It is widely accepted that microglial activation contributes to neurodegenerative disease, but the mechanisms by which this occurs remain elusive. The repeated observation that long-term use of nonsteroidal anti-inflammatory drugs (NSAIDs) protects against subsequent development of Alzheimer's disease (AD) (Vlad et al.,2008) and Parkinson's disease (PD) (Chen et al.,2005) has fuelled the idea that inflammation contributes to neurodegeneration and microglial cells have obviously been assumed to be the most likely culprit. However, although NSAIDs have generally been administered to counter peripheral inflammatory conditions such as rheumatoid arthritis, the likely contribution of systemic inflammation to the deterioration of brain function has been little considered. We have known for many years that systemic inflammation influences brain function. Our own behavioral responses to simple infections demonstrate that inflammatory mediators can signal to the brain to evoke significant changes in our behavior and in our metabolism. These changes are mostly adaptive: reorganizing our priorities and preserving energy to mount a fever and fight infection, while suppressing social and motor activity and isolating ourselves from the rest of the herd. However, there is now considerable evidence to suggest that systemic inflammation can have deleterious consequences for the brain if the inflammation is sufficiently severe, or if the brain shows vulnerabilities due to genetic predisposition, ageing, or neurodegenerative disease. This connection between the brain and systemic inflammation heralds new ways in which microglia may contribute to brain pathology. In this review, the author will discuss the accumulating evidence, from both preclinical and clinical studies, that systemic inflammation negatively impacts on chronic neurodegenerative disease. The focus of the review, in keeping with the theme of this issue of GLIA, will be the role of the microglia in these processes, but the clinical manifestations of systemic inflammatory insults will remain the “litmus test” of the relevance of the phenomenon of systemic influence on brain deterioration.
Microglial Activation in Chronic Neurodegeneration
It is obviously not possible to discuss here, en masse, the huge literature on microglial activation in neurodegeneration that has emerged in the last two decades. However, it is useful to identify commonalities in microglial responses in neurodegenerative pathologies before examining how systemic inflammation alters these. Macrophage biology nomenclature has moved in recent years from the designation of macrophages as simply classically (M1) or alternatively activated (M2), toward the recognition of macrophage plasticity and of a dynamic spectrum of activation states that can cover observed phenotypes including classically activated, wound-healing, and regulatory macrophages (Mosser and Edwards,2008). Likewise, microglial cells have now moved on from older, binary, classifications of “quiescent” and “activated.” In addition to macrophage activation simply by exogenous (pattern associated molecular patterns; PAMPs) or endogenous (danger associated molecular patterns; DAMPS) toll-like receptor (TLR) ligands, one can define classically activated macrophages (M1) as being activated by lipopolysaccharide (LPS), in combination with the Th1 cytokine interferon-γ (IFN-γ), to produce a robust proinflammatory profile including interleukin-1β (IL-1β), IL-12, tumour necrosis-α (TNF-α), and inducible nitric oxide synthase (iNOS). Alternatively activated macrophages (M2a) are driven by the influence of IL-4 and IL-13, typically from Th2 cells, resulting in an anti-inflammatory, profibrogenic profile. The M2b phenotype is driven by immune complex formation and TLR or IL-1β activation. Deactivated macrophages (M2c) are typically influenced by transforming growth factor-β (TGF-β), glucocorticoids, IL-10, or CD200 and express relatively low levels of MHC Class II and elevated prostaglandins, contributing to the suppression of proinflammatory cytokines. Given the influence of T cells on these peripheral macrophage phenotypes and the relative paucity of T cells in chronic neurodegenerative disease, it is reasonable to ask whether this nomenclature can be imposed on the brain: do microglial cells, conform to M1, M2 (and M2 a,b,c) phenotypes during neurodegenerative disease? In the experimental autoimmune encephalomyelitis (EAE) model of multiple sclerosis (MS), classically activated M1 phenotypes are well described and IL-4 has also been shown to induce an alternatively activated phenotype in microglial cells (Ponomarev et al.,2007). Since Th2 cells are known to be present in the brain in significant numbers in EAE, this is a plausible route to the M2a phenotype. However, though IL-4 can have a marked influence on microglial activation when directly applied to the brain (Lyons et al.,2007), there remains little direct evidence for IL-4 and IL-13 expression in the brain during other chronic neurodegenerative diseases. With respect to the M1 phenotype, studies with intracerebral LPS have shown that microglia can clearly adopt an innate activation/M1 phenotype. However, this phenotype is not similar to profiles during most models of chronic neurodegenerative disease in that levels of proinflammatory cytokines induced in chronic disease are much lower than those induced by LPS, despite the universal observation of significantly increased numbers of microglial cells during disease (Fig. 1). In that context one has to question the relevance of using substantial injections/infusions of LPS into the brain to model chronic neurodegenerative disease, as has been done for both Alzheimer's (Hauss-Wegrzyniak et al.,1998) (0.25 μg/h i.c.v. for 28 days) and Parkinson's diseases (Herrera et al.,2000) (2 μg LPS, intranigral). Likewise using severe sepsis/septic shock to produce pathological features of AD and PD, may have told us that particular regions of the brain are particularly sensitive to inflammatory damage (Qin et al.,2007), but they are unlikely to reflect the pathogenesis in those patients who suffer from this disease. It is significant that these models still provide the most unambiguous evidence of M1 microglial phenotypes and of a disease course that is amenable to anti-inflammatory treatment (Hirsch and Hunot,2009). Unlike the very substantial inductions of proinflammatory genes after LPS, studies in transgenic models of AD have reported expression of IL-1β and iNOS of approximately one to threefold, and protein levels have not often been reported (Schwab et al.,2010). There are many potential difficulties in the reliable assessment of cytokine expression in the CNS, as recently discussed (Ransohoff and Perry,2009). Notwithstanding this, there is a continuing perception that macrophages in neurodegenerative disease are of the M1, classically activated phenotype. A recent study of both human AD cases and AD mouse models showed evidence for limited M1 polarization (twofold increase in TNF-α mRNA, and no change in iNOS or IL-1β), alongside fourfold induction of the M2 marker YM1 and lesser inductions of arginase-1 and mannose receptor. These patterns were largely replicated in human AD frontal lobe samples (Colton et al.,2006). The demonstration that amyloid-β (Aβ) can activate the NLRP3 inflammasome (Halle et al.,2008) has sustained the argument that IL-1β is a key player in AD pathogenesis, but it is worth noting that even in that study the inflammasome was first “primed” by treating with LPS. Thus without the LPS-induced supply of pro-IL-1β, Aβ appears to show limited capability to produce significant IL-1β. Such studies probably oversimplify the relationship between Alzheimer's disease and IL-1β, but given the frequency of co-morbidity in the ageing population, the requirement for “multiple hits” (LPS + Aβ) to bring about an M1 phenotype in the brain is perhaps not surprising, and is a key point for discussion below.
One feature that has consistently been reported is that CCL2 (formerly MCP-1), with or without detectable TNF-α, appears consistently up-regulated in multiple models of chronic neurodegeneration: AD (Sly et al.,2001), PD (Sriram et al.,2006), ALS (Henkel et al.,2004), and prion disease (Felton et al.,2005), and this is entirely consistent with the inflammatory component of these diseases being dominated by myeloid lineage cells. MCP1/CCL2 is a key chemokine for microglial activation and monocyte chemoattraction (Fuentes et al.,1995). Though peripheral monocytes have been shown to infiltrate the brain and migrate to amyloid plaques (Simard et al.,2006) after whole body irradiation, it is apparent that when the brain is “protected” using a lead helmet during these irradiation experiments, bone marrow-derived macrophages do not infiltrate the brain in large numbers (Mildner et al.,2011). Irrespective of this, we remain unclear about the provenance of CCL2 expression and microglial activation in AD models: CCL2 overexpression increased microglial number and exacerbated disease (Kiyota et al.,2009) while CCR2−/− mice crossed with Tg2576 (El Khoury et al.,2007) or APP/PS1 (Naert and Rivest,2012) showed impaired microglial accumulation and accelerated Aβ deposition. However, microglial depletion experiments showed that removal of microglial cells for 4 weeks had no impact on Aβ deposits (Grathwohl et al.,2009) and it has recently been shown that CCR2 deficiency impairs the ability of PVMs, rather than microglia, to clear Aβ (Mildner et al.,2011), consistent with older reports that microglia are simply not efficient at Aβ clearance (Wisniewski et al.,1991). Thus microglial activation in AD models is not typified by a robustly activated and phagocytic phenotype. In this regard it is also significant that both inducible over expression of IL-1β in the APP/PS1 AD model (Shaftel et al.,2007) or intrahippocampal LPS injection (DiCarlo et al.,2001; Herber et al.,2004) significantly improve Aβ clearance, indicating that phenotypic switching occurs in these mice upon secondary stimulation. Thus, although there is some evidence that these phenotypes can shift from M2-type to more M1-like with advancing age in AD transgenics (Jimenez et al.,2008) it is clear that microglia are not operating in a fully M1 mode during chronic neurodegenerative disease per se. The microglia have potential that is somehow suppressed. Microglia in prion disease also show only limited evidence of IL-1β and TNF-α expression, but instead appear to be dominated by TGFβ1 and PGE2 (Cunningham et al.,2002; Minghetti et al.,2000; Walsh et al.,2001). This is consistent with the idea of a microglial cell engaged in the phagocytosis of apoptotic cells, as is known to occur in prion disease and other neurodegenerative diseases. Peripheral macrophages engaged in phagocytosis of apoptotic cells synthesize TGFβ1 and PGE2 and directly suppress expression of IL-1β and TNF-α (Fadok et al.,1998) and we have now demonstrated that microglial cells remain IL-1β negative while phagocytosing apoptotic cells (Fig. 1) in the hippocampus, in vivo (Hughes et al.,2010). It seems reasonable to suggest that these microglia show features of the M2c phenotype, but as will be discussed below, these phenotypes are not static but change with subsequent stimulation. The cells depicted in Fig. 1 are also consistent with the demonstration of a role for the inhibiting receptor TREM2 in noninflammatory phagocytosis (Takahashi et al.,2005), thus controlling inflammation in the midst of neurodegeneration. Recent data that prion disease progression is more rapid in mice lacking Mfge8, an important protein linking apoptotic cells to the phagocytic machinery (Kranich et al.,2010) would appear to underline the importance of effective clearance of apoptotic bodies.
Phenotypic Switching: Microglia are Primed by Neurodegenerative Disease
Microglial activation, per se, may contribute to damage during chronic neurodegeneration but it is essential to make the point that the profile of microglia in most models of chronic neurodegenerative disease is muted compared with what these cells can produce (Fig. 1), and it seems reasonable to suggest that their default mode is the clearance of debris with the minimal disturbance and damage to the tissue. There is clearly a level of control exerted by other factors that allows these cells to remain in a state of partial activation. We discovered that this state of activation in prion disease could be significantly altered by subsequent systemic inflammatory insults, and that this had functional and pathological consequences for the brain. These first experiments lead to the concept of microglial priming: that the brain is “primed” by chronic CNS disease, to show exaggerated responses to subsequent inflammatory stimulation, whether of a systemic or central origin (Combrinck et al.,2002) and we later demonstrated that it was the microglia that are primed and are responsible for this exaggerated response (Cunningham et al.,2005b). The terminology we adopted at that time (Perry et al.,2007) was derived from the original description of peripheral macrophage priming (Johnson et al.,1983) because our observations matched those of the original studies in that when the macrophage was primed, by IFN-γ in the original studies, the cells then responded to subsequent activation with LPS by expressing significant levels of iNOS (Cunningham et al.,2005b). Direct application of LPS to the normal brain induced detectable IL-1β, but iNOS was absent and subsequent neutrophil infiltration was very limited. However, application of LPS to the prion-diseased brain produced abundant microglial IL-1β and iNOS expression and overwhelming neutrophil infiltration (Cunningham et al.,2005b). We did not observe any consistent morphological change between “primed” and phenotypically switched microglia. Microglial cells of the diseased brain thus undergo phenotypic switching upon subsequent inflammatory stimulation, changing from perhaps an M2c phenotype to a more M1 phenotype, without obvious morphological change. We showed, in the same study, that this exaggerated microglial IL-1β response could also be demonstrated after systemic inflammatory activation with LPS, but this intracerebral, proof of concept, experiment remains the clearest in vivo demonstration that it is the microglial cell, rather than some other link in the chain linking the periphery to the brain, that is responsible for the exaggerated CNS inflammatory response. Since those experiments, exaggerated inflammatory responses to systemic inflammation have been reported in ageing and in models of Alzheimer's disease, Parkinson's disease, multiple sclerosis, ALS, stroke, and Wallerian degeneration. Thus it seems likely that priming is a generic phenomenon of the microglial response to pathology. Differences in the nature of activation during the primary pathology may lead to different manifestations postsystemic insult, but these insults certainly have consequences for the underlying disease process. Since these primed cells reside in areas of existing pathology, the deleterious effects of systemic inflammation are “targeted” to brain regions that are already vulnerable, thus exacerbating functions already impaired by disease.
Our use of the term priming to describe the nature of the microglial phenotype in the neurodegenerating brain arose independently from, but parallel with, the idea of priming of the inflammasome and this leads to some potential for confusion in the nomenclature. IL-1β synthesis and secretion is achieved via cleavage of an inactive form of the protein, pro-IL-1β, by cytosolic protein complexes called inflammasomes. Like the concept of microglial priming and phenotype switching, inflammasome activation is also a two step process involving “priming” and “activation,” but in inflammasome studies LPS has been typically used to “prime” the inflammasome (i.e. provide the signal for the transcription of pro-IL1β) and a second stimulus can then activate the inflammasome to allow caspase-1 to cleave this pro-IL-1β and allow secretion of the active cytokine (Eisenbarth and Flavell,2009). Conversely the author, and others, described that some aspect of neurodegeneration, ageing or amyloidosis “primed” the microglia, whereupon LPS could now trigger phenotypic switching toward an M1 phenotype. So, are there similarities between microglial priming and inflammasome priming (Fig. 2), or is this just an unfortunate coincidence of nomenclature? A study of Aβ peptide effects on primary microglial cultures showed that Aβ was capable of activating the NALP3 inflammasome to allow synthesis of active IL-1β and release of NO (Halle et al.,2008). This model proposed that Aβ was phagocytosed by microglia, and that cathepsin B was released from lysosomes and activated the inflammasome. Thus fibrillar amyloid would appear to be an activator of the inflammasome. However, it is important to note that microglial cells were “primed” with LPS in order to facilitate this IL-1β release and action: fibrillar amyloid by itself did not result in pro-IL-1β synthesis. This would suggest that if Aβ is to induce this robust IL-1β response, another stimulus would also be required. While experimental studies in inflammasome biology typically first “prime” with LPS and then “activate,” it is possible that the inflammasome could be assembled by a trigger such as Aβ, and that the pro-IL-1β signal could be supplied, at a later time in disease, by some other stimulus such as LPS, arising from a systemic infection. Inflammasome experiments have also been performed in cells isolated from the G93A SOD1 mutant mouse model of ALS, showing that endocytosed mutant SOD1 can activate caspase-1 activation and IL-1β secretion, dependent on ASC but independent of NLRP3 activation (Meissner et al.,2010). The authors state that this can occur without prior priming with LPS, but their data clearly show that a robust pro-IL-1β signal is already present in all microglial cultures, thus obviating the need for treatment with LPS to provide the pro-IL-1β signal. It will be necessary to test the hypothesis that mechanisms of microglial priming and phenotype switching overlap significantly with inflammasome priming and activation, but for the remainder of this article, the term priming will refer to microglial priming as originally described (Perry et al.,2007).
Danger-associated molecular patterns (DAMPs) such as HMGB1, nucleic acids, hyaluronon, α-synuclein, etc. may also be generated during chronic disease and may activate TLRs to transcribe pro-IL-1β via NF-κB activation, although in general the evidence for these processes is much stronger for acute insults such as stroke and indeed substantia nigra cell death induced by mitochondrial toxins (Gao et al.,2011b), where significant necrosis and release of cellular contents occurs (Chen and Nunez,2010).
So what, during chronic neurodegenerative disease, might be the priming factor or factors? There have been a number of interesting recent studies in this regard. It has recently been shown that deletion of the C3 convertase regulator complement receptor 1-related protein y (Crry) leads to microglial priming, as measured by dramatically enhanced responses to systemic LPS (Ramaglia et al.,2012). This priming was abolished in Crry/C3 double knockout animals, implicating expression of C3 as a key factor. The proposal arising from the study is that dysregulation of the complement system, as can occur in AD (McGeer and McGeer,2002), multiple sclerosis (Ramaglia,2012), and indeed in the ME7 model of prion disease (Cunningham et al.,2005a), leads to increased C3 and its cleavage products C3b and iC3b which bind to the microglial cell surface and consequently induce priming. This is an attractive explanation for microglial priming, because the complement mediated phagocytosis of apoptotic cells, via C1q, C3b, or iC3b opsonization, is an important mechanism of clearing debris in a relatively anti-inflammatory or “nonphlogistic” manner. As stated above, microglial activation is relatively muted in several models of neurodegenerative disease and we have observed that microglia engaged in phagocytosis of apoptotic cells in vivo, during prion disease, remain IL-1β-negative (Hughes et al.,2010). However, the cleavage of C3 that occurs during complement-mediated opsonization would appear to leave these phagocytes susceptible to further activation.
In the ME7 prion disease model a number of Fcγ receptors have been shown to be elevated and while disease progresses normally in γ chain knockout mice, the ability to synthesize CNS IL-1β in response to systemic LPS was significantly impaired (Lunnon et al.,2011). This indicates that the Fcγ chain has an important role in phenotypic switching of the microglial population.
In addition, it has been clear for some time that molecular signatures of the brain microenvironment contribute to suppressing microglial activation. There is an ever-growing list of molecules that can down-regulate microglial function via direct interaction between neuronal cell surface markers such as CD200, fractalkine (CX3CL1), and their corresponding receptors on microglia (CD200R, CX3CR1, and TREM2) among several others. The loss or deletion of these molecules has been shown in several studies to increase levels of microglial activation. Consistent with the observation that aged animals show primed microglial responses to secondary stimulation with LPS (Godbout et al.,2005), it has been shown that CD200 expression on neuronal dendrites decreases with age (Ojo et al., 2011) and glia prepared from CD200−/− mice show heightened responses to LPS stimulation (Costello et al.,2011). Likewise, decreased expression of CX3CL1 has been shown in aged mice (Wynne et al.,2010) and deletion of its receptor (CX3CR1) increases the susceptibility of these mice to LPS-induced CNS inflammation and sickness behavioral responses. However, it is important to note that systemic IL-1β was also enhanced in the CX3CR1 mice (Corona et al.,2010), suggesting that loss of these receptors may produce a heightened inflammatory responsiveness throughout the body.
It has been recognized for some time that neurotransmitters can exert tonic inhibition on inflammatory activation. Noradrenaline can exert anti-inflammatory effects on microglia via the β2 adrenergic receptor (Heneka et al.,2010; O'Sullivan et al.,2009), and the removal of noradrenaline's influence on cortical and hippocampal microglia in an animal model of AD, via lesioning of the locus ceruleus using the toxin dsp4, showed that Aβ pathology was exacerbated in the absence of NE's influence. Furthermore, in vitro characterization suggests that microglial proinflammatory cytokine responses to Aβ are suppressed by NE administration (Heneka et al.,2010). Acetylcholine can exert anti-inflammatory actions via the nicotinic α7 receptor, and although this is best characterized in the viscera, where vagal ACh outflow suppresses systemic inflammatory responses to LPS (Tracey,2009), there is evidence that microglia also bear this receptor and can be influenced by nicotine to suppress cytokine responses (De Simone et al.,2005). Whether endogenous ACh actually exerts this effect on microglia in vivo is not clear. We have recently failed to find evidence of microglial priming after selective lesioning of the basal forebrain cholinergic system (Field et al., 2012). While these data argue against a role for endogenous acetylcholine in microglial suppression, these lesions were deliberately limited in extent (approximately 20% ACh depletion) and more complete lesions may indicate an inhibitory role of endogenous ACh.
Continuing with the idea of suppression of microglial activation by constitutive factors in the brain, some potential players in Parkinson's disease have recently been shown to fulfill such a role. Nurr1, rare mutations in which are associated with familial PD, can inhibit expression of proinflammatory molecules by microglia and astrocytes in response to LPS, and Nurr1 suppression using intranigral injections of a lentiviral-encoded shRNA against Nurr1, showed considerable protection against LPS-induced nigral neuronal death (Saijo et al.,2009). Mechanistically, this is explained by the binding of Nurr1 to the NF-κB p65 subunit and recruitment of the coREST co-repressor complex and enhanced clearance of NF-κB with consequent transcriptional repression. Inflammation-induced down-regulation of another familial Parkinson's disease-related protein, parkin, also leads to exaggerated responses to LPS, albeit in peritoneal macrophages (Tran et al.,2011). However, in other studies by the same laboratory, Parkin−/− mice are more sensitive to inflammation-induced damage but do not appear to show exaggerated CNS inflammatory responses to the chronic LPS dosing regime in that study (Frank-Cannon et al.,2008). Thus, it remains unclear whether parkin deletion or suppression during inflammation leads to microglial priming. In contrast to these suppressions of constitutive proteins, the generation of extracellular α-synuclein, appears to be a significant priming factor for microglia as characterized in recent studies. Extracellular α-synuclein, injected directly into the substantia nigra provoked a robust proinflammatory response and when animals were challenged with systemic LPS, 18 h later, levels of proinflammatory mediator synthesis were equivalent to those induced by centrally administered LPS (Couch et al.,2011). Centrally administered 6-OHDA can also prime microglia, presumably secondary to nigral neuronal death (Depino et al.,2003). There are, therefore, several possible routes to priming in the Parkinson's diseased brain.
With respect to AD models, mice carrying a knock in of the familial AD mutation M146V in presenilin 1, showed exaggerated responses to systemic LPS, which included augmented IL-1β, TNF-α, and iNOS synthesis (Lee et al.,2002). Whether this was related to the aberrant PS1 or the increased Aβ deposition was not directly interrogated but exaggerated responses were apparent in isolated microglial cells but not in splenocytes, suggesting the latter. Other in vivo studies in AD transgenics showed that proinflammatory cytokine profiles were muted in disease per se and exaggerated by systemic LPS challenge (Sly et al.,2001), perhaps indicating that Aβ plaques are themselves sufficient to prime microglia.
In ALS, isolated microglia carrying the G93A SOD-1 mutation show exaggerated proinflammatory responses to LPS + IFN-γ stimulation and this is associated with increased expression of CCAAT/enhancer binding protein β (C/EBP/β), a transcription factor for which there are binding sites in the promoter regions of proinflammatory genes including TNF-α, IL-1β, IL-6, and iNOS (Valente et al., 2011). Increased nuclear translocation of C/EBPβ was seen in microglia of G93A SOD-1 mice challenged i.p. with LPS, albeit at the very high dose of 200 μg LPS per mouse.
It is well known that glucocorticoids have immunosuppressive and anti-inflammatory actions, but it has gradually become clear that they can in some instances, particularly in the brain, lead to heightened inflammatory responses to subsequent stimulation (de Pablos et al.,2006; Munhoz et al.,2010). Recent studies have suggested that psychological stress, in vivo, primes microglia for exaggerated responses to LPS ex vivo and that this is mediated by glucocorticoids, based on the ability of both adrenalectomy and inhibition of GC receptor function with the antagonist RU486 (Frank et al.,2012) to prevent this primed response. The dysregulation of the HPA axis is a frequent feature of neurodegenerative and neuropsychiatric disease and thus elevated glucocorticoids may well contribute to microglial priming in a number of settings, although this awaits verification in in vivo models of disease. Thus, while the microglial glucocorticoid receptor may be important for limiting inflammation in PD models (Ros-Bernal et al.,2011), this could still lead to a microglial state vulnerable to phenotypic switching.
Returning to the origins of the macrophage priming, it seems appropriate to raise the possibility of Type I interferons (IFN α/β) as microglial priming factors. The authors and others have shown that Type I IFNs are elevated by neurodegenerative insults (Field et al.,2010; Khorooshi et al.,2008; Stobart et al.,2007) and these molecules have previously been shown to be capable of inducing priming of peritoneal macrophages (Vadiveloo et al.,2000).
Finally it is worth considering whether it is sufficient, for priming, to have a heightened, albeit low grade, systemic inflammatory state on an ongoing basis. It is well established that chronic conditions such as obesity, diabetes, atherosclerosis, arthritis all have a systemic inflammatory component (Yaffe et al.,2004). We consider the impact of chronic co-morbidities on established disease later, but there is some evidence that these conditions themselves prime the brain (Drake et al.,2011).
Temporal Aspects of Priming
With respect to temporal aspects of microglial priming questions as to the transience or permanence of this state arise. Our experience of prion disease suggests that microglial already show primed responses as early as 12 weeks postinoculation with disease, before disease-associated cognitive changes are present (Murray et al.,2012) and as disease progresses this priming appears to become more robust (Cunningham et al.,2005b). In Tg2576 mice, priming is not apparent at 6 months but is demonstrable at 18 months (Sly et al.,2001). Notably C3 expression is also increased in these older mice, perhaps implicating C3 as a key priming factor. There are reports from other AD models of changes in phenotype with ageing and whether these different states are differentially susceptible to phenotype switching upon subsequent challenge has not been widely investigated. In the optic nerve crush model of Wallerian degeneration in which significant myelin degradation and phagocytosis occurs over a prolonged period, microglia remain primed for at least 28 days after the initial acute nerve injury (Palin et al.,2008), and that microglial priming persists in EAE lesions has recently been demonstrated out to 6 weeks post-resolution of monophasic disease onset (Moreno et al.,2011).
As for the time-course of cells after they become fully activated: we have used LPS or poly I:C to show that cells become activated to make IL-1β but this does not appear to last long (Field et al.,2010; Murray et al.,2012). There are, of course, downstream effects of these patterns of induction and these may mean that the “wave” effectively lasts longer. There is certainly evidence that Fcγ receptor changes are still evident at 24 h post-LPS (Lunnon et al.,2011). However, there is little evidence for a lasting M1 phenotype after either LPS i.c. or i.p. One recent study showed a decreased responsiveness of primed microglia from the aged brain to IL-4 (Fenn et al., 2011) but there is little doubt that the inflammatory wave post-LPS (at least at doses mimicking common infections) does resolve and microglia can return to their pre-LPS phenotype, with acute cognitive disruptions also resolving fully, at least at early stages of underling disease (Murray et al.,2012). How this situation changes when the systemic inflammatory insult is chronic or passes through innate and adaptive phases is not clear. Evidence from some PD models, suggesting that a single systemic insult with LPS is sufficient to induce lasting iNOS and NADPH oxidase activity in mice with α-synuclein mutations may reflect particularly potent microglial activating qualities of α-synuclein (Gao et al.,2011a).
Systemic Inflammation Contributes to Damage and Dysfunction in Neurodegenerative Disease: Evidence from Animal Models
Although the idea that systemic inflammation is a significant contributor to CNS pathology is a relatively new one, brain damage resulting from severe sepsis is well known to occur in humans and in rodents. Animal models of sepsis, using doses of LPS of the order of 10 mg/kg have shown evidence of robust CNS inflammation, microglial iNOS expression, neuronal death, and long-term cognitive decline (Semmler et al.,2005,2007). Transgenic studies with iNOS−/− mice demonstrated an important role for iNOS in both cognitive deficits and changes in presynaptic and postsynaptic molecules (Weberpals et al.,2009). However, there may be other mechanisms contributing to damage in such models: markedly decreased cerebral metabolism using micro-PET imaging of glucose uptake (18FDG) has been demonstrated and this was associated with decreased cerebral blood flow and decreased alpha activity in EEG (Semmler et al.,2008). Thus, robustly impaired brain function may occur acutely as a result of decreased tissue perfusion and oxygenation, but inflammation would appear to be a significantly contributor to subsequent neuronal death and denervation. There have been a number of studies in recent years in which these high “sepsis” doses have been used to replicate features of neurodegenerative diseases such as Alzheimer's disease and Parkinson's disease (Qin et al.,2007). It is important to draw a clear distinction between these high doses and those used to simulate systemic infection as generally experienced in the population.
Repeated dosing with LPS induces features of neurodegenerative disease
There have also been several studies with multiple doses of LPS administered to normal animals and animals with particular mutations. The group of Serge Rivest made repeated challenges with LPS (1 mg/kg every 2 weeks for a 3-month period) to induce markedly more severe axonal pathology and progression of disease in the G37RSOD1 mouse model of ALS (Nguyen et al.,2004). It is noteworthy that there was no evidence of an exaggerated acute inflammatory response in G37R mice compared with wild type: in this instance, there is not evidence of microglial priming, although LPS clearly exacerbated the chronic inflammation associated with disease per se. Another early study administered LPS weekly for 12 weeks (150 μg/mouse/week) and found increased APP expression and processing, leading to increased intraneuronal generation of amyloidogenic Aβ (Sheng et al.,2003). One study used a regime of 7 consecutive days of LPS challenge (250 μg/kg), in naive mice, to induce increased Aβ1–42 generation and deposition of plaque material, probably via increased β- and γ-secretase activities (Lee et al.,2008). In the triple transgenic model of AD, LPS treatment increased the severity of Tau tangle pathology when administered at 500 μg/kg twice weekly for 6 weeks (Kitazawa et al.,2005). LPS-induced Tau hyperphosphorylation could be blocked by inhibiting the activity of cyclin-dependent kinase 5 (cdk5). In the 3xTg model of AD, twice-weekly injections of LPS (250 μg/kg) for 4 consecutive weeks also increased preplaque APP accumulation (intraneuronal 6E10 labeling) and this was TNF-α-mediated (McAlpine et al.,2009). Thus, LPS has typically been shown to alter APP processing and Aβ deposition, but alternative mechanisms by which LPS may alter disease course, via microglial activation are discussed below.
A similar approach has been taken in the Parkinson's disease field. Although loss of function mutations in parkin are present in almost half of early onset PD cases, parkin−/− mice do not show typical features of Parkinson's disease such as nigral neuronal loss. However, parkin−/− mice were shown to be more vulnerable to inflammation-related nigral inflammation induced by administration of LPS (250 μg/kg) twice weekly for 3 or 6 months to parkin−/− mice (Frank-Cannon et al.,2008). It should be noted that prolonged treatment with LPS was sufficient to cause significant nigral loss even in wild-type animals. Despite the interaction between LPS and genotype with respect to nigral neuronal death and beam walking abilities, there were remarkably limited differences in the neuroinflammation observed in parkin−/− + saline, WT + LPS, and parkin−/− + LPS groups. The inflammatory markers in this study were examined only once, after 3 months of LPS treatment but the data suggest that there are not primed microglial responses in the parkin−/− mouse and neuronal death may be more to do with an inherent vulnerability of neurons of the parkin−/− mice. Interestingly, 1 mg/kg LPS administered once weekly for 4 months has exacerbated microglial activation but had no impact on neurological scores in an animal model of Huntingtons disease (Franciosi et al.,2012).
Though these repeated LPS models typically do not use LPS doses as high as those used in studies of severe sepsis, the dosing regime is nonetheless rather high or rather prolonged and in some cases, both. There has not typically been a rationalization of what these prolonged LPS exposures are designed to mimic: multiple systemic infections, chronic peripheral inflammatory disease, or some other source of inflammation? It is known that animals can develop tolerance (Ziegler-Heitbrock,1995) and/or hypersensitivity (Greer and Rietschel,1978) to LPS depending on the dose and timing regime and consecutive systemic LPS challenges (for 4 days) are sufficient to completely ablate systemic TNF-α responses to further systemic LPS, without inducing CNS tolerance (Faggioni et al.,1995). Similarly, when multiple LPS challenges are made in quick succession, CNS IL-1α, TNF-α, IL-6, and MCP-1 responses are considerably exacerbated (Erickson and Banks,2011). Given that the repeated LPS approach is now frequently used in AD and PD research, and has deleterious consequences for disease, it is important to interrogate the way in which the response to LPS changes upon multiple challenges.
Single LPS challenge exacerbates disease
Using a single challenge with LPS, to mimic a single episode Gram-negative bacterial infection, is a conceptually simple way to interrogate the interaction between systemic inflammation and chronic neurodegenerative disease. One early report showed that a single systemic challenge with LPS, at 500 μg/kg, could alter expression of APP but these authors did not examine disease per se (Brugg et al.,1995). More recently, a single challenge with LPS (500 μg/kg i.p., from Salmonella equine abortus), acutely increased neuronal apoptosis in the ME7 model of prion disease, but not in normal animals. (Cunningham et al.,2005b). This same dose, when administered just once at 15 weeks postinoculation with disease, was sufficient to cause acute and reversible neurological changes in ME7 and even after full recovery from acute deficits, the LPS-treated ME7 animals developed progressive and irreversible neurological impairments earlier and with greater severity than ME7 animals not treated with LPS (Cunningham et al.,2009). In studies using the human A53T α-synuclein mutation in mice, a single systemic challenge with 3 × 106 Endotoxin Units (approximately 1 mg/kg) LPS was sufficient to induce lasting iNOS and NADPH oxidase labeling and tyrosine hydroxylase-positive neuronal loss (Gao et al.,2011a). Also relevant to PD, using wild-type aged mice (20–22 months) a single challenge with LPS (200 μg/kg) was sufficient to induce TH-positive neuronal death in the substantia nigra (L'Episcopo et al.,2011), suggesting that the aging brain becomes progressively more vulnerable even to common and proportionate environmental stimuli.
Other systemic inflammatory stimuli
Though the majority of studies in this area have been performed with LPS, other stimuli have been assessed. In the 6-OHDA model of Parkinson's disease adenovirally-mediated systemic expression of IL-1β significantly exacerbated neuronal loss in the substantia nigra and exacerbated motor symptoms (Pott-Godoy etal.,2008). The inhibition of iNOS partially protected against IL-1β-induced neurotoxicity in the substantia nigra in these animals. Interestingly, active infection with the Gram-positive bacterium Streptococcus pneumoniae did not exacerbate features of disease in a number of disease models: Tg2576 AD model, the Thy1-A30P-α-SYN PD model, and the G93A SOD-1 ALS model (Ebert et al.,2010). Although those studies are partially confounded by the administration of antibiotics to infected animals within 24 h of infection, it is possible that Gram-positive bacteria, which lack LPS, constitute a less significant stimulus to the diseased brain. Deleterious effects have, however, been observed with other stimuli. The acute and longitudinal exacerbation of chronic neurodegenerative disease has been demonstrated after systemic challenge with the double-stranded RNA poly I:C (Field et al.,2010). These animals showed exaggerated CNS IL-1β and also Type I interferon (IFN α/β) responses to this mimic of systemic viral infection. Among the specific Type I interferon-responsive genes robustly elevated was RNA-dependent protein kinase (PKR). Increased activity of this kinase has been shown both to induce apoptosis (Balachandran et al.,1998) and to induce changes in eIF2α, which impairs LTP and memory consolidation (Jiang et al.,2010). The data arising from this poly I:C study suggest that one might think of disease progression as a consequence of the cumulative effects of multiple systemic insults (as opposed to one long-term peripheral inflammatory condition): poly I:C was administered three times, two weeks apart, with each successive challenge producing acute onset deficits that were progressively more severe and less reversible as the challenges were superimposed on disease at more progressed stages (Field et al.,2010) (Fig. 3). This mimics the fluctuating and variable rate of decline seen in Alzheimer's disease patients (Holmes and Lovestone,2003). Consistent with the idea that systemic viral infection may contribute to neurodegenerative disease, there is evidence that the H5N1 influenza virus can be disseminated and even enter the CNS and induce inflammation and neurodegenerative features such as α-synuclein aggregation and nigral neuronal loss (Jang et al.,2009), and indeed there is evidence for increased incidence of PD in the wake of an influenza epidemic, causing significant encephalitis, in the early 20th century (Ravenholt and Foege,1982).
Using another model of Parkinson's disease, induced by intranigral injection of 2 μg LPS, ulcerative colitis (induced by 5% sodium dextran sulphate; DSS) was also shown to significantly exacerbate dopaminergic neuronal loss (Villaran et al.,2010). DSS induced significant systemic levels of IL-1β and increased CNS expression of iNOS, TNF-α and ICAM-1 in the PD mice. There was also some evidence of blood brain barrier (BBB) breakdown, but interpretation of these data is complicated by the injection of LPS into the nigra during the period of ulcerative colitis. Thus the model examines the severity of LPS-induced nigral degeneration when performed in the presence of significant peripheral inflammation.
One recent study assessed the impact of osteoarthritis, a progressive disease associated with aging, on AD pathology in the APP/PS1 double transgenic model. Since IL-1β is known to contribute to osteoarthritis pathology, the Col1-IL1βXAT Cre inducible model was used to model osteoarthritis and when these animals were crossed with APP/PS1 mice and injected with Cre to induce IL-1β expression, there were significant exacerbations of Aβ deposition and associated microglial activation (Kyrkanides et al.,2011). Atherosclerosis also constitutes a systemic inflammatory disease, and is obviously very prevalent in western societies. Atherosclerosis is a major risk factor for stroke and recent studies show that the susceptible ApoE−/− mice, when fed on fat-rich diets such as the Paigen diet (Paigen et al.,1985), develop atherosclerotic plaques but also develop evidence of cerebrovascular and microglial activation (Drake et al.,2011). Whether this state predisposes these microglia to respond more robustly to LPS has not been examined, but there is evidence that stroke outcomes are more severe in these animals (Horsburgh et al.,2000). Systemic inflammation induced by LPS also exacerbates stroke outcomes, but in this case the mechanism appears to depend on LPS-induced systemic IL-1β and chemokine expression and mobilization of neutrophils, and may be independent of microglial activation (McColl et al.,2007). In a model of liver injury, induced by bile duct ligation and resection, animals with liver injury showed significant recruitment of monocytes to the brain, dependent on peripheral TNF-α signaling through TNFR1, to induce CNS MCP-1 (D'Mello et al.,2009). Though these findings showed that this signaling pathway was important for induction of sickness behavior rather than neurodegeneration, the recruitment of monocytes to the brain by peripheral organ disease has significant implications for chronic neurodegenerative disease.
Given the now numerous reports of systemic inflammation's impact on neurodegenerative pathology and symptomology, it is worth considering the impact of housing conditions on the inflammatory status in animal models. SPF animals may show different susceptibility to the effects of inflammation than conventionally housed animals. Similarly, undetected infections in conventionally housed animals may contribute to variability in makers of inflammation and pathology subsequently observed. There are, unsurprisingly, few studies on this subject, and apparently none that deal specifically with neurodegenerative disease. There is evidence that animals housed in specific pathogen free (SPF) conditions exhibited muted CNS inflammatory responses to administered adenovirus compared to those housed in conventional conditions (Ohmoto et al.,1999). Those mice previously exposed to adenovirus, peripherally, showed CD8+ T cell-skewed responses compared to the CD4+ T cell-dominated responses of naive mice. SPF mice also showed more severe relapses in EAE compared with conventionally housed animals (Birnbaum et al.,1998). In the periphery it has been shown that a single challenge with zymosan could induce rheumatoid arthritis in SKG mice under conventional conditions, but these animals were not susceptible under SPF conditions (Yoshitomi et al.,2005). It is of interest to know what influence housing and environmental factors may have had on the parameters of inflammation in animal models where systemic inflammatory stimuli were not deliberately applied.
Microglial-Mediated Mechanisms of Disease Exacerbation
The primary effect of further activation of primed microglia, after systemic LPS challenge, is an exaggerated proinflammatory response. This has been described in prion disease, AD, PD, Wallerian degeneration, and in aging. Some possible mechanisms whereby this switch can effect degenerative changes have been referred to, as they arose, in the above-mentioned sections. In this section, we discuss other possibilities for systemic inflammation-induced exacerbation of disease (Fig. 4).
Classical proinflammatory mechanisms: IL-1, TNF-α, and prostaglandins
The acute induction of IL-1β in the brain after systemic LPS challenges has been the hallmark of the previously primed microglial cell (Cunningham et al.,2005b) and is induced by proinflammatory stimuli such as LPS. IL-1 initiates cellular responses through its interaction with IL-1RI to activate NF-κB, as well as induction of mitogen-activated protein kinases (MAPK) p38, extracellular related kinase (ERK)1/2, and c-jun N-terminal kinase (JNK) pathways (Parker et al.,2002). Collectively these pathways are responsible for the induction of several effector mechanisms discussed below, including iNOS, COX, NADPH oxidase, and proteases including matrix metalloproteases and plasminogen activators. IL-1α has been much less studied, but seems to be released from necrotic cells, and may be a key step in the induction of postischemic inflammation, albeit one that appears to contribute to neuronal damage (Boutin et al.,2001). However, it can also activate the endothelium to facilitate peripheral leukocyte infiltration (Thornton et al.,2010) and may thus contribute to damage arising from systemic inflammation. TNF-α is also a robust inducer of NF-κB activation, but is more widely implicated in cell death processes than most cytokines since direct engagement of the TNFRI p55 can be sufficient to induce apoptosis. The TNF p55 receptor forms a complex containing TNF receptor associated death domain (TRADD), which can dissociate and recruit FADD (FAS-associated via death domain) and caspase-8 to initiate apoptosis (Micheau and Tschopp,2003). The balance between death domain-initiated and NFκB-initiated pathways is crucial in determining the apoptotic potential of TNF-α signaling (Beg and Baltimore,1996). However, there are now a very large number of studies suggesting that inhibition of TNF-α has beneficial effects in animal models of disease and thus, its acute up-regulation by systemic inflammatory events is likely to have a significant impact on existing brain pathology whether via direct apoptotic effects or by the up-regulation of effector enzymes such as iNOS and proteases.
The protective effects of NSAIDs on the development of AD and PD obviously suggest that prostaglandins generated from COX may be an important feature in these diseases. COX-2 has often been the pharmacological target of choice against neuroinflammation since it is induced by inflammatory stimuli (Cao et al.,1996; Ek et al.,2001) but the evidence for increased expression in chronic neurodegenerative diseases, including AD (Hoozemans et al.,2001; Yermakova et al.,1999), prion disease (Deininger et al.,2003), and HIV dementia (Griffin et al.,1994) is actually stronger for COX-1. It is clear that microglia can express both COX-1 and COX-2, although there appears to be some debate about the latter. However, the involvement of different COX isoforms (1 and 2), different specific synthases responsible for E, D, I, and F prostaglandins and different receptors for each of these prostaglandin classes have not received as much attention as one might expect given the protective effects of NSAIDs in AD and PD (see Cunningham and Skelly,2012 for review). There are studies suggesting that microglial EP2 contributes to amyloid load and oxidative damage in the APP/PS1 model (Liang et al.,2005), and consistent with the idea of both beneficial and deleterious roles of microglia there is evidence that deletion of the EP2 receptor improves microglial phagocytosis and decreases neurotoxicity (Shie et al.,2005). Conversely, EP4 has recently been described to have an anti-inflammatory role, limiting proinflammatory cytokine synthesis induced by LPS (Shi et al.,2010). While it is known that some NSAIDs can lower Aβ aggregation independent of their anti-inflammatory activity (Weggen et al.,2001), it has also been shown, in a large group of pooled prospective studies, that “amyloid-lowering” NSAIDs are not more effective in reducing the risk of AD than nonamyloid-lowering NSAIDs (Szekely et al.,2008). Since systemic inflammation clearly increases CNS prostaglandin concentrations with consequent effects on behavior and cognition (Hein et al.,2007; Teeling et al.,2010) the characterization of the actions of these molecules continues to be important.
Acute exacerbation of function: Delirium, depression, and cognitive impairment
It is now clear that systemic inflammation can induce acute working memory changes in aged animals (Chen et al.,2008) and those with prior neurodegenerative disease (Murray et al.,2012) or cholinergic neuronal loss (Field et al., 2012). These cognitive changes are comparable to the cognitive deficits observed in episodes of delirium in the elderly and demented according to DSM-IV and ICD-10 descriptions (American Psychiatry Association,1994; WHO,1992) and microglial priming as observed in these model systems has gained acceptance as a contributor to this clinical scenario (van Gool et al.,2010). Likewise both infection (Barrientos et al.,2006) and surgery (Cibelli et al.,2010) can impair the ability to form new contextual memories and this is especially true in the aged, in which microglia have been primed. Mechanisms for these acute deficits are not yet clear, but it appears that IL-1β, TNF-α, and PGE2, perhaps via direct effects on ACh or other neurotransmitters, are involved. At least in the case of IL-1, central administration of IL-1ra has been found to be protective (Chapman et al.,2010; Cibelli et al.,2010). IL-1β expression has been shown to impact on ACh outflow and to impair memory function (Taepavarapruk and Song,2010) and the impact of systemic LPS on working memory deficits in those with prior cholinergic neuronal loss can be protected against using the acetylcholinersterase inhibitor donepezil (Field et al., 2012). Understanding mechanisms of delirium will prove very important in dementia since these episodes are now known to significantly affect long-term cognitive function (MacLullich et al.,2009), to accelerate dementia (Fong et al.,2009) and to shorten the time to permanent institutionalization and death (Witlox et al.,2010). Priming of the microglial population in aged rodents also appears to predispose these animals to exaggerated and prolonged features of depression (Godbout et al.,2008). These are important complications that can occur frequently during the course of chronic neurodegenerative disease and further dissection of mechanisms could offer significant benefits to patients. It is important to consider the possibility that the same molecules may be involved in acute, but recoverable, neuronal dysfunction, and in neuronal death leading to permanent loss of function. Equally, reversible and irreversible neuronal dysfunction/damage may occur by distinct pathways (Figs. 3 and 4).
Reactive oxygen and nitrogen species: iNOS and NADPH oxidase
Microglia are capable, if sufficiently activated, of generating an oxidative burst involving the regulated induction of multiple enzymes/complexes including the NADPH oxidase, iNOS, and sometimes myeloperoxidase (MPO). These enzymes/complexes form superoxide (O2−), nitric oxide (NO), and hypochlorous acid (HOCl), respectively. Each of these enzymes have been described in both PD patients and models and preclinical studies have described a role for all three in toxicity to TH-positive cells of the substantia nigra (Choi et al.,2005; Hunot et al.,1996; Wu et al.,2002). The peroxynitrite anion ONOO− is particularly toxic to neurons. Among the most frequently cited proinflammatory and damaging aspects of microglial activation is the formation of iNOS. The induction of transcription of this NF-κB-regulated gene is extremely common in studies of neuroinflammation and the expression of iNOS protein and the synthesis of NO is a hallmark of the further activation of primed macrophages (Johnson et al.,1983) and microglia (Cunningham et al.,2005b). It is striking that elevation of iNOS is most frequently observed in animal models of PD, most of which employ either acute neurotoxins (MPTP, 6-OHDA) or rather severe LPS treatment regimes: LPS i.c. (Herrera et al.,2000), or i.p. at 5 mg/kg (Qin et al.,2007). However, despite reservations about the acute and severe nature of the initial stimulus in these studies, LPS has proved to induce apparently self-sustaining inflammation (Qin et al.,2007), and MPTP-induced inflammation has proved to have similar longevity in monkeys and in humans (Langston et al.,1999). Where robust iNOS induction has occurred after systemic inflammation, induced by LPS or adenovirally-induced IL-1β, inhibition (Pott-Godoy et al.,2008) or deletion (Weberpals et al.,2009) of iNOS has been sufficient to protect nigral or other affected neuronal populations.
Phagocytosis: Axon pathology, cell death, and protein aggregates
The degeneration of axons is something that is likely to happen in almost all brain pathologies including axon transection, stroke, and multiple progressive neurodegenerative diseases. Studies of Wallerian degeneration (Palin et al.,2008) showed that microglial cytokine profiles were muted after acute optic nerve crush, but the cells remain primed 28 days later and responded to systemic LPS with phenotype switching to a more IL-1β dominated profile and enhanced phagocytosis of neuronal debris. While this phagocytosis would appear to be beneficial in the context of successful clearance of myelin and axonal debris from functionally dead neurons, there is in vitro evidence that stimulating phagocytosis can result in clearance of neurons that would not otherwise have been committed to cell death: LPS or lipotechoic acid treatment of neuronal and microglial co-cultures, induced reversible neuronal display of phosphatidyl serine (PS), via ROS/RONS production. Mfge8 and the vitronectin receptor collaborate to facilitate microglial recognition of PS-positive cells and effect uptake of the cell. However, multiple blocking experiments suggest that many of these cells would have survived had they not been phagocytosed (Neher et al.,2011). There is recent evidence that systemic LPS-induced activation or reactivation of lesions in the EAE model of multiple sclerosis can lead to relapse and to new iNOS expression and axonal transection, as measured by APP end-bulbs (Moreno et al.,2011). These studies revealed significant heterogeneity of different lesions in the EAE brain, but significantly the authors showed increased microglial/macrophage iNOS consistently co-localized with APP, and those lesions that were iNOS negative were typically also APP negative and showed IL-10 and TGFβ1 expression persisting for several days post-LPS, consistent with the switching of some lesions to an M1 phenotype, but maintenance of others in an M2 phenotype. Though the EAE model has both innate and adaptive components, exacerbation of axonal pathology occurred significantly earlier than increased T-cell infiltration (Moreno et al.,2011). These studies provide clear evidence that brain lesions may lie relatively silent but later be reactivated by remote inflammatory stimuli and further activation of primed microglia in this scenario is clearly detrimental. In further EAE models, using fMOG or the delayed type hypersensitivity to BCG (DTH) model, LPS reactivated EAE lesions, resulting in increased demyelination and BBB breakdown (Serres et al.,2009). The degree to which primed microglia contributed to these effects is not clear, although macrophages were clearly recruited from the periphery. Consistent with the idea that complement activation can prime microglia, C3b-coated nerve fibers were found in close proximity to primed, IL-1β negative, microglia in MS brain, and EAE progresses more slowly in Crry−/− mice, in the absence of priming (Ramaglia et al.,2012). Since microglia are more abundant in white than in grey matter in the human brain and are increased in number in the aged brain, systemic insults may have significant effects on major axonal tracts in the brain and might be particularly disabling when reactivating white matter microglia.
Conversely, there is evidence that further stimulation of microglia in the AD transgenic brain, using intracerebral LPS, can enhance clearance of Aβ (DiCarlo et al.,2001; Herber et al.,2004). While Aβ clearance was improved, there was insufficient assessment of cellular infiltration in those studies to rule out a significant contribution of infiltrating monocytes and neutrophils, which surely would occur upon stimulation with LPS at the high doses used (4 and 10 μg LPS i.c.). We have performed similar experiments in the ME7 model of prion disease and, with just 0.5 μg LPS, found dramatic exacerbation of inflammation, marked neutrophil infiltration but surprisingly unaltered levels of extracellular PrPSc, the prion disease-associated amyloid species (Hughes et al.,2010). Consistent with this, there is evidence that phagocytic activity and proinflammatory cytokine synthesis are inversely related in macrophages and microglia (Fadok et al.,1998; De Simone et al.,2002; Takahashi et al.,2005), which is at variance with the prior proposal that more phagocytic microglia are also more pro-inflammatory (Streit et al.,1999).
Complement, Fc Receptors, IgG, and BBB breakdown
Experiments demonstrating the markedly up-regulated expression of activating Fcγ receptors FcγRII, FcγRIII, and FcγRIV upon systemic LPS challenge to ME7 animals (Lunnon et al.,2011) suggest other possibilities for neurodegeneration. Fc receptors bind IgG, and the same study shows evidence of BBB breakdown, leakage of IgG into the brain, and co-localization of IgG with CD68-positive microglia. Association between IgG and Fc receptors is described, in immune cells and in microglia (Ulvestad et al.,1994a), to promote phagocytosis, oxidative burst, and antibody-dependent cytotoxicity (see Okun et al.,2010, for review). A study of direct injection of IgG purified from Parkinson's disease patients into the substantia nigra of wild-type and FcgR−/− mice showed TH-positive neuronal death in the nigra only in animals carrying the FcγR (He et al.,2002). Increased expression of these Fc receptors has also been observed in both AD (Peress et al.,1993) and MS (Ulvestad et al.,1994b) brains.
The presence of increased IgG in the diseased brain may also be a trigger for complement activation, but there is also evidence for activation of complement even in the absence of IgG, via C1q binding to Aβ (Rogers et al.,1992). The full activation of complement pathways will eventually lead to formation of the C5b-9 membrane attach complex (MAC) that can induce bystander cell lysis. It is of interest that postmortem AD brains contain evidence for the C5b-9 membrane attack complex, while APP transgenic mice do not (McGeer and McGeer,2002; Reichwald et al.,2009), perhaps arguing for a role for secondary inflammatory stimulation to complete assembly of the MAC, as is observed at postmortem.
The discussion of IgG-mediated effects cannot be easily separated from the topic of BBB breakdown. Impaired BBB function with ageing has been shown in rodents and a meta-analysis of BBB function with ageing in humans showed that BBB permeability increased with ageing and was further increased in patients with vascular or Alzheimer's dementia (Farrall and Wardlaw,2009). In normal animals, LPS shows surprisingly little evidence of penetration of the brain (Banks and Robinson,2010; Singh and Jiang,2004), although at high doses it can induce robust BBB breakdown (Wispelwey et al.,1988). Recent intravital microscopy studies have shown increased extravasation of 70,000 MW dextran in real time after challenge with 1 mg/kg LPS (Ruiz-Valdepenas et al.,2011). Thus, even if LPS itself has limited access to the brain parenchyma, it is certainly plausible that there will be significant leakage of plasma proteins across the BBB in the sorts of regimes used in many of the studies cited above. Of course, in cases where underlying disease increases access of molecules to the brain parenchyma, via BBB breakdown, LPS might directly activate brain cells.
Apart from LPS and IgG, Increased BBB permeability also allows access of other microglial activating molecules to the brain. Thrombin is known to enter the brain during BBB breakdown, has been shown to activate microglia via the PAR-1 receptor (Suo et al.,2002) and, when injected into the substantia nigra, can result in nigral neuronal death (Carreno-Muller et al.,2003). Likewise fibrinogen and albumin “leak” into the CNS when the BBB is breached (Nishioku et al.,2009) and can activate proinflammatory and neurotoxic pathways in microglia (Hooper et al.,2005; Piers et al.,2011).
For all that has been said about microglia thus far, the distinction between expression of proinflammatory cytokines by the endothelium, the perivascular macrophages, and microglia or indeed astrocytes has not often been documented in the cited studies of systemic inflammatory activation. It is clear that multiple routes exist to convert a peripheral inflammatory signal into a CNS equivalent: activation of the cerebral endothelium, of the circumventricular organs, of the perivascular macrophages as well as transport of cytokines across the BBB and stimulation of vagal afferents arising in the viscera have all been well described after systemic inflammation and are discussed elsewhere (Banks and Erickson,2010; Konsman et al.,2002). While each of these routes is capable of transmitting a signal it is likely that there is some redundancy in these routes. We have recently found that blocking systemic cytokines using dexamethasone was insufficient to block CNS inflammatory cytokine synthesis, and that LPS was detectable in the blood even at a dose of 100 μg/kg (Murray et al.,2011). This means that in all of the “repeated LPS challenge” studies cited above, the brain endothelium will have been exposed to LPS irrespective of what systemic inflammatory mediators were also induced. Obviously in those studies where LPS dosing is sufficient to induce BBB breakdown, the brain parenchyma itself will be exposed to LPS. Similarly, it is well established that prostaglandins have an important role in transducing systemic inflammatory signals into the CNS (Cao et al.,1996; Ek et al.,2001), and since these are effectively targeted by NSAIDs, it is plausible that NSAID protection against the development of AD and PD may work in part by preventing the deleterious CNS consequences of systemic inflammatory activation.
Boosting immune responses to treat disease
It has been suggested (Yong and Rivest,2009) that boosting the immune response could be a beneficial strategy in neurodegenerative disease. This hypothesis is based on the idea that the microglial cell has both beneficial and deleterious roles in degenerative disease and that promotion of the beneficial and suppression of the deleterious could, for example, boost phagocytosis of plaque material without significantly disrupting neuronal integrity. While there is some preclinical data to support this, the only clinical data that can currently address this possibility are vaccination studies in AD, which have not successfully replicated (Holmes et al.,2008; Salloway et al.,2009) their resounding success in preclinical laboratories (Schenk et al.,1999). There has been strong evidence that vaccination against Aβ can clear amyloid, but this has not slowed disease progression (Holmes et al.,2008) and has lead to some significant adverse events (Nicoll et al.,2003). One possible reason for this may be the increased expression of Fc receptors occurring during chronic neurodegeneration (Lunnon et al.,2011; Peress et al.,1993). It is also proposed that infiltration of the brain by bone marrow derived myeloid cells can have beneficial effects on the diseased brain (Simard et al.,2006). Indeed frequent injections of the TLR9 ligand CpG can reduce amyloid pathology and improve working memory deficits in the Tg2576 model of AD (Scholtzova et al.,2009) suggesting that the systemic inflammatory mediator profile may be a key determinant of the CNS outcomes of these systemic challenges. Unfortunately these authors did not report the systemic and CNS inflammatory profile after this repeated CpG regime. Irrespective of these data, the possibility of beneficial effects have to be examined against the consistent clinical observation that patients who experience systemic infections or inflammation arising from surgery or injury do not fare well: they frequently suffer episodes of delirium, and these episodes typically lead to accelerated cognitive and functional decline (Witlox et al.,2010). Thus, while boosting aspects of systemic inflammation could theoretically be beneficial, this is a complex question and one that simply requires robust testing in multiple relevant models. Current evidence from patients suggests that generalized systemic inflammation is almost universally deleterious. The clinical evidence for disease exacerbation is discussed below.
Much has been made of inflammation as a double edged sword, but with regard to systemic inflammation, one thing seems relatively clear: when elderly individuals and patients with neurodegenerative disease experience systemic inflammatory episodes, such as infections, they typically do not show improvements of their condition: there is now good evidence that they get worse. One very clear example of this is delirium. It is now clear that dementia is the biggest risk factor for delirium and systemic inflammation is one the most frequent triggers and there is a strong association between episodes of delirium and subsequent cognitive decline, acceleration of dementia progression and shorter time to permanent institutionalization and death. Thus it is difficult to argue for anything other than deleterious effects of systemic inflammation in these patients. Indeed there are multiple clinical studies that indicate negative outcomes for patients after systemic inflammation. While several studies have examined the association of specific infectious agents with Alzheimer's disease, including HSV-1, Chlamydia pneumonia and spirochetes (see Holmes and Cotterell,2009, for review) observations have not been consistent. However, two or more infections, of any type, over a 4-year period increased the risk of AD by twofold in a general practitioner database review (Dunn et al.,2005) and general ill health was significantly associated with cognitive decline in two further studies (Strandberg et al.,2004; Tilvis et al.,2004). MMSE scores were shown to decrease with increasing viral burden and Herpes simplex virus and cytomegalovirus were of particularly high risk (Strandberg et al.,2003) although these pathogens have not been specifically implicated elsewhere. These data recall evidence of a major influenza epidemic in the early 20th century, which caused encephalitis and subsequently increased parkinsonism (Ravenholt and Foege,1982), but without robust evidence of viral antigens inside the CNS (Lo et al.,2003). In the absence of an ongoing association between particular infections and risk of disease, it seems reasonable to assume that the robust systemic acute phase response or ‘cytokine storm’ occurring during acute disease may affect the brain in similar ways for a large number of different infections. Periodontitis, a gum infection that is very common the community, caused most often by infection with the Gram-negative Porphyromonas gingivalis, has now been shown to be a significant risk factor for Alzheimer's disease (Kamer et al.,2009). In addition to these association studies, it has been shown that treating infection and vaccination against a number of common diseases can reduce the risk of development of Alzheimer's disease (Verreault et al.,2001) and antibiotic treatment of mild to moderate AD patients with doxycycline and rifampin can slow cognitive decline even in established disease (Loeb et al.,2004). While most of the above studies examined the possibility of infections as etiological factors, the hypothesis that acute systemic inflammation, superimposed on established disease, would exacerbate or accelerate disease was examined in a group of 85 AD patients. Though only a small fraction of these patients showed elevated serum IL-1β, this was significantly associated with increased cognitive decline across a 2-month period (Holmes et al.,2003). Further studies, in a cohort of 275 AD patients across 6 months, showed that elevated serum TNF-α was significantly correlated with accelerated cognitive decline and those patients with low serum TNF-α showed stable cognitive function across this period. While those with elevated serum TNF-α or with carer-reported infection showed greater decline, those with both of these features showed considerably greater decline (Holmes et al.,2009). All AD patients reaching criteria for delirium were omitted from this study, so one can say that even in the absence of delirium, these systemic inflammatory events are associated with increased pathological burden. It is of note that not all patients with elevated serum TNF-α suffered acute systemic inflammatory events and this TNF-α may arise from a number of other chronic co-morbidities such as obesity, atherosclerosis, diabetes and smoking, all of which have a systemic inflammatory component (Drake et al.,2011; Yaffe et al.,2004). As discussed above, there are now a number of preclinical studies demonstrating exacerbation of CNS inflammation/function resulting from inflammatory liver damage (D'Mello et al.,2009), osteoarthritis (Kyrkanides et al.,2011), atherosclerosis (Drake et al.,2011), and diabetes (McClean et al.,2011). While it is now relatively clear that patients suffering episodes of delirium show long-term cognitive impairments, even patients with altered mental status not fulfilling criteria for delirium (often classified as subsyndromal delirium) show evidence of long term cognitive impairment in geriatric (Cole et al.,2003) and critical care settings (Jones et al.,2006). Thus, acute systemic inflammatory episodes have a negative impact on CNS function, induce neuropathological changes and appear to accelerate neurodegenerative disease. The presence of chronic peripheral inflammatory disease may have even worse prognosis for the degenerating brain. The recent findings that systemic anti-TNF-α treatment for rheumatoid arthritis (Chou,2010) is protective against AD brings full circle, the original findings that NSAIDs taken for RA were also protective against the development of AD.
It is clear that severe systemic inflammatory episodes such as severe sepsis induce considerable CNS pathology and disability. Multiple preclinical studies have shown that robust and repeated systemic inflammatory activation can also produce neurodegeneration. These degenerative changes are particularly marked when animals have some underlying genetic predispostion or have existing degenerative pathology and/or primed microglia. Moreover, recent work in ageing shows that simply being exposed to an ageing bloodstream is enough to have significant impacts on neurogenesis and learning and memory (Villeda et al.,2011), with chemokines appearing to play a key role in this decline. There remains much to learn about the role of microglia and indeed the cerebrovascular endothelium and astrocytes in this periphery to brain communication, not to mention roles for infiltrating cells during inflammatory stimulation. It is reasonable to assume that this communication may be more complex than meets the eye and it will be important, in the coming years, to try to mimic both disease-associated inflammation and disease-exacerbating systemic inflammation in a proportionate fashion if we are to avoid making conclusions that are too simplistic and that will not translate to the human population. These investigations have considerable implications for our interpretation and further development of trials with anti-inflammatory drugs. Thus far, NSAIDs have provided protection when taken for many years but have been unable to treat established disease. This is consistent with the idea that a lifetime of peripheral inflammatory events contributes to brain pathology. One very promising aspect of the unmasking of systemic inflammation as a significant contributor to CNS disease is that if we can successfully identify key pathways to dysfunction, it may be possible to treat CNS disease with drugs that do not require access to the brain parenchyma.