in parallel, and GAPDH was amplified to ensure integrity of cDNA probes. We analyzed the 23 iTLE samples published in 2008 (see Table 1 in Eichler et al., 2008 for details on iTLE patients). Representative examples of the results of the PCR-RFLP experiments are shown in Figure 8B. Nineteen out of 23 samples showed Apobec-1 mRNA expression. In detail, 14 out of 19 expressed 80I-coding mRNA, and 5 out of 19 expressed 80M-coding mRNA. Thus, a fraction of 74% expressed the 80I-coding allele, which is lower than expected according to the data provided by our bioinformatics analysis (above, Figures 7B,C). With regard to anamnesis data of the iTLE patients analyzed in 2008, we found that the majority of patients (13 of 14) with Apobec-1 80I expression (93%) suffered from simple and/or complex partial seizure activity. In contrast, 2 out of 5 iTLE patients with expression of the 80M allele (40%) experienced secondarily generalized tonic clonic seizures, with 4 out of the 5 iTLE patients with 80M expression belonging to the W2-4 group (Eichler et al., 2008).

Changes in neural network homeostasis due to maladaptive changes in presynaptic plasticity and resulting persistent neuronal gain or loss of function can play a critical role in neuropsychiatric comorbidities of iTLE. In more detail, we showed that presynaptic expression of gain-of-function C-to-U RNA-edited GlyR leads to neuronal gain of function and, depending on the affected neuron type, elicits either cognitive dysfunction or persistence of contextual fear memory (Meier et al., 2014, 2016; Winkelmann et al., 2014; Çaliskan et al., 2016). Recent advances in sequencing technologies provide insights into global changes in RNA editing, but they are not able to draw conclusions on the cell/neuron type-specific changes involved in maladaptive plasticity in disease as they use bulk material. Thus, innovative technologies are required to assess changes in posttranscriptional RNA processing at the single cell level. Here, we present novel molecular and chemical tools for the investigation of changes in C-to-U RNA editing at the single cell level. Furthermore, our results highlight a possible link between Apobec-1 80I expression and simple or complex partial seizure activity.

Our molecular sensor tool indicates C-to-U RNA editing by nuclear translocation of a fluorescence protein, which provides a major advance compared to a recently published alternative tool that involves kinetics of EGFP protein turnover (Severi and Conticello, 2015). Our tool does not involve protein stability but fast kinetics of nuclear import at a time scale of seconds (Förstera et al., 2014). Thus, our sensor tool will be applicable to assess acute changes in Apobec-1 enzyme activity. Indeed, our new tool reliably indicated C-to-U RNA editing: Equimolar co-expression of the editing sensor and 2A-peptide-dependent Apobec-1 80M and 80I variants demonstrated efficient nuclear translocation of the sensor protein in the C-to-U RNA editing-deficient cell line HEPG2 (Figure 3). Moreover, we showed that our editing sensor works equally well in transfected rat primary hippocampal neurons (Figure 4). Our live cell imaging experiments furthermore revealed that spontaneous C-to-U RNA editing by the rat Apobec-1 80T variant (in the absence of co-expressed human Apobec-1) occurred in about 25% of sensor-expressing primary hippocampal neurons. In fact, our experiments revealed that expression of human Apobec-1 80I and the auxiliary protein ACF rendered almost all 80I-expressing neurons editing competent (Figure 6), and they showed that the 80I variant was significantly more effective in translocating the sensor to the neuronal nucleus compared to the 80M variant (Figure 4E).

The C-to-U RNA editing sensor tool presented here reliably indicated RNA editing in HEPG2 cells and primary neurons, but it did not allow to draw conclusions on GlyR C-to-U RNA editing. For this purpose, a combined approach involving NH₄ action on RNA-edited GlyRs and the sensor tool was required. Indeed, we showed that application of 10 mM NH₄ elicits strychnine-sensitive RNA-edited GlyR-dependent currents in transfected HEK293T cells. However, the fractions of C-to-U RNA editing-competent and NH₄-responsive neurons in the context of rat endogenous neuronal Apobec-1 80T expression or human Apobec-1 80M overexpression were rather low (31.3 vs. 37.5%). Overexpression of the human Apobec-1 80I variant increased this fraction to about 60% of neurons with nuclear sensor localization. This result reveals a preference of Apobec-1 80I for GlyR α2 C-to-U RNA editing and indicates that neuronal GlyR α2 C-to-U RNA editing is regulated, depending on the amino acid at position 80 of the polypeptide chain of Apobec-1. The ratio between current amplitudes generated by 10 mM NH₄ and 100 μM glycine in transfected HEK293T cells with RNA-edited GlyR α2-192L was about 10 times higher compared to neurons that overexpressed Apobec-1 80M or 80I in primary rat hippocampal neurons (Figure 5 vs. Figure 6). This supports neuronal endogenous regulation of GlyR C-to-U RNA editing and furthermore indicates that about 10% of endogenous rat neuronal GlyR α2 were edited in NH4-responsive and C-to-U editing-competent neurons. This fraction corresponds well with the fraction of RNA-edited GlyR α2 in iTLE patients with a severe course of disease (Eichler et al., 2008). However, all the other C-to-U editing enzymes expressed in primary rat hippocampal neurons may also contribute to neuronal regulation of GlyR α2 RNA editing, even irrespectively of ACF (Supplementary Figure 2). In agreement with a recent study (Snyder et al., 2017), co-expression of ACF with Apobec-1 80M or 80I did not affect the preference of Apobec-1 80I over 80M for GlyR α2 C-to-U RNA editing.

The preference of Apobec-1 80I over 80M for expression of GlyR α2 C-to-U RNA-edited protein in neurons led us to develop a new approach for the retrospective study of mRNA expression of Apobec-1 80M and 80I in iTLE samples (Eichler et al., 2008). Based on the fact that the 80M-coding sequence generates an NlaIII restriction site, we investigated mRNA preparations of previously characterized iTLE patients (Eichler et al., 2008). We chose investigation of mRNA derived from these iTLE patients because this approach unambiguously provides information about Apobec-1 mRNA expression even if a heterozygous genetic APOBEC1 background exists. In contrast to the rather low overall representation of the APOBEC1 80M-coding allele in our analysis of allele frequency distributions in Europe (11%, Figure 7A), assessment of heterozygosity in this population revealed a higher fraction (20%) of persons with at least one allele that codes for the 80M Apobec-1 variant (Figure 7C, see Supplementary Table 2 for details). Departing from this allele-distribution in our sample of the general European population, 26% of the iTLE patients analyzed in 2008 expressed the 80M allele, and 4 out of 5 iTLE patients with 80M expression belong to the W2-4 group of iTLE patients with increased expression of RNA-edited GlyR (Eichler et al., 2008). Although we can't retrospectively provide information regarding the ethnic origin of these patients, this result suggests an overrepresentation of iTLE patients with 80M expression compared to our bioinformatics analysis (26 vs. 20% in Europe), and the mismatch between increased GlyR RNA editing in the W2-4 group of iTLE patients with Apobec-1 80M expression suggests that other C-to-U RNA editing enzymes have contributed to GlyR RNA editing (including the GlyR α3 subunit) in these patients. Furthermore, pronounced hippocampal cell loss and the type of seizures in the W2-4 group of iTLE patients with Apobec-1 80M expression suggests that Apobec-1 80M targeted gene products are involved in neurodegeneration and secondary generalization of seizure activity (SGTCS). In fact, the success rate of our patch clamp analysis of neurons with 80M expression was rather low (38.1%), compared to 60.5% success rate in patch clamp experiments involving the 80I variant, and 50% of W2-4 classified iTLE patients with Apobec-1 80M expression experienced SGTCS. In contrast, the vast majority (93%) of the iTLE patients with Apobec-1 80I mRNA expression had simple or complex partial seizures. This fits well with the observed phenotypes of animals with increased expression of RNA-edited GlyR which exhibit cognitive dysfunction and impairment of spatial and contextual fear memory (Winkelmann et al., 2014; Çaliskan et al., 2016). Considering the rather low fraction of RNA-edited GlyRs in patients with iTLE and assuming a uniform receptor distribution in all neuronal compartments supports the proposed pathogenic presynaptic role of these receptors in response to low extracellular glycine concentrations, due to the low capacitance of the presynaptic compartment and that irrespectively of whether RNA-edited GlyR α2 or α3 is expressed (Meier et al., 2014, 2016), which is in agreement with a previous study (Chen et al., 2014). However, it is possible that only a few neurons or certain neuron types generate larger amounts of RNA-edited GlyRs in iTLE and contribute in a neuron type-specific way to different symptoms (Winkelmann et al., 2014; Çaliskan et al., 2016). We cannot rule out this possibility yet because bulk material was used for investigation of GlyR α2 or α3 RNA editing in iTLE (Eichler et al., 2008).

In conclusion, our study revealed human genetic APOBEC1 80M/I dimorphism as a new diagnostic marker in iTLE. The 80I variant can contribute to simple or complex seizure activities and increases protein expression of RNA-edited GlyR, which would preferentially affect the presynaptic compartment and hence trigger maladaptive and persistent dysregulation of neuron gain control, resulting in simple or complex partial seizures. In contrast, the 80M Apobec-1 variant is suggested to promote neurodegeneration and secondary generalization of seizure activity. Moreover, the results indicate that a complex interplay of different C-to-U RNA editing enzymes regulates GlyR RNA editing in iTLE, which remains to be characterized at the single cell level for example by applying the technologies established here to resected hippocampi from patients with iTLE. Finally, this study provides the basis for a new mapping of epidemiology and semiology of iTLE with regard to APOBEC1 80M/I alleles, which can open avenues for further characterization of the functional roles of Apobec-1 80M and 80I variants in healthy and diseased brains.

This study was carried out in accordance with the recommendations of guidelines of the Charité Ethics Commission (EA1/142/05) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Charité Ethics Commission.

SK, BF, AW, PK, XY, MS, EW, and FH performed experiments. SK, FH, and JM wrote the manuscript. JM designed the study.