We first attempted to determine if primary neuronal cultures of CbG and cerebral cortical cells were differentially sensitive to OS. Despite the caveat that studies performed with neurons grown in vitro may not reflect exactly the responses of neurons in vivo, primary neuronal cultures were used in these studies because they are composed of approximately 90–95% neurons. Thus, differential responses to OS could be directly attributed to specific characteristics of neurons from these two brain regions. Oxidative stress was induced by exposing the neuronal cultures to various concentrations of paraquat (0–50 μM). Paraquat induces OS by generating superoxide in neurons through the process of one-electron reduction by the flavoenzyme NADPH-cytochrome P450 reductase 22232425. Increases in paraquat concentration led to a dose-dependent increase in cell death in both types of neurons, but the effect of paraquat on neuronal survival was greater for CbG than cerebral cortical neurons. For example, treatment with the highest concentration of paraquat (50 μM) caused approximately 80% neuronal death in CbG but 40% in cerebral cortical neurons (Fig. 1A). Two-way ANOVA indicated that the effect of paraquat treatment on survival in the two types of cells was statistically significant (df = 4, F = 42.822, P < 0.001). The viability of CbG neurons was significantly lower than that of cortical neurons across all paraquat concentrations (t-test, P values < 0.001).

Since oxygen (O₂) tension in the ambient air surrounding the medium of primary neuronal cultures is ~20% (by volume), i.e., higher than that in the body, an important difference between neurons in culture as compared with those in the intact brain may be the fact that primary neurons in culture are continuously exposed to high O₂ concentrations and, possibly, some level of OS. Thus, the effects of paraquat on cell viability of primary neurons might have been a combined effect of exposure of these neurons to increased O₂ tension and to paraquat. In order to determine whether different levels of O₂ tension, in the absence of an ROS-generating chemical agent, were associated with differential neuronal viability, we compared neuronal survival over 21 days in vitro (DIV) for neurons in cultures maintained under either reduced (5%) or high O₂ (20%) atmosphere. While high O₂ tension had no significant effect on the survival of cerebral cortical neurons grown in culture for 7 and 14 DIV when compared with survival in low O₂ tension, the survival of cortical neurons grown for 21 days under reduced O₂ tension (5% O₂) exhibited significantly higher survival than neurons grown for 21 DIV under high O₂ (20% O₂) tension (Fig. 1B). The effect of high O₂ tension on the survival of CbG neurons was greater than that on cerebral cortical neurons. CbG neurons grown for either 14 or 21 DIV under 5% O₂ tension had significantly higher survival than those grown under 20% O₂ tension (Fig. 1C). Two-way ANOVA testing the effects of neuronal type and O₂ tension on cell viability showed that, overall, cortical neurons had significantly higher viability than CbG neurons (df = 1, F = 8.646, P = 0.006). Three-way ANOVA testing time of exposure, O₂ tension, and cell type showed significant interactions (all P values < 0.001). Post-hoc analyses using the Holm-Sidak method indicated significant effects of time of exposure (df = 2, F = 130.825, P < 0.001), O₂ tension (df = 1, F = 27.296, P < 0.001), and cell type (df = 1, F = 79.106, P < 0.001) on cell viability. These data demonstrated that, even in the absence of agents that generate intracellular ROS, CbG neurons were more sensitive to changes in environmental O₂ tension than cortical neurons.

The conditions of primary cultures of cerebral cortical neurons differed from those used in primary cultures of CbG neurons and may thus have affected the overall response of these two types of neurons to paraquat-induced OS. In order to determine the effects of OS under uniform conditions of cell growth and maintenance in vitro, we also examined the effects of paraquat on neuronal viability in organotypic slice cultures of cerebral cortex and cerebellum. The conditions of growth of slice cultures were the same for both cortical and cerebellar cultures. Organotypic slice cultures would also allow us to determine if the differential vulnerability of CbG and cortical neurons observed in primary neuronal cultures would be observed in brain sections that contain both neurons and glia. Organotypic slice cultures maintain a relatively normal brain cyto-architecture while cells are growing in culture. Such cultures were previously used by us and others to document the differential vulnerability to OS of hippocampus neurons of the CA1 and CA3 regions 520. In these previous studies, the superoxide-generating agent employed was duroquinone. But, in the present studies, the investigations were expanded to include a different agent whose metabolism leads to superoxide formation, paraquat, and thus assess whether the cell death produced was related to the common property of these two chemical agents, i.e., superoxide generation in cells.

In the present study, we employed organotypic slice cultures of cerebral cortex, cerebellar cortex, and hippocampus, including the CA1 and CA3 regions, all grown and exposed to either paraquat or vehicle under identical conditions. But, since the intent was to obtain neurons from freshly prepared brain slices for the purpose of transcriptomic analyses, i.e., fresh tissue at near in vivo conditions, we also examined the effects of paraquat on neuronal viability in hippocampus, cerebral and cerebellar cortex in acute brain slice preparations. The acute slice preparations were obtained from adult rats whereas those maintained in vitro for prolonged periods were obtained from 6–7 day-old pups. In the slice cultures maintained under prolonged incubation conditions in vitro, paraquat (final concentration 100 μM) was applied to the brain slices after they were in culture for one week and the effects on neuronal survival assessed 24 h after exposure to paraquat or vehicle. In the case of acutely prepared slices, the slices were incubated in culture medium for only a period of two hours before the same paraquat treatment. Although the cell survival rate was lower, in general, in cultures established from adult animals in comparison to those from young pups, greater neuronal death following paraquat exposure was observed in the CA1 and CbG regions as compared with the CA3 and cerebral cortex, and that was the case in both prolonged organotypic slice cultures and acute slice preparations (Fig. 2). Thus, in terms of sensitivity to OS, CbG neurons were similar to CA1 hippocampus neurons, whereas cortical neurons were similar to CA3 neurons. The presence of glial cells in slice cultures did not seem to alter the relative vulnerability of CbG and cortical neurons to OS.

The observed differential sensitivity of CbG and cortical neurons to paraquat and superoxide-induced OS, regardless of the culture type, culture conditions, or the relative chronicity of in vitro incubation of the cultures, afforded a good opportunity to conduct preliminary studies to determine whether differences in either the metabolism of paraquat or in the handling of superoxide by neurons might account for the differential vulnerabilities between these two neuronal populations. As mentioned above, paraquat generates superoxide anions through one-electron transfer reactions catalyzed by NADPH-cytochrome P450 reductase. Therefore, we first focused on potential differences in the levels of NADPH-cytochrome P450 reductase between CbG and cortical neurons. The protein levels of this enzyme were higher in cerebral cortical than CbG neurons (Fig. 3A). Based on these observations, it might have been expected that cortical neurons would be more susceptible to superoxide generation than CbG neurons. However, this was not the case. Thus, enhanced superoxide formation by cytochrome P450 reductase could not account for the greater sensitivity of CbG neurons to OS.

We next considered the possibility that the vulnerable neurons expressed lower levels of key anti-oxidant enzymes. Cu/Zn-SOD (SOD1), Mn-SOD (SOD2) and glutathione peroxidase (GPX1), were selected as oxidoreductases whose differential levels of expression might determine differential vulnerabilities to OS. The levels of SOD1 and GPX1 did not differ significantly between cerebral cortical and CbG neurons (Fig. 3B). However, the protein levels of SOD2 were 2.5 times higher in CbG than cortical neurons (Fig. 3B). Thus, based on endogenous SOD2 levels, it might have been predicted that CbG neurons would be more resistant than cortical neurons, which was not the case. It is possible that because these studies were performed using primary cultures maintained at 20% oxygen atmosphere, the elevated SOD2 levels in CbG neurons might have reflected an enhanced response of these neurons to OS induced that might have been induced by the high O₂ tension present under the culture conditions used.

To determine if exposure of cerebral cortical and CbG neurons to OS that was produced through increased superoxide generation could affect the expression of the three oxidoreductases, the levels of SOD1, SOD2, and GPX1 were measured in neurons treated with paraquat (Fig. 3C). Statistical analyses showed that the levels of SOD1 were not significantly altered following treatment of either CbG or cortical neurons with paraquat. However, the levels of both SOD2 and GPX1 were significantly increased in CbG but not cortical neurons following exposure to paraquat. Results of two-way ANOVA indicated that SOD2 content was in general significantly higher in CbG vs. cerebral cortical neurons (df = 1, F = 113.958, P < 0.001). Therefore, the higher levels of SOD2 in CbG as compared with cortical neurons grown in primary cultures could reflect an adaptive response to exposure to elevated levels of O₂ tension and the resulting OS.

Overall, these results indicated that the OS-vulnerable CbG neurons had the capacity to respond to increases in OS through the expression of higher SOD2 levels as compared with cortical neurons. Furthermore, CbG neurons showed much greater up-regulation of GPX1 than did cerebral cortical neurons following exposure to paraquat. This would suggest that CbG neurons would be equally or more protected from OS-induced cell death than cortical neurons. In conclusion, CbG neurons neither had reduced capacity to generate superoxide from paraquat nor reduced levels of anti-oxidant enzymes that protect neurons from OS. These relatively straightforward explanations of possible mechanisms of differential vulnerabilities of neurons to OS failed to reveal a pathway that might account for these differential properties of neurons. For this reason, the subsequent studies were focused on a more global approach to characterization of differentiating molecular pathways among the neurons derived from each of the four regions of rat brain that exhibited differential susceptibility to OS.

In order to obtain a global view of molecular processes that might predispose CA1 and CbG neurons to greater sensitivity to OS compared with CA3 and cerebral cortical neurons, we performed transcriptomic analyses of neurons from these four brain regions. These studies were performed under basal conditions, i.e., no paraquat-induced exogenous OS. In order to minimize non-target noise that might result from variations in sample collection and preparation, we employed LCM for the collection of highly homogeneous neuronal populations 2627. The homogeneity in the neuronal populations collected from the four designated brain regions is shown in the representative sections in Fig 4A. The cryosections used for LCM were prepared from rat brains frozen immediately after sacrifice, thus preserving, as much as possible, their native state in the brain and allowing us to assess the molecular differences among neurons under conditions that are likely to exist in vivo. The conditions for LCM dissection and recovery of neurons from the four brain regions were similar to those employed in the preparation of the acute adult brain slices that were used to detect the differential responses of neurons to paraquat-induced OS. Thus, uncovering differences under basal, near in vivo conditions, in the transcriptomic patterns between neurons that were vulnerable and those that were resistant to OS, might be indicative of the molecular pathways that are involved in determining the differential sensitivities to OS. Because the transcriptomic analyses planned were focused on differences detected under basal conditions, the rats used in the microarray studies were not subjected to any treatment that would induce OS.

Excellent reproducibility among replicate samples in GeneChip analysis was achieved. Non-target, systematic noise was very low. The mean correlation coefficient between replicate chips was 0.96 (range 0.91–0.99). Condition Tree and Principal Components Analysis (PCA) further supported the high reproducibility and low noise achieved in these studies (Fig. 4B, C). These analyses were based on genome-wide gene expression profiles and showed that samples from each brain region formed a cluster distinguishable from those from the other regions, regardless of the experimental animal from which the neurons were derived. This pattern indicated that each homogeneous neuronal population, e.g., the CA1 region neuronal population, had its own distinct transcriptome profile.

For the planned comparative transcriptome analyses, however, we were interested not in the distinguishing characteristics between one neuronal population and another, as such studies have been performed previously 14151617181920, but in those characteristics that defined vulnerability vs. resistance to OS. Therefore, we classified the neuronal samples into two groups, i.e., neurons vulnerable to OS (CA1 hippocampus and CbG neurons) and those resistant (CA3 hippocampus and cerebral cortex neurons). This approach was expected to focus the analyses of differential gene expression on the common characteristics that might determine sensitivity to OS. Based on the volcano plot analysis shown in Fig 5, there were 994 genes that were differentially expressed in OS-vulnerable as compared with OS-resistant neurons (t-test P ≤ 0.05 and fold difference ≥ 1.5). After adjusting for multiple testing, the Benjamini and Hochberg False Discovery Rate was 0.09. Of the 994 genes, 481 were expressed at significantly higher levels in the OS-resistant neurons (region "A" of Fig. 5), while 513 genes were expressed more highly in OS-vulnerable neurons (region "B" of Fig. 5). Additional file 1 [see Additional file 1] lists the subpopulation of genes in the volcano plot whose expression was ≥ 2.0-fold in vulnerable as compared with resistant neurons, or the reverse. Uncharacterized genes were not included in Additional file 1.

In an effort to confirm that the identified genes were a core set of genes whose expression under basal conditions could differentiate between neurons vulnerable and those resistant to OS, an alternative target-gene identification approach was performed. In this alternative approach, we compared gene expression profiles between CA1 and CA3 and identified differentially expressed genes between them. Likewise, gene expression data generated from CbG and cerebral cortical neurons were also compared and differential gene expression patterns between them were identified. A comparison of convergent genes from the two independent comparisons and the target genes identified using the group-based approach described above (i.e., the 481 genes expressed at higher levels in resistant neurons and the 513 at higher levels in vulnerable neurons) revealed the following: a) 91% of the 481 resistant-neuron-higher genes (a total of 437) were convergent genes that were expressed at higher levels in CA3 and cortex in the two independent comparisons; and b) 94% of the 513 vulnerable-neuron-higher genes (a total of 484) were convergent genes that were expressed at higher levels in CA1 and CbG neurons in the two independent comparisons. These findings suggested that the target genes identified from our group-based approach can be used to differentiate neurons vulnerable to OS from those that are resistant with a high degree of accuracy.

In order to evaluate the potential functional differences between OS-resistant and OS-vulnerable neurons, we performed a comparative analysis on key biological functions associated with the two lists of genes that were expressed more highly either in OS-vulnerable or OS-resistant neurons. The Core Comparison Analysis module in the Ingenuity Pathways Analysis (IPA) package allowed us to gain an overview of key differences between the two types of neurons in terms of their respective gene transcriptomes (Fig. 6). Overall, neurons in the vulnerable group showed higher transcriptional activity for genes related to gene expression, nucleic acid damage repair, RNA trafficking and post-transcriptional modification, and immune response, while those in the resistant group had higher activities in cell signaling, cell function and maintenance, lipid metabolism, and energy production.

We subsequently conducted a series of in-depth bioinformatic analyses of the genes that were differentially expressed in the two populations of neurons. The first of these was the Gene Ontology (GO) characterization, i.e., the determination of biological processes, molecular functions, and cellular components linked to groups of genes with differential expression patterns 28. In this study, we used a statistical model based on the hypergeometric distribution implemented in MAPPFinder 29. GO terms enriched with genes which were differentially expressed in vulnerable and resistant neurons are shown in Table 1. A close examination of the results from this analysis and a comparison with those of the IPA analysis showed that there were many common biological functions or processes identified by the two types of analyses. The major functional groups identified by the GO analyses among genes expressed more highly in vulnerable as compared with resistant neurons were: 1) DNA repair and response to DNA damage; 2) Regulation of transcription; and 3) Intramolecular oxidoreductase activity. The GO category of regulation of transcription included a transcription factor that functions as a sensor of hypoxia, hypoxia-inducible factor 1 (Hif1), a glucocorticoid receptor, nuclear receptor subfamily 3/group c/member 1 (Nr3c1), that controls inflammatory responses in cells, and a transcriptional co-activator of steroid receptors, Rnpc2. In addition, one of the transcription factors that transactivates Nr3c1, the POU family homeodomain transcription factor Pou2f1 30, was also expressed at higher levels in vulnerable compared with resistant neurons, possibly indicating the importance of inflammatory responses and glucocorticoid receptor signaling in OS-vulnerable neurons. One transcription factor shown in Table 1, chromobox 7 (Cbx7), and another listed in Additional file 1 as more highly expressed in OS-vulnerable neurons, MAD homolog 5 (Smad5), are involved in maintaining cells in a transcriptionally repressed state and repressing cyclin-dependent cell cycling 3132. Elevated expression of genes involved in suppression of cell cycling might be a cell defense against neuronal de-differentiation and cell death 33. In the last GO category enriched with genes more highly expressed in vulnerable than resistant neurons, there were only two genes, protein disulfide-isomerase A3 precursor (Pdia3) and emopamil binding protein (Ebp). Both PDIA3 and EBP are isomerases, both may control Ca²⁺ homeostasis in cells, and both are located in the endoplasmic reticulum (in the lumen or the membrane, respectively) 3435. EBP is 3-β-hydroxysteroid-Δ7-Δ8-isomerase that binds drugs that have anti-ischemic properties, such as emopamil and ifenprodil 35. Based on the differential expression of the genes described above, the general GO category of stress-defense-response genes was identified by GO analysis as being more highly expressed in OS-vulnerable compared with OS-resistant neurons. This GO category is populated by genes in categories of "DNA repair", "response to DNA damage stimulus", and "intramolecular oxidoreductase activity". This is in agreement with the results from the IPA comparative analysis which identified "DNA damage and repair" and "immune system development and function" among the key functions more prevalent in neurons vulnerable to OS.

The GO categories that contained genes with higher expression in OS-resistant vs. OS-sensitive neurons were those of energy generation and carbohydrate metabolism, as well as action potential, neurotransmission, calcium signaling, and regulation of signal transduction. Related GO terms such as "calcium- and calmodulin-dependent protein kinase activity", "regulation of signal transduction", "neurotransmitter transport", "synaptic vesicle", "regulation of neurotransmitter secretion", and "regulation of action potential" were enriched with genes expressed at higher levels in OS-resistant than OS-vulnerable neurons. The GO terms "energy derivation by oxidation of organic compounds" and "main pathways of carbohydrate metabolism" were also enriched with genes that were expressed at higher levels in resistant as compared with vulnerable neurons. These results matched some of the key categories identified through the use of the IPA analysis, in particular, the two major categories of genes expressed more highly in OS-resistant than OS-vulnerable neurons, "cell-to cell signaling and interaction" and "energy production".

In the present studies, OS-resistant neurons had significantly lower transcription levels of genes that fell in the GO categories of "defense response" (GO:0006952, Z = -2.14, P = 0.027) and "immune response" (GO:0006955, Z = -2.037, P = 0.043) compared with OS-vulnerable neurons. These results are not shown in Table 1 as only GO terms with positive Z scores are included in that table. The negative Z scores associated with the GO terms "defense response" and "immune response" might indicate that either the genes in these two categories were down-regulated in resistant neurons or they were more highly expressed in vulnerable neurons.

To explore further the biological significance of differences in gene expression between OS-vulnerable and OS-resistant neurons, the patterns of differential gene expression were analyzed to determine the role of such genes in defined biological pathways. Using a statistical model similar to the GO analysis, biological pathway analysis assembles genes (or their corresponding proteins) into categories of functions that fit certain biological tasks or processes. Our analysis covered nearly all the currently known biological pathways deposited in various databases, including the Kyoto Encyclopedia of Genes and Genomes (KEGG) 36, BioCarta 37 and GenMAPP 38. Although this approach is different from that of the IPA and GO analyses, the results of pathway analysis (Table 2) showed substantial agreement with those of IPA and GO analyses.

The pyrimidine metabolism pathway, enriched with genes expressed more highly in OS-vulnerable compared with OS-resistant neurons, and the citrate (or TCA) cycle biological pathway, enriched with genes more highly expressed in OS-resistant than OS-vulnerable neurons corresponded, to some extent, to the theme of DNA damage/repair and energy derivation/carbohydrate metabolism shown in Fig 6 and table 1. The biological pathway analyses, however, revealed some new pathways, such as the alanine/aspartate metabolism pathway that was enriched with genes more highly expressed in OS-vulnerable vs. OS-resistant neurons, and the signal transduction, prostaglandin synthesis, and actin cytoskeleton regulation pathways that were enriched with genes expressed more highly in OS-resistant vs. OS-vulnerable neurons (Table 2). The signal transduction pathways included the MAPK cascade, Wnt signaling, insulin signaling, and G-protein signaling pathways. Some of the genes in the biological pathways listed in table 2 may also be involved in cell signaling and may have a neuroprotective function, for example, the annexin IV (Anxa4), annexin V (Anxa5), and annexin VI (Anxa6) genes 394041.

Pathway analysis identified four genes belonging to the protein ubiquitination pathway as expressed more highly in OS-vulnerable than OS-resistant neurons. This pathway did not reach statistical significance due to the large number of genes involved. Nevertheless, the four genes whose expression was high in OS-sensitive neurons, Usp1 (ubiquitin specific peptidase 1, VUL/RES = 10.08, P = 0.029; VUL: vulnerable group [CbG + CA1], RES: resistant group [Cortex + CA3]), Usp3 (ubiquitin specific peptidase 3, VUL/RES = 4.63, P = 0.040), Usp48 (ubiquitin specific peptidase 48, VUL/RES = 1.63, P = 0.001) and Ube2e2 (ubiquitin-conjugating enzyme E2E 2, VUL/RES = 1.54, P = 0.049), may be indicators of high stress in vulnerable neurons.

The biological function of energy generation was populated by genes expressed more highly in OS-resistant than OS-vulnerable neurons. We performed gene networking analyses on genes related to cell energy generation using information from currently available literature data [see Additional file 2]. This in silico network construct identified several of the energy-generating genes in the network that were expressed more highly in neurons from regions resistant to OS. These genes included Eno1, Idh1, Idh3a, Ldha, Me1, and Pfkp. Genes that interact with Eno1, Idh1, Idh3a, Ldha, Me1, and Pfkp were identified by gene networking analyses and included Cap1, Egf, Grem1 and Tgfb1. The function of these five genes is related to cell development, including differentiation and proliferation, and cell-to-cell signaling and interaction. The gene network analysis suggested a link between energy production and the high energy demand for cell signaling in OS-resistant neurons. Direct experimental evidence on differential levels of energy reserves, i.e., ATP levels, in OS-vulnerable and OS-resistant neurons is presented below.

The expression of genes for cellular oxidoreductases known to protect neurons and other cells from OS, i.e., those for Sod1, Sod2, or Gpx1, was examined for possible differential patterns of transcriptional regulation of these enzymes. The microarray data on Sod1, Gpx1, and Sod2 indicated no significant differences in expression between CbG and cerebral cortical neurons. The lack of differential expression of the genes for Sod1 and Gpx1 was anticipated based on the lack of any differences in SOD1 and GPX1 protein levels between CbG and cerebral cortical neurons in culture in the absence of added OS stimuli. But, whereas the protein levels of SOD2 were 2.5 times higher in CbG compared with cerebral cortical neurons in cultures without the added stimulus of paraquat-induced OS, the Sod2 gene levels were only marginally higher in OS-vulnerable compared with OS-resistant neurons captured from fresh brain sections by LCM (VUL/RES = 1.16, P = 0.063). Thus, the higher protein expression levels of SOD2 in CbG neurons in culture might have been the result of exposure of primary neurons to the increased O₂ tension in culture.

The results of the microarray analyses were validated by real-time quantitative PCR (qPCR) measurements of the expression of select genes. These genes were: Nefl, neurofilament, light polypeptide; Tf, transferrin; Mt1a, metallothionein 1a; and Nfe2l2 (or Nrf2), nuclear factor, erythroid derived 2, like 2. The selection of these genes was based on two criteria: 1) their expression profiles were representative of most genes on the array in terms of the range of transcript abundance levels and expression ratios between VUL and RES neurons; and 2) the genes were considered important in terms of differential vulnerability of cells to OS 20. For all selected genes, the qPCR data confirmed the relative expression ratios in VUL vs. RES neurons (Fig. 7). Among the four genes, three (Mt1a, Nfe2l2, and Tf) showed similar ratios of expression as determined by microarray and qPCR. For Nefl, the array analysis showed a ratio of 0.30 (VUL/RES) whereas the ratio estimated by qPCR indicated only slightly lower expression in VUL as compared with RES neurons (VUL/RES = 0.89). This difference might have been due to the fact that the targeted gene regions for detection by the two approaches were different. The GeneChip probe set targeted the last exon (exon 4) of Nefl, while the qPCR analysis, because of constraints in primer design, targeted exons 1–2. Despite this difference, the patterns of expression of the four genes estimated by qPCR were consistent with those estimated by microarray analyses.

The results of genomic analyses indicated that neurons vulnerable to OS had a lower expression of genes involved in energy generation. To determine if this would be true in neurons maintained in culture, we measured the levels of ATP in primary CbG and cerebral cortical neurons, both before and after OS induction. The ATP levels of CbG neurons under basal conditions were approximately 25% lower than those of cerebral cortical neurons, but this difference was not statistically significant. However, when exposed to OS, ATP levels in both cell types dropped, but they did so more precipitously and significantly in CbG than cortical neurons (Fig. 8). Two-way ANOVA on the combined data of all paraquat concentrations showed that, overall, CbG neurons had significantly lower ATP levels than cortical neurons (df = 1, F = 16.677, P < 0.001). In addition, the ANOVA results demonstrated that paraquat had a significant effect on ATP levels (df = 3, F = 9.513, P < 0.001). To explore further the relationship between OS and ATP levels in these two populations of neurons, we measured changes in the levels of ATP in CbG and cortical neurons under two different levels of O₂ tension (5% and 20% O₂). While the ATP levels in cerebral cortical neurons were not significantly altered by growing them for 7 and 14 DIV under 20% O₂ tension, the levels dropped to significantly lower values (P < 0.05) in neurons maintained for 21 DIV under high O₂ tension when compared with those maintained under 5% O₂ (data not shown). CbG neurons were more sensitive to the different levels of oxygen and showed significant decreases in ATP levels at both 14 and 21 DIV for cells exposed to 20% O₂ tension compared with those maintained at 5% O₂ (P < 0.05 and P < 0.01 for 14 and 21 DIV neurons, respectively; data not shown). Two-way ANOVA on combined data of cerebral cortical and CbG neurons demonstrated, once again, that cerebral cortical neurons had significantly more ATP than CbG neurons (df = 1, F = 12.089, P = 0.001).

Dissociated cerebral cortex and CbG neuron cultures were established as described before 636465. All animal procedures were performed in accordance with guidelines established by the University of Kansas IACUC. For preparations of cortical neuron cultures, the cortical lobes from 18-day old Sprague Dawley fetuses were dissected and cells dissociated by gentle trituration with trypsin and harvested by centrifugation. Cortical cells were re-suspended in fresh DMEM/F-12 with 10% fetal calf serum (FCS), plated at densities ranging from 0.3 to 3 × 10⁶ cells/35 mm dish, coated with poly-D-lysine, and after 24 h, the FCS-containing medium was replaced by a defined medium with DMEM/F12 containing N₂ supplements, KHCO₃ (15 mM), and 20% glial conditioned medium. CbG neurons were prepared from 7–8 day old Sprague Dawley pups. The dissociated cells were re-suspended in Krebs-Ringer-buffered medium. They were then treated with trypsin (50 μg/ml) and subsequently trypsin inhibitor and DNase (80 μg/ml), plated at densities ranging from 0.3 to 3 × 10⁶ cells/35 mm dish (coated as above) and maintained in basal Eagle's medium supplemented with 10% FCS, 25 mM KCl, 2 mM glutamine, and 100 μg/ml gentamicin. Growth of glial cells was inhibited by the addition of 10 μM cytosine arabinoside. These cultures were maintained at 37°C in ambient air mixed with 5% CO₂.

OS was induced by exposing primary cortex and CbG neurons to increasing concentrations of paraquat (1,1^'-dimethyl-4,4^'-bipyridynium dichloride; 0–100 μM). Neurons plated at 6 × 10⁵ cells/35 mm dish were exposed to freshly prepared paraquat added to the culture medium. After 24 h, neuronal viability was determined by the calcein AM and propidium iodide (PI) labeling method 64. The medium was removed, replaced with 1 ml PBS containing 150 nM calcein AM and 20 μM PI for 30 min, after which the cells were rinsed with PBS and visualized in a Nikon fluorescence microscope. Viability was determined by counting the number of green (alive) and red (dead) cells. Neurons were classified as vulnerable if after treatment their survival rate dropped below 40%.

To test the effects of O₂ tension on CbG and cortical neurons, primary neuron cultures established in the same way as described above were maintained in either 5% or 20% O₂, in 90% or 75% N₂, respectively, and in a constant 5% CO₂ atmosphere. Cell viability was measured by the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reduction assay 66. The assay was performed by incubating neuronal cultures with MTT (0.5 mg/ml) for 4 h, solubilizing the formazan crystals formed by the addition of an equal volume of 0.04 N HCl in isopropanol, sonicating the samples (2 min), and measuring the absorbance at 570 nm. Background absorbance was determined in cells pre-treated with iodoacetamide (10 mM), an inhibitor of succinate dehydrogenase.

For immunoblot studies, proteins were extracted from primary neurons 24 h after exposure to either paraquat or buffer. Cells were lysed in 3 mM Tris-Cl, pH 8.0, containing 10 μl/ml of a protease inhibitor cocktail (Calbiochem, San Diego, CA, USA) and the extracted proteins were separated by SDS-PAGE, transferred to PVDF membranes, and probed with antibodies to cytochrome P450 reductase (1:200) (Nventa, San Diego, CA, USA), Cu/Zn-superoxide dismutase (SOD1) (1:1000) (Chemicon, Temecula, CA, USA), Mn-SOD (SOD2, 1:1000) (Stressgen, Victoria, Canada), and glutathione peroxidase 1 (GPX1, 1:500) (Cortex Biochem, San Leandro, CA, USA). Equal amounts of protein sample were loaded into each lane of SDS-PAGE gels. Incubation with alkaline phosphatase-conjugated secondary antibody (1:1000) and color development were as described 67. Immunoblots were scanned and densitometric analyses performed using Photoshop 5.0.

The levels of cellular ATP were measured using the ATP Determination Kit (Molecular Probes, Carlsbad, CA, USA) according to the manufacturer's instructions. The assay is based on the generation of light by the ATP-dependent reaction between luciferin and luciferase. The reaction mixture in a final volume of 100 μl contained: 25 mM Tricine buffer, pH 7.8, 0.1 mM EDTA, 0.5 mM luciferin, and 1.25 μg/ml luciferase. The reaction was initiated by the addition of 10 μl of neuronal lysate prepared from control and paraquat-treated cells, and the resultant bioluminescence (emission maximum 560 nm) measured in a luminometer. The concentration of cellular ATP was calculated by a standard curve generated using known concentrations of exogenous ATP.

Two types of brain slice preparations were employed in the studies: acute slice preparations from adult rat brains and organotypic slice cultures from newborn rats. The acute slice preparations were obtained from 3 mo-old male rats, whereas the organotypic slice cultures were established from 10-day old male rats 202168. For the organotypic cultures, slices (300 μm thick) of hippocampus, cerebellum, and cerebral frontal cortex were prepared and maintained in culture for one week in 50% MEM, 25% Hanks' balanced salt solution, and 25% heat-inactivated horse serum, plus 4.5 mg/ml glucose, 1 mM glutamine, and 20 ml/L penicillin-streptomycin. The culture medium was removed 24 h prior to treatment with paraquat and was replaced with serum-free medium. Slices were either treated with 100 μM paraquat for 1 h (OS) or buffer (controls). Neuronal death was determined at 24 h after initiation of treatment by staining with PI. PI was added to the treatment medium at a final concentration of 5 μg/ml. PI uptake was used to monitor cell death since it is excluded from live cells. Fluorescence from the PI staining was excited at 515–560 nm using a Zeiss microscope fitted with rhodamine filter. Fluorescence as well as visible-light (to visualize the whole area) images were captured, and the percentage of dead neurons was calculated. For the acute slice preparations, the slices from the same brain regions as those used for organotypic cultures were incubated in serum-free medium (described above) for 2 h to allow for recovery from the dissection. Subsequent paraquat treatment and estimation of cell death by measuring PI uptake, were performed the same as for the above organotypic cultures.

An optimized protocol 69 from Arcturus (Mountain View, CA, USA) was followed for LCM. Six-months-old Fisher 344/BN F1 rats that were not subjected to any prior treatment were sacrificed, the brain dissected and immediately frozen in liquid N₂, and cryosections (8 μm thick) obtained. In order to preserve RNA quality, frozen sections were quickly stained with Arcturus HistoGene LCM Frozen Section Staining Kit. The Arcturus PixCell^® IIe Laser Capture Microdissection System was used for capturing the cells. The four populations of target neurons were collected from the hippocampal CA1 and CA3 regions, the CbG cell layer, and the frontal cortex (layers IV-VI).

Total RNA was extracted from the microdissected neurons with the Arcturus PicoPure RNA isolation kit. Three replicate samples representing each of the four neuronal populations from each animal were obtained. The quality of the RNA samples was checked with Agilent Bioanalyzer 2100 with RNA 6000 Pico Chips (Agilent Technologies, Santa Clara, CA, USA). For GeneChip target preparation with signal amplification, the GeneChip Two-Cycle cDNA Synthesis Kit from Affymetrix (Santa Clara, CA, USA) was used according to the manufacturer's instructions. The Affymetrix Rat Expression 230A (RAE230A) oligonucleotide arrays containing 14,280 Unigene clusters were used for the hybridization at 45°C for 16 h. Subsequent washing and staining steps were performed on an Affymetrix GeneChip System running GeneChip Operating Software (GCOS, ver 1.1.1). In order to minimize experimental variability, all steps throughout the entire process, from cryosectioning to array data collection, were performed by the same investigator. The microarray data generated from all chips met the quality control criteria set by Affymetrix, including low background and noise, positive detection of QC probesets such as bioB, percentage of genes called present in normal range (generally between 40–60%), similar scaling factors across all chips, and 3'/5' ratio. All microarray data were deposited in NCBI's Gene Expression Omnibus 70, with series accession number GSE7139.

The microarray data generated from GCOS, scaled to the same signal intensity of 500, were imported to GeneSpring (version 7.2) (Agilent Technologies, Santa Clara, CA, USA). In GeneSpring, the data were normalized on a Per Chip (to 50^th percentile) and Per Gene (to median) basis. Gene Tree and PCA were subsequently conducted to examine the reproducibility of samples from each brain region. Both analyses were based on all probesets on the array and Pearson correlation was used to measure similarity.

For identification of differentially expressed genes between vulnerable and resistant neurons, genes showing expression (i.e., called "present" by GCOS) in at least two thirds of all samples were selected as candidate genes, while the rest were filtered out from further analysis. For subsequent data analysis, samples from hippocampal CA1 and CbG neurons were classified as the vulnerable group (VUL), and those from CA3 and cortex neurons as the resistant group (RES). In order for a gene to be identified as differing significantly in expression between the two groups of neurons, two criteria had to be met: 1) average VUL/RES ratio of gene expression had to be either ≥1.5 or ≤0.667, i.e., genes had to have at least a 1.5-fold higher expression in the VUL over RES, or a 1.5-fold higher expression in the RES over VUL group; and 2) differential gene expression had to reach statistical significance of P ≤ 0.05 in paired t-test statistical comparisons. Genes fulfilling both criteria were plotted in a volcano plot. Multiple testing corrections were based on the Benjamini and Hochberg False Discovery Rate approach 71.

Genes showing significantly higher expression in VUL or RES neuronal populations were subjected to further comparative analyses of their overall biological functions. GO, Biological Pathway and gene networking analyses were used to uncover their functional profiles. Comparative analyses of the biological functions of differentially expressed genes in VUL vs. RES neurons were conducted using IPA (Ingenuity Systems, Mountain View, CA, USA). For these analyses, all differentially expressed genes that were associated with specific biological functions in the Ingenuity Pathways Knowledge Base were considered. Fischer's exact test was used to calculate a P value determining the probability that each biological function assigned was due to chance alone.

For GO and Biological Pathway analyses, the lists of probesets identified to be differentially expressed between the VUL and RES groups were fed into MAPPFinder, a component of GenMAPP (Gene Map Annotator and Pathway Profiler, version 2.0) 29. The gene and GenMAPP databases used for the analyses were those of August 2006. Hypergeometric distribution was used in MAPPFinder. A Z score > 1.96 or < -1.96, and a permute P < 0.05, were considered to be statistically significant. Gene networking analyses were performed using IPA on genes related to cell energy generation. For these analyses, the list of probesets for the genes was uploaded into IPA. Each probeset was mapped to its corresponding gene object in the Ingenuity Pathways Knowledge Base. These mapped genes were overlaid onto a global molecular network developed from information contained in the Ingenuity Pathways Knowledge Base. A gene network was then algorithmically generated based on their connectivity.

The same RNA samples used for the microarray analyses were also used for the qPCR analyses. The RNA was first reverse-transcribed to cDNA using oligo(dT) primers and SuperScript™ II Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA). TaqMan^® Gene Expression Assays with the following Assay IDs (Applied Biosystems, Foster City, CA, USA) were used: Rn00821759_g1 (Mt1a), Rn00582365_m1 (Nefl), Rn00582417_g1 (Nfe2l2), Rn00690607_m1 (Tf). TaqMan^® Fast Universal PCR Master Mix was employed for the amplification on an Applied Biosystems 7500 Fast PCR System using Fast Mode. The thermal cycling conditions were: 20 sec at 95°C (initial denaturation), followed by 40 cycles of 95°C for 3 sec and 60°C for 30 sec. β-Actin (Actb, TaqMan^® Gene Expression Assay ID Rn00667869_m1) was used as endogenous control since its expression was found to be constant across neuronal samples from the various regions. Quantification of target genes was based on the relative standard curve method 72.