Patients were retrospectively included between May 2021 and September 2023 from outpatient or inpatient presentations in accordance with the inclusion and exclusion criteria at the Department of Neurology, Kepler University Hospital, Linz. Four patient groups were predefined. The first group included patients who had been diagnosed with PSE. The second group included patients with ischemic stroke without the development of epileptic seizures within at least 2 years after stroke. The third group included patients with epilepsy without seizures in the last 2 weeks, and the fourth group was healthy controls. Patients aged between 18 and 99 years were included. Patients with severe comorbidity that can significantly influence the clinical presentation, such as known diseases from the spectrum of autoimmune diseases, patients with immunosuppressive medication, current drug abuse associated with epileptic seizures, previous structural gliotic cerebral lesions with cortical involvement visible on imaging other than the stroke lesions, vasculitis, previously known moderate to advanced dementia [Mini Mental State Examination (MMSE) ≤15 points] and malignant neoplasia were excluded. Furthermore, patients with pre-existing epilepsy before ischemia were excluded from the stroke and PSE group.

Patients were split into two cohorts. A total of 24 patients were used for miRNA-sequencing (epilepsy n = 9, stroke n = 7, PSE n = 8), and differentially expressed miRNAs were initially validated with these samples. To further validate our findings in a bigger cohort, we used the rest of the samples (epilepsy: n = 6, stroke: n = 8, PSE: n = 9, healthy controls: n = 5) to repeat the analysis of differentially expressed miRNAs and ELISA analysis. Details of which patient was allocated to which cohort can be found in Supplementary Table 1.

This study was approved by the Ethics Committee of the Johannes Kepler University, Linz, Austria (EK1066/2020; EK1151/2019, 1150/2019). All patients gave written informed consent.

Clinical data were collected during the clinical examination, from medical history, and from existing hospital documentation. The currently existing modified rating scale (mRS) is determined for patients in all groups on the basis of a clinical examination.

The age, gender, and mRS of all patients were recorded during check-ups. In the PSE group, as in the stroke group, the acute treatment of the ischemia, the mRS, the National Institutes of Health Stroke Scale (NIHSS) on admission and discharge, and the localization of the ischemia were recorded. In the PSE group, the duration from ischemia to the first epileptic seizure, as well as the seizure semiology and number of needed antiseizure medications, was recorded. In four patients, the interval could not be precisely determined, as the brain infarction occurred abroad (according to the patient, the seizure occurred within 1 year) or the patients were mute. The epilepsy diagnosis, with the localization of the suspected epileptogenic focus and the seizure semiology were recorded for epilepsy and PSE patients. The number of needed antiseizure medications was also included.

For serum collection, blood was drawn from each patient and left at room temperature for 30 min before centrifugation at 1,000 × g for 15 min at 4 °C. Serum was then aliquoted in 500 μL and stored at −80 °C.

To identify samples with a high level of hemolysis, we quantified the levels of hemoglobin in serum with NanodropOne by measuring absorbance at 414 nm. We used the custom plug-in provided by the company (Thermo Fisher; MA, United States) to directly calculate the concentration of hemoglobin. We excluded all samples with a calculated concentration of hemoglobin above 0.1 mg/mL, which is considered to be the cutoff for non-hemolyzed samples (Supplementary Figure 1A) (22).

miRNA was isolated from 200 μL of serum with the miRNeasy Serum/Plasma Advanced kit from Qiagen (Hilden, Germany), following the manufacturer’s instructions, and eluted in 16 μL of water.

To evaluate the successful isolation of miRNAs from serum, we used the Bioanalyzer RNA 6000 Pico Kit (Agilent, CA, United States) according to the manufacturer’s instructions. A peak between 25 and 200 nt was assumed to correspond to small RNA species (Supplementary Figure 1B).

After miRNA isolation, we used qPCR for highly sensitive detection of hemolysis in the serum samples. To this end, we used 1.14 μL of eluate from the isolated RNA (equivalent to 16 μL of serum) and performed reverse transcription with the miRCURY LNA RT Kit from Qiagen (Hilden, Germany) according to the manufacturer’s instructions. For qPCR, cDNA was diluted 1:60, and the miRCURY LNA SYBR Green Kit (Qiagen, Hilden, Germany) was used with primers targeted against miR-23a-3p and miR-451a (miRNA qPCR Assay from Qiagen, Hilden, Germany). qPCR was performed on a BioRad CFX96 cycler under the following conditions: 95 °C for 2 min once, 95 °C for 10 s and 56 °C for 60 s, cycled 40 times in total. For the quantification of hemolysis, the difference between the two miRNAs was calculated. A difference below five is considered low hemolysis, between five and seven moderate, and above seven high hemolysis (23) (Supplementary Figure 1C).

For library prep, the TruSeq small RNA Library Prep Kit from Illumina (CA, United States) was used. We used 5 μL of isolated miRNA as input for library prep and followed the manufacturer’s instructions. As the miRNA concentration in biofluids is usually low, we performed 15 PCR cycles for library amplification and added the optional library concentration step after gel purification. 2 nM library was used for sequencing.

All samples were sequenced together on an Illumina NextSeq 2000 device on a single NextSeq 1000/2000 P1 XLEAP-SBS Flowcell (100 Cycles) (# 20100983). The samples were loaded with a concentration of 650 pM and denatured on board. The following cycle settings were used: 6 cycles index read and 36 cycles insert read.

The basecall was performed by the DRAGEN BCL Convert v4.2.7 pipeline of the DRAGEN server, locally integrated in the Illumina NextSeq 2000 device. The following analysis steps were then performed using CLC Genomics Workbench v24.0.2 (QIAGEN): trimming of reads by Phred score <13, trimming by ambiguous nucleotides >2, trimming by length <8, adapter trimming, quantification of miRNAs by aligning to miRBase v22 (Homo sapiens), and computation of differential expressions between the samples and sample groups with TMM normalization and FDR correction-filtering. This way, between 27,517 and 1,108,725 reads per sample could be associated with miRNAs described in miRNABase v22. The quality of sequencing results was evaluated with mirnaQC (24). A volcano plot of differentially expressed miRNAs (±1.5-fold change, p-value <0.05) was created with VolcanoseR for better graphical presentation of sequencing results but not used for selection of differentially expressed miRNAs (25). A Venn diagram was created with BioVenn (26). A heat map and hierarchical clustering of z-scores of selected miRNA expression were created with SRplot (27). For plotting single miRNA sequencing results, the read count of each miRNA of interest was normalized to the total read count of each sample.

To validate differentially expressed miRNAs by qPCR, we used a custom-designed miRCURY LNA PCR panel, which included two spike-in controls (UniSp6 and UniSp3) and SNORD68 (Mus musculus) as a negative control. As there is no clear consensus on reference genes for miRNAs in biofluids, we selected several potential reference miRNAs based on literature and sequencing data (miR-93-5p, miR-23a-3p, and U6snRNA) (28–31). miRNAs of interest were selected based on miRNA-seq data (miR-486-5p, miR-182-5p, miR-10b-5p, miR-10a-5p, 192-5p, miR-92-3p, miR-25-3p, miR-148-3p, miR-186-5p, and miR-145-5p). We used the previously generated cDNA and diluted it 1:40 according to the manufacturer’s protocol. miRCURY LNA SYBR Green Kit (Qiagen, Hilden, Germany) was used for qPCR. qPCRs were performed on a BioRad CFX96 cycler with the following conditions: 95 °C for 2 min once, followed by 40 cycles of 95 °C for 10 s and 56 °C for 60 s, followed by a standard melting curve analysis. All data were analyzed with Gene Globe (Qiagen, Hilden, Germany). We used the global mean to normalize Cq values of all samples.

We performed in silico analyses of miRNA targets. To this end, we investigated targets of miR-10b-5p based on three different tools: miRDB (32), TargetScan (33), and miRTargetLink 2.0 (34). We found that BDNF was one of the main targets of miR-10b-5p. In the next step, we investigated which of the differentially expressed miRNAs modulated BDNF expression in each group and manually counted them.

We performed an ELISA analysis of all serum samples for BDNF (BioLegend Cat. No 446604). Samples were diluted 1:10 or 1:20 for measurement. Absorbance was measured at 450 nm with a background correction at 570 nm with a standard plate reader (Tecan Sunrise). Frozen human brain tissue was used as a positive control, and cell culture medium from fibroblasts and keratinocytes as a negative control. Analysis was performed using a four-parameter logistic curve.

Statistical analysis was performed with GraphPad Prism (v.8). Analysis for the difference between the patient cohorts of clinical patient data was performed with one-way ANOVA for age and disease duration. For differences in scores (mRS and NIHSS), the Mann–Whitney U-test was performed. For the analysis of differential expression of normalized read counts, the Kruskal–Wallis test with Dunn’s correction for multiple testing was used. For qPCR results and ELISA, statistical testing was performed using one-way ANOVA with Holm–Šídák’s correction for multiple testing. To identify the required sample size (alpha value: 0.05, power: 0.95) for a validation cohort, we performed a power calculation between PSE and stroke based on the initial qPCR data and determined the required samples to be 13 per group. Correlation analyses were performed with Pearson’s correlation for all analyses except for the correlation between BDNF levels and mRS and NIHSS, where a Spearman’s correlation was used.

A total of 53 patients were included in the analysis, with a median age of 61.5 years (IQR 45.0–72.8), and 32% of the cohort were female. These patients were categorized into four groups: 16 with epilepsy, 15 with stroke without subsequent epilepsy, 17 with PSE, and 5 healthy controls. In the epilepsy group, the median age was 35.5 (IQR 24.17–59.03), in the stroke group, 73.0 (IQR 63.50–76.80), and in the PSE group, 61.55 (IQR 57.40–76.80), and in healthy controls, 51.24 (37.62–67.47, Table 1, more detailed information in Supplementary Table 1).

In the epilepsy group, 12 patients had focal epilepsy, of whom 8 also had secondary generalized seizures, and 4 patients had generalized seizures. One patient received triple antiseizure medication (ASM), 14 patients received monotherapy, and 1 patient no longer required ASM after years of seizure freedom (more details on seizure semiology and therapy in Supplementary Table 1).

The stroke group presented with a median mRS of 3.00 (IQR 1–5) and NIHSS of 3.0 (IQR 2.0–10). At discharge, functional outcomes had improved, with a median mRS of 1 (IQR 0.0–1.0) and a median NIHSS of 1 (IQR 0.0–2.0), reflecting largely favorable recovery (Table 1). Detailed information on lesion location and therapeutic measures for ischemia can be found in Supplementary Table 1.

The PSE group exhibited similar disability and neurological deficit scores both at admission and discharge. The mRS at admission was 5 (IQR 3–5), and the median NIHSS score was 13 (IQR 7–17), indicating moderate to severe neurological impairment. In contrast to the stroke group, the PSE group had a significantly worse functional status at discharge, with a median mRS of 4.0 (IQR 2.0–5.0), and the NIHSS decreased to 9 (IQR 1.0–11.0). The median latency between stroke and the first unprovoked seizure was 7.34 months (IQR 3.2–19.4). All patients had focal seizures, and three also experienced secondary generalized seizures. A total of 15 patients had a single ASM, 1 patient had dual therapy, and 1 was untreated after years of seizure freedom. Detailed information on lesion location, therapeutic measures for ischemia, and ASM can be found in Supplementary Table 1.

We were able to isolate small RNA species from serum samples, as indicated by Bioanalyzer profiles showing a peak between 25 and 200 nt long RNA fragments (representative profile in Supplementary Figure 1B). As hemolysis can drastically alter the miRNA profile in serum samples, we measured the concentration of hemoglobin in all samples and used 0.1 mg/mL of hemoglobin as a cutoff (22). Our samples were all below the cutoff value (median epilepsy: 0.04, stroke: 0.03, PSE: 0.02; Supplementary Figure 1A). As a more sensitive measure for hemolysis, we also performed qPCR for miR-23a and miR-451a. Our samples were all below 7 (median epilepsy: 5.11, stroke: 4.79, PSE: 3.73; Supplementary Figure 1C) (23).

Of the 24 samples sequenced, two did not pass the quality control. One sample had a raw read count of 7.14 *10⁵, and in one sample only 42% could be mapped to the reference genome (Supplementary Figures 1D,E). Excluding those two samples, the median number of raw reads per sample was 3.79 × 10⁶. The median number of detected miRNAs was 254.5, with an average length of 21.3 nt. A mean of 72.90% of the reads was identified as miRNAs, and 76.60% could be mapped to the reference genome. The average PhredScore was 29.6.

A total of 41 miRNAs were differentially expressed between stroke and epilepsy (Figures 1A,B), 30 miRNAs between epilepsy and PSE (Figures 1A,C), and 23 miRNAs between stroke and PSE (±2-fold change and a p-value of ≥0.05; Figures 1A,D). However, we only considered miRNAs with a ±2-fold change and an FDR value smaller than 0.05. When applying these stricter filtering criteria, only eight miRNAs were differently expressed among the three groups (annotated in Figures 1B–D). Between stroke and epilepsy, a total of five miRNAs were differentially expressed (miR-101-3p, miR-99-5p, miR-3651, miR-194-5p, and miR-340-5p; Figure 1B). In the comparison of PSE and epilepsy, only miR-486-5p was downregulated in PSE (Figure 1C). When comparing stroke with PSE, miR-10b-5p and miR-182-5p were upregulated in stroke (Figure 1D and Table 2; Supplementary Table 2). Based on the differentially expressed miRNAs, we performed hierarchical clustering of the samples and could show that most samples cluster well within each group, except for two PSE and one epilepsy case. Moreover, PSE samples cluster more closely to epilepsy samples rather than stroke samples, indicating a higher degree of similarity (Figure 2A). In a K-Medoids analysis, we found several miRNAs distinguishing the stroke and the PSE group, most prominently miR-10b-5p (Figure 2B). We then analyzed the miRNAs significantly changed in PSE individually. miR-486-5p was significantly upregulated in epilepsy compared to PSE (p-value = 0.0015), with no significant difference between stroke and PSE or epilepsy (p-value = 0.13 and 0.7; Figure 2C). miR-182-5p was significantly decreased in PSE compared to stroke (p-value = 0.008), with no significant difference between epilepsy and stroke (p-value >0.99) and a p-value of 0.05 between epilepsy and PSE (Figure 2D). miR-10b-5p was significantly increased in stroke and in epilepsy compared to PSE (epilepsy vs. PSE p-value = 0.04, stroke vs. PSE p-value = 0.02) with no significant difference between epilepsy and stroke (p-value >0.99; Figure 2E).

To validate the results from miRNA sequencing, we performed qPCR for differentially expressed miRNAs in cohort 1 (sequencing cohort). We found that miR-486-5p is significantly increased in epilepsy compared to PSE (p-value = 0.03), with no differences between PSE and stroke (p-value = 0.43) or epilepsy and stroke (p-value = 0.36), confirming results from sequencing (Figure 3A). We were not able to validate results for miR-182-5p and found no significant difference between any of the three groups (PSE vs. epilepsy: p-value = 0.8, PSE vs. stroke: p-value = 0.62, epilepsy vs. stroke: p-value = 0.24; Figure 3B). However, we could validate miR-10b-5p as a main candidate to differentiate stroke patients from PSE patients, as also by qPCR, miR-10b-5p was significantly decreased in PSE compared to stroke (p-value = 0.013). There was no significant difference between PSE and epilepsy (p-value = 0.6) or epilepsy and stroke (p-value = 0.06; Figure 3C). To rule out differences in miR-10b-5p expression due to the differences in age between the three groups, we calculated the correlation between the age and dCt values for miR-10b-5p and found no significant correlation (R² = 0.01, p = 0.59, data are not shown).

As a next step, we pooled all patients from cohorts 1 and 2 to validate the differential expression of miR-10-p between stroke and PSE in a larger cohort by qPCR. We were able to show the differential expression of miR-10b-5p between PSE and stroke (p-value = 0.015), whereas we did not find a difference between PSE and healthy controls (p-value = 0.53, Figure 3D).

In the in silico analyses, we found that miR-10b-5p targets BDNF. Additionally, we found that of the 23 initially differentially expressed miRNAs between stroke and PSE, 6 (26%) bind to BDNF mRNA, namely miR-182-5p, miR-10a-5p, miR-192-5p, miR-16-5p, and let-7b-3p. In contrast, 4 of 30 (13%) of differentially expressed miRNAs were found to interact with BDNF in the comparison between PSE and epilepsy (miR-182-5p, miR-381-3p, miR-10b-5p, and miR-26b-3p), and only 3 of 41 (7%) in the comparison between stroke and epilepsy (miR-210-3p, miR-323a-3p, and miR-103a-3p; Supplementary Table 2).

We did not find a difference between any of the four groups in the ELISA analysis for BDNF (cohorts 1 and 2 pooled, Figure 3E, healthy control vs. epilepsy p = 0.98, healthy control vs. stroke p = 0.99, healthy control vs. PSE p = 0.99, PSE vs. epilepsy p = 0.82, PSE vs. stroke p = 0.99, epilepsy vs. stroke p = 0.83). However, we correlated BDNF levels with clinical parameters, namely age, NIHSS, mRS, disease duration, and time until the first seizure. We could show that in PSE, there was a significant direct correlation between BDNF levels and the disease duration (Figure 3F, R² = 0.56, p = 0.002). However, this correlation was specific to PSE and could neither be found in epilepsy (R² = 0.001, p = 0.9) nor in stroke (R² = 0.13, p = 0.19). Furthermore, we found a correlation between BDNF levels and the time until the first seizure occurred after stroke (Figure 3G, R² = 0.4, p = 0.02). Neither NIHSS nor mRS were correlated with BDNF levels, neither at admission nor at discharge for stroke (NIHSS admission: R = 0.26, p = 0.35; NIHSS discharge: R = 0.009, p = 0.97; mRS admission: R = 0.1, p = 0.72; mRS discharge: R = −0.2, p = 0.48) nor for PSE (NIHSS admission: R = −0.33, p = 0.19; NIHSS discharge: R = −0.26, p = 0.32; mRS admission: R = −0.35, p = 0.17; mRS discharge: R = −0.28, p = 0.28, Supplementary Figures 2B,C). We found no significant correlation between BDNF levels and age (Supplementary Figure 2A).