APP is cleaved by α-secretase in the centre of the Aβ domain. Three related metalloproteases of the ADAM (a disintegrin and metalloprotease) family, ADAM-9, ADAM-10 and ADAM-17, also termed TACE (tumor necrosis factor converting enzyme), appear to exert α-secretase activity 1657. A confirmation that ADAM10 is involved in α-secretase activity came from studies in transgenic mice. A moderate neuronal over-expression of ADAM10 in mice transgenic for human APP([V717I]) showed increased secretion of the neurotrophic soluble α-secretase-released N-terminal APP domain (α-APPs), reduced formation of Aβ peptides, and prevention of their deposition in plaques 58. Functionally, in these mice impaired long-term potentiation and cognitive deficits were alleviated. Expression of mutant catalytically inactive ADAM10 led to an enhancement of the number and size of amyloid plaques in the brains of double-transgenic mice. These results provided the first in vivo evidence for a proteinase of the ADAM family as an α-secretase of APP, revealed activation of ADAM10 as a promising therapeutic target, and supported the hypothesis that a decrease in α-secretase activity contributes to the development of AD 58. Another candidate with α-secretase activity is BACE2, which likewise cleaves within the Aβ domain and abrogates Aβ formation 57 (Figure 1).

ADAM proteinases have emerged as the major protein family that mediates ectodomain shedding, the proteolytic release of extracellular domains from their membrane-bound precursors. Proteolytic cleavage or ectodomain shedding is an additional mechanism whereby cells can regulate the proteins expressed on their cell surface. In many cases, soluble ectodomains are biologically active as mediators of functions ascribed to their transmembrane counterparts 59. ADAM-like sheddases activate, for instance, growth factors and cytokines, thus regulating signalling pathways that are important in development and pathological processes such as cancer 60. ADAM17 and 10 appear to play a particularly prominent role in ectodomain shedding of inflammatory proteins at all stages of leukocyte recruitment. Soluble TNF-α is released from its membrane-bound precursor by shedding through ADAM17 61. In vivo studies have shown that many, but not all, of the inflammatory effects of TNF-α require cleavage and shedding from the cell surface. Besides TNF-α, ADAM17 cleaves ectodomains of other receptors and ligands, such as TNFR2 and L-selectin 62. In addition, it was recently shown that IL-1R2 can be proteolytically processed in a manner similar to APP 63. IL-1R2 undergoes ectodomain shedding in an α-secretase manner, resulting in the secretion of the IL-1R2 ectodomain and the generation of a C-terminal fragment.

That α-secretase is involved in inflammation is supported by the observation that ADAM-10 is expressed constitutively by astrocytes in the normal and inflamed human CNS 64. ADAM10 and ADAM-17 are also enriched in microglia 65. The functions of ADAMs in microglia are complex and they do not only have pro-inflammatory and neurotoxic properties but also reparative ones 66. In fact, IL-1α has been reported to increase ADAM-17 and ADAM-10 levels and α-secretase activity in human astrocytic cultures 67. More recently, it has been shown that the proinflammatory cytokines TNF-α, IFN-γ and IL-1β, as well as TGF-β and LPS, are able to increase ADAM10 activity, leading to a loss in E-cadherin expression, which is another substrate for this secretase 68.

On the other hand, it has been also reported that various NSAIDs (including nimesulide, ibuprofen and indomethacin) stimulate the non-amyloidogenic secretion of sAPPα from neuroblastoma cells 69. Shedding of sAPPα induced by nimesulide and thalidomide was modulated by inhibitors of PKV and Erk MAPK, indicating that NSAIDs activate the Erk MAPK signalling cascade. However, these results have not been reproduced by other groups 7071.

β-Secretase (BACE1 for β-site APP cleaving enzyme) was cloned and identified as a type I transmembrane aspartyl protease 72. BACE1 cleaves APP at the N-terminal position of Aβ. BACE1 deficiency precludes Aβ formation in transgenic mice, and does not cause or promote any neurological or phenotypic abnormalities. However, it has been recently demonstrated that BACE1 knockout mice exhibit a number of schizophrenia-like behavioural traits, most likely because BACE1 is involved in the cleavage of neuregulin-1, which has been linked to the pathogenesis of schizophrenia and related psychiatric disorders 73. Because BACE1 inactivation rescues memory deficits in transgenic mice 74, this strongly supports the importance of BACE1 as a therapeutic target in AD.

BACE1 is primarily expressed in neurons 727576, but it can be also expressed in astrocytes under conditions of chronic stress 77 and in old transgenic mice 7632. In addition, in young transgenic mice, neuronal BACE1 was induced in the proximity of activated microglia and astrocytes 32. These observations lead to the conclusion that BACE1 expression is regulated by inflammatory events. BACE1 has also been found up-regulated in neuronal cultures upon exposure to pro-inflammatory cytokines 7078, under oxidative stress in NT2 neurons 79, and in the hippocampus of rats after experimental traumatic brain injury 39. In addition, BACE1 protein levels, activity and the β-secretase product (β-CTF) are increased in brain of sporadic AD patients 8081 as well as in platelets and CSF from AD and MCI patients 828384. These alterations in BACE1 expression and activity indicate a transcriptional and/or translational regulation of BACE1 expression in the brain 85. Interestingly, consensus binding sites for various transcription factors that are known to be regulated by inflammation (such as NFκB, PPARγ and STAT1) are present in the BACE1 promoter.

NFκB sites are present in the promoters of APP 86, presenilin and BACE1 87. In neurons exposed to soluble Aβ peptides and in TNFα-activated glial cells the mutation of the BACE1 promoter NFκB site led to significant decreases in promoter activity, indicating an activating role for NFκB in BACE1 expression in Aβ 87. In addition, some NSAIDs such as flurbiprofen and indomethacin, which target NFκB, have been shown to be effective at decreasing amyloid load in vitro and also in APP transgenic mice 545588. A recent report showed that deletion of TNFα1 death receptor (TNFα1R) in APP23 transgenic mice inhibited Aβ generation reducing BACE1 levels and activity via the NFκB pathway 89.

The effect of NFκB on BACE1 promoter could be direct or through changes in PPARγ, because PPARγ agonists can antagonize the activity of transcription factors such as NFκB 50.

PPARγ is a transcription factor that is involved in the regulation of the metabolism of glucose and lipids, in cellular differentiation as well as in the control of transcription of a wide range of inflammatory genes. A consensus binding site for PPARγ was found in the BACE1 promoter. PPARγ activation by agonists such as thiazolidinediones (TZD) and certain NSAIDs such as ibuprofen, indomethacin and naproxen results in a decrease of BACE1 transcription, expression and activity 70. Furthermore, lack of PPARγ led to an increase of BACE1 promoter activity 90, which suggested that PPARγ could be a repressor of BACE1.

PPARγ levels are decreased in AD brain, indicating that inflammatory events may decrease PPARγ transcription. Furthermore, in vitro experiments have shown that inflammatory cytokines and oxidative stress decrease PPARγ levels. Therefore, these findings suggest the existence of a down-regulation of PPARγ under inflammatory conditions, which would result in an increase in BACE1 transcription and Aβ generation. This effect could be prevented by modulation of PPARγ activity by NSAIDs, which have been shown to increase the levels of PPARγ in adipocytes and neurons 9053.

STAT-1 can bind directly to a putative STAT1 binding sequence in the BACE1 promoter. A recent study showed that when STAT1 becomes phosphorylated by IFNγ-mediated activation of JAK2 and ERK1/2, STAT1 binds to BACE1 promoter region to increase BACE1 protein expression in astrocytes 91. On the other hand, another report has demonstrated that IFNγ-receptor knockout mice crossed with Tg2576 mice expressing the human Swedish mutant APP had reduced gliosis and amyloid plaque compared with Tg2756 animals, apparently by decreasing TNFα secretion and the number of reactive astrocytes expressing BACE1 in cortex and hippocampus 78.

γ-Secretase is a protein complex of four essential membrane proteins called aph-1, pen-2, nicastrin and presenilin (PS). While aph-1, pen-2 and nicastrin function in the assembly and subcellular transport of γ-secretase and in the recognition of protein substrates 959697. PS proteins are members of an aspartyl protease family and represent the catalytically active components of the γ-secretase complex 98. PS1 and PS2 proteins are homologous polytopic membrane proteins critically involved in intra-membraneous cleavage of APP and a number of other type I membrane proteins 9899100. Accordingly, PS proteins have been implicated in different biological processes, including the regulation of cell differentiation and death, calcium homeostasis, cell adhesion, and subcellular trafficking of several membrane proteins. Mutations in PS are associated with familial forms of early onset Alzheimer's disease (AD) and increase the ratio of Aβ42/Aβ40 by either elevating production of the elongated, highly fibrillogenic Aβ42 or by decreasing the generation of Aβ40 57101102.

It was recently published that activated microglia and astrocytes have enhanced expression of PS and nicastrin following brain damage 42. Although glial cells express PS1, it is not known if PS1 mutations alter glial cell functions. Carriers of PS FAD associated mutations not only show earlier deposition of Aβ in β-amyloid plaques, but also inflammatory processes in the brain 90. On the other hand, it is unclear whether PS FAD associated mutations cause neuroinflammation by promoting formation and deposition of Aβ42 or by triggering other processes. In PS1 associated FAD brain, distinct 'inflammatory plaques' have been described, that lack reactivity for apolipoprotein E and Aβ in the core, but revealed association with reactive microglia and astrocytes, suggesting that mutations in PS1 could also induce Aβ independent neuroinflammation in Early Onset Alzheimer's disease (EOAD) 103. In addition, studies using knock-in mice for PS1 FAD mutations have revealed an enhanced inflammatory cytokine response to immune challenge with bacterial lipopolysaccharide (LPS). LPS-induced levels of mRNAs encoding TNFα, IL-1α, IL-1β, IL-1 receptor antagonist, and IL-6 were significantly greater in the hippocampus and cerebral cortex of PS1 mutant mice as compared to wild-type mice. These findings demonstrate an adverse effect of PS1 mutations on microglial cells that results in their hyperactivation under pro-inflammatory conditions, which may, together with direct effects of mutant PS1 in neurons, contribute to the neurodegenerative process in AD 104.

Additional evidence for an Aβ independent role of PS proteins in inflammation comes from studies with conditional knockout mice lacking both PS genes in the postnatal forebrain. These mice display strong age-dependent neurodegeneration and impairment of cognitive function 105. Gene profiling revealed up-regulation of several pro-inflammatory genes, including glial fibrillary acidic protein, complement component C1q, and cathepsin S. Also, activated microglia were detected in the brain of these mice. Since Aβ production is strongly reduced upon deletion of PS in forebrain neurons, these data indicate that impairment of PS function could also trigger inflammation independent of Aβ 105. Analogously, the cleavage of other substrates for PS, such as Notch, could also have some effect on brain inflammation. It was shown that mice transgenic for antisense Notch and normal mice treated with inhibitors of γ-secretase showed reduced damage to brain cells and improved functional outcome in a model of focal ischemic stroke. These mice had a reduced number of activated microglial cells in the brain after ischemic perfusion 106.

As described above several NSAIDs decrease the risk for development of AD. Although the molecular mechanism(s) underlying this protective activity remain to be identified, certain NSAID could directly modulate γ-secretase activity 107. Thus, besides suppression of inflammatory processes via inhibition of COX dependent biosynthesis of pro-inflammatory prostaglandins, the protective role of NSAIDs could also involve altered generation of Aβ. Indeed, several NSAIDs, including ibuprofen, sulindac sulphide, and indomethacin could decrease the levels of secreted Aβ42 10854109. Importantly, generation of Aβ40 is largely unaltered by these compounds, indicating that certain NSAIDs modulate rather than inhibit γ-secretase activity. The decrease in Aβ42 production was associated with increased generation of Aβ38 suggesting a shift in the cleavage specificity of the protease 107. However, more recently it has been shown that the generation of Aβ38 and Aβ42 is independent and differently affected by FAD mutations in PS as well as by modulators of γ-secretase 110. After the initial identification of ibuprofen, sulindac sulphide, and indomethacin as selective Aβ42 lowering drugs, several other NSAIDs, including flurbiprofen and fenoprofen have been shown to exert similar effects 54. Because Aβ42 lowering activity was also observed with several derivatives of NSAIDs that do not inhibit COX, and was also present in COX1/2 deficient cell lines, the modulation of γ-secretase activity is unlikely to be mediated via COX enzymes 107.

Surprisingly, the PPARα antagonist fenofibrate and some COX-2 specific inhibitors could even increase Aβ42 production 56111112. Whether additional effects of NSAIDs on NFκB and PPARα or PPARγ (see above) contribute to the Aβ42 lowering activity remains to be investigated in more detail. However, certain NSAIDs decrease Aβ42 production in vitro with purified γ-secretase complex, indicating NFκB and PPARγ independent effects 113114. Interestingly, FAD associated mutations in PS1 could cause insensitivity of γ-secretase to Aβ42 lowering NSAIDs 115. Because direct interaction of NSAIDs with the presenilins or other components of the γ-secretase complex has not been shown so far, the molecular mechanisms for the action of NSAIDs are unclear. Since NSAIDs reveal Aβ42 lowering activity at relatively high doses (IC₅₀ values of 25 – 500 μM) at which they could also affect biophysical properties of biological membranes, the modulation of γ-secretase specificity might also involve interference with membrane fluidity and/or accessibility of substrate to the enzyme 116. Nonetheless, since NSAIDs, even at higher concentrations did not inhibit cleavage of Notch and ErbB4 or alter AICD formation, this class of inhibitors might be valuable compounds to selectively decrease Aβ42 production without affecting important intracellular signaling pathways, thereby reducing potential side effects in clinical applications.

In preclinical studies, long-term treatment of APP transgenic Tg2576 mice with ibuprofen and indomethacin were effective in reducing the plaque pathology 51117. Also, acute treatment with additional NSAIDs, sulindac sulfide, and flurbiprofen, selectively reduced Aβ42 levels without changing the concentration of Aβ40 54108. However, other studies did not reproduce these results 111118. In clinical trials some positive effects on cognitive performance of AD patients were observed with indomethacin and the (R)-enantiomer of flurbiprofen, while other NSAIDs without Aβ42 reducing activity did not reveal beneficial effects 107 (see Table 1 for summary). Although this indicates that it would be relevant to evaluate Aβ42 lowering NSAIDs in further clinical trails, recent epidemiological studies revealed that the protective effect of NSAIDs seems to be independent of the Aβ42 suppressing activity of the NSAID 119120.