AGO-bound miRNAs represent a major pool of circulating miRNAs (Arroyo et al., 2011), and measuring AGO-bound miRNAs may enhance the ability to distinguish brain disease subtypes (Raoof et al., 2017), and identify post-seizure from baseline samples in plasma from epilepsy patients (Raoof et al., 2018). Moreover, there is disruption of the blood–brain barrier after seizures sufficient to allow passage of macromolecules between the blood and brain (Van Vliet et al., 2007; Reschke et al., 2021).

To obtain evidence that seizures and the pathophysiologic changes that accompany the development of epilepsy are associated with changes to the circulating AGO pool of miRNAs, we first sequenced Ago2-bound miRNAs from the plasma of mice following status epilepticus (SE) triggered by intraamygdala microinjection of kainic acid (Mouri et al., 2008; Brennan et al., 2020). The model displays typical features of the pathophysiology of TLE including select neuron loss and gliosis within the hippocampus and the emergence of recurrent, spontaneous seizures which are resistant to various anti-seizure medicines (Mouri et al., 2008; West et al., 2022). Figure 1A shows a schematic of the study design, including the time points at which blood was collected. In order to obtain sufficient Ago2 yield for sequencing, 500 μL of plasma was used, from pooling samples from n = 10 mice per time point. After confirming immunoprecipitation of Ago2 from mouse plasma (Supplementary Figure S1A), plasma samples at various time points were processed for Ago2 elution followed by small RNA sequencing.

A total of 23.7 million reads were mapped from 7 samples across four time points in control and SE mice. The numbers of normalised miRNA counts per sample varied from 2,038,033 to 5,388,335, with an average of 3,390,383 (Supplementary Figure S1B). Numbers of unique miRNA detected per sample showed minimal variation (range 141–154; average: 150) (Figure 1B). Likewise, the average number of miRNA was quite similar between seizure compared to control mice (Figure 1B). The highest number of counts was observed in the 4 h post SE time-point (5,388,335) (Supplementary Figure S1B). While the lack of replicates prevents a statistical analysis, miRNA counts at the 8 h and 24 h post SE time-points were at intermediate levels (3,151,945 and 3,709,772, respectively) and the lowest numbers of counts (2,532,271) in a seizure sample were observed at the 2-week post SE time-point. These findings suggest that SE may have a small effect to increase the amount and diversity of circulating Ago2-bound miRNAs, which subsides over time.

To explore the likely tissue origins and diversity (i.e., the number of different miRNAs) of the circulating Ago2-bound miRNAs, we compared the miRNAs across all samples (SE and control) to tissue-enriched miRNA datasets (Lagos-Quintana et al., 2002; Landgraf et al., 2007; Liang et al., 2007; Lee et al., 2008; Bargaje et al., 2010; Guo et al., 2014; Ludwig et al., 2016). Nearly half of the detected miRNAs were predicted to be enriched in a specific tissue type (Figure 1C). MiRNAs enriched in a number of different tissue types included heart (4%), liver (1%), digestive system (3%), spleen (2%), kidney (2%) and blood (3%). The highest proportion of circulating miRNA for which specific tissue expression was known were brain-enriched miRNA (19%).

Next, we used available data on brain cell-enriched miRNAs across all samples (SE and control) to assign the detected miRNAs to cell types, including neurons and glia (see Methods and Supplementary Table S1; Figure 1D). Where there were discrepancies in assigning brain cell types, priority of assignment was based on a ranking (see Methods). For miRNAs reported as brain-enriched but where there was insufficient information on cell type, these were termed unspecific brain enriched (Ludwig et al., 2016). Of all miRNAs detected in the Ago2 fraction of mouse plasma, 22% had known brain or brain cell type enrichment (Figure 1D). MiRNAs of neuronal origin were the most abundant in the Ago2-eluted plasma of mice, followed by astrocytic, oligodendrocytic and microglial. Numbers of miRNA enriched in these brain cell types are as follows; unspecific (15), neuron (62), astrocyte (26), oligodendrocyte (19) and microglia (18) (Figure 1D). Counts per million of the 10 most abundant brain cell type enriched miRNA observed in SE across all time points include miRNAs with reported enrichment in a variety of brain cell types; astrocytes, neurons, microglial and oligodendrocytes (Figure 1E).

We next explored time-dependent changes to the Ago2-bound miRNAs in mouse plasma after SE (Figures 1E–G and Supplementary Figures S2, S3). The largest differences were observed at early time-points. At the 4 h, 8 h and 24 h time-points after SE, changes were observed for miRNA enriched in brain, heart, liver, digestive system, spleen, kidney and the blood (Supplementary Figures S2, S3). At the two-week time-point, when mice typically have spontaneous seizures in this model, miRNA enriched in brain, heart, spleen, kidney, and the blood were dysregulated. In addition to these organ-enriched miRNA, miRNA with broad expression across organs were also dysregulated (see Supplementary Table S1).

The largest differences between SE and control samples for Ago2-bound miRNA were found at the earliest time point (4 h; 82 miRNAs) and decreased over time (Figure 1F). Differences were also found at later time points, although numbers declined (35, 34 and 16 miRNAs at the 8 h, 24 h and 2-week time-points, respectively). At the 4 h time-point, 22 and 5 brain/brain cell type-enriched miRNA were higher or lower, respectively, in SE samples compared to control (Figure 1G). Among those showing higher levels were 7 neuron-enriched (e.g., miR-132–5p), 5 astrocyte-enriched (e.g., miR-31-5p) and 1 microglia-enriched (miR-19a-3p). Neuron-enriched miRNAs, including miR-139-5p and miR-135-5p were also among those at lower levels 4 h after SE. At the 8 h time-point, 19 and 3 brain-enriched miRNAs were higher or lower, respectively (Figure 1G). The proportion of the elevated miRNAs that were of likely neuronal origin all increased (e.g., miR-132-3p and miR-218-5p). Brain-enriched miRNAs for which cell-type-enrichment was uncertain were also detected at increased levels compared to controls. By 24 h, a more even mix of higher and lower miRNA levels was observed (Figure 1G). A high proportion of those at higher levels were neuronal (e.g., miR-124-3p and miR-434-3p) whereas several glial-enriched miRNAs were among those at lower levels (e.g., miR-31-3p and miR-29b-3p). At the two-week time-point, a similar ratio of higher and lower miRNAs were observed between SE and control. Again, a majority of those increased being of likely neuronal origin (e.g., miR-669a-5p and miR-541-5p) (Figure 1G). These findings suggest SE adjusts the abundance and cell type- contribution of the circulating pool of Ago2-bound miRNAs in a time-dependent manner in the plasma of mice.

To validate the RNA sequencing findings, we measured a sub-set of miRNAs in additional plasma samples, either from individual animals or a pool of samples, using individual Taqman miRNA assays. Here, we focused on miRNAs that were likely brain-enriched and potential biomarkers of TLE, selecting let-7i-5p (present data and (Lagos-Quintana et al., 2002)) and miR-19b-3p, which we previously identified as an Ago2-enriched, cerebrospinal fluid biomarker of TLE in humans (Raoof et al., 2017). This revealed upregulation of let-7i-5p and miR-19b-3p in Ago2 elutions from the plasma from mice at the two-week time point after SE compared to controls (Supplementary Figure S4).

To determine if some of the circulating miRNA biomarkers of epilepsy originate from the brain, and to establish whether any of these are from neurons, we generated mice that express a FLAG-tagged Ago2 protein under a neuron- or microglial-specific promoter. To achieve this, mice expressing a FLAG-tagged Ago2 with a loxP flanked stop codon under the control of the ubiquitously expressed ROSA26 promoter (Schaefer et al., 2010; Tan et al., 2013), were crossed with transgenic lines with a tamoxifen-inducible cre-recombinase driven by a neuronal promoter, Thy1 (Young et al., 2008), or the microglial Cx3cr1 promoter (Parkhurst et al., 2013) (Figure 2A). Attempts to generate an astrocyte-specific FLAG-Ago2 line using a cre-Gfap promoter were unsuccessful (data not shown).

Using western blotting, we first confirmed that tamoxifen treatment (+) of the resulting FLAG-Ago2Thy1 and FLAG-Ago2Cx3cr1 lines resulted in detectable FLAG at the expected molecular weight within the mouse brain (Figure 2B). In contrast, only the endogenous Ago2 was detected in cortical lysates from non-tamoxifen treated FLAG-Ago2Thy1 and FLAG-Ago2Cx3cr1 mice (Figure 2B). Levels of endogenous Ago2 appeared normal in the cortex of both FLAG-Ago2 mouse lines (Figure 2B). We also confirmed that the FLAG-tagged Ago2 from both lines was expressed in lysates from the brain of tamoxifen-treated mice but not a selection of other organs tested (Figure 2C).

We then used confocal microscopy to confirm that the FLAG-Ago2 was correctly restricted to either neurons (FLAG-Ago2Thy1+) or microglia (FLAG-Ago2Cx3cr1+). Hippocampal tissue sections from FLAG-Ago2Thy1+ mice were immunostained with antibodies against FLAG and neuronal nuclear protein (NeuN). Analysis of double-stained sections revealed expression of FLAG in a subset of neurons of various morphologies in the hippocampus (Figure 2D). This suggests that the FLAG-tagged Ago2 is likely expressed by both excitatory and inhibitory neurons, consistent with the expected labelling pattern using a Thy1 promoter (Young et al., 2008).

Next, we stained sections from FLAG-Ago2Cx3cr1+ mice with antibodies against FLAG and the microglial marker Iba1. Staining for Iba1 had the expected appearance of microglia, with a small cell body from which multiple diffuse processes extended, and there was extensive co-localisation of FLAG staining with the microglial marker (Figure 2E). These experiments confirm the two different mice lines express neuronal and microglial cell type-restricted expression of FLAG-Ago2. Furthermore, cellular patterns of staining for the endogenous Ago2 protein did not appear affected by the tamoxifen-induced expression of FLAG-Ago2 in either line (Figure 2F; Supplementary Figure S5).

Macroscopic inspection of the brains of the FLAG-Ago2Thy1+ and FLAG-Ago2Cx3cr1+ mice revealed normal morphology (data not shown). Nevertheless, we next assessed the severity and effects of SE in the transgenic mice. Seizures were induced by intra-amygdala kainic acid and recorded using skull-mounted EEG, comparing the two different tamoxifen-treated FLAG-Ago2 lines to untreated transgenic lines.

Intra-amygdala kainic acid triggered prolonged seizures in all mice, characterised by high amplitude, high frequency discharges (Figure 3A). Semi-quantitative analysis of EEGs revealed there were no significant differences in seizure severity between the different mouse lines (Figure 3B). FLAG-Ago2 mice also displayed similar EEG power to controls across delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz) and beta (12–30 Hz) frequency band (Figure 3C). To extend these findings, we analysed neuronal death in the ipsilateral hippocampus 24 h after SE in a subset of the mice. In wildtype animals, seizure-induced damage was observed within the ipsilateral CA3 subfield as expected (Mouri et al., 2008), whereas the bordering seizure-resistant CA2 subfield was spared (Figure 3D). Seizure-induced damage was similar for both FLAG-Ago2Thy1+ and FLAG-Ago2Cx3cr1+ mice and their untreated littermates subject to SE (Figure 3D). Together these data indicate that expression of the FLAG-Ago2 does not significantly alter the electrographic and pathologic features of the intra-amygdala kainic acid model.

We next sought evidence that SE causes blood–brain barrier disruption in this model sufficient to allow passage of a large protein such as Ago2 to pass between blood and brain. For this, we injected wildtype mice with a biotin tracer, 2 h after injection of kainic acid, a time-point when the blood–brain barrier had previously been found to be maximally disrupted in this model (Reschke et al., 2021) (Figure 3E). Staining tissue sections from mice injected with the tracer revealed staining within the CA3 subfield, the major site of damage in the model, but not in control mice (Figure 3E). Similar results were obtained from two-week post-SE mice (Figure 3E). This indicates that changes to the integrity of the blood brain barrier in epilepsy are likely sufficient to allow direct passage of a protein the size of Ago2 into the blood.

Last, we sought proof-of-principle that miRNAs from the brain can be detected associated with the FLAG-Ago2 in the plasma of mice after seizures. We subjected FLAG-Ago2Thy1+ and FLAG-Ago2Cx3cr1+ mice to prolonged seizures induced by kainic acid and then immunoprecipitated FLAG from plasma collected 2 weeks later (Figure 4A). RNA was extracted from the eluted FLAG and subjected to miRNA assays for either let-7i-5p or miR-19b-3p (Figures 4B–E). We observed significantly elevated levels of Flag-Ago2-bound miR-19b-3p in both the FLAG-Ago2Thy1+ and FLAG-Ago2Cx3cr1+ mice which experienced SE compared to those which did not. Let7i-5p was elevated in FLAG-eluates from the FLAG-Ago2Cx3cr1+ which experienced SE but not the FLAG-Ago2Thy1+ mice (Figures 4B–E). These findings indicate that a portion of the circulating pool of miRNA has originated from the brain, including a contribution from both microglia and neurons.

All animal experiments were performed in accordance with the European Union Directive (2010/63/EU) and were reviewed by the HPRA (19,127/P001) and the Research Ethics Committee, RCSI (REC 842). Adult male C57BL/6JOlaHsd mice were bred in house in the Biomedical Research Facility (BRF) in RCSI unless otherwise indicated. All animals were housed 2-5/cage in the climate-controlled environment on a 12 h light/dark cycle. Food and water were provided ab libitum.

EEG traces were analysed using custom MATLAB (vR2022a) scripts. Data were converted to the frequency domain by fast Fourier transform (FFT size 4,096) and the power of individual frequency bands extracted. Power in each band was measured from artefact-free periods during baseline (before SE) and during SE. Normalised power was expressed as the ratio of power during SE:basline.

Heterozygous transgenic mice expressing Rosa-Stop^fl/fl-Flag-Ago2 were kindly provided by Anne Schaefer, Icahn School of Medicine, NYC (Schaefer et al., 2010; Tan et al., 2013). Homozygous FLAG-tagged Ago2 mice were crossed with inducible cre-recombinase mouse lines driven by brain cell type restricted promoters. Upon treatment with the inducing agent, tamoxifen, neuron (Thy1, Stock 012708) and microglia (Cx3cr1, Stock 021160) restricted expression of FLAG-tagged Ago2 resulted. New mouse lines produced as part of this study have genotypes Thy1-Cre^tg/+; Rosa-Stop^fl/fl-Flag-Ago2 (FLAG-Ago2Thy1) and Cx3cr1-cre^tg/tg; Rosa-Stop^fl/fl-Flag-Ago2 (FLAG-Ago2Cx3cr1). Tamoxifen (0.1 mL of 10 mg/mL solution, Sigma-Aldrich, Ireland) was administered via intraperitoneal injection as follows; once a day for five days, two day rest, then once a day for five days.

SE was induced as previously described (Reschke et al., 2021) with some minor modifications. Briefly, adult male mice were anesthetized. A guide cannula and three surface EEG electrodes were implanted. Kainic acid (0.3 μg) was injected into the ipsilateral amygdala in awake mice to trigger SE. To reduce mortality and morbidity, midazolam or lorazepam (8 mg/kg) was administered after 40 min (Diviney et al., 2015). Blood was collected from mice using the submandibular bleed technique (Brennan et al., 2020), into a 0.5 mL tube coated with K2 EDTA (0.5 M, Sigma-Aldrich, Ireland) at various time points following SE. Samples were centrifuged at 1300 x g for 10 min at 4°C to separate the plasma from blood cells. The plasma supernatant was removed and placed in a fresh tube and immediately placed on ice. Haemolysis levels were evaluated using a Nanodrop 2000 spectrophotometer and only samples with a reading of <0.25Abs at 414 nm were used (Brennan et al., 2020).

Ago2 protein was extracted as described previously (Raoof et al., 2018). Briefly, pre-cleared plasma was incubated with Ago2 antibody (C34C6; Cell Signalling Technology) overnight at 4°C. Protein A/G beads were added, and samples were incubated at 4°C for 2 h. Bead-Ago2 complexes were pelleted by centrifugation, washed and RNA was extracted using the phenol-chloroform technique. 50 μL of plasma from 10 mice were pooled (final volume 500 μL) and the Ago2-bound fraction was extracted as outlined above. The FLAG-tagged Ago2 protein was precipitated from transgenic mouse plasma using the technique outlined above except the pre-cleared plasma was incubated with 10 μL of the Anti-FLAG M2 Magnetic Beads (Cat#: M8823 Sigma-Aldrich, Ireland) overnight at 4°C.

The time-points for Ago2 sequencing were 4 h, 8 h, 24 h and 2 weeks after SE. Each SE sample was compared to a plasma pool collected from a control (intraamygdala PBS) mouse from the sample time point (n = 10/each). An exception was the 2 week time-point where the control sample failed to generate a library that passed QC and therefore the 2 week SE sample was compared to the 24 h control. Small RNA libraries from plasma were prepared using the Illumina TruSeq Small RNA Library Prep Kit (Illumina, UK) with reagent volumes halved to account for lower concentrations of miRNA to prevent excessive formation of adapter dimers. Small RNA libraries were separated by electrophoresis on Pippin Prep (Sage Science, US) using a 3% agarose gel cassette (Lab Tech, UK), selecting for RNA in the size range of 90–250 bp (adapted-ligated miRNA). Libraries were sequenced on an Illumina HiSeq. The sequencing data have been deposited to the Gene Expression Omnibus (GSE233890).

FastQ files were downloaded from the Illumina BaseSpace website.¹ Chimera software was used to trim adapters, size-select miRNA reads and to align identified reads to the reference genome of interest. Reads are aligned to miRBase using BLASTn. This software allows 2-mismatched nucleotides. Downstream analyses to identify differentially expressed miRNA were performed using edgeR and Limma packages from Bioconductor. Graphics were produced using the glimma package. For mouse analysis, based on the fact that the mice are genetically identical, a square root value (BCV) of 0.1 was set to allow an estimated p value based on dispersion values.

An extensive literature search was performed to identify reported enrichment of miRNA identified by RNAseq. The entire list of miRNAs that mapped to the mouse miRBase V21 database was compared against those reportedly enriched in different organs and brain regions (Lagos-Quintana et al., 2002) and enriched in different brain cell types in vitro (Jovicic et al., 2013) and in vivo (Butovsky et al., 2014). miRNAs were searched in PubMed to identify in situ hybridization studies (Schratt et al., 2006; Thompson et al., 2007; Cheng et al., 2009; Huse et al., 2009; Pena et al., 2009; Aronica et al., 2010; Nelson et al., 2010; Iyer et al., 2012; Ouyang et al., 2013; Kaalund et al., 2014; Van Scheppingen et al., 2016) demonstrating colocalization of miRNAs with known brain cell type markers and using techniques such as miRAP (He et al., 2012).

Brain cell type of origin was investigated using Taqman small-scale miRNA assays let7i-5p (002221) and miR-19b-3p (000396) (Applied Biosystems) as described (Heiland et al., 2023). Plasma from two male animals of the same genotype were pooled per sample and all amplifications were performed in triplicate in a 96-well plate and the comparative Ct values were measured on the 12 K-Flex. For validation of the pooled RNA-seq from individual samples ~100 μL of plasma collected from individual animals or combined from 2 animals. Samples were transferred to a new microcentrifuge tube and 20 μL of the sample was set aside for RNA isolation and use as input for normalisation. On the remaining 80 μL Ago2 immunoprecipitation was performed as described above, RNA was then isolated from immunoprecipitates and subjected to targeted cDNA synthesis and Taqman assay on the 12 K-Flex. The 2-DDCt method was used for relative quantification.

Tissue samples were homogenised in 500 μL protein lysis buffer (0.1 M NaCl, 20 mM Tris–HCl, pH7.6, 1 mM EDTA, pH8.0, 1% NP-40). Samples were centrifuged to pellet any cellular debris. FLAG-Ago2 was immunoprecipitated from 800 μg of tissue lysate using an Anti-FLAG M2 coated 96-well plates (P2983, Sigma-Aldrich, Ireland) as per the manufacturer’s guidelines. Lysates were removed and protein antibody-conjugates linked to the bottom of the plate were washed three times with 300 μL of ELISA wash buffer (100 mM Tris buffered saline, pH8.2, 2 mM MgCl₂, 0.5% Tween-20 – all Sigma-Aldrich, Ireland). Protein was eluted by incubating wells in 2 X Laemelli buffer for 10 min, shaking at 900 rpm. FLAG-Ago2 immunoprecipitates and 30 μg of whole brain lysate were analysed using standard western blotting techniques using antibodies against FLAG and Ago2.

Paraformaldehyde-fixed, free-floating sections were blocked in 10% normal goat serum, before incubation in antibodies against FLAG, NeuN and Iba1 overnight at 4°C. Sections were washed and were washed and incubated in suitable secondary in suitable secondary antibody (Alexa Fluor 488 or Alexa Fluor 568) diluted in PBS containing 0.1% Triton-X 100 for 2 h at room temperature. Sections were investigated under Leica LSM confocal microscope at 40X magnification.

For the endogenous Ago2 staining, mouse brain slices were cryosectioned at a thickness of 12 μm and mounted on slides and formalin fixed. Sections were blocked in 1% BSA for 1.5 h followed by incubation overnight with the primary antibody against Ago2 (Cell Signaling; 1:100). On the next day, slices were washed and then incubated with fluorescent secondary antibody (Invitrogen, Ireland) for 2 h. Then, sections were rinsed again, stained with DAPI and mounted and imaged using a Leica DM4000 epifluorescence microscope.

Fluorojade B staining was performed to assess SE-induced neuronal death (Mouri et al., 2008). Briefly, brain slices were cryosectioned at a thickness of 12 μm and tissue was fixed in formalin. Slices were passed through graded ethanol, incubated in a 0.006% potassium permanganate and finally incubated in 0.001% Fluorojade B. Neuronal damage was assessed by manual counting of Fluorojade B positive cells in the CA1, CA2, CA3 and dentate gyrus hippocampal regions under a Leica DM4000 epifluorescence microscope.

BBB disruption was assessed by systemically injecting Biotin dextran amine (BDA, 70 kDa) following SE and staining for the presence of BDA in the mouse brain. For the acute experiments, BDA at a concentration of 90 mg/kg was administered via intraperitonial injection 2 h following KA injection. Then, 72 h following BDA injection, mice were PBS perfused and brains were snap frozen. To confirm disruption to the BBB in the chronic phase, BDA was injected 11 days after SE and then brains collected 72 h later. Mouse brain slices were cryosectioned at a thickness of 12 μm and mounted on slides and formalin fixed. Staining was performed as above, however streptavidin-AF594 (1:500) was added to the secondary antibody and incubated 2 h at room temperature. Sections were investigated under a Leica DM4000 epifluorescence microscope.