Background

Hypoxic-ischemia or asphyxia, whether it occurs pre-, peri- or postnatally, is still a major cause of neonatal mortality and morbidity. It is frequently associated with permanent neurological deficits, such as motor disabilities, learning and cognitive problems. Nowadays, post-asphyctic hypothermia is the only available evidence-based therapeutic strategy for treating term asphyxiated infants. However, only a subset of patients benefit from this strategy. Therefore, there is an urgent need to develop additional neuroprotective strategies that may, whether or not combined with hypothermia, provide an even better neurological outcome [1].

A promising approach for studying neuroprotection is investigating the physiological phenomenon of preconditioning, which was first described in the brain in 1964 [2]. The underlying mechanisms governing this phenomenon have not been fully elucidated yet. Insight into these mechanisms could provide us with directions for future neuroprotective strategies. Genomic reprogramming could explain many of these mechanisms [3] and with genome-wide micro-array technology it is now possible to investigate this neuroprotective reprogramming in experimental models.

So far, seven studies have investigated large scale gene expression with micro-array techniques after preconditioning in the newborn or adult brain [4–10]. Several of these studies did not adopt a genome-wide approach but used limited and therefore biased transcripts on the array. Furthermore, different experimental paradigms were used and, considering that no paradigm incorporated the fetal-to-neonatal transition, none of these truly resembled the global impact of perinatal asphyxia. Therefore, unique physiological mechanisms specific to the time of birth are missed, although these may play an important role in the development of post-asphyctic brain injury and/or neuroprotection.

Here we present a whole genome micro-array study in a previously validated model where we combined a global perinatal asphyctic insult at the time of birth, with fetal asphyctic (FA) preconditioning on embryonic day 17 (E17). We previously reported that animals subjected to fetal preconditioning on E17, 4 days before suffering severe perinatal asphyxia at birth, had better survival and less brain apoptosis postnatally [11]. Moreover, preconditioned animals performed similar to control animals in behavioral testing at 6 months of age [12]. Consequently, the time-point we chose to investigate the genomic response in the brain is E21, which is just before birth and 96 hours after FA. In these fetal brains we expect to find the neuroprotective mechanisms that are in place perinatally. Accordingly, we hypothesize that the preconditioned animals show a neuroprotective gene expression pattern different from control animals. Furthermore, we chose to take our micro-array data-analysis beyond the single-gene approach, and subjected our data to Gene Set Enrichment Analysis in order to derive results with maximum biological relevance [13].

Results

Changes in mRNA expression following preconditioning

Whole genome micro-array technology was used to evaluate differential gene expression in preconditioned animals compared to controls, 96 hours after FA preconditioning. Single-gene analysis with the Bioconductor Limma package yielded 88 transcripts that were differentially expressed with a p-value <0.01, fold changes ranged from 0.69 to 1.43 (see Additional file1). We found 53 transcripts that were down-regulated and 35 transcripts that were up-regulated after FA preconditioning. The majority of down-regulated genes are involved in signal transduction (expi, gjb6, itgbl1, itpka, lims2, slc22a2, slc35f4), synaptic transmission (hpcal4, htr1b, myrip, tomt, vamp5), and metabolism (abhd3, adh1, aplnr, osbp2, pdia5). Most up-regulated genes exert their function in the cell nucleus (dbx2, elf2, emx2, id1, prim1, rbmx, ste2, zfp862), regulating transcription factors and proteins involved in DNA and RNA binding. Furthermore, we observed down-regulation of dbx2, nav1, and mpzl2 as well as up-regulation of fgfr4, leptin and smad6 which seem to be involved in brain development.

In order to verify micro-array results we randomly chose several genes from the Limma analysis to validate with Real-Time qPCR (RT-qPCR) (see Figure 1A-E). We found a significant difference, confirming our micro-array results, for adh1 (p = 0.01), edn1 (p = 0.004), rdh2 (p = 0.03), and smad6 (p = 0.005). A trend towards significance was found for leptin (p = 0.07). Moreover, RT-qPCR was used to evaluate possible regional differences in expression for leptin, rdh2 and smad6 (see Figure 2A-C). Analysis of leptin mRNA expression revealed a significant up-regulation in the FA group in CPU (p = 0.01) and hippocampus (p = 0.02), and no significant difference in PFC. For rdh2 we observed no significant differences in PFC and CPU, but we found a significant up-regulation in FA animals in the hippocampus (p = 0.04) Finally, analysis of smad6 mRNA expression in prefrontal cortex (PFC), caudate-putamen (CPU), and hippocampus revealed no significant difference between control and FA animals.

Figure 1
figure 1

RT-qPCR validation of micro-array results in whole hemisphere. A-E show RT-qPCR validation results in whole cerebral hemisphere represented as mean + SEM normalized to control (n = 4-6). Adh1 (p < 0.05), Edn1 (p < 0.01), Rdh2 (p < 0.05) and Smad6 (p < 0.01) demonstrate a significant difference. Leptin shows a trend towards significance with a p = 0.07. (White bars = Control, grey bars = Fetal Asphyctic Preconditioning, significance was tested with unpaired two-tailed Student’s t-test: * = p < 0.05, ** = p < 0.01).

Full size image
Figure 2
figure 2

RT-qPCR validation of micro-array results in brain regions. A-C show RT-qPCR validation results in prefrontal cortex (PFC), caudate-putamen (CPU), and hippocampus (HIPPO) represented as mean + SEM normalized to control (n = 4-6). Leptin reveals a significant difference in CPU and HIPPO (p < 0.05), Rdh2 demonstrates only a significant difference in HIPPO (p < 0.05), and Smad6 does not show significant differences in the investigated brain regions. (White bars = Control, grey bars = Fetal Asphyctic Preconditioning, significance was tested with unpaired two-tailed Student’s t-test: * = p < 0.05).

Full size image

Changes in biological pathways: Gene Set Enrichment Analysis

In order to derive results with maximum biological relevance we decided to subject our data to a pathway based approach. The method we used was Gene Set Enrichment Analysis (GSEA) for which the entire data set was ranked according to moderate t-statistics (see Figure 3).

Figure 3
figure 3

Heatmap of ranked data-sets for Gene Set Enrichment Analysis. This heatmap, based on the moderated t-statistics generated with the Bioconductur Limma package, visualizes the differential expression between both phenotypes. Horizontal lines represent the 10 individual arrays, 5 control (C) on top, and 5 fetal asphyctic (FA) preconditioning in the bottom, with red indicating high expression and green indicating low expression. Vertical lines represent the expression of the individual genes. Expression in phenotype FA is higher than C on the top of the ranked list (left), expression in phenotype C is higher than FA at the bottom of the ranked list (right).

Full size image

GSEA analysis revealed that, out of a total of 737 gene sets, 10 gene sets were significantly enriched (FDR q-value <0.05) in the FA group and therefore up-regulated in the preconditioned animals (see Additional file 2). Also, 19 gene sets were significantly enriched (FDR q-value < 0.05) in the control animals and therefore down-regulated in the preconditioned animals (see Additional file 3). The majority of down-regulated gene sets play a role in signal transduction, and the remaining gene sets are important in synaptic transmission. The majority of up-regulated gene sets have their gene products located in the cell nucleus, and the remaining gene sets are important for ribosomal structure.

To identify which genes contributed most to the enrichment score within different gene sets, we performed a Leading Edge Analysis in the GSEA environment. In such an analysis enriched gene sets are examined for genes that occur before the maximum of the running enrichment signal, because these genes are the core of the gene set that drive the enrichment signal [13]. Our Leading Edge Analysis revealed that there were 367 individual transcripts present in the leading edge of up-regulated gene sets, and 377 transcripts in the down-regulated gene sets (all annotated genes can be found in Additional files 2 and 3). The genes in up-regulated gene sets include many histone clusters, ribosomal proteins and transcription factors, but interestingly also several key epigenetic players such as: hdac 1, hdac2, hdac3, myst3, mecp2, pcgf2, dnmt1 and dnmt3l. Among the Leading Edge genes in down-regulated gene sets were many different neurotransmitter receptors such as: glutamate receptors, GABA receptors, serotonin receptor, and a dopamine receptor. However, initial GSEA analysis showed that ‘regulation of glutamate secretion’ was the only significant gene set specific to one neurotransmitter (NES 1.94, FDR q-value 0.019). Besides genes related to neurotransmitter receptors we also found many genes that play a role in the pre-synaptic phase of neurotransmission. These genes (cplx2; cplx1; pclo; slc18a2; snap25; snap91; stxbp2; stx3; syn1-2; syt1-2; syt10; sv2b; vamp2) play different roles in the process of synaptic vesicle exocytosis such as: vesicular transport, docking, priming, and ultimately vesicular fusion.

Discussion

Here we present the first whole genome expression data in fetal brain tissue after fetal preconditioning. In concordance with our hypothesis we found that preconditioned animals have a gene expression pattern that is different from control animals. We chose to take our micro-array data-analysis beyond the single-gene approach, and subjected our data to Gene Set Enrichment Analysis in order to derive results with maximum biological relevance. Our most interesting finding is that up-regulated gene expression seems to involve epigenetic mechanisms.

Gene Set enrichment analysis

Analyzing micro-array results is typically done by comparing genes on a gene-by-gene basis and assessing if they are differentially expressed between experimental groups. In this approach the focus is on the genes that show the largest difference in expression between the experimental groups. However, this approach has some vital limitations. Most importantly it assumes that all genes act independently of one another, although biologically this is not the case. It is well known that biological processes often affect sets of genes that act simultaneously. Therefore, a small increase in all genes that belong to a certain pathway is likely to be more biologically relevant then a high increase in a single gene in that pathway [13]. Consequently, we chose Gene Set Enrichment Analysis which is a pathway-oriented approach, in order to obtain results that resemble physiological circumstances. Interestingly, GSEA results revealed a similar, but more extensive perspective than our Limma analysis results. Both indicate that the most up-regulated genes were involved in processes within the cell nucleus, and that the most down-regulated genes were involved in signal transduction and synaptic transmission. However, GSEA provided more information on the pathways that are involved.

Involvement of epigenetic mechanisms

With Limma analysis we found that the majority of up-regulated transcripts have gene products located in the cell nucleus. Their function is related to DNA binding, replication, and transcription.

Further GSEA analysis on our data revealed the majority of significantly enriched gene sets in the preconditioned animals are also mainly concerned with the cell nucleus, such as chromatin, the nucleosome, and histones. Ultimately, investigation of the Leading Edge Genes in these gene sets revealed several well known epigenetic players involved in histone acetylation (hdac1, hdac2, hdac3, myst) and DNA methylation (mecp2, dnmt1, dnmt3l). It is well known that mecp2 interacts with methylated DNA and, together with histone deacetylases, is able to cause transcriptional repression [14]. Since we observe marked down-regulation in several other functional categories we now wonder if the observed down-regulation is a consequence of the up-regulation of epigenetic players. Although the involvement of genes that have their gene products in the cell nucleus, such as DNA binding proteins or proteins involved in cell cycle control, was previously demonstrated in preconditioning studies, the pathways involved were not clear and in particular the link to key epigenetic players has not been previously described in a whole transcriptome approach [5, 22]. Furthermore, a long-lasting reduced expression of the glutamate receptor NR1 subunit was previously described in neonatal rats after fetal hypoxia-ischemia [23]. On the other hand we know that physiologic changes in glutamate receptor levels are an important part of brain maturation because of glutamate-mediated neuroplasticity [24]. Therefore, it is possible that this preconditioning induced change in glutamate signaling interferes with normal brain development. Further studies are needed to establish if there is a negative effect of down-regulated neurotransmission on neonatal brain development.

Signal transduction

Both Limma and GSEA reveal a marked down-regulation in this functional category, with the exception of posphodiesterase 9a (pde9a) and slc22a13 up-regulation in Limma analysis. From all phosphodiesterases pde9a has the highest affinity for cyclic guanosine monophosphate (cGMP).In the brain cGMP synthesis is increased after NMDA-receptor activation, on the other hand pde9a is known to modulate the response to dopaminergic, serotonergic and cholinergic neurotransmission [25, 26]. The asphyctic preconditioning stimulus is likely to have caused NMDA-receptor activation due to excessive glutamate release, and possibly activated other neurotransmitter receptors as well, which could explain the observed up-regulation in pde9a.

GSEA revealed a marked down-regulation of many gene-sets related to signal transduction: PFAM Ion channel family, PFAM organic anion transporter polypeptide, GO ion channel activity, GO voltage gated ion channel activity, GO cation channel activity, GO metal ion transmembrane transporter activity, KEGG calcium sigalling pathway, KEGG long term potentiation, KEGG salivary secretion, KEGG gastric acid secretion, KEGG pancreatic secretion, and finally the Biocarta Nos1 pathway. In literature, the down-regulation of genes related to signal transduction after preconditioning has been compared to neuroprotective strategies in hibernation [Tissue preparation

FA and control pups were delivered on E21 by Caesarean section and immediately decapitated. Control animals have not undergone any intervention prior to birth (see Figure 4). To prevent litter effects only one pup per dam was used for micro-array analysis, with a total of five males per condition. After removing the cerebellum left hemispheres were dissected and submerged in RNA stabilizing reagent (Qiagen, Benelux BV, Venlo, The Netherlands). Samples were kept at 4°C for four days, before being snap frozen in liquid nitrogen, and ultimately stored at −80°C. For RT-qPCR analysis a maximum of two pups per litter were used, with a total of five pups per condition. Right hemispheres were dissected, then snap frozen in liquid nitrogen, and ultimately stored at −80°C. Additionally, in six males per condition we dissected three different brain regions: prefrontal cortex (PFC), caudate-putamen (CPU), and hippocampus for analysis of regional expression by RT-qPCR. Dissection of these brain regions was performed in situ under 4x magnification immediately after sacrificing the pups.

Figure 4
figure 4

Experimental design. Two experimental groups were used: Control (C) and Fetal Asphyctic (FA) preconditioning. On embryonic day 17 (E17) FA animals were subjected to a 30-minute preconditioning stimulus by clam** the uterine circulation, and 96 hours later both experimental groups were sacrificed. Control animals are wild-type.

Full size image

RNA-isolation

For micro-array analysis total RNA extraction and purification were performed on mini RNeasy columns (Qiagen Benelux BV, Venlo, The Netherlands), according to the manufacturer’s instructions. Quantity and purity of total RNA was determined by spectrophotometer analysis using the Nanodrop ND-1000 (Thermo Fisher Scientific Inc., Waltham, USA). Only samples with a 260/280 ratio between 1.8 and 2.1, and a 260/230 ratio between 1.5 and 2.0 were selected for micro-array analysis. Additionally, RNA quality measurements were performed with Bioanalyzer 2100 (Agilent Technologies Netherlands B.V.). Samples with an RNA integrity number (RIN) below 8 were excluded.

For RT-qPCR total RNA was extracted with Trizol® reagent (Invitrogen, Paisley Scotland, UK) according to manufacturer’s instructions. Next, cDNA was generated with RevertAid First Strand cDNA synthesis kit (Fermentas GMBH, St. Leon-Rot, Germany).

RT-qPCR

RT-qPCR reactions were carried out using SYBR green PCR master mix and the LightCycler 480 (Roche Diagnostics, Almere, The Netherlands). To evaluate relative expression we used Gapdh, Hprt, and ß-actin as internal controls. Sequences of primers used can be found in Table 1.

Table 1 Primer design for RT-qPCR
Full size table

Micro-array analysis

Using the Ambion WT Expression Kit, per sample, an amount of 100 ng of total RNA spiked with bacterial poly-A RNA positive controls (Affymetrix Inc., Santa Clara, USA) was converted to double stranded cDNA in a reverse transcription reaction. Next the sample was converted and amplified to antisense cRNA in an in vitro transcription reaction which was subsequently converted to single stranded sense cDNA. Finally, samples were fragmented and labeled with biotin in a terminal labeling reaction according to the Affymetrix WT Terminal Labeling Kit. A mixture of fragmented biotinylated cDNA and hybridization controls (Affymetrix Inc., Santa Clara, USA) was hybridized on Affymetrix GeneChip Rat Gene 1.0 ST Arrays followed by staining and washing in a GeneChip® fluidics station 450 (Affymetrix Inc., Santa Clara, USA) according to the manufacturer’s procedures. To assess the raw probe signal intensities, chips were scanned using a GeneChip® scanner 3000 (Affymetrix Inc., Santa Clara, USA). According to MIAME requirements data were submitted the NCBI GEO database, and are available under accession number: GSE42676.

Gene Set enrichment analysis (GSEA)

For GSEA, a total of 737 rattus norvegicus gene sets were assembled, including 196 KEGG pathways (release 59.0), 81 Biocarta pathways (accessed August 18th 2011), 184 Gene Ontology terms (AmiGO version 1.8), and 276 Pfam protein families database (Pfam 25.0). Each gene-set contained a minimum of 15 genes and a maximum of 500 genes in accordance with GSEA recommendations. The analysis was conducted using the GSEA software v2.07, provided by the Broad Institute (Cambridge, MA, USA) [13].

Statistics

For RT-qPCR all data were distributed normally as tested with Kolmogorov-Smirnov test. Statistical significance was tested with the unpaired, two-tailed Student’s t-test. Results are presented as means + SEM, normalized to control and p-values <0.05 were considered statistically significant.

Analysis of the micro-array data was performed in the R programming environment (version 2.12.0), with the packages developed within the Bioconductor project [39]. The analysis was based on the RMA expression levels of the probe sets. Differential expression was assessed with the Limma package using moderated t-statistics [40]. Results are presented as fold changes and p-values < 0.01 were considered statistically significant.

For GSEA the micro-array dataset was pre-ranked using moderated t-statistics [40]. A gene set enrichment score (ES) was calculated based on the Kolmogorov-Smirnov statistic and for each gene set the ES was normalized to account for difference in gene set size. Finally, a false discovery rate (FDR) was calculated relative to the normalized enrichment score (NES) values to determine the probability of type I errors. To control for multiple testing we used the false discovery rate (FDR) as described by Benjamin and Hochberg [41]. Enriched gene-sets with an FDR q-value <0.05 were selected. Ultimately we performed a ‘Leading Edge Analysis’ in GSEA on significantly enriched gene-sets, to identify the genes that contribute most to the enrichment signal.

Author’s contributions

KC participated in design of the study, carried out the animal experiments and RNA extraction, performed Gene Set Enrichment Analysis, and drafted the manuscript. JV and AG participated in study design and coordination and helped to draft the manuscript. JS carried out the qPCR analysis. LZ and ES participated in study design and helped to draft the manuscript. All authors have read and approved the final manuscript.