All experimental procedures were approved by the Committee on Animal Health and Care of Upper Bavaria (Regierung von Oberbayern), Germany (approval ID-AZ: 55.2-1-54-2531-41-09) and were conducted according to the current regulations for animal experimentation in Germany and the European Union (European Communities Council Directive 86/609/EEC). Twenty-three days old male C57BL/6NCrl mice purchased from Charles River GmbH (Sulzfeld, Germany) were housed in groups in the animal facility of the Max Planck Institute (MPI-P) for 6 weeks under an inverse 12:12 h light-dark cycle (lights off: 09:00 a.m.) with food and water ad libitum.

Experiments were performed during the activity phase of the mice, i.e., between 9:30 a.m. and 6:00 p.m., employing our established PTSD mouse model which we described in detail previously (28, 29). Briefly, 10-week-old male C57BL/6NCrl mice were subjected to a single 1.5 mA electric footshock for 2 s or mock treatment (exposure to shock chamber, the latter is termed “shock context” or “shock chamber” in the following). Beginning the day after footshock or mock treatment, half of the footshocked and half of the mock-treated mice received oral fluoxetine treatment (n = 6 per group). Thus, we compared four groups of mice, i.e., footshocked and mock-treated mice, which were either fluoxetine (Ratiopharm, Ulm, Germany) or vehicle-treated; these groups are termed “no-shock-vehicle,” “no-shock-fluoxetine,” “shock-vehicle,” and “shock-fluoxetine” in the following. Fluoxetine was administered in drinking water in a dose of 20 mg/kg/day for 28 days. The control group received drinking water only. On day 28 after footshock or mock treatment, fluoxetine efficacy was assessed by evaluation of their generalized fear response for 60 s during the presentation of a neutral tone (80 dB, 9 kHz) in a neutral context. Subsequently, the dose of fluoxetine was halved (10 mg/kg/day) and treatment was further continued for 3 days until discontinuation on day 31. Then, 59–60 days after footshock or mock treatment, hyperarousal was assessed by evaluation of their acoustic startle response. In addition, their generalized fear response was analyzed by monitoring their freezing behavior upon subsequent exposure to a neutral experimental context and to an experimental context similar to the shock chamber. Finally, the conditioned fear response of the mice was assessed by measuring their freezing behavior during (re-)exposure to the shock chamber. Video-taped animal behavior was rated off-line by a trained observer who was blind to the experimental conditions. Statistical analysis of behavioral data was performed using two-way ANOVA and Bonferroni post hoc tests.

Seventy four days after footshock or mock treatment, mice were sacrificed by cervical dislocation and PFCs were dissected (n = 6 per group). Total RNA was extracted employing the TRIzol^® protocol following the manufacturer’s instructions (Invitrogen, Paisley, UK). Extracted total RNA was resolved in nuclease free water. Concentrations of total RNA were assessed spectrophotometrically with a Nanophotometer (Implen GmbH, Munich, Germany). RNA integrity was assured by Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA) both in our laboratory and in the laboratory of the microarray service provider (Exiqon A/S, Vedbaek, Denmark). RNA integrity numbers were ≥8.90 throughout the samples and all samples exhibited clear 18S and 28S RNA peaks in Bioanalyzer profiles.

RNA samples (6 × 4 = 24 samples) were shipped from the MPI-P in Munich to the microarray service provider Exiqon (Exiqon A/S, Vedbaek, Denmark) where all miRNA microarray experiments were performed. Accordingly, the chapter at hand (description of miRNA microarray procedure) is based on information provided by Exiqon (Exiqon A/S, Vedbaek, Denmark): 600 ng of total RNA extracted from samples were labeled with fluorescent Hy3™ and 600 ng of total RNA from reference probe with fluorescent Hy5™ using the miRCURY LNA™ miRNA Hi-Power Labeling Kit (Exiqon A/S, Vedbaek, Denmark) according to the manufacturer’s protocol. The Hy3™ -labeled samples were mixed pairwise with a Hy5™ -labeled reference probe and hybridized to the miRCURY LNA™ miRNA Array 7th Gen (Exiqon A/S, Vedbaek, Denmark) which contains capture probes targeting all miRNAs registered in the miRBase 18.0 (human, mouse, or rat)¹ as well as viral miRNAs related to these species. Hybridization was performed according to the miRCURY LNA™ miRNA Array Instruction manual (Exiqon A/S, Vedbaek, Denmark) using a Tecan HS4800™ hybridization station (Tecan Austria GmbH, Salzburg, Austria). After hybridization, microarray slides were scanned and stored in an ozone free environment (ozone below 2.0 ppb) in order to prevent potential bleaching of fluorescent labels. The miRCURY LNA™ miRNA Array slides were scanned using the Agilent G2565BA Microarray Scanner System (Agilent Technologies, Inc., USA). Image analysis was carried out with ImaGene^® 9 miRCURY LNA™ miRNA Array Analysis Software (Exiqon A/S, Vedbaek, Denmark).

Pre-processed microarray data was provided by Exiqon (Exiqon A/S, Vedbaek, Denmark). Accordingly, the description of microarray data processing is based on information provided by Exiqon: Signal intensity was the basis of data filtering. Background correction of quantified signals was performed via subtraction of the median global background from the median local background from the intensity of signals (Normexp with offset value 10) and resulted in the exclusion of two samples of the experimental group “no-shock-fluoxetine” (Figures 3B and 4). Normalization of data was performed with the global Lowess (locally weighted scatterplot smoothing) regression algorithm (37). All calculations have been performed using the software R/bioconductor employing mainly the limma package. Comparisons of miRNA expression values between experimental groups were performed using moderated t-statistics with standard errors moderated across genes, i.e., shrunk toward a common value, using a simple Bayesian model. This has the effect of borrowing information from the ensemble of genes to aid with inference about each individual gene (38). P-values were corrected for multiple testing by the Benjamini and Hochberg adjustment method to control for false positive results.

With the corrected p-values delivered by Exiqon (Exiqon A/S, Vedbaek, Denmark), we performed unsupervised hierarchical clustering analyses (HCA) in which we included the top 50 miRNA candidates with the lowest corrected p-values. HCA results are depicted in heatmaps which we generated by a web-based tool provided by the Los Alamos National Laboratory HIV sequence database². For HCA, the complete-linkage method together with the Euclidean distance measure was employed. Complete-linkage clustering (by Euclidean distance) between sample subsets is represented by dendrograms (Figures 2–4).

For reverse transcriptase quantitative polymerase chain reaction (RT-qPCR), which was performed at the MPI-P, we employed either pre-designed LNA™ PCR primer sets for miRCURY LNA™ Universal RT microRNA PCR or Custom LNA™ PCR primers (UniRT) (Exiqon A/S, Vedbaek, Denmark). A list of all primer sets and their respective target sequences used is provided in Table 1. We used the miRCURY LNA™ miRNA PCR system first strand synthesis kit for poly-adenylation (poly-A-tailing) and reverse transcription (input of total RNA: 100 ng) according to the manufacturer’s protocol (Exiqon A/S, Vedbaek, Denmark). Then, 1 μl of 1:80 diluted cDNA was amplified by RT-qPCR in 5 μl SYBR Green PCR master mix containing 0.25 mM of LNA™ miRNA specific primer sets (Table 1). The total reaction volume was 10 μl. RT-qPCR was performed on the LightCycler^® 480 instrument (Roche Diagnostics, Penzberg, Germany). Each sample was analyzed in duplicate in every run, i.e., for each miRNA candidate tested. Cycling conditions were as follows: denaturation step 95°C 5 min followed by 45 loops of a two-segment amplification step (95°C, 30 s, 62°C, 1 min). A standard curve was generated for each individual plate assay with 1:10, 1:100, and 1:1000 dilutions and PCR efficiencies were calculated. C_p values were obtained with the software provided by the manufacturer (Roche Diagnostics, Penzberg, Germany). MiRNA entities for normalization were selected via NormFinder analysis based on microarray data (39) and mmu-miR-100-5p was used for normalization. Relative expression was calculated by the ΔΔC_t method (40).

Materials and methods for miRNA target prediction and gene ontology (GO) analysis are described in detail in the Section “Results.”

To analyze the impact of traumatic stress on miRNA profiles in mouse PFC, we employed a well-established mouse model for PTSD that we (28, 29, 41–43) and also other research groups used, at least in slightly modified ways (44–46), for previous experiments. The electric footshock-elicited murine PTSD-like syndrome can be effectively counteracted by the orally administered SSRI antidepressant fluoxetine (28, 29) and, as we published recently, lasts at least until day 60 after shock application (28). First, we had to re-establish the behavioral syndrome-inducing effect of footshock and the relieving action of fluoxetine in the mouse cohort studied here: in contrast to our expectations (28, 29), and despite a significant effect of shock (F_1,20 shock = 5.696, p = 0.027), the relative increase of the generalized fear response of shocked mice in the neutral context was not statistically significant after Bonferroni correction on day 60 (Figure 1D) but at least on day 28 (Figure 1B, t = 6.461, p < 0.001). The results of the other behavioral experiments turned out as expected: on day 59/60 after their subjection to shock, in comparison to mock-treated control mice, shocked mice exhibited a significantly stronger generalized fear response (in a context similar to the shock chamber) (Figure 1E, t = 4.058, p < 0.01) as well as more pronounced acoustic startle (Figure 1C, t = 4.468, p < 0.001) and conditioned fear responses (Figure 1F, t = 3.609, p < 0.01). Moreover, fluoxetine treatment drastically reduced trauma-mediated behavioral changes (Figure 1B: t = 5.630, p < 0.001, Figure 1C: t = 3.939, p < 0.01, Figure 1E: t = 4.193, p < 0.001, Figure 1F: t = 3.505, p < 0.01).

The behavioral consequences of stress exposure make this mouse model an animal model of PTSD, not the type or intensity of the stressor applied. The relatively increased conditioned and generalized fear responses in footshocked mice mirror the PTSD-associated avoidance behavior in humans: In most PTSD patients, the aversive avoidance of trauma-related reminders generalizes over time in sense that someday also trauma-unrelated cues suffice to elicit an intense avoidance anxiety. Moreover, the relatively increased startle response in footshocked mice has been repeatedly described also in PTSD patients (47–49). Hence, it reflects trauma-elicited nervous hyperexcitability in both men and mice. Other PTSD animal models employ more intense stressors in order to better model their traumatizing nature (50, 51).

To avoid molecular analyses to be influenced by acute effects of the behavioral testing procedure, we harvested mouse brains 2 weeks after behavioral analyses. For preparation of total RNA and subsequent miRNA profile analyses, PFCs were dissected from six mice per group. With the aim to identify miRNA candidates regulated by traumatic stress and/or by fluoxetine treatment, we subjected all of these 24 PFC total RNA samples to miRNA microarray analysis. After background correction and normalization, expression values were subjected to pairwise t-testing (no-shock-vehicle versus shock-vehicle; shock-vehicle versus shock-fluoxetine; no-shock-fluoxetine versus shock-fluoxetine; no-shock-vehicle versus no-shock-fluoxetine) and the resulting p-values were Benjamini–Hochberg corrected. Then, miRNAs were ranked by corrected p-values and the resulting top 50 candidates, i.e., the miRNAs with the lowest p-values, were subjected to unsupervised HCA. We performed four HCAs in total (Figures 2–4).

First, we looked for miRNAs regulated by traumatic footshock: unsupervised HCA of footshocked versus non-shocked groups (both vehicle-treated) revealed that samples clustered, with one exception, according to treatment by their miRNA expression values (Figure 2). However, pairwise comparison of miRNA expression profiles showed that no miRNA was significantly differentially expressed between these two groups. Thus, traumatic footshock causes a long-lasting PTSD-like syndrome in mice (Figure 1) but does not significantly alter long-term miRNA expression in mouse PFC (Figure 2).

Then, we looked for the influence of fluoxetine treatment on miRNA profiles of shocked mice: unsupervised HCA of shocked fluoxetine-treated (shock-fluoxetine) versus shocked vehicle-treated (shock-vehicle) groups revealed that samples clustered perfectly according to treatment (Figure 3A). Moreover, comparison of these two groups, i.e., the shock-vehicle versus the shock-fluoxetine group, revealed, that therapeutic (Figure 1) fluoxetine treatment significantly reduced the relative expression of two miRNA candidates analyzed, namely of rno-miR-3559-3p [fold change (FC) 0.29, corrected p (corr. p) < 0.003] and of mmu-miR-1971 (FC 0.82, corr. p < 0.05) (Figure 3A, highlighted in bright pink) and furthermore decreased the expression of two other miRNAs [at least on the level of a trend toward statistical significance (i.e., with p < 0.1)], namely the expression levels of ebv-miR-BART8-3p (FC 0.53, corr. p < 0.06) and of mmu-miR-1947-3p (FC 0.67, corr. p < 0.06) (Figure 3A, highlighted in bright blue).

Next, to further dissect the individual contributions of traumatic stress and fluoxetine treatment on mouse PFC miRNA signatures, we compared the shock-fluoxetine group to the no-shock-fluoxetine group (Figure 3B) and to the no-shock-vehicle group (Figure 4). Two samples of the no-shock-fluoxetine group had to be excluded from microarray analysis during array data processing. HCA of the no-shock-fluoxetine group versus the shock-fluoxetine group showed that the two groups did not cluster correctly according to treatment. However, pairwise comparison of these two groups revealed that, the relative expression of mmu-miR-33-5p was enhanced, at least with a statistical trend (FC 1.26, corr. p < 0.07) and the relative expression of rno-miR-3559-3p was decreased, also with a statistical trend (FC 0.80, corr. p < 0.07) in PFC of shock-fluoxetine mice (Figure 3B, both highlighted in bright blue). Although the HCA of the no-shock-fluoxetine group versus the no-shock-vehicle group illustrates that, with one exception, the respective samples clustered according to treatment, pairwise comparisons of these two groups revealed no significant differences in miRNA profiles (Figure 4). Notably, mmu-miR-3559-3p emerged as a regulated miRNA candidate in two different pairwise comparisons since its expression was altered in shock-fluoxetine mice in comparison to both shock-vehicle (Figure 3A) and no-shock-fluoxetine mice (Figure 3B).

Taken together, microarray analyses revealed that, in shocked mice, on day 74 after subjection of mice to footshock, the therapeutic effect of fluoxetine (Figure 1) went along with a significant decrease in prefrontal cortical rno-miR-3559-3p and mmu-miR-1971 expression as well as with a trend of reduction in prefrontal cortical ebv-miR-BART8-3p and mmu-miR-1947-3p expression (Figure 3A). Finally, our analyses revealed that none of the miRNA candidates tested was altered by traumatic stress per se (Figure 2) or by fluoxetine treatment per se (Figure 4) which suggests that fluoxetine treatment interacts with traumatic stress to alter the expression levels of the mentioned miRNA candidates.

Two out of the five array-identified miRNA candidates (rno-miR-3559-p, mmu-miR-1971, ebv-miR-BART8-3p, mmu-miR-1947-3p, mmu-miR-33-5p) could be validated by miRCURY LNA™ RT-qPCR: calculation of the statistical significance of RT-qPCR results with two-way ANOVA followed by Bonferroni post hoc correction confirmed a statistical trend toward a fluoxetine-mediated increase in prefrontal cortical mmu-miR-33-5p expression in shock-fluoxetine mice comparison to no-shock-fluoxetine mice (Figure 5C: Bonferroni posttest of shock-fluoxetine versus no-shock-fluoxetine: t = 2.205, p = 0.055). Most important, we observed a statistically significant reduction in mmu-miR-1971 expression in the PFC of shock-fluoxetine mice in comparison to shock-vehicle mice (Figure 5A: Bonferroni posttest of shock-vehicle versus shock-fluoxetine: t = 2.509, p < 0.050). RT-qPCR data do not allow the conclusion that fluoxetine rescues the footshock-induced increase in mmu-miR-1971 expression, since the latter failed to survive statistical correction (Figure 5A). Furthermore, despite a significant treatment effect in the two-way ANOVA (F_{1,20 shock} = 4.494, p = 0.030), Bonferroni posttests did not detect any significant fluoxetine-mediated changes in relative expression of mmu-miR-1947-3p. Thus, we cannot consider mmu-miR-1947-3p as a fully validated candidate (Figure 5B). Finally, despite repetitive tries and employment of optimized LNA™ -technology based miRNA primer sets, expression of rno-miR-3559-3p and ebv-miR-BART8-3p could not be detected by RT-qPCR. Given that, we speculate that the array-detected rno-miR-3559-3p and ebv-miR-BART8-3p signals might possibly represent technical artifacts.

In summary, the most important conclusion of this study is that in the PTSD mouse model studied here, the therapeutic action of fluoxetine (Figure 1) is accompanied by a significant reduction in prefrontal cortical mmu-miR-1971 expression on day 74 after shock exposure (Figures 3A and 5B).

Finally, to get an idea of the potential role of mmu-miR-1971 and mmu-miR-33-5p in PTSD and of their general function, we performed an in silico analysis of target genes regulated by these two miRNA candidates: analysis performed with the miRWalk database³ (52) revealed several validated target genes of mmu-miR-33-5p (Table 3), but none of mmu-miR-1971. Then, we used computational methods to predict potential target genes of mmu-miR-1971: we applied TargetScanMouse 6.2⁴ (53) and MirTarget2⁵ (54, 55); we included only those predicted target genes that were identified with both approaches into subsequent GO analysis by employing GenericGeneOntologyTermFinder⁶ (56) and REViGO⁷ (57) (Table 2). The molecular functions of predicted mmu-miR-1971 target genes are mainly associated with small molecule and nucleic acid binding (Table 2). Moreover, most of them are involved in metabolic processes such as organic compound and RNA metabolism (Table 2).

Then, to characterize possible molecular functions of validated mmu-miR-33-5p regulated targets, we performed GO analysis as described above. We found that most of the validated mmu-miR-33-5p targets are associated with protein, miRNA, and organic compound binding (Table 3). Furthermore, many of the validated mmu-miR-33-5p target genes are involved in cellular and developmental processes, and most important, in epigenetic regulation of gene expression and lipid metabolic processes (Table 3).

Taken together, target prediction and GO analyses revealed several predicted target genes of mmu-miR-1971 and several validated target genes mmu-miR-33-5p which might possibly be involved inter alia in PTSD pathobiology or fluoxetine-mediated alterations of molecular pathways. Interestingly, amongst these target genes we found none which had previously been repetitively associated with PTSD, like for instance FKBP5 (8), CDK5 (58), or synapsin (28).