Address correspondence and reprint requests to Dr. L. Vitkovic at CNRS-INSERM Centre de Pharmacologie-Endocrinologie, 141 rue de la Cardonille, 34094 Montpellier Cedex 5, France. E-mail: firstname.lastname@example.org
Abstract: If cytokines are constitutively expressed by and act on neurons in normal adult brain, then we may have to modify our current view that they are predominantly inflammatory mediators. We critically reviewed the literature to determine whether we could find experimental basis for such a modification. We focused on two “proinflammatory” cytokines, interleukin (IL)-1 and tumor necrosis factor-α (TNFα) because they have been most thoroughly investigated in shaping our current thinking. Evidence, although equivocal, indicates that the genes coding for these cytokines and their accessory proteins are expressed by neurons, in addition to glial cells, in normal brain. Their expression is region- and cell type-specific. Furthermore, bioactive cytokines have been extracted from various regions of normal brain. The cytokines’ receptors selectively are present on all neural cell types, rendering them responsive to cytokine signaling. Blocking their action modifies multiple neural “house-keeping” functions. For example, blocking IL-1 or TNFα by several independent means alters regulation of sleep. This indicates that these cytokines likely modulate in the brain behavior of a normal organism. In addition, these cytokines are likely involved in synaptic plasticity, neural transmission, and Ca2+ signaling. Thus, the evidence strongly suggests that these cytokines perform neural functions in normal brain. We therefore propose that they should be thought of as neuromodulators in addition to inflammatory mediators.
Cytokines are potent, multifunctional, pleiotropic proteins first discovered in the context of cellular activation and cell-to-cell communication in the immune system. Some cytokines are thought of as “proinflammatory” and others as “antiinflammatory,” depending on the sum total of their effects on immunocytes. The first direct evidence of the presence and activity of cytokines in the brain appeared a decade or so ago. It concerned the “proinflammatory” cytokines interleukin (IL)-1β immunoreactivity (ir) in human and IL-1β and tumor necrosis factor-α (TNFα) biological activity in rat brain, respectively (Breder et al., 1988; Plata-Salaman et al., 1988). Most of the ensuing experimentation was predominantly directed at understanding pathophysiological processes often involving cells of the immune system. This led to a view that cytokine expression in the brain only occurs in response to an infection, insult, or other perturbations and often, if not always, involves immunocytes or nervous-immune system interactions. The view that inflammatory cytokines are only expressed in the brain in disease conditions is currently being revised with the accumulation of data indicating that cytokines are expressed in “normal” adult brain in the absence of pathophysiological stimuli. Usually “normal” implies healthy, but what is a healthy brain? Healthy does not mean inactive or nonreactive. Specific physiological activities appear to be accompanied by cytokine expression, indeed dependent on this expression, although these activities cannot be considered pathological. Constitutive expression does not mean necessarily permanent. Cytokine expression may be low or even undetectable, can moderately rise after some specific physiological stimulation, and may become highly elevated in a disease.
This article reviews published data about constitutive expression and activity of inflammatory cytokines in the brain and evaluates this evidence in the context of well-known up-regulation/induction of cytokine expression in disease conditions. All experimental procedures perturb the brain and may thereby cause up-regulation/induction of cytokine expression. For example, isolation and culturing change the phenotype of glial cells, and preparation of slices causes tissue trauma. Unfortunately, noninvasive methods for detection and quantification of cytokines have not been devised. Thus, the data reviewed here must be interpreted with caution. We place the most weight on the observations corroborated both in vivo and in vitro and the least weight on observations obtained with cell lines in the absence of corroborating evidence from other paradigms. We conclude that genes coding for inflammatory cytokines and proteins required for their activity are expressed in some regions of the normal adult brain. This expression is not only cytokine- and region-specific; it is also cell type-specific. This indicates that some cytokines function in normal brain. Whereas few working in this field now doubt the significant neuromodulatory actions exerted by the inflammatory cytokines, some believe that there is little or no evidence that inflammatory cytokines are expressed in normal brain. The purpose of this review is to critically examine relevant published data and, based on experimental evidence, to foster thinking about inflammatory cytokines in the brain as neuromodulators.
ARE INFLAMMATORY CYTOKINES CONSTITUTIVELY EXPRESSED IN NORMAL BRAIN?
The IL-1 family of cytokines is composed of proteins encoded by different genes: IL-1α, IL-1β, IL-1 receptor (IL-1R) antagonist (IL-1Ra), and two IL-1R accessory proteins (IL-1R AcP I and II). Their expression by neural cells has been demonstrated in vitro, but the results concerning their expression in normal brain are not in complete agreement with each other.
IL-1R AcP I and II mRNAs were detected by RNase protection assay in normal rat brain in all regions tested (hypothalamus, cortex, hippocampus, and cerebellum) at approximately equal levels. The amount of IL-1R AcP I mRNA far exceeded that of IL-1R AcP II mRNA (Ilyin et al., 1998; Plata-Salaman et al., 1998a,b,c)). IL-1R AcP I mRNA was also detected by PCR in mesencephalon/pons in addition to the regions cited above (Taishi et al., 1998).
In summary, nine reports cite detection of mRNAs of IL-1 family members by PCR, and four do not (Table 1). In contrast, the detection by ISH is more even: Three reports cite detection, and three do not. This suggests that these mRNAs are either present at a limit of detection or absent in normal brain.
Table 1. Selected references referring to the presence or the absence of IL-1 in normal CNS
IL-1 proteins. Breder et al. (1988) were the first to demonstrate in a landmark article that IL-1 ir is present in normal human hypothalamus, which we confirmed and extended to the cortex (Fig. 1) (da Cunha et al., 1993a,b). We actually showed that ir for both isoforms of IL-1, α and β, was reproducibly detected in normal human cerebral cortex in glial cells and not neurons. The detection was with two antibodies, each from different sources (da Cunha et al., 1993a,b). Others, using various technical approaches, also have detected ir for both IL-1 isoforms in brain of the mouse, rat, and human. IL-1α ir was detected in neurons in rat hypothalamus by immunohistochemistry (IHC) (Rettori et al., 1994). This suggests that neurons and glial cells express IL-1α protein and that cell type specificity of its expression depends on brain region. Detection of the IL-1β ir in undisturbed brain by IHC has been reported in the pioneering articles of Lechan et al. (1990) and Bandtlow et al. (1990). The IL-1β ir also was detected in neuronal cell bodies and processes of porcine hypothalamus (Molenaar et al., 1993). The presence of IL-1β revealed by IHC was confirmed by a sensitive immunoassay of both rat and mouse hypothalamus (Hagan et al., 1993). In addition, the cytokine itself was detected and quantified by a bioassay and ELISA in extracts from various regions of the rat brain (Quan et al., 1996). This was recently independently confirmed by ELISA in extracts from rat brain (Murray and Lynch, 1998; Nguyen et al., 1998). Collectively, these results indicate that IL-1β protein is (a) present in normal brain of several species, (b) present in neurons in the hypothalamus, and (c) biologically active.
In contrast, Fontana et al. (1984) did not detect IL-1β biological activity, and two groups also did not detect ir in normal rat brain except in a few perivascular cells (Van Dam et al., 1995; Holmin et al., 1997). IL-1α ir was not detected in normal mouse brain, although it was readily detectable after a stab wound to the hippocampus (Fig. 2) (Tchélingérian et al., 1993). Thus, the evidence indicates that IL-1β protein is present in some areas of normal brain, such as hypothalamus, hippocampus, cerebral cortex, and thalamus, in both neurons and glial cells, depending on the region, and that it may be biologically active. The evidence for the presence of IL-1α in normal brain is weaker perhaps because fewer studies have been reported. IL-1α binds to the same receptors as IL-1β but with lower affinity. It is thought to act more as an autocrine rather than a paracrine effector. Expression and function of IL-1α in the brain need to be better characterized.
The third member of the IL-1 family of cytokines, IL-1Ra, was investigated less than the others because reagents for its detection have only recently become available. However, IL-1Ra ir was detected in neurons of human cortex and hippocampus (Yasuhara et al., 1997) and rat hypothalamus (Diana et al., 1999). In rat hypothalamus, IL-1Ra ir and IL-1RI were colocalized in vasopressin-containing magnocellular neurons of the paraventricular nucleus and supraoptic nucleus. IL-1Ra null mice were retarded in their growth after weaning (Horai et al., 1998). Collectively, these studies suggest that IL-Ra protein is constitutively expressed in the brain. Selected references concerning the expression of IL-1α, IL-1β, and IL-1Ra genes are listed in Table 1.
In summary, six studies (two of which used the most sensitive detection method, PCR) did not and six studies (five of which used PCR) did detect TNFα in normal brain tissue (Table 2). This suggests that TNFα mRNA is either present at a low level or absent in normal brain tissue.
Table 2. Selected references referring to the presence or the absence of TNFα in normal CNS
TNFα protein. Breder et al. (1993) detected TNFα ir in the brain of healthy, but colchicine-treated, mice and Ignatowski et al. (1997) detected constitutive expression of TNFα in neurons in the brain of untreated rats. The use of an ELISA technique allowed Saito et al. (1996) to detect TNFα in the striatum, thalamus, and cortex but not in the hippocampus of the gerbil brain. These results were confirmed and extended by Floyd and Krueger (1997). These investigators reproducibly measured TNFα biological activity by bioassay in a soluble fraction extracted from (in order of decreasing concentration) hippocampus, hypothalamus, cerebellum, cortex, midbrain, and pons of the rat brain (n = 8). At the onset of the light period, the TNFα concentration was 275 pg/g of protein in hippocampus and 75 pg/g in pons; at the middle of the dark period, the TNFα concentration was ≈25 pg/g throughout the brain. The TNFα protein concentration displayed a diurnal rhythm that was in phase in all brain regions where biologically active protein was detected. In addition, the mRNA and protein rhythms were in phase with each other (Bredow et al., 1997; Floyd and Krueger, 1997). This indicates that the diurnal translation of TNFα mRNA is temporally coordinated across brain regions. This is consistent with TNFα mRNA being present and available for translation at all times rather than needing to be synthesized before the translation can occur.
The data on diurnal variation of IL-1β and TNFα in the brain indicate that the highest probability of detecting them and their mRNAs is during peak sleep periods, when few experiments are done.
DO CONDITIONS FOR ACTION OF INFLAMMATORY CYTOKINES EXIST IN NORMAL, ADULT BRAIN?
Cytokines can exert their effects when (a) they are biologically active and (b) their receptors are present on target cells. What is the evidence that these conditions are met in normal adult brain? Whereas only few studies have determined the first condition, many have investigated the second. IL-1 is the most investigated cytokine in this regard.
Presence of biologically active proteins
Inflammatory cytokines are produced in a proprotein form and are processed by converting enzymes before they can bind to their receptors. IL-1 converting enzyme has been detected in rodent brain at both mRNA and protein levels (Layé et al., 1996; Tingsborg et al., 1996). This is consistent with the presence of biologically active IL-1β extracted from several regions of normal rat brain (Quan et al., 1996) and supports the view that a biologically active IL-1β exists in rodent brain.
TNFα converting enzyme cleaves membrane-bound TNFα to release soluble cytokine (Black et al., 1997; Moss et al., 1997). Whether or not TNFα converting enzyme is expressed in the brain has apparently not been reported.
Expression of cytokine receptors in brain
In contrast to cytokine activation, “receptorology” of inflammatory cytokines is emerging, and for IL-1 it has been extensively investigated.
IL-1Rs. We briefly summarize the major findings about the presence of IL-1Rs in the brain because this subject has recently been reviewed (Loddick et al., 1998). Two types of IL-1Rs are presently recognized. Type I is the major receptor on cell surface, binds both isoforms of IL-1, and transduces a signal. Type II IL-1R does not transduce the signal. Its soluble form acts as a decoy by sequestering the active ligand (Colotta et al., 1993). Neutralizing this receptor potentiates the effects of exogenously administered IL-1β binding to IL-1RI. Thus, IL-1RII, together with IL-1Ra, negatively regulates the biological effects of IL-1 (Arend, 1991). Farrar et al. (1987) first reported IL-1 binding sites in rodent brain. These results were confirmed and extended by several groups using binding techniques (Takao et al., 1990; Ban et al., 1991), ISH (Cunningham et al., 1992; Wong and Licinio, 1994; Yabuuchi et al., 1994; Ericsson et al., 1995), RNase protection assay (Gayle et al., 1998), or PCR amplification (Parnet et al., 1994; Gabellec et al., 1996). IL-1RII mRNA was not detected in normal rat brain (Nishiyori et al., 1997). However, both types of IL-1Rs were detected in the mouse brain, mainly in hippocampus and hypothalamus (Parnet et al., 1994). The expression of these receptors by neurons has never been unequivocally demonstrated (by, for example, double labeling for a receptor and a cell type-specific marker). However, neuronal expression is strongly suggested by their distribution in the hippocampus and dentate gyrus (Cunningham et al., 1991) and also by their expression in neuronal cell lines (Parnet et al., 1994). The distribution of IL-1RI in rat brain seems to be quite different from that in the mouse (Ericsson et al., 1995). IL-RI mRNA is present mainly in different populations of neurons and notably absent from granule cells in the dentate gyrus. The functional significance of IL-1RI in the brain has been demonstrated with behavioral readouts. For example, blocking IL-1RI, but not RII, specifically abrogated IL-1β-induced decrease in social exploration (Cremona et al., 1998a) and feeding (Swiergiel et al., 1997). Similarly, blocking IL-1RII potentiated IL-1β-induced anorexia (Cremona et al., 1998b). These data strongly suggest that the signal-transducing IL-1R, IL-1RI, is constitutively expressed on some neurons and possibly glial cells in rodent brain.
TNFα receptors (TNFRs). In the immune system, TNFα acts via two types of receptors. Type I (TNFRI) is involved in cell death of some immune cell types (Tartaglia et al., 1991, 1993). Death induction was the property that initially served to define TNFα and may hold a key to understanding its neural effects. The precise function of the type II receptor (TNFRII) is still obscure. Binding sites for TNFα were first identified in brain homogenates (Kinouchi et al., 1991; Wolvers et al., 1993). TNFRI was detected on dopaminergic neurons in autopsied brains from parkinsonian and control subjects (Boka et al., 1994), cultured rat neurons from different cerebral regions (Cheng et al., 1994), and neuronal cell lines (Chambaut-Guérin et al., 1995). In normal mouse brain TNFRII was only detected on oligodendrocytes (Tchélingérian et al., 1995), whereas in culture both types of TNFRs were seen on oligodendrocytes (Tchélingérian et al., 1995; Wilt et al., 1995). Both types of TNFRs were demonstrated on neurons in the cortex (Tchélingérian et al., 1995), as well as in the hippocampus, thalamus, mesencephalon, and cerebellum of the mouse (J. Tchélingérian and C. Jacque, unpublished data). Collectively, these data indicate that TNFRs are constitutively expressed on neurons and glial cells in (some) regions of human and rat brain.
Signaling by IL-1Rs and TNFRs. A thorough discussion of signaling pathways activated by stimulation of IL-1Rs and TNFRs is outside the scope of this review. However, emerging data on intracellular integration of signals elicited by these two cytokines are particularly relevant to understanding their interweaving actions in the brain, e.g., in regulating sleep. Thus, we briefly highlight the current excitement in this area. The effects of both cytokines are mediated by two transcription factors: nuclear factor κB (NF-κB) and activating protein 1 (AP-1). AP-1 consists of two subunits, Jun and Fos. Expression of c-fos, an immediate early gene, is extensively characterized in the brain under many conditions and signifies to most neuroscientists “cellular activation.” However, consequences of Fos induction, if any, are rarely investigated. For example, Jun, a requisite partner of Fos in regulating gene expression, is rarely quantified. Thus, the term “cellular activation,” as reflected by induction of Fos, is often devoid of biological content. Characterizing downstream effects of Fos induction in neural cells will bring insight into neural functions of cytokines in addition to other benefits. The interest of neuroscientists in NF-κB is much more recent than that in Fos, but NF-κB has already been hailed as “crucial for glial and neuronal cell function” (O’Neill and Kaltschmidt, 1997). It also consists of two subunits, p50 and p65, and interacts with a protein that regulates its translocation from the cytoplasm to the nucleus, inhibitory factor κB (I-κB). The p65 subunit of NF-κB is apparently constitutively synthesized in the cytoplasm of some hypothalamic neurons as judged by IHC (Fig. 3). It is interesting that these neurons were not but others were stained with an antibody against the p50 subunit (M. Lerner-Natoli et al., unpublished data). NF-κB was present, as judged by IHC, in mouse cortex and hippocampus and in cultured neurons from these areas (Kaltschmidt et al., 1994). It was active, as judged by electrophoretic mobility shift assays, in a nuclear extract from rat cortex (Kaltschmidt et al., 1994). These results are consistent with the mapping of NF-κB in rat brain by IHC (Joseph et al., 1996) and indicate that this transcription factor is constitutively expressed in (some) hypothalamic neurons. In addition, I-κB was detected also in hypothalamus (Joseph et al., 1996). This is consistent with the presence of biologically active IL-1 and IL-1RI in this region and suggests that IL-1 signaling operates in normal hypothalamus. A review on this subject concluded that NF-κB “may participate in normal brain function” (Grilli and Memo, 1999). NF-κB plays an apparently important role in apoptosis (Van Antwerp et al., 1998) and therefore may be a key to understanding apparently contradictory effects of TNFα on neural cell survival. NF-κB and AP-1 signal transduction pathways apparently are linked to (some) mitogen-activated protein kinase kinases by a new class of proteins known as TNF receptor-associated factors. Thus, IL-1 and TNFα-elicited signal transduction pathways are integrated by mitogen-activated protein kinase kinases “into an array of parallel and common signaling cascades” (Eder, 1997). This suggests that IL-1 and TNFα signals are integrated within the cell, which may be relevant to understanding how they interact in regulating some neural functions such as sleep.
Published experimental results consistently indicate that the receptors for inflammatory cytokines considered here are constitutively expressed on neurons and glial cells in normal brain. Constitutive expression of cytokine receptors by neural cells renders these cells sensitive to the presence of cytokines, even at a very low level. Thus, these cytokines, whatever their cellular source—immune cells, glia, or neurons—are likely to act on neural cells in general and neurons in particular. What is then the evidence for a direct cytokine-neuron interaction?
Evidence for neural functions of inflammatory cytokines at the systems level
This evidence is abundant, strong, and convincing, especially for IL-1. IL-1 is involved in the regulation of the hypothalamic-pituitary-adrenal axis (Turnbull and Rivier, 1999), fever (Kluger et al., 1998), feeding (Plata-Salaman, 1998), sickness behavior (Dantzer et al., 1998), and sleep (Krueger et al., 1998). The mechanisms underlying these behaviors and involving IL-1β differ from each other. Among these processes, sleep without a doubt occurs in the normal, healthy, adult organism. We therefore briefly summarize, as an example, evidence for the role of the two proinflammatory cytokines in the regulation of sleep (for review, see Krueger et al., 1998). Complex biochemical cascades regulate sleep. Exogenous IL-1β or TNFα increases non-rapid eye movement (NREM) in five species, including rodents and primates. Inhibiting their activity reduces spontaneous sleep (Krueger et al., 1998). For example, the somnogenic effects of IL-1β can be completely blocked by IL-1Ra. Sleep rebound in deprivation experiments is affected by intracerebral but not peripheral administration of a soluble fragment of IL-1RI used as a decoy (Takahashi et al., 1997). That IL-1RI plays an important role in the modulation of sleep is evident from the results obtained with IL-1RI null mice (Fang et al., 1998). IL-1RI and TNFRI null mice sleep less than the control animals (Krueger et al., 1998). TNFRI null mice do not respond to exogenous TNFα but do respond to exogenous IL-1β (Fang et al., 1998). This is important because it suggests that IL-1 and TNFα act via parallel somnogenic pathways (Krueger et al., 1998) and may explain why inhibiting IL-1 or TNFα in normal animals produces small effects. The sooner we learn how to measure and interpret such “small” effects, the sooner we will appreciate physiological functions of IL-1/TNFα in normal brain. Diurnal variations in concentrations of IL-1β mRNA and TNFα mRNA and protein, in various brain regions, are associated with the sleep-wake cycle: They reach maxima at peak NREM periods (Krueger et al., 1998). The IL-1β mRNA concentration is higher after sleep deprivation than in controls (Mackiewicz et al., 1996); Taishi et al., 1998). These and other data indicate that “IL-1β and TNFα are key regulatory components of physiologic NREMS” (Krueger et al., 1998). The data on the role of inflammatory cytokines in sleep strongly suggest that these cytokines affect neuronal functions for the purpose of integrating autonomic responses.
Actions at neuronal level
Plata-Salaman and his collaborators were among the first to show that IL-1β and TNFα affect activity of neurons in the brain and that these effects were biologically relevant (Plata-Salaman et al., 1988). In this section, we review the known actions of these two cytokines at the neuronal level and how neuronal activity affects their production. We restrict our review to the data judged to be relevant to understanding the action of these cytokines in normal rather than perturbed brain.
IL-1. Recombinant human IL-1β and TNFα electrophoretically applied to the lateral hypothalamus of the rat suppressed activity of glucose-sensitive neurons (Plata-Salaman et al., 1988). Glucose-insensitive neurons were little affected. (This technique allows direct application of a cytokine onto individual neurons with the simultaneous monitoring of their activity.) Heat-inactivated cytokines and bovine serum albumin had no effect. Peripherally administered cytokines in doses greater than or equal to doses administered centrally also had no effect. These results indicate that IL-1β and TNFα specifically act on glucose-sensitive neurons in the hypothalamus (Plata-Salaman et al., 1988).
This effect was accompanied by inhibition of feeding consistent with the well-documented function of these cytokines in regulation of feeding. IL-1β administered intracerebroventricularly (0.5-0.8 ng/24 h for 72 h) decreased levels of the G protein Goα in ventromedial nucleus of the hypothalamus as it induced anorexia (Plata-Salaman et al., 1998c). This protein was previously shown by the same authors to be involved in the control of normal feeding, suggesting a mechanism for IL-1β action.
In addition, IL-1 likely stimulates vasopressin synthesis and release via IL-1RI in magnocellular neurons in the paraventricular nucleus and supraoptic nucleus of the hypothalamus (Diana et al., 1999). This is consistent with the role of IL-1β in temperature regulation because arginine-vasopressin released from bed nucleus of the stria terminalis neurons acts as an antipyretic (Wilkinson et al., 1993).
IL-1β administered either intracerebroventricularly (10-100 ng per rat) or intraperitoneally (1 μg per rat) increased norepinephrine turnover in hypothalamus. TNFα had no such effect. Norepinephrine turnover was unaffected in cortex and medulla oblongata. Corticotropin-releasing factor (CRF) and eicosanoid-cyclooxygenase products mediated this effect (Terao et al., 1993). IL-1β administered intravenously activates CRF-producing neurons in the hypothalamus, and this is controlled by medullary catecholaminergic neurons projecting to the paraventricular nucleus (Ericsson et al., 1994). This suggests that IL-1β specifically affects both catecholaminergic and noradrenergic neurons projecting to the hypothalamus.
In addition to specific neurons in the hypothalamus, IL-1β markedly excited for a long time 24% of the neurons in the bed nucleus of the stria terminalis as judged by single unit recordings of identified neurons in urethane-anesthetized rats (Wilkinson et al., 1993). IL-1β (5 and 10 ng/ml) perfused into rat cerebellar slices rapidly exerted an effect on Purkinje cells (Pringle et al., 1996). Stimulation of the area of passage of paralleled fibers in this preparation containing intact neuronal circuitry produced a pure GABAA inhibition of the spontaneous firing of Purkinje cells. IL-1β reduced the duration of inhibition 10 min after beginning the perfusion. It also reduced the effect of exogenous GABA (0.1 mM) with the same time course as in the absence of exogenous GABA. This effect reversed within 15 min of washing out IL-1β. Indomethacin, an inhibitor of prostaglandin second messenger synthesis, had no effect on IL-1β action. Thus, IL-1β can reduce cerebellar GABAA responses, and prostaglandin does not mediate this effect (Pringle et al., 1996).
IL-1β also directly acts on neurons from the hippocampus. An increase in the concentration of endogenous IL-1β, e.g., with age, in rat dentate gyrus was inversely proportional to the long-term potentiation (LTP) in perforant path granule cell synapses. This effect was (proposed to be) indirectly mediated by reactive oxygen species leading to lipid peroxidation leading to a decrease in membrane arachidonic acid content leading to a decrease in LTP (Murray and Lynch, 1998). At concentrations of ≥0.1 pg/μl, the pathophysiological concentration in CSF, IL-1β decreased voltage-gated Ca2+ channel currents in acutely dissociated neurons from the CA1 region of the guinea pig hippocampus (Plata-Salaman and ffrench-Mullen, 1994). IL-1βRa blocked, and IL-6, epidermal growth factor, basic fibroblast growth factor, and lipopolysaccharide had no effect. Together, these studies suggest that IL-1β directly acts on some neurons modulating their electrophysiological activity.
IL-1 production and therefore its actions are, in turn, regulated by neurons. Acute rat hypothalamic slices rapidly released IL-1 ir into the medium (Tringali et al., 1997). The release was inhibited by acetylcholine and histamine and stimulated by dopamine and CRF. This effect was specific in that these slices did not release IL-10 or IL-4. CRF stimulated the IL-1 release in a dose-dependent manner. Thus, CRF not only mediates the action of IL-1 but also stimulates its release by a positive feedback loop. These data indicate that IL-1 may exist in a pool in cells (neurons?) of the hypothalamus and can be released in a manner regulated by neurotransmitters and neurohormones.
LTP stimulated IL-1β gene expression in hippocampus of freely moving rats and in hippocampal slices (Schneider et al., 1998). This effect was long-lasting, specific to LTP, and diminished when LTP was blocked. The effect was measured by PCR but could not be detected by ISH. ISH detected an increase in IL-1β mRNA 2 h but not 5 h after slicing. This transient increase in IL-1β mRNA was probably caused by tissue injury. Thus, all measurements on slices were done 5 h after slicing and 1 week after surgery in animals (Schneider et al., 1998). (This reflects awareness of these investigators that cytokine expression is very susceptible to perturbations and shows that such artifacts can be avoided.) These results indicate that IL-1 gene expression may be stimulated by a sustained increase in the activity of a discrete neuronal population. In turn, IL-1Ra (administered intracerebroventricularly or in vitro) caused “reversible impairment of LTP maintenance” if administered during the precise time after induction of LTP when an increase in IL-1β mRNA was detected! This effect was independently confirmed and further explored in vitro. The results suggest that IL-1β, at presumed physiological concentrations, acts via mitogen-activated protein kinase p42/44 (Coogan et al., 1999). Furthermore, it has been proposed that IL-1β functions by increasing endogenous production of adenosine (Luk et al., 1999). This is based on the observation that blocking adenosine A1 receptors abrogates “profound decrease of glutamatergic transmission, but not GABAergic inhibition, in hippocampal CA1 pyramidal neurons” in brain slices, elicited by IL-1β at subfemtomolar concentrations (Luk et al., 1999). How exactly all this fits together is yet to be determined, but the results clearly indicate that IL-1 may be involved in some forms of synaptic plasticity (Zhao and Schwartz, 1998). The LTP model of neuronal activation is of particular value as it correlates with learning and memory processes. These processes occur in so-called normal, healthy brain.
TNFα. The actions of IL-1β and TNFα interweave at the neuronal, in addition to the systems, level. TNFα also suppressed activity of glucose-sensitive neurons in the lateral hypothalamus (Plata-Salaman et al., 1988). This effect was similar to the effect of IL-1β and consistent with suppression of food intake affected by these cytokines when administered intracerebroventricularly. TNFα inhibited neuronal norepinephrine release in slices of median eminence (Elenkov et al., 1992) and hippocampus (Ignatowski and Spengler, 1994; Ignatowski et al., 1996, 1997). The latter effect was dose-dependent and was potentiated by the α2-adrenergic receptor antagonist idazoxan. However, TNFα had exactly the opposite effect when added after long-lasting (14-day) treatment with izipramine, an α2-adrenergic receptor agonist (Ignatowski and Spengler, 1994). These results suggested to the authors that “TNFα alters presynaptic α2-adrenergic receptor responsiveness.” TNFα also may act on neurons through modifying ion channel permeability. Specifically, it enhanced the transient outward K+ current in cultured rat cortical neurons (Houzen et al., 1997) and modulated the Ca2+ current in sympathetic neurons (Soliven and Albert, 1992). In cultured rat hippocampal neurons, TNFα increased Ca2+ current density through L-type voltage-gated channels but decreased current elicited by stimulation of glutamate receptors (Furukawa and Mattson, 1998). These latter effects were likely indirect because they occurred in 24-48 h rather than 2.0 h. Thus, the effects of TNFα on neurons are cell type-specific and analogous, although not identical, to the effects of IL-1β.
The data together suggest that TNFα exerts pleiotropic effects on neurons dependent on the state of target cell in addition to other factors. This should be considered in evaluating apparently contradictory reports of TNFα activity in in vitro models of pathophysiology. The following effects have been reported: toxic (Gelbard et al., 1993; Talley et al., 1995), protective (Cheng et al., 1994; Barger et al., 1995), and none on survival but some on neural transmission (Barone et al., 1997). The toxic effect of TNFα on human neurons was mediated by glutamate α-amino-3-hydroxy-5-methylisoxazole-4-propionate (AMPA) receptors (Gelbard et al., 1993) and blocked by the antioxidant N-acetylcysteine and bcl-2 and cruA gene products (Talley et al., 1995). The neuroprotective effect was mediated through induction of a calcium binding protein (Cheng et al., 1994) and/or Mn2+-dependent superoxide dismutase (Mattson et al., 1997). By injecting TNFα into the brain or blocking the effects of endogenous TNFα, Barone et al. (1997) found that this cytokine is toxic during ischemia. In contrast, TNFR null mice were more vulnerable to ischemia and excitotoxicity than their wild-type counterparts (Bruce et al., 1996). A possible participation of TNFα in excitotoxic events is suggested by the observations of the Bock et al. (1996, 1998) showing production and release of this cytokine in the hippocampus following epileptic seizures and intraamygdala injection of kainate. Neuronal death induced by kainate in organotypic hippocampal cultures did not, whereas the death caused by endotoxic shock did, require TNFα (de Bock et al., 1996, 1998). These apparent contradictions between deleterious and beneficial effects of TNFα on neurons exemplify the complexity of its actions.
Evidence for direct action of inflammatory cytokines in normal brain
The evidence reviewed here, on balance, strongly suggests but does not prove that the “proinflammatory” cytokines IL-1 and TNFα are constitutively expressed and modulate neuronal activities. The evidence of action at the systems level, i.e., regulation of autonomic and behavioral functions, is the most unequivocal, and the presence in tissue is the most equivocal. We discuss first the most equivocal evidence.
Reasons for inconsistencies in detection of cytokine gene expression in normal brain
Inconsistencies in the detection of constitutive expression of inflammatory cytokines in the brain may be due to at least four causes: (a) differential sensitivity of the techniques and sensitivity and specificity of the reagents, (b) overall health status of the subjects, (c) species differences, and (d) bias of investigators.
(a) Absence of detection of a molecule, protein, or mRNA may simply be attributed to a lack of sensitivity of a reagent or of the technique itself. Detection of a molecule by one group and not by another may be attributed either to higher sensitivity of a technique and/or a reagent or to a false-positivity. Diurnal variations of IL-1β and TNFα (Krueger et al., 1998) may contribute to the discrepancies among groups who likely assayed these molecules at different times of a day. (b) False-positivity can be due to an undetected latent or chronic low-level infection, other “silent” pathologies, stress caused by handling (Shintani et al., 1995; Murray and Lynch, 1998), or aging (Sheng et al., 1998). (c) Species-related differences are well known but poorly understood. For example, differences in regional distribution of IL-1Rs in the brain of rats and mice are well known, but their functional significance has not been investigated. Strain differences also may account for these inconsistencies. Compare, for example, detection of IL-1β mRNA in male Lobund-Wistar (Plata-Salaman et al., 1998a) but not in Wistar (VAF) rats (Plata-Salaman et al., 1998b). (d) Most studies of cytokine expression in the brain have been done to determine the role of cytokines in pathophysiology. Large changes (usually increases) in cytokine expression in a pathological as compared with a control, usually normal condition were easiest to measure and interpret. This approach discouraged careful and sensitive determination of basal expression. When investigators focused on basal expression, they found not only mRNA but also biologically active IL-1β and TNFα (Quan et al., 1996; Floyd and Krueger, 1997). Furthermore, the spatiotemporal variations of these proteins in normal brain were consistent with their biological functions in a normal organism, e.g., sleep, arguing against artifacts. Thus, the “negative bias” is giving way to more thoughtful and careful treatments.
Overcoming the inconsistencies in detection of cytokine gene expression in normal brain
As reagents improved and the interest of investigators increased, detection and quantification of IL-1 in the brain became more rigorous and definitive. The careful work of quantifying IL-1β protein in rat brain (Quan et al., 1996) has been corroborated by equally careful work on its mRNA (Nguyen et al., 1998). However, the mRNA still remains elusive even for Quan and his subsequent collaborators (Quan et al., 1998). Recent publications by Plata-Salaman and his collaborators illustrate the variability in measuring IL-1β mRNA by RNase protection assay. In three publications in 1998, IL-1β mRNA was absent in hippocampus, hypothalamus, and cerebellum (Plata-Salaman et al., 1998b), absent in hippocampus but present in hypothalamus and cerebellum (Ilyin et al., 1998), and present in hypothalamus, cerebellum, and cortex (Plata-Salaman et al., 1998a). Good science does not lie: The basal concentration of IL-1β mRNA is at the detection limit of this assay, and the differences among reports are well within the margin of experimental error (C. R. Plata-Salaman, personal communication). When they could not detect IL-1α mRNA with the RNase protection assay, they detected it with competitive PCR (Gayle et al., 1998). However, PCR should be done on perfused brain tissue to avoid false-positivity from blood cells trapped in cerebrovascular endothelium. Low basal concentration of IL-1 and TNFα mRNAs may be due to several reasons. The first and foremost reason is that their mRNAs are short-lived owing to a nuclease-sensitive consensus sequence in the 3′-noncoding region that is common to several inflammatory mediators (Caput et al., 1986). Thus, once translated, these mRNAs may rapidly disappear from tissue. They may be synthesized only when IL-1 has been recently released or used (Nguyen et al., 1998). They may be detectable only when at diurnal peak concentrations (Krueger et al., 1998) and/or when sufficiently induced by each other (Horai et al., 1998). Thus, lack of detection does not exclude low-level expression of a cytokine or its mRNA. Cytokines usually have high specific activity and therefore are present in low concentrations, making detection difficult. In addition, they are highly labile and rapidly turn over, especially in environments permitting diffusion over long distances, e.g., circulation. In addition, proteins such as IL-1α and TNFα may be difficult to reach in situ because they are bound to a cell or extracellular matrix. These properties make cytokines difficult to detect and even more difficult to quantify. [For further discussion of this topic see Watkins et al. (1999).] In summary, inconsistencies and contradictions in the data on the presence of inflammatory cytokines in normal brain are, at least in part, due to their low concentration (and/or high specific activity) and high turnover, making them very difficult to detect, let alone quantify, in the brain. These inconsistencies are, however, being rapidly overcome by increasing attention to and interest in constitutive expression of cytokines in the brain.
Spatiotemporal and cell type specificity of cytokine expression and action in brain
The evidence indicates exquisite and widespread cell type, regional, spatial, and temporal specificity of expression and action of the cytokines and their accessory molecules in the brain parenchyma. Current data permit only a glimpse at this specificity. In human cerebral cortex, IL-1Ra ir is in neurons (Yasuhara et al., 1997), and both IL-1α ir and IL-1β ir are in glial cells (da Cunha et al., 1993a). This suggests that IL-1Ra acts in a paracrine manner on glia cells and an autocrine manner on neurons. Cell type specificity of cytokine expression needs to be better characterized and corroborated to shed light on their multiple functions. Although microglia make cytokines (da Cunha et al., 1993a; Van Dam et al., 1995; Konsman et al., 1999), so do neural cells (da Cunha et al., 1993a). Thus, the focus on microglial cells may limit our vision of neural, in addition to immune, functions of an inflammatory cytokine in the brain.
Diurnal rhythm in the expression of IL-1β and TNFα genes is probably the most regular and therefore the easiest temporal variation to understand. Superimposed on this rhythm are the mutual control of gene expression of the two IL-1 isoforms, normal physiological variation in CRF levels affecting IL-1 release in parts of hypothalamus, and variation imposed by LTP in hippocampus. Spatiotemporal variation in the synthesis and activity of cytokines is likely to be very complex.
We still lack strong and sufficient evidence for the precise roles of cytokines in normal brain. This may be due to their very low concentration and lack of proper experimental paradigms. Their effects in the brain may be difficult to assess because they are subtle and have not been appropriately looked for. For example, the absence of neurological changes, at least at the level of gross examination, in TNFα null mice (Marino et al., 1997) suggests that this cytokine either plays a subtle role in the CNS or can be replaced by some other molecule(s). The overall effect of this cytokine may result from the addition of numerous convergent and/or divergent effects on different cascades of molecular events. It may depend on local environmental conditions at the moment. Among these conditions is the activation status of the cell governing which receptors, transcription factors, or other molecular intermediates are temporarily available (Pan et al., 1997). These include also the metabolic status of the cell itself and of its neighboring cells. It probably means that discrete local electrical events may modulate its primary effect. Spatiotemporal variation in inflammatory cytokine expression and function is consistent with their neuromodulatory activities in normal brain. Pathophysiological stimuli rapidly elevate cytokine concentrations probably because, in part, the genes are already expressed.
Data reviewed here suggest that the proinflammatory cytokines IL-1 and TNFα are neuromodulators in normal adult brain. This conclusion is based on four lines of evidence. (a) At either mRNA and/or protein level, these cytokines, accessory proteins involved in the regulation of their activity (except TNFα converting enzyme), and their receptors have been detected. Their expression in normal adult brain is in neurons and glial cells, in addition to microglia, and it is regionally specific. Regional distribution is consistent with their neuromodulatory functions. (b) Their neuromodulatory functions are documented, in vivo and in vitro, in several autonomic and behavioral processes that are subserved by central neurons, e.g., sleep, feeding, etc. Modulation of neuronal activities includes the whole spectrum of biological functions from electrophysiological activity to gene expression. (c) They have a direct and/or indirect action on the electrophysiology of central neurons. (d) The activity of central neurons mediates, in part, regulation of their production. For example, their gene expression is regulated, in some regions of the brain, by the circadian clock and for IL-1 by electrical activity of some hippocampal neuronal cell populations.
Thus, neural cells in the brain constitutively synthesize the “proinflammatory” cytokines IL-1 and TNFα, and these cytokines likely perform “neural” functions. This has been previously suggested by others and ourselves (Blatteis, 1990; Goetzl et al., 1990; Plata-Salaman, 1991; Krueger and Majde, 1994; Sei et al., 1995; Turnbull and Rivier, 1999), but it has not yet been widely accepted. Viewing cytokines solely as inflammatory mediators that are exclusively involved in pathophysiological processes needs to be broadened. IL-1 and TNFα are neuromodulators in addition to proinflammatory factors. This is consistent with their pleiotropic nature and tissue-specific effects. Determining how these and other cytokines affect neural cells, especially neurons, in vivo and what is the biological significance of these actions are the most important tasks ahead of us.
Note added in proof: Several papers relevant to this review have appeared since its acceptance. We are drawing the readers’ attention to Carrie A., Jun L., Bienvenu T., Vinet M. C., McDonell N., Couvert P., Zemni R., Cardona A., Van Buggenhout G., Frints S., Hamel B., Moraine C., Ropers H. H., Strom T., Howell G. R., Whittaker A., Ross M. T., Kahn A., Fryns J. P., Beldjord C., Marynen P., and Chelly J. (1999) A new member of the IL-1 receptor family highly expressed in hippocampus and involved in X-linked mental retardation. Nat. Genet. 23, 25-31.
We thank the following individuals and institutions: J. J. Jefferson, M. Lerner-Natoli, J.-L. Tchélingérian, F. Le Saux, and C. Colin for experimental data; C. R. Plata-Salaman and G. Rondouin for comments on an earlier version of the manuscript; M. Passama for Figs. 1 and 3; and INSERM, ARSEP (to C.J.), and Elf Aquitaine, FRM (to L.V.) for financial support. L.V. was a Professor of the French Academy of Sciences, Chair Elf Aquitaine.