NFκB in Neurons? The Uncertainty Principle in Neurobiology

Abstract

Nuclear factor κB (NFκB) is a dynamically modulated transcription factor with an extensive literature pertaining to widespread actions across species, cell types, and developmental stages. Analysis of NFκB in a complex environment such as neural tissue suffers from a difficulty in simultaneously establishing both activity and location. But much of the available data indicate a profound recalcitrance of NFκB activation in neurons, as compared to most other cell types. Few studies to date have sought to distinguish between the various combinatorial dimers of NFκB family members. Recent research has illustrated the importance of these problems, as well as opportunities to move past them to the nuances manifest through variable activation pathways, subunit complexity, and target sequence preferences.

NFκB is name given to a class of transcription factors that mediate diverse biological processes, from inflammation to apoptosis. While there are more extensive reviews on the variety to be found in NFκB form and function (Hayden and Ghosh 2004), a cursory introduction is necessary for the discussions here. Active binding to specific DNA sequences is performed by hetero- or homodimers of NFκB subunits; the names of vertebrate subunits are RelA (p65), RelB, c-Rel, p50 and p52. The most prominent and extensively studied dimer is that of RelA and p50, which we will refer to as NFκB_can. Under basal conditions, this moiety is held inactive in the cytoplasm by an inhibitory subunit (IκBα through IκBγ); the precursors of p50 and p52—p105 and p100, respectively—can also serve inhibitory functions. In the “canonical” activation scheme, the IκB is phosphorylated by an IκB kinase (IKK) complex (below), leading to ubiquitination and proteasomal degradation of the IκB. This frees NFκB_can to translocate into the nucleus and induce transcription of genes containing κB cis elements in their promoters. RelB and p52 form a dimer we will refer to as NFκB_non, and this moiety participates in the “noncanonical” scheme. This alternative activation is roughly analogous to the canonical except that a single polypeptide, p100, is responsible for providing both the IκB (p100 in its full-length form) and one of the subunits of the active transcription factor (p52, a proteolytic derivative of p100); kinases activating the noncanonical pathway stimulate the conversion of p100 to p52. Details of the canonical and noncanonical pathways differ by 1) the kinases that trigger phosphorylation-instigated, ubiquitin-directed proteolysis of the IκB, 2) the time-frame for maximal activation, and 3) the specific DNA sequences to which their products (NFκB_can vs. NFκB_non) bind with highest affinity (Bonizzi et al. 2004). RelA/c-Rel dimers have been described in some neural systems (Pizzi et al. 2005), and robust transactivating effects can result from various homodimers of subunits other than p50 or p52 (these lack a transactivation domain and generally stimulate transcription only when paired with a “Rel” subunit). However, the natural regulation of dimers other than NFκB_can and NFκB_non is poorly understood. Additional complexity is added by the varying requirements that have been reported for post-translational modification of RelA by phosphorylation and acetylation.

A wide variety of genes with different and sometimes opposite functions respond to NFκB_can. For example, proapoptotic genes such as p53 (Kirch et al. 1999), c-myc (La Rosa et al. 1994), Fas and FasL (Kimura et al. 1997; Matsui et al. 1998) can be regulated by NFκB_can; but so can prosurvival genes such as IAP (inhibitors of apoptosis) (Chu et al. 1997; Yabe et al. 2005), Bcl-2 (Tamatani et al. 1999), Bcl-x and SOD2 (Mattson et al. 1997; Tamatani et al. 1999). Because of this apparent role of NFκB_can in diverse phenomena, many investigators have been interested in determining whether or how NFκB_can participates in neurobiology.

The highly differentiated cell types and subtypes in the CNS create a particularly difficult challenge for studies of NFκB. Traditional techniques for establishing cellular localization, such as immunohistochemistry, are rather limited in their ability to reflect the bioactivity of a set of proteins that are regulated by binding partners and multiple post-translational modifications. Likewise, activity assays that depend on in vitro binding assays typically require homogenization of relatively large (and cellularly complex) tissue samples. Therefore, difficulty arises in simultaneously determining both activity and location of NFκB in the nervous system—a biological analogy to the Heisenberg Uncertainty Principle. Until recently, rigorous studies of NFκB in neurons (as opposed to other CNS cell types) have required the reductionist utility of cell culture, where additional activity assays like reporter-gene transfection can be more readily conducted, as well.

Is NFκB Responsive to Glutamatergic Stimuli?

One of the most potent and consistent activators of NFκB is tumor necrosis factor (TNF). Under some circumstances TNF can be cytotoxic (particularly, for tumor cells). So, guilt by association originally indicted NFκB as a potential mediator of this toxicity. Other reports demonstrated that antioxidants could block activation of NFκB (Schreck et al. 1991), leading to speculation that NFκB mediated the untoward effects of reactive oxygen species (ROS). Eventually, it was reported that glutamate could activate NFκB (Guerrini et al. 1995; Kaltschmidt et al. 1995) or p50 homodimers (Grilli et al. 1996) in cerebellar cultures, and NFκB was assumed to contribute to excitotoxicity, despite the facts that 1) cerebellar neurons cannot be enriched with mitotic inhibitors (Seil et al. 1992), 2) glutamate promotes survival of cerebellar granule cell neurons, and 3) p50 homodimers alone are not transcriptionally competent (Schmitz and Baeuerle 1991).

The hypothetical role for NFκB in glutamate toxicity was revised when reports of survival enhancement by NFκB began to appear in the literature. NFκB was shown to ameliorate the conditional toxicity of TNF in epithelial and mesenchymal cells (Beg and Baltimore 1996; Van Antwerp et al. 1996; Wang et al. 1996); to mediate the trophic effects of activity-dependent neurotrophic factor (Glazner et al. 2000), depolarization, and IGF-1 (Koulich et al. 2001); to induce expression of the “inhibitor of apoptosis” (IAP) genes (Wang et al. 1998); and to contribute to neuroprotective inductions of manganese superoxide dismutase (SOD2) (Mattson et al. 1997). No longer relegated to the harmful side of the equation, NFκB and its attendant phenomena took on a new light. Rather than participating in the toxicity of TNF or glutamate, NFκB was interpreted to be a compensatory factor that might elevate expression of anti-oxidant and anti-apoptotic genes.

The possibility that a glutamate → NFκB pathway contributed to conditioning or compensatory responses inspired attempts to replicate the glutamatergic induction of NFκB that had been reported for cerebellar cultures, instead using highly enriched cultures of cortical neurons (both neocortical and hippocampal cultures) (Mao et al. 1999; Moerman et al. 1999). Cortical cultures documented to be approximately 99% neurons were exposed to glutamate, and nuclear extracts were analyzed by electrophoretic mobility shift assays (EMSA) utilizing probes that contained a κB sequence. Surprisingly, the only consistent effect of glutamate under these conditions is a rapid reduction in the DNA-binding activity (Mao et al. 1999; Moerman et al. 1999). Results consistent with these are obtained when cortical neurons are transfected with a κB-responsive reporter plasmid (Barger et al. 2005). When resolved electrophoretically under denaturing conditions, the proteins constitutively binding κB sequences in neurons show apparent molecular weights inconsistent with those of NFκB family members (Moerman et al. 1999). Furthermore, the κB-binding proteins in such neuronal cultures are insensitive to antibodies against p50, p52, RelA, RelB, or c-Rel; instead, this neuronal κB-binding activity is attributable to Sp1-related proteins (below). However, NFκB_can is readily detectable after glutamate treatment of mixed neuron-glia cocultures (Mao et al. 1999; Moerman et al. 1999). In other experiments, neurons and glia have been grown in separate but communicating chambers to permit independent extractions of the nuclei from each cell type. In this paradigm, glia grown in the presence of neurons show a robust activation of NFκB by glutamate; neither neurons from these cultures nor pure glial cultures respond to glutamate (Moerman et al. 1999). These results have been observed in both hippocampal and neocortical cultures.

Besides documenting glia as the cellular source of NFκB activated by glutamate, the coculture experiments described above exclude the possibility that technical artifacts of experimental conditions obviated the detection of active NFκB in neurons. Nonetheless, they do not exclude the possibility that cell-cell contact between neurons and glia in vivo permits cortical neurons to respond to glutamate with an activation of NFκB. To address this caveat, a mouse line carrying a β-galactosidase transgene driven by a promoter with κB elements was employed. Cortical neurons from these mice were placed in culture with nontransgenic glia, thus creating conditions in which NFκB activity could be detected specifically in neurons, even while permitting direct physical interactions between glia and neurons. But this scenario also failed to detect a neuronal activation of NFκB in glutamate-treated cultures (Figure 1). Other reports have claimed that glutamate can activate NFκB in cortical neurons (e.g., (Qiu et al. 2001; Meffert et al. 2003; Pizzi et al. 2005). However, most of these reports used culture conditions likely to permit glial contamination of the neuronal cells when they reported DNA-binding activity. And other assays, such as the translocation of immunocytochemically detected RelA, do not measure the actual activity of NFκB factors. One approach that mitigates these issues is the use of a reporter gene that can be assessed on a cell-by-cell basis, such as β-galactosidase. Two recent reports use such a system to provide compelling evidence that neurons contain constitutively active κB-dependent transcription in vivo (Bhakar et al. 2002; Fridmacher et al. 2003); both used auxillary transgenic expression of IκBα to demonstrate that the β-gal signal was sensitive to an NFκB inhibitor. Caveats include the fact that IκBα can inhibit other transcription factors, including Sp1 (Algarte et al. 1999; Heckman et al. 2002). The p105 promoter used by Fridmacher et al. (Fridmacher et al. 2003) to create a NFκB-responsive reporter construct relies upon a κB element (GGGGGCTTCCC) that would appear to bind Sp1-related factors robustly (below).

Figure 1. Glutamate suppression of NFκB activity in primary neurons.

Cortical neurons were prepared from the neocortex of mice transgenic for a κB-reporter β-galactosidase gene; these neurons were cultured alone (with a transient AraC exposure to kill mitotic cells) or plated onto a confluent monolayer of wildtype astrocytes. At T₀, glutamate was applied at 50 μM transiently (10 min) or at 20 μM continuously. β-gal activity was measured in lysates at the subsequent times indicated (h). CD40L was also applied to cocultures for 18 h at the concentrations indicated (μg/mL). Values represent activity relative to untreated cultures ± SEM. Comparison between control and each treatment yielded p < 0.02. (Activity was undetectable in pure neuronal cultures 24 h after glutamate application.)

Unique Control of NFκB by Canonical and Translational Pathways In Neurons

Neurons appear to either silence or activate NFκB in a tissue-specific fashion consistent with unique adaptive stress responses in these cells (Ward and Massa 1995; Jarosinski et al. 2001). Thus, a key question is how NFκB may be uniquely regulated in neurons to effect appropriate expression of NFκB-responsive genes in health and disease.

Class I major histocompatibility (MHC) molecules have held fascination for neuroimmunologists as a component of hypotheses about CNS immunoprivilege during viral infections. MHC class I molecules are responsible for the presentation of an infected cell's viral antigens to cytotoxic T cells, essentially sacrificing the infected cell in order to mitigate viral replication. As this would be a disadvantageous strategy with regard to post-mitotic neurons, it was hypothesized that the CNS enjoyed an “immunoprivilege” (Oldstone et al. 1986; Streilein 1993). MHC class I were connected to the first mechanistic explanations of this phenomenon (Schachner and Hammerling 1974; Schnitzer and Schachner 1981; Lampson et al. 1983; Lampson 1987; Drew et al. 1993). These studies described the conspicuous lack of MHC class I molecules in the brain, primary neurons, or neuroblastoma cells grown in culture. It was also noted that MHC class I genes could be induced by proinflammatory cytokines in glia but not neurons in vivo or in vitro (Wong et al. 1984; Wong et al. 1985; Mauerhoff et al. 1988; Massa 1989; Massa et al. 1993; Ward and Massa 1995). These studies were followed by a number of elegant investigations in which the functional relevance of MHC class I exclusion from neurons was demonstrated using animal models of CNS viral infection (Joly et al. 1991; Rall et al. 1995; Oldstone 1997).

Speculation about mechanisms for regulating MHC class I genes have naturally focused on transcription. A number of conserved regulatory elements appear to be responsible for either constitutive or cytokine-induced expression (Burke et al. 1989; Burke and Ozato 1989). In response to cytokines, at least two adjacent elements activate MHC class I transcription: one for interferon-mediated induction (interferon regulatory factor element, IRF-E) and the other for expression induced by tumor necrosis factor (TNF) (Sugita et al. 1987; Kieran et al. 1990; Dey et al. 1992; Massa et al. 1992; Drew et al. 1995). This latter element (MHC class I κB site, MHC-κB) is a NFκB-binding site; in some cell types, the two sites appear to act synergistically in response to IFN-γ and TNF to induce high levels of MHC class I (Drew et al. 1995). TNF or double-stranded RNA stimulates binding of the transactivating NFκB_can to the MHC-κB, replacing the repressive p50 homodimer bound to this site under basal conditions (Yano et al. 1987; Kieran et al. 1990; Drew et al. 1995; Massa and Wu 1995). Similarly, the IRF-E is bound by a stimulatory transcription factor (IRF-1) or a repressor (IRF-2) (Yano et al. 1987; Kieran et al. 1990; Drew et al. 1995; Massa and Wu 1995). IRF-1 expression is activated by NFκB (Harada et al. 1994; Ohmori et al. 1997), and this relationship means that MHC class I genes are predominantly controlled by the IRF-E in astrocytes, secondary to induction of IRF-1 by NFκB (Massa and Wu 1995; Jarosinski and Massa 2002).

If neurons are to avoid transcriptional activation of MHC class I genes during viral infections, the above data indicate that squelching NFκB would be effective. In neurons, one observes neither constitutive MHC class I promoter activity nor induction of this promoter by IFN-γ, consistent with a severe attenuation of NFκB_can binding to the MHC-κB in neurons compared to glia (Massa et al. 1993). And this inactivity is not limited to IFN-γ: multiple Toll-like receptor (TLR) ligands, including lipopolysaccharide and dsRNA, as well as proinflammatory cytokines TNF and interleukin-1β, fail to induce NFκB_can in primary neurons, even after prolonged treatment (Jarosinski et al. 2001). In all cases, this refractoriness of neuronal NFκB_can was compared to positive controls in primary astrocytes, which responded to these same stimuli within 30 minutes. Clearly, MHC class I promoter activity, NFκB_can activity, and IRF-1 expression are profoundly suppressed in neurons in response to various inducers (Ward and Massa 1995).

Given this confirmed recalcitance of NFκB_can in neurons, it came as a surprise when subsequent studies found that dsRNA or Sendai virus (ligands of TLR-3) produced convincing activation of the interferon-β (IFN-β) gene in neurons under the same conditions in which MHC class I genes remained silent (Ward and Massa 1995). Induction of the IFN-β expression by dsRNA and virus infection is critically dependent on a κB cis element in the IFN-β promoter (Thanos and Maniatis 1992, 1995a). TLR-3 ligands also induce transcription of IRF-2 in cerebellar granule cell cultures (Ward and Massa 1995), probably via a κB site (Harada et al. 1994). The treatment of these cultures with dsRNA evokes a selective binding to the IFN-β promoter cis element (PRDII/IFNβ-κB) but not the MHC-κB site. This specificity of the binding activity results from RelA homodimers (henceforth designated RelA₂) rather than NFκB_can (Ward and Massa 1995), consistent with a higher affinity of RelA₂ for the PRDII/IFNβ-κB site versus the MHC-κB (Fujita et al. 1992; Ganchi et al. 1993). Similar findings were obtained with LPS treatment of neuronal cultures (Figure 2). Thus, TLR3 appears to be connected to a specific activation of RelA₂ rather than NFκB_can in neurons, permitting responses to dsRNA and LPS through a set of genes distinct from those that create vulnerability to cytotoxic T-cells.

Figure 2. Like dsRNA, LPS induces NFκB_can in astrocytes but RelA homodimers (RelA₂) in neurons.

DNA-binding activity in nuclear extracts from mouse astrocytes or cerebellar granule cell cultures was specifically detected with an iNOS-κB site and IFNβ-κB site, respectively.

NFκB Induction by Translational Inhibition: An Alternative “Noncanonical” Pathway in Neurons?

The above considerations of immunoprivilege notwithstanding, evidence suggests that NFκB is activated in neurons by non-immunological stimuli, including developmental signals or stress in adult brain (Bethea et al. 1998; Kaltschmidt et al. 2001; Mattson et al. 2001; Qiu et al. 2001; Bhakar et al. 2002; Aleyasin et al. 2004; Pizzi et al. 2005). In some cases, these data appear to be confounded by glial contamination in neural cultures (Barger et al. 2005). But reporter constructs that allow visualization of their products in individual cells indicate κB-dependent transcription in neurons. What pathways participate in activation of NFκB and NFκB-responsive genes during these instances?

Although NFκB_can is not appreciably induced in primary neurons by glutamate, TLR ligands, or proinflammatory cytokines, p50, RelA, and IκBα are expressed at similar levels in neurons and glia (Jarosinski et al. 2001). Expression of these molecules is consistent with a potential role for NFκB in neurons. To wit, the levels of IκBα can be reduced pharmacologically by treatment with inhibitors of translation (Jarosinski et al. 2001). Multiple translational inhibitors lead to a relatively slow decline in IκBα levels over a four-hour period. This is in sharp contrast to astrocytes, where a rapid degradation of IκBα occurs within fifteen minutes of application of TNF and IL-1β, presumably through the canonical pathway (Jarosinski et al. 2001). Concomitant with the slow decline of neuronal IκBα caused by translation inhibitors, a gradual increase is observed in a DNA-binding activity containing RelA; an increase in IRF-1 mRNA is also observed under these conditions. Therefore, the potential for activation of NFκB exists in neurons, particularly in response to severe translation inhibition.

Natural pathways to translational inhibition of IκBα and activation of NFκB in neurons

There are natural stress conditions in which translation may be inhibited, including irradiation, unfolded protein response, and essential amino acid starvation (Anderson and Kedersha 2002). Protein translation is rapidly inhibited by amino acid starvation, for instance, via the activity of the eIF2α kinase GCN2 (Narasimhan et al. 2004). IκBα may be particularly sensitive to this process, leading to induction of NFκB (Jiang and Wek 2005). Similarly, removal of the essential amino acid arginine from growth medium of primary astrocytes resulted in a slow decay in IκBα content, peaking at two hours (Figure 3). Concomitant with this decrease, DNA binding by NFκB_can and trace amounts of RelA₂ were observed in the nucleus in astrocytes, detected by EMSA using an iNOS-κB probe (Figure 3). When neurons and astrocytes were analyzed in parallel samples, four hours of arginine depletion depressed IκBα levels and induced κB DNA-binding activity similarly in both cell types (Figure 4). Interestingly, the migration and composition of the binding activity differed between the cell types, with astrocytes showing activation of NFκB_can but neurons showing activation of RelA₂ only. Thus, the amino acid starvation pathway may lead to preferential activation of NFκB in astrocytes but of RelA₂ in neurons, as was seen with dsRNA and LPS.

Figure 3. Induction of IκBα degradation and NFκB nuclear binding activity in astrocytes by amino acid starvation.

Astrocytes were treated with standard medium or with arginine-free medium for 0.5 to 4 h. Nuclear extracts were analyzed for NFκB binding activity with the iNOS-κB probe (5'-AACTGGGGACTCTCCCTT-3'). This probe binds primarily to p50:RelA heterodimers (NFκB_can) but also shows some affinity for RelA₂.

Figure 4. Preferential induction of different κB binding activities by amino acid starvation in neurons and astrocytes.

Purified astrocyte and neuron primary cultures were treated with either standard or arginine-free medium for 4 h. Cytosolic extracts were tested for IκBα degradation western blot analysis, and nuclear extracts were tested for NFκB induction by EMSA. Neurons show activation of RelA₂ while astrocytes show activation of NFκB_can.

In addition to the activation of RelA₂, neurons may control transcription of some κB elements through Sp1-related factors. Sp1, Sp3, and Sp4 are present at relatively high levels in the CNS and bind DNA sequences rich in poly(G) and poly(C), as are most κB elements. Recent immunofluorescence analysis of hippocampal cultures localized Sp1 and Sp3 to astrocytes; neurons, on the other hand, appeared to express only Sp3 and Sp4 abundantly (Figure 5). Corroborating evidence was obtained with western blot analysis and reverse-transcriptase polymerase chain reaction (Mao and Barger 2005). The first indication of neuronal Sp-family proteins binding to κB elements came from EMSA, where these proteins were identified as the most prominent factors constitutively binding a κB element from the immunoglobulin and HIV promoters (Mao et al. 2002). Functional relevance was provided by reporter-gene assays, where manipulation of Sp-family factors was shown to influence expression dependent on a κB element (Mao et al. 2002; Barger et al. 2005).

Figure 5. Sp1 is glial and Sp4 is neuronal.

Mixed neuron/glia primary cocultures from rat hippocampus were fixed and subjected to immunofluorescence with FITC (green)-conjugated antibodies to Sp1 (right) or Sp4 (left), combined with Texas red-conjugated antibodies to GFAP (top) or MAP2 (bottom). Colocalization appears yellow. Note the restriction of Sp1 from neurons (MAP2⁺) and the restriction of Sp4 from astrocytes (GFAP⁺). Both cell types expressed Sp3 (not shown).

Sequence selectivity of κB binding factors

Sp1 and related factors from neuronal nuclei were initially found to bind the κB element contained in the immunoglobulin/HIV promoter: AGTTGAGGGGACTTTCCCAGGC (NFκB target underlined) (Mao et al. 2002). Because this widely used, commercially available probe also shows extraneous binding by RBP-Jκ (this factor's target sequence is partially in the flanking region and not an essential component of the NFκB consensus) (Mao et al. 2002), it was important to characterize the components of the probe that permitted Sp-factor binding. Through substitution of individual nucleotides in competition assays, it was determined that neuronal Sp1-related proteins bind to a subset of κB elements: those containing four guanosines, a spacer sequence of three to five nucleotides, followed by three cytosines (Mao et al. 2005). Alteration of either of the guanines or even the 3' cytosine abrogated binding by Sp1-related factors (Table I). However, alterations of the two most distal nucleotides (the 5' G or the 3' C) still permit binding by NFκB_can.

TABLE I.

Selectivity for Sp1-related factors or NFκB

Name of element	Sequence of κB element	Sp1 binding ?	NFκB binding ?
Ig/HIV-κB	AGTTGAGGGGACTTTCCCAGGC	+++	+++
Ig/HIV-κB-mut1	AGTTGAcGGGACTTTCCCAGGC	+	+++
Ig/HIV-κB-mut2	ATTGGGGACTTTCCAGGC	+	+++
APP1	GAGACGGGGTTTCACCGTGTT	−	+++
APP2	GCATGGGGCTCCTCCCACCG	+++	+
Bcl-x	GCGGGGGGGACTGCCCAGGGAG	+++	+++
COX-2	GGGAGAGGGGATTCCCTGCGCC	+	+++
SOD2 (human)	AGACTGGGGAATACCCCAGTTGT	+++	+++

It has become clear that different dimer combinations of NFκB family members recognize somewhat different κB sites fitting a loosely palindromic consensus decamer. Thus, p50 homodimers (KBF1) bound to the consensus decamer GGGGATYCCC, and RelA₂ recognized a notably unique sequence with high AT rich sequence at the center of the decamer (GGGRNTTTCC) where Y is a pyrimidine, R is a purine, and N is any nucleotide (Kunsch et al. 1992) (Table II). Binding of NFκB heterodimers was proposed to be most avid to elements with a combination of these features on each half site. Perhaps more important was for the center of the palindrome to be AT-rich (Thanos and Maniatis 1992; Falvo et al. 1995; Thanos and Maniatis 1995b). Interestingly, a number of RelA-responsive anti-apoptotic genes including IAP1, IAP2, Bcl-xL, and Bf1-1/A1 (Burstein and Duckett 2003; Kucharczak et al. 2003) have AT-rich κB sites (Chen et al. 1999) which may preferentially bind to RelA₂ in neurons. It will be important to determine whether these RelA₂-responsive genes are preferentially activated in neurons and are important for neuronal survival under stressful conditions in the CNS.

TABLE II.

Preferred NFκB heterodimer and RelA homodimer κB sites

p50:RelA heterodimers (NFκBcan) and  p50 homodimers (KBF1)	p50:RelA heterodimers (NFκBcan) and  RelA homodimers (Rel2)
MHC class I-κB (GGGGATTCCC)	IFN-β-κB(GGGAAATTCC)
IRF-1-κB (GGGGAATCCC)	iNOS-κB (GGGACTCTCC)
iNOS-κB (GGGACTCTCC)	IRF-2-κB(GGGGATTTCC)

Role of NFκB in Neuronal DNA Damage: Friend or Foe

Neuronal death, central to neurodegenerative and pathological disorders, is initiated by numerous factors. For example, during stroke the death signalling cascade is initiated by lack of oxygen and glucose. In other pathologies, such as Alzheimer's or Parkinson's disease, the initiators are less well defined and likely involve multiple death/stress cues (Hutchins and Barger 1998).

DNA damage is one such death initiator implicated in a number of pathological neurodegenerative conditions such as Parkinson's disease (Alam et al. 1997), Huntington's (Dragunow et al. 1995), Alzheimer's (Overmyer et al. 2000), and amyotrophic lateral sclerosis (Bogdanov et al. 2000). Importantly, evidence from some experimental models shows DNA strand breaks well advanced of DNA fragmentation caused by the apoptotic process (Cui et al. 2000). Impairment of DNA repair mechanisms has also been implicated in neurodegeneration, as well as in developmental defects of the central nervous system (Barnes et al. 1998; Deans et al. 2000). Such evidence supports the idea that DNA damage is among the primary events which cause neuronal death.

Death pathways initiated by DNA damage can be examined in cell culture models of neuronal death initiated by the topoisomerase I inhibitor, camptothecin. Camptothecin induces single-stranded DNA breaks in post-mitotic neurons that initiates a cascade of events ultimately leading to apoptosis. Previous studies have shown that mitochondrial translocation of Bax and release of cytochrome c, followed by activation of the apoptosome, define the irreversible stage of neuronal death in this model (Cregan et al. 1999; Keramaris et al. 2000). The tumor suppressor p53 as well as cell-cycle regulatory pathways involving CDK/E2F/Rb seem to play pivotal role in determining whether the mitochondrial pathway to death is activated (Park et al. 1998). Effective inhibition of each of these upstream pathways substantially delays neuronal death. How these signaling pathways work together to orchestrate neuronal death remains to be elucidated. In addition, these two pathways are clearly not the only sets of events which regulate the conserved mitochondrial pathway of death. A network of cross-talking pro-death and pro-survival signals decides the ultimate survival/death fate of a neuron.

Several lines of evidence suggest activation of the NFκB pathway following DNA damage. Camptothecin evokes degradation of IκBα and elevated DNA binding by NFκB_can in cortical cultures (albeit much lower than that of the Sp1-related factors) (Aleyasin et al. 2004). Inhibition of NFκB with pharmacological agents considerably delays camptothecin-induced neuronal death. Two inhibitors of NFκB—helenalin, a chemical agent that causes alkylation of RelA (Lyss et al. 1998), and caffeic acid phenethyl ester (CAPE)—both showed protection against camptothecin. One must always be cautious about interpretation of data obtained with first-generation pharmacological inhibitors of NFκB. For instance, CAPE suppresses NADPH-dependent events (Hiipakka et al. 2002), lipoxygenases, cyclooxygenase, glutathione Stransferase, xanthine oxidase, matrix metalloproteinase-9, and lipid peroxidation (Chung et al. 2004); helenalin inhibits thioredoxin, glutaredoxin, IMP dehydrogenase, the ribonucleotide reductase complex, and DNA polymerase-α (Hall et al. 1988). However, protection was also afforded by over-expression of a degradation-resistant IκBα mutant (IκBα “super-repressor”); similar protection was observed after reduction of RelA levels through RNA interference techniques. Another pharmacological agent BAY11-7082—an IKK inhibitor—failed to show any protection against camptothecin. But this may simply indicate that camptothecin activates NFκB by an IKK-independent mechanism. Indeed, phosphorylation of serine 32 in IκBα (a critical site for IKK action) was not detected in camptothecin-treated cultures. These data suggest that an unconventional activation of NFκB plays a central role in death of neurons evoked by camptothecin.

A key player in neuronal death induced by DNA damage is p53. The mRNA and protein levels of p53 increase in cortical neurons when treated with camptothecin (Aleyasin et al. 2004). Embryos from mice genetically ablated for p53 are substantially resistant to camptothecin-induced death (Morris et al. 2001). All the NFκB inhibitors capable of promoting neuronal viability in the face of camptothecin also inhibit p53 transcript and protein (Aleyasin et al. 2004). Moreover, transcription of noxa and puma, two downstream targets of p53, are suppressed by helenalin. Together, these data document an effect of NFκB on p53 and its downstream pro-death targets in this model. The IKK inhibitor BAY11-7082, which is not neuroprotective, also fails to prevent activation of noxa and puma. Finally, over-expression of p53 in the presence of helenalin reverses the protective effect of this NFκB inhibitor, favoring a model in which p53 is downstream of NFκB. These results are consistent with previous reports indicating that p53 can be a target of NFκB (Hellin et al. 1998; Kirch et al. 1999).

While the above evidence indicates a pro-death role for NFκB, final interpretations await elucidation of the pro-survival functions of this family of transcription factors. In fact, several pharmacological NFκB inhibitors tend to be neurotoxic in cortical cultures over time. While some of this toxic effect could be due to nonspecific effects on molecules other than NFκB, it is also possible that some tonic level of RelA₂ plays a survival role, as well. Forced expression of IκBα also promotes death of cortical neurons; this approach should inhibit RelA₂ as well as NFκB_can and is less likely be confounded by the nonspecific effects that plague pharmacological approaches. Suppression of RelA by helenalin, which delays neuronal death, also causes morphological damage to neurites (Aleyasin et al. 2004). These data suggest a role for a RelA-containing factor in neuronal survival and neurite maintenance.

The above studies underscore the complexity of NFκB function and suggest that the ultimate biological effect will be dependent on the mechanism by which NFκB is activated/repressed, the subunit composition of the “NFκB” activated, and the incumbant DNA target sites modulated. Thus, it is not surprising that differing technical approaches can color the interpretation of experiments. Culmsee et al. (Culmsee et al. 2003) found that camptothecin inhibits NFκB activity, as measured by association with the transcriptional cofactor p300 or activation of a transgenic luciferase reporter; they also found that inhibition of NFκB has no effect on camptothecin toxicity and actually blocks the action of a neuroprotective agent. Bahkar and colleagues (Bhakar et al. 2002) found that RelA over-expression leads to increased levels of IAPs and protection from camptothecin-induced death. This paradigm likely leads to RelA₂ rather than NFκB_can, and this undoubtedly influences distinct subsets of κB cis elements residing in the promoters of different genes (above). Neurons from animals genetically ablated for RelA (or p50, for that matter) survive well in culture and are no more sensitive nor resistant to neuronal death induced by camptothecin (Aleyasin et al. 2004). While the ever-present concerns about compensatory responses to germ-line deficiency apply, it is also possible that different ways of interfering or activating the NFκB pathway lead to activation of different target genes. Finally, NFκB will certainly have different consequences for neurotoxicity when activated in glia versus in the neurons themselves. This suggests that care must be taken with data concerning NFκB, the interpretation of which may change depending upon how or where one modulates this complex signaling pathway and which dimers of the Rel family are active.

Relative Induction of NFκB and RelA₂ in Neurons: Appropriately Disparate Effects on Viability?

Here, we have described various unique responses of neurons with regard to activation of nuclear κB-binding activities dependent on particular stimuli. These ranged from 1) poor induction of NFκB by proinflammatory cytokines or excitotoxins; 2) preferential induction of RelA₂ by TLR ligands or amino acid starvation; and 3) the induction of NFκB_can by translational blockade or DNA damage. If translational blockade and DNA damage induce NFκB_can but milder physiological stresses induce RelA₂ in neurons, then is there a functional difference in response to these insults? It seems likely that transient induction of RelA₂ in response to amino acid starvation or through other related pathways that impact translation, such as an unfolded protein response, is neuroprotective. However, translational blockade may have a number of detrimental consequences, including activation of NFκB_can and induction of multiple pro-inflammatory and pro-apoptotic genes like p53. Mechanistically, this scenario could involve pathways related to the eIF2α kinase known as PERK, which mediates translational inhibition during the ER stress response. Transient activation of PERK, like GCN2, can lead to NFκB_can activation during ER stress and is thought to be important in the induction of anti-apoptotic responses (Jiang et al. 2003a; Deng et al. 2004). Conversely, chronic PERK-mediated phosphorylation of eIF2α has been implicated in neuronal death in Alzheimer's disease or in response to transient focal ischemia in the CNS (Kumar et al. 2001; Mengesdorf et al. 2002; Owen et al. 2005). These untoward effects of PERK could involve persistent translational inhibition, but the role for NFκB_can activation was not addressed in these models. As summarized in Figure 6, translational inhibition and other stress pathways may be a double-edged sword in neuron-specific survival and cell-death responses; modest, survivable stresses could shore up cellular defenses through induction of Rel₂-responsive genes, while irrecoverable errors such as widespread DNA strandbreaks or severe translation poisons would initiate apoptosis (and possibly even immunological elimination) of the affected neuron through autonomous activation of NFκB_can.

Figure 6. Induction of RelA homodimers via physiologic translational pathways in neurons: A preferential pathway in neurons?

Several stimuli, coupled to distinct initial signal transduction events, may converge on the phosphorylation of eIF2α, irrespective of cell type. The phosphorylation of eIF2α can act as a switch, preferentially inhibiting the translation of specific mRNAs; one of those dramatically affected is IκBα. The reduction of this inhibitor tends to be permissive for NFκB activity, but the particular dimer activated depends in some manner on cell type. Distinct dimers also have preferential affinity for different DNA sequences, and particular DNA sequences may be grouped in functionally related genes. For example, sequences with preferential affinity for Rel₂ may be present in important prosurvival genes, whereas NFκB_can binds the promoters of genes connected to a more generalized immune system activation.

An important unresolved issue concerns the mechanism of selective activation of RelA₂ homodimers in neurons following exposure to dsRNA, LPS, or amino acid starvation. In nonneuronal cells, both dsRNA and LPS are known to induce high levels IFN-β through Toll/IL-1 receptor (TIR) domain-containing adapter proteins (Beutler et al. 2005; Sen and Sarkar 2005). The TIR-containing adapter used by TLR3 and TLR4 in IFN-β induction is called “TIR domain-containing adapter inducing IFN-β” (TRIF; aka, TICAM-1) (Yamamoto et al. 2002) (Figure 6). TRIF is able to effect the rapid, IKK-dependent phosphorylation/degradation of IκBα, activating NFκB_can. However, dsRNA binding to TLR3 can also lead to activation of the eIF2α kinase protein kinase R (PKR) and activate κB binding (Jiang et al. 2003b). The identity of the proteins responsible for this κB binding was not determined; nor was its reliance on IKK. One intriguing possibility is that activation of PKR preferentially activates RelA₂ by acting on eIF2α to bring about translational inhibition of IκBα, thus initiating an IKK-independent pathway similar to amino acid starvation. Nevertheless, the ability of eIF2α kinases (PKR, PERK, or GCN2) to trigger activation of NFκB_can in nonneuronal cells suggests that an additional level of regulation is involved in bringing about the neuron-specific activation of RelA₂ in response to physiological stimuli.